**Cardiovascular and Thoracic Imaging: Trends, Perspectives and Prospects**

Editor

**Micka ¨el Ohana**

MDPI • Basel • Beijing • Wuhan • Barcelona • Belgrade • Manchester • Tokyo • Cluj • Tianjin

*Editor* Mickael Ohana ¨ Strasbourg University Hospital France

*Editorial Office* MDPI St. Alban-Anlage 66 4052 Basel, Switzerland

This is a reprint of articles from the Special Issue published online in the open access journal *Journal of Clinical Medicine* (ISSN 2077-0383) (available at: https://www.mdpi.com/journal/jcm/ special issues/Cardiovascular Thoracic Imaging Perspectives Prospects).

For citation purposes, cite each article independently as indicated on the article page online and as indicated below:

LastName, A.A.; LastName, B.B.; LastName, C.C. Article Title. *Journal Name* **Year**, *Volume Number*, Page Range.

**ISBN 978-3-0365-3371-1 (Hbk) ISBN 978-3-0365-3372-8 (PDF)**

© 2022 by the authors. Articles in this book are Open Access and distributed under the Creative Commons Attribution (CC BY) license, which allows users to download, copy and build upon published articles, as long as the author and publisher are properly credited, which ensures maximum dissemination and a wider impact of our publications.

The book as a whole is distributed by MDPI under the terms and conditions of the Creative Commons license CC BY-NC-ND.

## **Contents**


**Salim Aymeric Si-Mohamed, Lauria Marie Restier, Arthur Branchu, Sara Boccalini, Anaelle Congi, Arthur Ziegler, Danka Tomasevic, Thomas Bochaton, Loic Boussel and Philippe Charles Douek**

Diagnostic Performance of Extracellular Volume Quantified by Dual-Layer Dual-Energy CT for Detection of Acute Myocarditis

Reprinted from: *J. Clin. Med.* **2021**, *10*, 3286, doi:10.3390/jcm10153286 ................ **111**

## **About the Editor**

**Mickael¨ Ohana** is a Professor of Radiology at Strasbourg University Hospital, specializing in cardiovascular and chest imaging. His main research interests are low-dose CT, coronary CTA, CT angiography, artificial intelligence and its applications to medical imaging.

### *Review* **Chronic Lung Allograft Dysfunction Post Lung Transplantation: A Review of Computed Tomography Quantitative Methods for Detection and Follow-Up**

**Trieu-Nghi Hoang-Thi 1,2,3, Guillaume Chassagnon 1, Thong Hua-Huy 3, Veronique Boussaud 4, Anh-Tuan Dinh-Xuan <sup>3</sup> and Marie-Pierre Revel 1,\***


**Abstract:** Chronic lung allograft dysfunction (CLAD) remains the leading cause of morbidity and mortality after lung transplantation. The term encompasses both obstructive and restrictive phenotypes, as well as mixed and undefined phenotypes. Imaging, in addition to pulmonary function tests, plays a major role in identifying the CLAD phenotype and is essential for follow-up after lung transplantation. Quantitative imaging allows for the performing of reader-independent precise evaluation of CT examinations. In this review article, we will discuss the role of quantitative imaging methods for evaluating the airways and the lung parenchyma on computed tomography (CT) images, for an early identification of CLAD and for prognostic estimation. We will also discuss their limits and the need for novel approaches to predict, understand, and identify CLAD in its early stages.

**Keywords:** lung transplantation; image processing; computer-assisted; pulmonary disease; chronic obstructive; bronchiolitis obliterans

### **1. Introduction**

Compared to other organ transplantation, lung transplantation remains associated with a poorer prognosis, with a median survival of only six years after transplantation [1]. The main cause of death beyond the first year is chronic lung allograft dysfunction (CLAD). This term refers to the deterioration of lung function as a result of chronic rejection and is not limited to an exclusively obstructive phenotype, referred to as bronchiolitis obstructive syndrome (BOS), but also includes a restrictive phenotype of chronic allograft dysfunction, termed restrictive allograft syndrome (RAS). The classification of CLAD phenotypes updated in 2019 by the International Society of Heart and Lung Transplantation (ISHLT) included BOS, accounting for 70% of allograft dysfunctions, and three other entities, RAS, mixed, and undefined phenotypes, representing the other 30% of phenotypes [2]. Their respective definitions are summarized in Table 1.

Hota et al. provided a comprehensive description of the different CLAD phenotypes on high-resolution CT [3]. RAS is associated with persistent parenchymal opacities and pleural thickening showing upper lung predominance, with pleuro parenchymal fibroelastosis on pathological examination, explaining the restrictive pattern [2,3]. Complementing pulmonary function tests (PFTs) with CT is thus mandatory not only to exclude other causes of lung function degradation but also to classify CLAD phenotypes. This classification is important for prognosis, with patients with a restrictive pattern of CLAD having worse survival [4]. Furthermore, Levy et al. [5] demonstrated that patients with

**Citation:** Hoang-Thi, T.-N.; Chassagnon, G.; Hua-Huy, T.; Boussaud, V.; Dinh-Xuan, A.-T.; Revel, M.-P. Chronic Lung Allograft Dysfunction Post Lung Transplantation: A Review of Computed Tomography Quantitative Methods for Detection and Follow-Up. *J. Clin. Med.* **2021**, *10*, 1608. https://doi.org/10.3390/ jcm10081608

Academic Editor: Takashi Ohtsuka

Received: 6 March 2021 Accepted: 8 April 2021 Published: 10 April 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

undefined/unclassified CLAD phenotypes who had RAS-like opacities on chest imaging had significantly worse allograft survival than patients with the BOS phenotype. Suhling et al. [6] performed visual scoring of reticulations and consolidations in lung transplanted (LTx) patients and found poorer survival in patients with severe reticular changes. Imaging follow-up is especially important to detect CLAD in patients with single lung transplantation, because native lung disease may confound the interpretation of physiology [7].


**Table 1.** Definition of different CLAD phenotypes.

A switch from visual descriptive evaluation to quantitative automatic measurements has recently emerged in the field of medical imaging. Quantitative imaging has been mainly used for the evaluation of BOS, the obstructive phenotype of CLAD. The implementation of such imaging analysis tools for CLAD evaluation in LTx patients either directly confirms airway thickening and vascular remodeling in the allograft lung [8,9] or indirectly suggests their occurrence through air-trapping quantification [10,11]. Applying machine learning to imaging data analysis may further improve the management of CLAD [12]. In this review article, we will provide an overview of the CT quantification methods used to investigate CLAD of various phenotypes (see Figure 1 and Table 2), and discuss novel approaches to identify CLAD in its early stage. As sub-clinical chronic rejection compromises allograft integrity, there is a need to develop physiological and imaging tests allowing early detection of CLAD, with a potential role for quantitative imaging to serve as a biomarker, allowing early identification of CLAD [13].

**Figure 1.** Overview of CT quantification methods for CLAD. CT: computer tomography, CLAD: chronic lung allograft dysfunction. Airway measurement methods [8,9,12,14], Vascular measurement methods [14], Parenchyma based methods [10,11,15–21].


4

**Table 2.**

Overview of CT

quantification

 methods for CLAD.


**Table 2.** *Cont.*

5


**Table 2.** *Cont.*

between expiration and inspiration; A/V ratio: airway vessel ratio; ΔWT and ΔWA%: temporal changes of WT and WA%; PRMfSAD: Parametric Response Mapping representing functional small airways disease; PRMPD: Parametric Response Mapping representing parenchymal disease; PRMNormal: Parametric Response Mapping representing normal pulmonary parenchyma; EXP-790 HU to -950 HU: percentage of voxels with attenuation values from -790 HU to -950 HU on expiration; DM: density mapping; E/I-MLD: expiratory-to-inspiratory mean lung density ratio; E/I- Volumes: expiratory-to-inspiratory volume ratio; RV/TLC: residual volume/total lung capacity qCT: quantitative CT.

### **2. Airway Measurement Methods: Computer-Assisted Airway Morphometry**

Since the most common phenotype of CLAD is that of BOS, quantitative methods allowing the direct assessment of the airways have been developed. BOS primarily affects the small airways, with diameters of less than 2 mm. Due to the limited spatial resolution of clinical CT scanners, new measurement methods had to be developed to allow the measurement of the density and wall thickness of the small airways [22]. Such a method made it possible to obtain values of the wall thickness of the small airways correlated with FEV1 in COPD patients [23]. Several software now available commercially make it possible to obtain an automatic segmentation of the bronchial tree and sections strictly perpendicular to the bronchial axis considered at different levels, as well as an automatic measurement of the bronchial wall thickness (WT) and the wall area ratio (WA%), the ratio of the airway wall area to the whole airway area.

Using the YACTA software [24], Doellinger et al. retrospectively examined a total of 2190 airway cross sections on CT scans performed within 11 years of follow-up of 26 LTx patients with at least one measurement in every lobe of the transplanted lung [8]. They showed significant differences between patients with and without BOS in term of changes over time of WA% and WT and concluded that bronchial wall thickening and luminal dilatation observed in lung transplant allograft rejection can be detected and quantified using computer-assisted airway morphometry. Gazourian et al., in a retrospective analysis of 22 patients, observed and increase in the air airway internal lumen perimeter (PI), and airway lumen area (AI) in the 13 patients developing BOS [14]. They used Airway Inspector (now renamed Chest Imaging Platform) for evaluating the third and fourth generation B1 and B10 airways. They also evaluated the pulmonary vessels cross-sectional areas (CSA) after upper thresholding (cut-off level of −500 HU) and use of a connected components technique on volumetric CT pulmonary angiogram to isolate the vessel cross sections. These authors also measured the airway/vessel ratios (A/V ratio) on non-volumetric CT scans. They found a statistically significant decrease in the vessel CSA in BOS patients and a significant increase in the airway/vessel ratio in BOS progressors. This observation is in line with known physiology, namely vasoconstriction in poorly ventilated areas due to small airways disease. These two studies (Doellinger [8] and Gazourian [14]), although both based on a limited number, showed differences between BOS and controls, but automated quantification of airway dimension was limited by the need for manual adjustments. In addition, the methods used do not translate into early identification or prognostic tools for CLAD.

Dettmer [9] evaluated 25 patients with BOS and 116 controls using MeVis Airway Examiner. WA% on inspiration was significantly greater in patients with BOS, but the variability of bronchial wall measurements was high and the values for the WA% on inspiration in patients with and without BOS overlapped considerably, due to variable underlying lung volumes. Even though this limitation could be overcome by performing spirometry-controlled CT acquisitions, the authors concluded to a limited value of WA% for establishing a diagnosis of BOS in individual patients.

Barbosa et al. [12] conducted a retrospective analysis of LTx patients who had paired inspiratory and expiratory CTs, with an objective to detect BOS 0-p stage, as indicator of early disease (≥10% decline in FEV1 or ≥25% decline in forced expiratory flow 25–75%). Baseline prediction of BOS was performed in a cohort of 41 patients, of which 15 developed BOS. They measured the airway volumes and airway resistances after 3D analysis of the airways and lung lobes using three different commercially available software, Mimics, TGrid 14.0 and Fluent 14.0, for computational fluid dynamics simulations. The authors found that BOS patients experienced an increase in central airway volume on expiratory CT, whereas patients with other cause of FEV1 decline had a decrease in the central airway volume on inspiratory CT. According to the authors, these findings reflected an increase in the extent of air trapping and peribronchiolar fibrosis which characterizes the BOS phenotype of CLAD. Image post-processing included segmentation of the tracheobronchial tree down to the level of airways with a diameter of 1–2 mm which required manual

correction and took 2 to 6 h per scan, which is the main limitation to consider the clinical use of this approach.

### **3. Lung Parenchyma Methods: Assessment of Lung Volume and Attenuation, and Their Variations between Inspiration and Expiration**

Barbosa et al. [16], using the Advanced Normalization Tools (ANTs) software package, calculated the aerated lung volume in inspiration and expiration via exclusion of any voxel >−50 HU and then evaluated the volume change between inspiration and expiration as the difference between these two values, in a retrospective patient cohort excluding the RAS phenotype. In addition to volume change, two other quantitative parameters reflecting air trapping were evaluated: the volume of voxels with attenuation <−856 HU on expiration and the volume of voxels with an increase in attenuation <75 HU from inspiration to expiration following non-rigid registration between the inspiratory and expiratory imaging datasets. Only 59% of quantitative CT metrics were significantly correlated with BOS status in this study including 176 LTx patients, and none of the variables alone was a good predictor of BOS. However, a support vector machine (SVM) model based on the quantitative CT variables outperformed models using visual scoring of CT anomalies or PFT for BOS prediction after unilateral LTx.

Dettmer et al. [17] conducted a prospective evaluation with the objective to detect the BOS phenotype of CLAD in a cohort of 122 LTx patients. They performed CT acquisition with a spirometry-controlled technique at full inspiration and end of expiration and used Mevis Pulmo software, allowing volume measurement and histogram analysis (mean lung density/MLD, peak, percentiles) for the whole lung and separately for each lobe. They demonstrated that in patients with early-stage BOS, the lower lobe volume increased and the MLD decreased between the baseline and follow-up examinations, whereas the volume and MLD in the upper lobes remained nearly constant. The histogram parameters showing the highest accuracies for early BOS detection were the 10th percentile on expiration (AUC: 0.903) and the expiratory-to-inspiratory (E/I) MLD ratio (AUC: 0.886).

E/I MLD ratio is one of the multiple quantitative parameters used to evaluate the volume of air trapping in BOS. Solyanik et al. [15] evaluated two other automated methods to quantify air trapping, which had only slight differences with those previously mentioned in the study by Barbosa [16]: the lung volume having attenuation values ranging from −950 to −790 HU on expiratory CT scans (attenuation < −856 HU for Barbosa) and the lung volume with less than 80 HU change (<75 HU change for Barbosa) between inspiration and expiration. The latter requires non-rigid registration of inspiration-expiration CT-data and voxel-to-voxel mapping, which Solyanik described as "density mapping". Among the three evaluated methods, density mapping showed the best correlation with the ratio of residual volume to total lung capacity (RV/TLC).

Parametric response mapping, similar to density mapping, involves assessing the attenuation of each voxel on inspiration and expiration after applying an elastic registration algorithm. Voxels with attenuation ≥950 HU and <−810 HU at inspiration and <−856 at expiration are considered to represent functional small airways disease (PRMfSAD). Belloli et al. [11] evaluating a retrospective cohort of 52 LTx patients reported that PRMfSAD values ≥ 30% were the strongest predictor of survival in a multivariable model including BOS grade and baseline FEV1% predicted. Verleden et al. [10] used PRM to monitor BOS progression in a retrospective study including 20 BOS and 20 controls (no restrictive CLAD phenotype). They observed an increase at time of BOS diagnosis. They also reported that patients who died from BOS had significantly higher PRMfSAD than living patients. Galban et al. investigated the use of PRM as an imaging biomarker in the diagnosis of BOS. They found that PRMfSAD > 28% was highly indicative of BOS occurrence, whether a concurrent infection was present or not [19].

Other authors have developed quantitative methods only requiring inspiratory CT data. This was the case for Horie et al. [18], who evaluated CT lung density histograms on a single inspiratory CT in CLAD patients after lung segmentation. Indeed, lung fibrosis associated with the RAS phenotype of CLAD is likely to increase lung density, whereas

mosaic perfusion secondary to BOS has an invert effect, both being detectable when analyzing the lung density histogram: the right-sided tail reflects processes increasing lung density whereas the left-sided tail reflects processes decreasing lung density, such as mosaic attenuation in areas of BOS. Their objective was threefold, to evaluate prognosis after CLAD onset, distinguish between BOS and RAS phenotypes, and evaluate the prognosis within BOS and RAS patient groups separately. They evaluated the quantitative density metrics (QDM) defined as ratios of the right and left quantile weights of the histogram, reflecting the left vs. right asymmetry of the histogram. There was a statistically significant difference in QDMs between RAS and BOS patients. Higher QDM values were significantly associated with decreased survival. The authors concluded that this quantitative analysis of CT images was associated with survival after the onset of CLAD and was able to differentiate RAS and BOS phenotypes. QDM measurement has been used by the same authors [25] to predict the risk of subsequent CLAD in patients with CLAD-0p, defined as a drop in FEV1 to 81–90% of baseline. Higher QDM values were associated with a shorter time between CLAD-0p-CT and CLAD.

Fewer quantitative CT methods have been developed for the RAS phenotype of CLAD. Measurement of inspiratory and expiratory CT lung volumes can be easily performed and CT volumetry has been shown to differentiate RAS patients from BOS patients due to the strong positive correlation between inspiratory lung CT volumes and TLC [20]. Computed tomography is particularly useful for detecting CLAD affecting a single lung [7,21], because the unaffected contralateral lung compensates for the deficit in lung function on PFTs. A decrease over time in the difference between inspiratory and expiratory lung volumes has been reported for obstructive and restrictive CLAD phenotypes [21].

PRM is another method allowing quantitative evaluation of patients with the RAS phenotype of CLAD, who have persistent parenchymal opacities and pleural thickening. Voxels with attenuation values ≥−810 HU at inspiration represent parenchymal disease (PRMPD). In the study by Belloli et al. [11] evaluating 22 LTx patients, those with concurrent declines in both FEV1 and FVC (e.g., RAS phenotype) were found to have significantly more PRMPD than their control group, even after adjusting for age, baseline FEV1% predicted, and baseline FVC % predicted. The prognostic value of PRMPD was not mentioned and should probably be evaluated in a larger patient group.

To be clinically useful, quantitative methods should not require multiple complex algorithms or manual steps to correct segmentation, which is the case with computerassisted airway morphometry. In addition, these methods only assess the BOS phenotype of CLAD and have limited diagnostic value. Simpler methods based on CT volumetry are more accessible, with the assessment of lung volume and lobe attenuation on inspiration and expiration of interest for all CLAD phenotypes.

Regarding the processing of imaging data by artificial intelligence for CLAD diagnosis, the current literature is scarce. Classical machine learning methods such as the support vector machine (SVM) using quantitative CT metrics were of interest to diagnose BOS following unilateral LTx, as previously mentioned [16]. SVM was also used to evaluate multiple combinations of functional respiratory indexes (FRI) for BOS prediction at six months after transplantation [12]. A maximal accuracy of 85% was obtained by combining three baseline FRI features: the right middle lobe volume at total lung capacity (inspiratory CT scan), the right upper lobe airway resistance and the central airway surface at functional residual capacity (expiratory CT). Deep learning represents a major advance in the field of artificial intelligence applied to medical imaging, but requires a large amount of data [26,27]. As previously highlighted, most series on quantitative imaging for CLAD evaluation post LTx are based on a limited number of patients, because lung transplantation is not a common procedure. In 2015, only 14 centers reported performing 50 or more LTx per year [28]. Therefore, the standardization of imaging follow-up and data sharing through a common registry is crucial before considering the development of deep learning algorithms for the prediction and early diagnosis of CLAD.

### **4. Conclusions**

Quantitative imaging methods offer the opportunity to perform reader-independent assessment of the airways and lung parenchyma in LTx patients. Although said to be automatic, most methods still require significant time-consuming manual corrections, and require the availability of segmentation and elastic registration algorithms, which limits their direct use in clinical routine. However, the results obtained in the studies published to date demonstrate the prognostic impact of methods such as parametric response mapping quantifying functional small airway disease (PRMfSAD) or quantitative density metrics (QDM). The performance for the prediction or early detection of CLAD needs to be strengthened, which could be an objective of deep learning-based methods. Developing prediction models is indeed important to improve outcomes for patients who are developing CLAD and better understand the underlining pathophysiology.

**Author Contributions:** Conceptualization, T.-N.H.-T. and M.-P.R.; methodology, G.C.; Literature research T.-N.H.-T. and T.H.-H.; writing—original draft preparation, T.-N.H.-T.; writing—review and editing, all authors.; supervision, M.-P.R. and A.-T.D.-X. All authors have read and agreed to the published version of the manuscript.

**Funding:** This review article received no external funding.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** Not applicable.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


### *Review* **Spectral Photon-Counting CT Technology in Chest Imaging**

**Salim Aymeric Si-Mohamed 1,2,\*, Jade Miailhes 2, Pierre-Antoine Rodesch 1, Sara Boccalini 1,2, Hugo Lacombe 1, Valérie Leitman 1, Vincent Cottin 3, Loic Boussel 1,2 and Philippe Douek 1,2**


**Abstract:** The X-ray imaging field is currently undergoing a period of rapid technological innovation in diagnostic imaging equipment. An important recent development is the advent of new X-ray detectors, i.e., photon-counting detectors (PCD), which have been introduced in recent clinical prototype systems, called PCD computed tomography (PCD-CT) or photon-counting CT (PCCT) or spectral photon-counting CT (SPCCT) systems. PCD allows a pixel up to 200 microns pixels at iso-center, which is much smaller than that can be obtained with conventional energy integrating detectors (EID). PCDs have also a higher dose efficiency than EID mainly because of electronic noise suppression. In addition, the energy-resolving capabilities of these detectors allow generating spectral basis imaging, such as the mono-energetic images or the water/iodine material images as well as the K-edge imaging of a contrast agent based on atoms of high atomic number. In recent years, studies have therefore been conducted to determine the potential of PCD-CT as an alternative to conventional CT for chest imaging.

**Keywords:** thorax; lung; diagnostic imaging; computed tomography; photon-counting detectors

### **1. Introduction**

Chest imaging is constantly evolving, and the demand for diagnostic performance is growing in many areas, such as lung cancer, detection of pulmonary nodules, chronic obstructive pulmonary disease, as well as evaluation and follow-up of interstitial lung diseases (ILD). While CT is the most powerful tool currently available for this purpose, it has many limitations such that accurate diagnosis often requires additional histological analyses or other non-radiological examinations.

Photon-counting detector computed tomography (PCD-CT) technology is a new CT modality that has gained increasing interest in all areas of imaging, including musculoskeletal, digestive, cardiovascular, pulmonary, and molecular imaging [1–6]. Contrary to energy integrated detectors (EID) found in current standard CT scanners, PCD are made of compounds, such as silicon, cadmium-telluride, or cadmium-zinc-telluride, to directly convert each X-ray photon into an electric pulse, allowing resolution of the energy of the photons. This characteristic allows better intrinsic technical performance by reducing noise, increasing spatial resolution, decreasing beam hardening, and decreasing X-ray doses [7,8]. In addition, the energy-resolving capabilities of these detectors allow to generate spectral basis images, such as the monoenergetic images or water/iodine material images as well as the K-edge images of a contrast agent based on atoms of high atomic number. Taken together, PCD-CT provides new diagnostic perspectives compared to the conventional EID found in current CT systems. Studies have therefore been conducted in recent years,

**Citation:** Si-Mohamed, S.A.; Miailhes, J.; Rodesch, P.-A.; Boccalini, S.; Lacombe, H.; Leitman, V.; Cottin, V.; Boussel, L.; Douek, P. Spectral Photon-Counting CT Technology in Chest Imaging. *J. Clin. Med.* **2021**, *10*, 5757. https://doi.org/10.3390/ jcm10245757

Academic Editor: Mickaël Ohana

Received: 8 November 2021 Accepted: 2 December 2021 Published: 9 December 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

mainly limited to non-contrast lung imaging, to determine the potential of PCD-CT as an alternative to conventional CT. In the present review, we will discuss these studies and explore the perspectives of PCD-CT for chest imaging.

### **2. Technical Aspects**

### *2.1. Photon-Counting Detector Technology*

PCD are a recently developed technology seen as the future of X-ray measurement; the design of these detectors differs from that of EID, which are commonly used in clinical practice (Figure 1). The EID use a scintillator material transforming the incoming X-ray into visible light that is then captured by a photo-diode. The deposited energy in the pixel is integrated by an application-specific integrated circuit (ASIC) that produces a current proportional to the energy of the incoming photon. In contrast, the PCD are made of a semi-conductor material that directly converts the incoming photon to electrical charges, which migrate into a counting ASIC (Figure 1). This process is called direct conversion (by opposition to indirect conversion). The ASIC shapes a voltage pulse proportional to the incoming photon energy. From the amplitude of the pulse the PCD can differentiate photons according to their energy (Figure 2). PCD-CT can operate in two different modes: conventional or spectral. In the conventional mode, photons are not discriminated according to their energy but are only summed; "conventional images" are those computed from the attenuation of all counted photons. This corresponds to the single energy CT (SECT). The direct conversion design fundamentally improves three major aspects of X-ray SECT imaging: the size of the detector pixel, the reduction of the electronic noise, and the energy weighting in the analogical output signal. The resulting benefits of these aspects are detailed in the next section.

**Figure 1.** Schematic representation of photon detection technologies: energy-integrating (**top**) and photon-counting detectors (**bottom**).

**Figure 2.** Example of a 500 ns signal output of a photon-counting detector pixel.

### *2.2. Benefits for Conventional Images*

### 2.2.1. Spatial Resolution

Because the scintillator can produce visible photons in each direction, the EID pixels require separation by a reflector material (Figure 1). This leads to an interseptal gap between pixels that critically decreases the ratio of effective sensitive area to detector area when the pixel size becomes too small. The subsequent major loss of efficiency with respect to exposure limits the size of EID to 0.5 mm at iso-center in normal resolution for most of the current clinical system. PCD pixel pitch is not technically limited and can reach 0.19–0.225 mm at the iso-center in the latest full-body prototypes [9–12].

The most recent ultra-high-resolution (UHR) EID-CT uses EID made by a different manufacturing process than conventional ones [13]; the septa gap is thinner and closer to the thickness of the anti-scatter collimators (Figure 1). This leads to 0.25-mm voxel element at iso-center, which fills the gap between standard EID and PCD in terms of reconstructed voxel size for lung imaging [14]. However, some methods have shown promising results to correct scatter without a pre-detector collimator [15], even for PCD [16]. This anti-scatter grid removal would increase the sensitive area of PCD and not EID. Moreover, interseptal gaps prohibit the possibility for subpixel resolution, which is a process that better estimates the photon reaching point for a pixel based on the adjacent pixels [17].

To the best of our knowledge, a comparison of the dose efficiency between EID and PCD for a large detector panel with 0.25-mm voxel element size is not yet available. Even with the same this geometry configuration, the PCD still benefit from technical advantages described in the next subsections.

### 2.2.2. Reduction of Electronic Noise

According to the number of available thresholds (between 3 and 8 according to PCD manufacturer), several measurement bins can be defined (Figure 2) [9,11,12]. The first threshold is set just above the electronic noise level and suppresses it from the final counts, which cannot be done for EIDs, and therefore, electronic noise is added to the integrated energy. This means the noise in the reconstructed image is lower with PCD than with EID for the same dose.

### 2.2.3. Contrast Improvement

In addition to noise reduction, another aspect improving contrast-to-noise ratio (CNR) is the energy weighting of each incoming photon. As the linear attenuation coefficient decreases with energy, lower energy provides greater contrast, for example, when differentiating water from calcium [18] (Figure 3). Whereas EID produce a signal proportional to the incoming energy for one photon (linear weight), PCD will generate only one count irrespective of the photon measured (constant weight).

**Figure 3.** Energy weighting of energy integrating detectors (EID) and photon-counting detectors (PCD). The optimal weight for differentiating calcium from water is derived from the attenuation coefficients subtraction (dashed lines).

### 2.2.4. Reduction of Beam Hardening

The constant weight also reduces beam hardening artifacts in conventional images. Beam hardening is due to a difference in mean energies between the incident and the attenuated spectra, and the proportional weighting of EID increases this difference, which is not the case for the constant weighting of PCD [8,19].

### 2.2.5. Dose Efficiency

The advantages described above of PCD over EID lead to a better dose efficiency, as demonstrated by several studies [9,20–23]. The effect of this dose efficiency can be either to produce the same image quality for the same dose or to investigate new protocols. In addition, iterative reconstruction can decrease noise levels in reconstructed CT images. Combined with the emergence of iterative reconstruction, PCD-CT enables either the reconstruction with a larger matrix (1024 × 1024 or 2048 × 2048) or the efficiency of ultra-low dose protocols [9,22].

### *2.3. Spectral Mode*

Recently, progress in informatics (high storage capacity, high-speed computation reconstruction) has allowed for spectral reconstructions in clinical practice through dualenergy CT (DECT) devices. However, DECT systems require major modification in their architecture compared to single energy CT [1]. While PCD-CT has the advantage to be able to run in both conventional and spectral mode, using a traditional X-ray source and CT geometry allows different types of images that are described in following subsections.

### 2.3.1. Virtual Monoenergetic (VMI) Images

PCD have the possibility to provide spectral measurement bins by differentiating incoming photons according to their energy (Figure 2). This spectral information can be used to reconstruct VMI [23]. VMIs are computed from spectral information at a (virtual) selected energy leading to beam hardening artifact reduction. Conversely, conventional images are the reconstruction of the attenuation coefficient averaged over the measured spectrum. VMIs can be reconstructed from either a PCD or a DECT acquisition. At low energies, VMIs allow a boost of the photoelectric effect such as the iodine boost in the presence of an iodinated contrast agent. This may allow a better depiction of the vessel lumen [24], while at high energies, VMI are robust to metal artifacts [25]. Another metal artifact-reduction technique is the reconstruction of an image using only photons measured

over 70 keV. The resulting high-threshold image presents an increased CNR compared to conventional images in presence of metal artifacts [26].

### 2.3.2. K-edge Imaging

K-edge materials are a type of material with a discontinuity in their attenuation coefficient (Figure 4). This spectral singularity allows for an accurate contrast agent identification in the reconstructed volume [27]. Imaging of a K-edge material, known as K-edge imaging, requires the measurement of at least three energy bins, which is not feasible with DECT [1]. To be detected for clinical imaging, this effect must be within the energy range of the diagnostic X-ray that is between 40 and 140 keV. Unfortunately, iodine is not a good candidate because of the photon starvation around its K-edge, which is 33.3 keV. Hence, new contrast agents for CT should be designed and validated for K-edge imaging using high atomic-number atoms, such as gadolinium, gold, or bismuth (Figure 4).

**Figure 4.** Attenuation coefficients of contrast agents with a K-edge in the medical energy range.

### *2.4. Conclusions*

In this section detailing technical aspects, we introduced the three main advantages of PCD over EID for conventional imaging: smaller detector pixel, electronic noise reduction, and more efficient energy weighting of photons. These advantages have been demonstrated to improve conventional image quality in pre-clinical and preliminary clinical studies. However, this is not the only strong point of PCD-CT, as it can perform simultaneously spectral mode to study VMIs and K-edge contrast agents, which may open to new and promising applications.

### **3. Pre-Clinical and Clinical Applications**

### *3.1. Parenchyma Imaging*

Despite the recent availability of high-resolution CT (HRCT), evaluation of lung parenchyma integrity is a daily challenge for the radiologist. The principal reason is that radiological evaluation is sometimes limited by the fine semiology of the anatomical structures of the lung lobule [28]; pathological analysis therefore remains the technique of choice in case of diagnostic uncertainty [29]. However, due to its high spatial resolution, PCD-CT systems could revolutionize the evaluation of the lung. Accordingly, numerous studies have investigated PCD-CT imaging of the parenchyma by studying the influence of the reconstruction parameters, such as the matrix size, the slice thickness, and the iterative algorithms recently made available for this new technology.

For instance, Bartlett et al. [30] demonstrated in patients the value of using a 1024 matrix with a high-frequency filter (Q65) with a limited field-of-view (FOV) PCD-CT system, in comparison with a 512 and 1024 matrix and standard detailed filter (B46) with PCD-CT and EID-CT. These reconstruction parameters were used to convey the high spatial resolution of the PCD-CT system, which improved the depiction of the lung parenchyma structures, such as the bronchi and morphological features of lesions (e.g., nodule, groundglass nodules). Recently, on a different PCD-CT system [9], we demonstrated in human volunteer with a large FOV and a 1024 matrix, that lung structures, such as fissures, distal airways, and vessels, were of greater conspicuity and sharpness with better overall image quality than standard CT, despite a 23% flux reduction (Figure 5). Interestingly, the noise was also better rated by three experienced chest radiologists who noted a particular texture that characterizes high-frequency noise. In this study, we showed that the UHR parameters allowed to depict a 178-μm line width on a line pair phantom, which was enabled by a spatial resolution at 22.3 lp/cm at 10% modulation transfer function. Similarly, Leng et al. [10] reported in patients with a different PCD-CT system the value of an UHR mode for small lung structures using a sharp filter with a spatial frequency cutoff of 32.4 lp/cm, a 1024 matrix, and 0.25 slice thickness in comparison to a standard (macro) and sharp mode. The sharp and UHR modes showed similar modulation transfer functions, both better than for macro mode; 10% modulation transfer function value was 9.48 lp/cm for macro, 16.05 lp/cm for sharp, and 17.69 lp/cm for UHR modes.

**Figure 5.** Comparison of high-resolution lung imaging between a conventional CT (Brilliance 64; Philips Haifa, Israel; (**A**,**B**)), a clinical prototype PCD-CT (SPCCT; Philips; (**C**,**D**)), and in a human volunteer. Close-up views on the left pulmonary hilum found a greater overall image quality as well as better depiction and a greater number of small structures (bronchial wall, vessels; white arrows).

Taken together, these studies show the potential for more accurate parenchyma evaluation in vitro as well as in human volunteers and patients with the UHR mode of the PCD-CT technology that requires confirmation in a prospective clinical study.

### *3.2. Nodule Imaging*

Lung nodule imaging is a challenging task that necessitates an accurate characterization of the morphological features of a nodule [31]. In recent years, several studies have proven the performance of PCD-CT in the detection and morphological characterization of pulmonary nodules, thus demonstrating its potential for this public health problem. For example, Zhou et al. [32] studied the impact of UHR mode for lung nodule volume evaluation and shape characterization using a PCD-CT system. They used a 512 matrix with a small FOV (11 cm) resulting in a pixel size of 215 μm and compared the UHR mode to the macro mode using two reconstruction algorithms (s80f and b46f). The UHR mode used an effective detector pixel size of 0.25 mm by 0.25 mm at the iso-center, while the conventional macro mode was limited to 0.5 mm by 0.5 mm. For small nodules (diameter ≤ 5 mm), UHR mode was able to achieve more accurate volume measurements than macro mode due to higher spatial resolution, which is an advantage for routine use. Moreover, all acquisitions showed comparatively lower bias in the evaluation of the volume of spherical nodules. Additionally, for irregularly shaped (star) nodules, particularly of small size (diameter ≤ 5 mm), UHR acquisitions with the sharp kernel provided substantially better accuracy in volume measurements compared to the other three acquisition mode/reconstruction kernel combinations. Furthermore, receiver-operating-characteristics analyses showed a clear benefit of the UHR mode, demonstrating an increased ability to differentiate small, smooth, spherical nodules from small, irregular-shaped nodules. These finding are particularly important, as studies have shown that nodules with irregular or spiculated margins, particularly with distortion of adjacent vessels, are likely to be malignant.

In a second study, Zhou et al. [33] performed a similar study in phantoms with similar parameters and proved that for all nodules, the volume estimation was more accurate, as there was a lower mean absolute percent error (6.5%) compared with macro mode (11.1% to 12.9%). The improvement of volume measurement with the UHR mode was more obvious for small nodule size (3 mm, 5 mm) or star-shaped nodules. Thus, the authors demonstrated that the UHR mode of the evaluated PCD-CT system was able to improve measurement accuracy for nodule volume and nodule shape characterization.

In line with these studies, Kopp et al. [34] demonstrated a better morphological and volumetric evaluation of different nodule phantoms presenting spikes with a different PCD-CT system. The volume estimation indicated better accuracy for PCD-CT compared to CT and HR-CT (with a root mean squared error of 21.3 mm<sup>3</sup> for SPCCT, 28.5 mm3 for CT, and 26.4 mm3 for HR-CT). The Dice similarity coefficient was greater for PCD-CT considering all nodule shapes and sizes (mean, PCD-CT 0.90; CT: 0.85; HR-CT: 0.85). These findings were explained by the higher spatial resolution of the system as demonstrated by a higher modulation transfer function (10% modulation transfer function at 21.7 lp/cm).

Recently, using a clinical prototype [22], we demonstrated the enhanced performances of PCD-CT for solid and ground-glass lung nodule detectability in a phantom study using a task-based model observer assessment and a subjective image quality evaluation. The comparison between a dual-layer CT and the PCD-CT system from the same manufacturer with optimized parameters for each platform found a noise reduction with PCD-CT, as demonstrated by the noise power spectrum analysis and an improved spatial resolution as demonstrated by the task transfer function. As a consequence, the task-based model observer (*d'*) evaluation demonstrated a greater detectability for solid and ground-glass nodules (Figure 6). It is noteworthy that the difference in detectability between CT systems was more pronounced for the ground-glass nodules, which are known to be more difficult to detect due to their low contrast with conventional CT systems. Finally, experienced chest radiologists determined that the best compromise for optimal image quality, noise reduction, and increased spatial resolution was reached by using the iterative reconstruction at level iDose 6, which may enable a clinical use.

**Figure 6.** Comparison of high-resolution CT lung imaging of ground-glass nodules (2, 4, 6 mm) between a dual-layer EID-CT (iQon; Philips, Haifa, Israel) (**A**) and a clinical prototype PCD-CT (SPCCT; Philips) (**B**) in an anthropomorphic thorax phantom with an extension ring simulating an obese patient, using a standard dose protocol (120 kVp, 40 mAs).

Finally, Jungblut et al. [35] evaluated the image quality of pulmonary nodule at different low radiation dose for PCD-CT and EID-CT systems. Their analysis was based on a subjective evaluation by observers and on a commercial artificial intelligence-based, computer-aided detection system. They analyzed an anthropomorphic chest phantom containing 14 pulmonary nodules of different sizes with three vendor-specific scanning modes at decreasing matched radiation dose levels. By comparing PCD-CT to EID-CT images, they found using the HR mode a better image quality with PCD-CT in the subjective analysis as well as the lowest image noise in favor of a better objective analysis. Despite the improved performance, the artificial intelligence-based, computer-aided detection system delivered comparable results for lung nodule detection and volumetry between PCD- and dose-matched EID-CT. Nevertheless, the mean sensitivity PCD-CT images was higher than with EID-CT with, for example, a rate of 95% versus 86% for dose-matched EID-CT with a volume CT dose index of 0.41 mGy.

Taken together, these studies show the potential for more accurate lung nodule evaluation in vitro and in vivo with PCD-CT technology that requires confirmation in a prospective clinical study.

### *3.3. Lung Cancer Screening Imaging*

Lung cancer is the leading cause of cancer deaths worldwide [36], and despite the conclusions of the National Lung Screening Trial that CT screening significantly reduces mortality associated with lung cancer, among other diseases [37], to date, there is no worldwide agreement on the screening recommendations. This can in part be explained by the fact that the diagnostic benefits of lung cancer screening have to be balanced with the inherent risks of ionizing radiation as well as overdiagnosis and work-up of false-positive findings [38].

Symons et al. [39] studied the impact of a low-dose acquisition with PCD-CT and a conventional CT with a 512 matrix on two phantoms by comparing attenuation accuracy, noise, and CNR values between different dose settings. Hounsfield unit stability and accuracy for lung, ground-glass nodules, and emphysema were greater with PCD-CT than EID-CT, particularly at lower doses where the attenuation decreased significantly with EID-CT. In addition, a reduction of noise was consistently found for PCD-CT at 80, 100, and 120 kVp in comparison to EID-CT. As a result, CNR were improved compared to EID for ground-glass nodules and emphysema, particularly for 100 and 80 kVp, without significant difference for other objects (acrylic, water, lung foam, air, etc.). These results support the technical benefits of PCD making PCD-CT a promising tool in radiation dose optimization that is critical in further improving the risk-benefit ratio of CT lung cancer screening; however, they require confirmation in a prospective clinical study.

### *3.4. Low and Ultra-Low Dose Imaging*

Ultra-low and low-dose CT has been increasingly used in low-radiation risk assessment of lung cancer, but this is not the only pulmonary disease where the risk-benefit ratio seems positive; for instance, the follow-up of fibrosing disease at low doses of radiation could be of great interest because of the iterative CT scans. However, the main limitation for use of these protocols is the impairment of the objective and subjective image quality because of the photon starvation and increase in artefacts, such as beam hardening using EID-CT. With PCD-CT, numerous studies have suggested an improvement of the image quality (Figures 7 and 8). For instance, Symons et al. [39] proved on phantoms using a 512 matrix the capacity of PCD technology to be more stable in measuring the parenchyma attenuation at low dose. This may also provide better reproducibility and accuracy for quantitative imaging in the future, such as performed for lung density. The latter is necessary to monitor disease, such as alpha-1 antitrypsin deficiency, emphysema, and idiopathic pulmonary fibrosis, but can also be used for the treatment strategy, such as performed in alpha-1 antitrypsin deficiency [40]. More recently, the same group reported in 30 healthy volunteers the feasibility of a CT lung cancer screening protocol (120 kVp, 20 mAs) with a 512 matrix and an iterative reconstruction algorithm [41]. They showed that experienced readers identified the PCD-CT images as having better diagnostic quality for assessment of lung tissue, lung nodules, soft tissue, and bone with lower subjective image noise when compared to those obtained with the EID system. They also found a better image quality in areas with known beam hardening, e.g., around the vertebrae and in the apical lobes. Quantitative measurements showed that PCD-CT images had 15.2–16.8% lower noise at two different dose levels, with 21.0% higher lung nodule CNR. In addition, we demonstrated the feasibility of using low dose (CTDI: 1.11 mGy) and ultra-low dose (CTDI: 0.10 mGy) protocols in lung nodule evaluation with phantoms and one human volunteer while preserving UHR parameters [9]. We used a 1024 matrix, a large FOV, and slice thickness of 0.25 mm and found a satisfactory detection with low X-ray dose protocol for solid, mixed, and ground-glass nodules. Using ultra-low dose parameters (80 kVp, 3 mAs), ground-glass nodules were not clearly defined because of increasing noise. However, iterative reconstruction (i.e., iDose 5 and 9) enabled a satisfactory nodule visualization. Moreover, improved image quality was found with low-dose PCD-CT images compared to standard-dose EID-CT for lung imaging in a human volunteer despite 87% of flux reduction. Additionally, the use of newly developed iterative reconstruction for PCD-CT data led to a decrease in noise by up to 68% when using iDose reconstruction algorithm (iDose 9). In conclusion, this study showed the feasibility of significant dose reduction with PCD-CT with or without iterative reconstruction while conserving the high-resolution parameters for lung analysis.

Taken together, these studies highlight the feasibility of PCD-CT imaging for improving lung nodule detection with a good compromise between increased spatial resolution, decreased dose, and equivalent or reduced noise.

**Figure 7.** Comparison of high-resolution low dose lung imaging (120 kVp, 10 mAs) between a duallayer EID-CT (iQon, Philips, Haifa, Israel; (**A**): filter YB, (**B**): filter F) and a clinical prototype PCD-CT (Philips; (**C**): filter detailed 1) in an anthropomorphic phantom CT Torso CTU-41 (Kyoto Kagaku, Tokyo, Japan). Both 1024 matrix and field-of-view of 350 mm were matched. Beam hardening was greatly reduced on the PCD-CT images, greatly improving the image quality.

**Figure 8.** Comparison of high-resolution low dose lung imaging (120 kVp, 10 mAs) between a conventional dual-layer CT (iQon; Philips, Haifa, Israel; (**A**): filter YB) and a clinical PCD-CT (Philips; (**B**): filter detailed 1) in an anthropomorphic phantom CT Torso CTU-41 (Kyoto Kagaku, Tokyo, Japan). Both 1024 matrix and field-of-view of 350 mm were matched. Noise was significantly reduced on the PCD-CT images in the upper and basal lobes, greatly improving the image quality.

### *3.5. Interstitial Lung Disease Imaging*

Interstitial lung diseases are complex pathologies and HR-CT evaluation is essential for accurate diagnosis and determination of the best treatment. However, this requires very high spatial resolution because of the microsemiology of their key signs, e.g., intralobular reticulations, bronchiectasis, and honeycombing, which are frequently indeterminate and may lead to invasive procedures, such as biopsies for histological evaluation [42]. In addition, current low-dose protocols, i.e., lower than 1 mSv, are not recommended due to the impaired image quality that may lead to misclassification of the interstitial lung disease [29]. Therefore, PCD, because of its high resolution and noise reduction capacity, could be a promising tool for ILD evaluation.

In an attempt to evaluate the feasibility of PCD-CT in interstitial lung diseases, a recent study conducted by Ferda et al. [12], with a clinical prototype large FOV PCD-CT system, identified a few cases in a cohort of 60 patients addressed for various reasons. They investigated the diagnostic quality of PCD-CT lung imaging using a standard and a lowdose protocol by subjectively assessing critical structures, such as bronchi and bronchial walls. When PCD-CT images were compared to those obtained with a similar dose using EID-CT, the background noise was significantly lower, and the signal-to-noise ratios were significantly higher in PCD-CT images. An important point is that the subjective quality

score between full and low-dose lung imaging was comparable. In patients with interstitial lung disease, the PCD-CT demonstrated a better visualization of higher-order bronchi and third-/fourth-/fifth-order bronchial walls, such as previously suggested with a PCD-CT system from a different manufacturer [9]. However, the authors did not evaluate the diagnostic performance of PCD-CT compared to EID-CT for key signs of fibrosis. Nevertheless, these initial findings indicated an important potential for further radiation dose reduction in interstitial lung disease imaging diagnosis and follow-up.

### *3.6. Distal Airways and Bronchial Imaging*

Bronchial diseases are difficult to evaluate with conventional CT imaging. For patients with chronic obstructive pulmonary disease, the initial morphological evaluation of the bronchial tubes is difficult as well as their follow-up under treatment due to the lack of spatial resolution of conventional scanners. This limitation is also observed for interstitial lung disease, where the differentiation between traction bronchiectasis and honeycombing leads to numerous controversies in the literature [42], leaving a room for image quality improvement using PCD-CT.

For example, Bartlett et al. [30] included 22 patients with pulmonary condition (pneumonia, pulmonary nodule, etc.) and showed improvement in the detection of higher-order bronchi compared with the EID-CT system, irrespective of the image reconstruction kernel used (readers detected more seventh- and eighth-order bronchi). The addition of a sharp reconstruction kernel designed to convey the higher spatial resolution of the PCD-CT system (Q65) further improved the visualization of small bronchi and bronchial walls (walls of the third- and fourth-order bronchi in both lungs). The results of this study demonstrated the potential benefit of PCD-CT lung HR-CT, particularly in assessing airway diseases. Similarly, Kopp et al. [34] compared in-vivo images of a rabbit and a patient acquired with PCD-CT and HR-CT. With HR-CT, bronchi and bronchioles down to a diameter of 1.5–2 mm could be identified. Comparing images of HR-CT and PCD-CT adjusted to the same size, vessels, and walls of bronchioles could be visualized more distinctly with PCD-CT. In addition, we showed a better conspicuity and sharpness of the distal airways as rated by three experienced chest radiologists on a human volunteer, which allowed to see the origin of a 0.5-mm diameter bronchiole that was not visible on an EID-CT [9]. Taken together, these studies support the potential of PCD-CT to propose a new evaluation of bronchial diseases (Figure 9).

### *3.7. Pulmonary Vascularization Imaging*

Pulmonary vascular involvement is highly prevalent in various lung diseases, such as in chronic obstructive pulmonary disease, emphysema, chronic thromboembolic pulmonary hypertension, as well as infectious disease, such as COVID-19 [43–47]. Current standard CT can allow the diagnosis of vessel structure abnormalities, such as thrombus or caliper disparity, while DECT enables the specific imaging of the iodine distribution in the lung, allowing the microcirculation imaging known as a surrogate marker of lung perfusion [48]. Considering the probable improved performance of PCD-CT compared to EID-CT, there is hope for better monitoring the proximal and distal pulmonary vascular involvement in the diseases cited above (Figure 10). For example, this would be of great interest in chronic thromboembolic pulmonary hypertension for the characterization of the involvement of microvasculopathy in the sub-pleural area that might contribute to severe hemodynamics, as suggested by Onishi et al. [49]. In addition, the virtual monoenergetic capabilities of PCD-CT would allow the use of low monoenergetic images, e.g., 40 or 50 keV, to boost iodine attenuation that would help decrease the amount of iodinated contrast agent for a pulmonary CT angiography.

**Figure 9.** Comparison of the airways imaging in a 72-year-old patient with a respiratory bronchiolitis, which was on the same day as a clinical prototype PCD-CT (first row; Philips, Haifa, Israel) and a dual-layer EID-CT (second row; iQon, Philips). Sagittal views (**A**,**B**) demonstrated thickening of the bronchial wall as well as a mosaicism probably due to air trapping. Close-up views of the airspaces (**C**,**D**) showed a better sharpness and conspicuity of the bronchial wall (white arrowheads) with PCD-CT (**D**). Greater sharpness of the bronchial calcifications were noticed on PCD-CT images (**F**) compared to EID-CT images ((**E**); white empty arrowheads).

**Figure 10.** Comparison of the distal arterial pulmonary tree with a dual-layer EID-CT (iQon, Philips, Haifa, lsrael; (**A**)) and a clinical prototype PCD-CT (Philips; (**B**)) and) after injection of iodinated contrast agent. The improvement in quality is visible, notably of the distal vessel lumen and calipers that are depicted until the pleural space.

### *3.8. K-Edge Imaging*

Through its multi-energy CT capabilities, PCD technology enables a spectral tissue characterization and differentiation with or without the use of contrast agents, explaining the use of the term spectral photon-counting CT (SPCCT) in recent studies [1,50–66]. Among these capabilities, K-edge imaging is one of the new features of SPCCT enabled by the energy-resolving characteristics of PCD. It is based on the recognition of the specific K-edge signature of an atom, i.e., the binding energy between the inner electron shell and the atom [67]. Indeed, by dividing the spectrum into well-chosen energy-based datasets, it is possible to detect multiple elements, such as gadolinium, gold, and other atoms that have a K-edge within the relevant energy range of the X-ray spectrum used (e.g., ~40–100 keV; Figure 11) [27]. The main advantage of this technique is to permit a specific and quantitative imaging of the agent without any signal arising from the background, which would allow an improved contrast of the tissue as compared to conventional CT imaging. This type of imaging requires the use of contrast agents adapted for SPCCT, which opens the door for research and development. Among the candidate contrast agents for K-edge imaging, nanoparticles are becoming the agents of choice because of their numerous advantages, such as high payload of atoms, potential for blood pool effect, tunable physicochemical properties, potential for targeting imaging, cell tracking, and theranostic applications [50,68,69].

**Figure 11.** Spectral photon-counting CT (SPCCT) images of a phantom containing tubes with clockwise decrease concentrations of gadolinium from 15 mg/mL to 1 mg/mL of atoms using a clinical prototype (SPCCT; Philips, Haifa, Israel). Conventional image (**A**) and material decomposition of the K-edge image was obtained by reconstructing three material bases (**B**): K-edge image, (**C**): water image, (**D**): iodine image. Only the K-edge images showed the specific signal in the tubes according to the gadolinium concentrations, while water image showed signal from the plastic phantom made of water and the water in the tubes. Iodine image showed no signal accordingly to the absence of iodine.

In 2018 [51], we showed in vitro the feasibility of monocolor and bicolor imaging in order to detect and quantify different mixed contrast agents. Three contrast agents (iodine-, gadolinium-, and gold-based) distributed in tubes at varying proportions were tested by reconstructing the specific images of each material, i.e., based on a two-basis material decomposition (MD) for iodine, three-basis MD for gadolinium or gold with or without iodine, and four-basis MD for gadolinium and gold. The mixtures were prepared such that the solutions could not be differentiated in conventional images. However, distinction was observed in the material images within the same samples, and the measured and prepared concentrations were strongly correlated confirming that SPCCT enables multicolor quantitative imaging in vitro.

Additionally, in 2017 [56], we used a first of its kind system with high-count rate capabilities to demonstrate in vivo in rabbits the feasibility of a dynamic monocolor K-edge imaging of a blood pool agent using PEGylated gold nanoparticles. In the organs analyzed, we showed the persistent visualization of lung vascularization during the first hours after injection. In addition, we presented for the first time the feasibility of bicolor K-edge imaging in animals, i.e., simultaneous differentiation of two different contrast agents, using gold nanoparticles and an iodinated contrast agent (Figure 12). Quantification of these

contrast agents in the lungs demonstrated persistent concentration for gold, while for iodine, an arterial first-pass was noticed with rapid clearance in the lungs, as expected. Taken together, these features may potentially allow a new form of functional imaging, where multiple contrast agents with different pharmacokinetics are used simultaneously in the same biological system.

**Figure 12.** Bicolor imaging of the lung perfusion blood volume in a rabbit after injection of a standard iodinated contrast agent and a contrast agent on a K-edge material (gold nanoparticles) using a spectral photon-counting CT (SPCCT; Philips, Haifa, Israel). Conventional image (**A**) showed the enhancement of the chest vessels as well as the underlying tissue, while the gold K-edge image showed only the signal of gold (**B**). The iodine image showed the signal of iodine as well as misclassification of the bone (**C**).

### **4. Limitations and Perspectives**

The studies cited above provide insight for the future potential of PCD-CT imaging, particularly in-vivo studies that confirm the improvement in spatial resolution and noise management. However, they remain preliminary, and larger patient cohort studies are still needed to identify all the clinical benefits of this new technology. Despite this, it is already possible to consider the clinical potential of SPCCT thanks to studies such as that reported by Yanagawa et al. [70], who proved the value of studying lung adenocarcinomas with a high spatial resolution (2048 matrix on a conventional CT) to define local invasion of the tumor before the surgery.

With respect to the K-edge imaging feature of SPCCT, there are still many hurdles to overcome before considering a clinical translation. These are related to the detection chain and spectral model performance but also by the current contrast agents. It should be noted that, for example, the gadolinium doses needed for in-vivo imaging are far higher than currently used and recommended in MR, and other contrast agents are in an experimental stage that do not allow any kind of clinical human use in the nearer future.

In addition, the current technical limitations of the PCD-CT systems that depend on the progress made by manufacturers should be noted, such as the detector array number, the rotation time, and the energy-resolving capabilities. However, these limitations are expected to be addressed in the near future.

### **5. Conclusions**

Photon-counting detector computed tomography (PCD-CT) represents an emerging medical imaging modality. The key features are an improved spatial and contrast resolution and a significant noise reduction in comparison to the standard energy-integrating detectors CT (EID-CT) as well as spectral capabilities, such as K-edge imaging, that may contribute to the improvement of current CT practice in chest imaging.

**Author Contributions:** Writing—original draft preparation, S.A.S.-M., J.M. and P.-A.R.; Writing review and editing, S.B., H.L., V.L., V.C. and L.B.; Supervision, S.A.S.-M. and P.D. All authors have read and agreed to the published version of the manuscript.

**Funding:** This work was supported by European Union Horizon 2020 grant No 668142 and FLI imaging.

**Acknowledgments:** We are deeply grateful to Philip Robinson for his help in editing the manuscript, Adja Diaw, Daniel Bar-Ness for their help in experiments, Marjorie Villien, Philippe Coulon, Yoad Yagil, Klaus Erhard for their support in the SPCCT project.

**Conflicts of Interest:** The authors declare no conflict of interest.

### **References**


### *Article* **Right Ventricular Volume and Function Assessment in Congenital Heart Disease Using CMR Compressed-Sensing Real-Time Cine Imaging**

**Benjamin Longère 1,\*, Julien Pagniez 2, Augustin Coisne 1, Hedi Farah 2, Michaela Schmidt 3, Christoph Forman 3, Valentina Silvestri 2, Arianna Simeone 2, Christos V Gkizas 2, Justin Hennicaux 2, Emma Cheasty 4, Solenn Toupin 5, David Montaigne <sup>1</sup> and François Pontana <sup>1</sup>**


**Abstract:** Background and objective: To evaluate the reliability of compressed-sensing (CS) real-time single-breath-hold cine imaging for quantification of right ventricular (RV) function and volumes in congenital heart disease (CHD) patients in comparison with the standard multi-breath-hold technique. Methods: Sixty-one consecutive CHD patients (mean age = 22.2 ± 9.0 (SD) years) were prospectively evaluated during either the initial work-up or after repair. For each patient, two series of cine images were acquired: first, the reference segmented multi-breath-hold steady-state free-precession sequence (SSFPref), including a short-axis stack, one four-chamber slice, and one long-axis slice; then, an additional real-time compressed-sensing single-breath-hold sequence (CSrt) providing the same slices. Two radiologists independently assessed the image quality and RV volumes for both techniques, which were compared using the Wilcoxon test and paired Student's *t* test, Bland–Altman, and linear regression analyses. The visualization of wall-motion disorders and tricuspid-regurgitation-related signal voids were also analyzed. Results: The mean acquisition time for CSrt was 22.4 ± 6.2 (SD) s (95% CI: 20.8–23.9 s) versus 442.2 ± 89.9 (SD) s (95% CI: 419.2–465.2 s) for SSFPref (*p* < 0.001). The image quality of CSrt was diagnostic in all examinations and was mostly rated as good (*n* = 49/61; 80.3%). There was a high correlation between SSFPref and CSrt images regarding RV ejection fraction (49.8 ± 7.8 (SD)% (95% CI: 47.8–51.8%) versus 48.7 ± 8.6 (SD)% (95% CI: 46.5–50.9%), respectively; *r* = 0.94) and RV end-diastolic volume (192.9 ± 60.1 (SD) mL (95% CI: 177.5–208.3 mL) versus 194.9 ± 62.1 (SD) mL (95% CI: 179.0–210.8 mL), respectively; *r* = 0.98). In CSrt images, tricuspid-regurgitation and wall-motion disorder visualization was good (area under receiver operating characteristic curve (AUC) = 0.87) and excellent (AUC = 1), respectively. Conclusions: Compressed-sensing real-time cine imaging enables, in one breath hold, an accurate assessment of RV function and volumes in CHD patients in comparison with standard SSFPref, allowing a substantial improvement in time efficiency.

**Keywords:** cardiac; heart; magnetic resonance; CMR; compressed sensing; congenital heart disease; GUCH; real-time imaging

**Citation:** Longère, B.; Pagniez, J.; Coisne, A.; Farah, H.; Schmidt, M.; Forman, C.; Silvestri, V.; Simeone, A.; Gkizas, C.V; Hennicaux, J.; et al. Right Ventricular Volume and Function Assessment in Congenital Heart Disease Using CMR Compressed-Sensing Real-Time Cine Imaging. *J. Clin. Med.* **2021**, *10*, 1930. https://doi.org/10.3390/jcm10091930

Academic Editor: Mickaël Ohana

Received: 16 March 2021 Accepted: 20 April 2021 Published: 29 April 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

### **1. Introduction**

The advent of heart surgery and percutaneous cardiac procedures has considerably improved outcomes in patients born with congenital heart disease [1]. It has led to a growing number of adult survivors with complex congenital heart diseases, with a concomitantly increasing need for imaging follow-up in this clinical context.

Right ventricular (RV) function and volume assessment is of paramount importance in many of these patients, such as in post-repair tetralogy of the Fallot population, as treatment decisions and outcomes mainly rely on RV parameters according to the European Society of Cardiology guidelines for the management of adult congenital heart disease [2]. Although echocardiography remains the first-line investigation, cardiac magnetic resonance (CMR) is a method of choice for RV morphological and functional evaluation in congenital heart disease due to its complex geometry. CMR is considered superior to echocardiography for the evaluation of RV and should be regularly used when the information is essential for patient management, i.e., for quantification of RV volume and ejection fraction, quantification of pulmonary regurgitation, evaluation of RV outflow-tract and pulmonary arteries, detection of myocardial fibrosis or scar, and tissue characterization [2].

However, one major limitation of such extensive CMR examinations is currently the acquisition time, which can be difficult to tolerate in this population, as well as the iterative breath holds, which can be difficult to maintain, leading to poor-quality examinations because of breathing artifacts. To reduce this limitation, the development of acceleration techniques in MR imaging is crucial, and compressed sensing (CS) represents a promising technique in this category. Schematically, CS is a technique that combines a strong and random *k*-space subsampling, thus enabling a very high scan speed, and it uses non-linear iterative reconstructions to make the final image look as close as possible to that if the *k*-space had been fully sampled. The use of CS for CMR cine imaging theoretically enables real-time acquisition with whole-ventricle coverage in a single breath hold, and its reliability has been successfully tested in previous studies for left ventricular (LV) or sometimes right ventricular (RV) functional assessment in healthy volunteers and in patients with various extra-congenital pathologies [3–11].

We thus aimed at evaluating the reliability of real-time cine imaging using the CS technique for quantification of RV and LV function and volumes in congenital heart disease patients in comparison with conventional multi-breath-hold segmented steady-state freeprecession cine imaging.

### **2. Materials and Methods**

### *2.1. Study Population*

From January to April 2019, 61 consecutive patients were prospectively included. All patients were clinically scheduled for CMR in the context of congenital heart disease for either the initial work-up or after repair. A single-ventricle anatomy was considered as an exclusion criterion. The protocol was approved by our institutional Ethics Committee, and the patients gave informed consent.

### *2.2. CMR Protocol*

All CMR examinations were performed on a 1.5 T magnetic resonance scanner (MAG-NETOM Aera, Siemens Healthcare, Erlangen, Germany). For each patient, two series of two-dimensional cine images were systematically acquired: prospectively triggered segmented multi-breath-hold steady-state free-precession sequence (SSFPref) was considered as the reference technique, including a conventional short-axis stack, one LV and one RV two-chamber slice, and one four-chamber slice with an 8 mm slice thickness and a 2 mm gap; an additional prospectively triggered real-time CS sequence (CSrt) in a single breath hold. CSrt cine images were acquired with the same slice number, position, and thickness as those used in the reference technique. An additional phase-contrast imaging sequence was acquired on the pulmonary trunk to assess the RV stroke volume and the severity of

the encountered tricuspid regurgitation. The details of the imaging parameters are listed in Table 1.

**Table 1.** Imaging parameters of the reference steady-state free-precession cine imaging and real-time compressed-sensing cine imaging.


Data are expressed as mean ± standard deviation in the absence of any indication. <sup>a</sup> Constant value. <sup>b</sup> Interpolation was performed to provide a constant frame rate of 20 cardiac phases per cycle for post-processing. <sup>c</sup> The number of views per frame was set according to the shorter R–R interval in order to acquire 20 cardiac phases. <sup>d</sup> The first cardiac cycle is required for signal preparation and the second one for signal acquisition. Abbreviations: SSFPref, reference steady-state free-precession cine; CSrt, real-time compressed-sensing cine; ECG, electrocardiogram; *n*, data represented as numbers; NA, not applicable.

### *2.3. Functional Evaluation*

The quantitative assessment consisted of the evaluation of RV functional parameters with both the SSFPref and CSrt sequences, i.e., ejection fraction (EF), end-diastolic volume (EDV), end-systolic volume (ESV), and stroke volume (SV). The same parameters were measured for the left ventricle, as well as the LV mass. For these quantitative measurements, endocardial and epicardial contours were segmented on the conventional short-axis stacks of cine images using a dedicated analysis software (Cardiac MR analysis workflow, Syngo.via VB30A, Siemens Healthcare, Erlangen, Germany). According to our CMR practice, RV trabeculations were included in the RV volume. Four-chamber and long-axis slices were used as reference images to trace the atrio-ventricular valve planes to ensure an optimal delineation of the heart base for an accurate volume calculation.

### *2.4. Image Quality Assessment*

The overall subjective image quality of the SSFPref and CSrt cine images was rated on the basis of a four-point Likert scale as follows: 4 = excellent, 3 = good, 2 = fair image quality, and 1 = non-diagnostic examination.

In addition, the objective RV image quality was assessed using previously published criteria, which are mostly based on artifact rating, and they were adapted to the RV [12]. Schematically, 1 point was given if an artifact (fold-over, respiratory ghost, cardiac ghost, image blurring/mistriggering, metallic, or shimming) hampered the visualization of the RV border at the end-systole and/or end-diastole; if such an artifact involved 2 or ≥3 slices, 2 or 3 points were given, respectively.

The depictions of the regional RV wall-motion abnormalities (i.e., hypokinetic, akinetic, or dyskinetic wall) were also rated at 4 anatomical levels (base, mid-cavity, apex, and RV outflow tract), and the depictions of tricuspid-regurgitation-related flow artifacts were assessed on the four-chamber slice.

### *2.5. Conditions of Image Analysis*

The acquired SSFPref and CSrt cine images were independently analyzed offline by a CMR radiologist (HF) with 3 years of experience. After anonymization, the images from both sequences were randomized and mixed. The two types of cine sequences from one patient were not read in the same session. A radiologist (JP) with 10 years of experience performed the functional RV assessment on a 20-patient sample for the determination of interobserver agreement with the new CS cine technique [13].

### *2.6. Statistical Analysis*

Categorical data are represented as numbers (percentages). Continuous variables are represented as mean ± standard deviation (SD) (95% confidence interval (CI)) in the case of normal distribution and median (range: minimum–maximum) in other cases. SSFPref and CSrt were compared using paired Student's *t* test, Bland–Altman, and linear regression analyses. The interobserver agreement of CSrt was determined by calculating the intra-class correlation coefficient. An analysis of variance was performed to compare the RV stroke volumes assessed with both cine sequences with the forward pulmonary volume assessed with PCI. Differences in quality scores between SSFPref and CSrt were assessed using the Wilcoxon test. Values of *p* < 0.05 were considered statistically significant. For the depictions of valvular regurgitations and wall-motion disorders, a receiver operating characteristic (ROC) curve was used. The statistical analysis was performed with dedicated software (MedCalc 18.11, MedCalc Software bvba, Ostend, Belgium).

### **3. Results**

### *3.1. Population Description*

The 61 patients (29 men, 32 women; mean age: 22.2 ± 9.0 (SD) years; 95% CI: 19.9–24.5 years) underwent CMR for: tetralogy of Fallot (*n* = 33/61; 54.1%), pulmonary atresia with a ventricular septal defect (*n* = 7/61; 11.5%), cardiac shunt (*n* = 7/61; 11.5%), transposition of great arteries (*n* = 3/61; 4.9%), aortic coarctation (*n* = 2/61; 3.3%), congenital pulmonary stenosis (*n* = 2/61; 3.3%), cor triatriatum sinister (*n* = 2/61; 3.3%), congenitally corrected transposition of the great arteries (*n* = 2/61; 3.3%), pulmonary atresia with intact ventricular septum after biventricular repair (*n* = 2/61; 3.3%), and congenital aortic stenosis (*n* = 1/61; 1.6%). Table 2 summarizes further details of the characteristics of the study population.



Abbreviations: SD, standard deviation; 95% CI, 95% confidence interval.

### *3.2. Cine Acquisitions*

The mean duration for single-breath-hold CSrt acquisition was 22.4 ± 6.2 (SD) s (95% CI: 20.8–23.9 s) versus 442.2 ± 89.9 (SD) s (95% CI: 419.2–465.2 s) for SSFPref (*p* < 0.001). The mean acceleration factor provided by CSrt was 20.8 ± 5.6 (95% CI: 19.3–22.2) as compared with SSFPref. A mean number of 13.3 ± 2.9 slices (95% CI: 12.5–14.1 slices) was acquired with each sequence.

### *3.3. Quantitative Evaluation*

Detailed results regarding the SSFPref and CSrt segmentations for the RV and LV functional parameters are presented in Table 3. There was no statistically significant difference between mean SSFPref and CSrt for RVEDV (192.9 ± 60.1 (SD) mL (95% CI: 177.5–208.3 mL)

versus 194.9 ± 62.1 (SD) mL (95% CI: 179.0–210.8 mL), respectively; *p* = 0.169). The RVEF was slightly underestimated in the CSrt images (CSrt: 48.7 ± 8.6 (SD) % (95% CI: 46.5–50.9%); SSFPref: 49.8 ± 7.8 (SD) % (95% CI: 47.8–51.8%); *p* = 0.006) as a result of a statistically significant but not clinically relevant underestimation of the RVESV in CSrt. The analysis of variance did not demonstrate any significant differences with respect to the RV stroke volume, regardless of the measurement method (SSFPref: 93.6 ± 25.7 (SD) mL (95% CI: 87.0–100.2 mL); CSrt: 92.3 ± 26.0 (SD) mL (95% CI: 85.7–99.0 mL); PCI: 88.6 ± 27.1 (SD) mL (95% CI: 91.6–95.5 mL); *p* = 0.605). No statistically significant differences were visible between SSFPref and CSrt for LVEF and LVEDV. The LV mass was slightly overestimated in CSrt. The linear regression yielded good agreement between both acquisition techniques for all RV functional parameters (Figure 1), and the *r* values were excellent for all parameters. On the other hand, graphical analysis of the Bland–Altman plot demonstrated up to five (tetralogy of Fallot, *n* = 5/5; 100%) paired measurements out of the limits of agreement (LOA) depending on the RV parameter considered (LOA in RVEF bias: −13.7 to +9.3%).

**Table 3.** Functional parameters segmented on both the reference steady-state free-precession and real-time compressedsensing cine.


ICC was used to evaluate the interobserver agreement for the RV segmentation. The significance of Student's *t* test is defined by *p* < 0.05. Abbreviations: SSFPref, reference steady-state free-precession cine; CSrt, real-time compressed-sensing cine; SD, standard deviation; 95% CI, 95% confidence interval; RV, right ventricular; LV, left ventricular; EDV, end-diastolic volume; ESV, end-systolic volume; SV, stroke volume; LVM, left ventricular mass; ICC, intraclass correlation coefficient; Inter, interobserver; Intra, intraobserver.

### *3.4. Qualitative Evaluation*

Figure 2; Figure 3; Video S1; Video S2 (Supplementary Materials) provide representative examples of the image quality achieved with CSrt images in various clinical situations. The image quality of CSrt was diagnostic in all examinations (Table 4). There was a significantly lower overall image quality score for CSrt images (*p* = 0.0001) because most of the examinations were rated as excellent with SSFPref and good with CSrt. However, qualitative artifact presence was statistically lower in the CSrt images than in SSFPref (*p* = 0.0016). Considering SSFPref as the gold standard, there were no diagnostic losses for regional RV wall-motion abnormalities in CSrt images, demonstrating a 100% sensitivity and specificity (normokinetic: *n* = 157/244 (64.3%); hypokinetic: *n* = 39/244 (16.0%); akinetic: *n* = 1/244 (0.4%); dyskinetic: *n* = 47/244 (19.3%)). The tricuspid-regurgitation-flow void depictions in CSrt images had a sensitivity and specificity of 74.2% and 100%, respectively (predictive positive value = 100%; predictive negative value = 78.9%; area under ROC = 0.87). Using

SSFPref, 23/61 (37.7%) tricuspid regurgitations were depicted (mild: 20/23 (87.0%); moderate: 3/23 (13.0%)). Of the 8/61 (13.1%) tricuspid regurgitations that were not depicted with the CSrt cine, all were quantified as mild with the reference technique (the difference between RVSV and anterograde pulmonary volume was measured with the phase-contrast sequence) [14].

**Figure 1.** Bland–Altman plots and linear regression trendlines for quantification of the right ventricular functional parameters. Left column: Linear regression trend lines for (**a**) RVEF, (**c**) RVEDV, (**e**) RVESV, and (**g**) RVSV, representing the correlation between parameters measured on the SSFPref and CSrg sequences. Right column: Bland–Altman plots for the (**b**) RVEF, (**d**) RVEDV, (**f**) RVESV, and (**h**) RVSV. Solid blue lines are the mean differences and dashed green lines are the 95% limits of agreement. Abbreviations: SSFPref, reference steady-state free-precession cine; CSrt, real-time compressed-sensing cine; SD, standard deviation; RVEF, right ventricular ejection fraction; RVEDV, right ventricular end-diastolic volume; RVESV, right ventricular end-systolic volume; RVSV, right ventricular stroke volume.

**Figure 2.** Four-chamber cine slice acquired with both sequences in a 31-year-old male patient referred for transposition of great arteries after a Senning repair follow-up. SSFPref view in the diastole (**a**) and systole (**c**); overall image quality score = 4/4; RVEF = 42%; EDV = 345 mL. The same slices acquired with CSrt in the diastole (**b**) and systole (**d**); overall image quality score = 3/4; RVEF = 40%; EDV = 346 mL. The tricuspid regurgitation flow artifact remains conspicuous with both sequences (blue arrow). Abbreviations: SSFPref, reference steady-state free-precession cine; CSrt, real-time compressed-sensing cine; RVEF, right ventricular ejection fraction; EDV, end-diastolic volume.

**Figure 3.** *Cont.*

**Figure 3.** Short-axis cine slice acquired with both sequences in a 26-year-old female patient referred for a tetralogy of Fallot post-repair follow-up, demonstrating an irregular heart rate. Mean heart rate = 78 ± 14 (SD) bpm (range: 51 to 107 bpm). SSFPref view in the diastole (**a**) and systole (**c**); overall image quality score = 2/4; RVEF = 51%; EDV = 148 mL. The same slices were acquired with CSrt in the diastole (**b**) and systole (**d**); overall image quality score = 3/4; RVEF = 48%; EDV = 154 mL. The fair image quality is due to the mis-triggering of artifacts in SSFPref, while CSrt provided both accurate segmentation and good image quality. Abbreviations: bpm, beats per minute; SD, standard deviation; SSFPref, reference steady-state free-precession cine; CSrt, real-time compressed-sensing cine; RVEF, right ventricular ejection fraction; EDV, end-diastolic volume.

**Table 4.** Qualitative assessment of the reference steady-state free-precession cine and real-time compressed-sensing cine.


Considering SSFPref as the gold standard, CSrt demonstrated the following diagnostic performances for the depiction of tricuspid-regurgitation-flow-related artifacts: sensitivity = 74.2%; specificity = 100%; positive predictive value = 100%; negative predictive value = 78.9%; area under ROC = 0.87. Abbreviations: SSFPref, reference steady-state free-precession cine; CSrt, real-time compressed-sensing cine; TR+, conspicuous tricuspid-regurgitation-flow-related artifact; TR–, no tricuspid-regurgitation-flow-related artifact depicted; ROC, receiver operating characteristic.

### **4. Discussion**

Our prospective monocentric study based on a cohort of 61 pediatric and grown-up CHD patients, including 33 tetralogies of Fallot, demonstrated that the quantification of RV function and volumes yields similar results for CSrt and for the standard SSFPref cine techniques, while the former allows a drastically shorter acquisition time. The agreement between CSrt and SSFPref regarding the RV volume assessment is in line with the findings of previous studies performed on smaller cohorts of healthy volunteers and non-CHD patients [5,10]. The *t* test comparisons performed in our study demonstrated a statistically significant trend towards a 1.07% RVEF underestimation (relative mean difference = −2.14%), a 3.51 mL RVESV overestimation (relative mean difference = 3.54%), and a 1.28 mL RVSV underestimation (relative mean difference = −1.36%) with CSrt. The segmented steady-state free precession cine is currently considered the gold-standard technique for the measurement of ventricular volumes, including in CHD patients [2,15–17]. However, these trends must be balanced with the clinically relevant LOA demonstrated in the Bland–Altman plots, mainly regarding the relative bias of RVEF (−13.7; +9.3%), which was also the case in other studies evaluating CS cine for RV assessment in non-CHD populations, where the LOA was reported to be from −10.5 to +11.6% [5,10,18,19]. A more recent study also found a similar performance of a novel real-time steady-state free precession spiral sequence reconstructed using CS in a pediatric-only CHD population, but did not evaluate the regional wall-motion abnormalities or the depiction of tricuspid-regurgitation-related flow artifacts [19]. It must also be highlighted that both intra- and interobserver agreement was excellent for all CSrt-evaluated RV functional parameters. Additionally, our findings demonstrate a strong agreement between both SSFPref and CSrt for functional LV parameters, despite the slight LV mass overestimation with CSrt, as previously reported [5].

The best image plane required for post-processing of RV volumes has long been debated, especially in the clinical context of CHD. However, it has been shown that despite a trend favoring the axial plane rather than the short axis in terms of reproducibility, there were no clinically significant differences between these two contouring methods [20]. Thus, we drew RV endocardial contours on short-axis stacks, as this is widely performed and is easier to set up in routine practice, but care was taken to trace the tricuspid valve on reference four-chamber and RV long-axis slices to delimit the right ventricular basis as precisely as possible. Despite controversies about segmentation methods, we included trabeculations in RV volumes according to our CMR center's habits [21]. The justification for this choice lies in the need for consistency in our practice in order to preserve reproducibility in patient follow-ups [22,23]. Nevertheless, we acknowledge that excluding trabeculations from the blood volume could be more accurate [24]. This could explain the lower dispersion of the differences in parameters measured with SSFPref and CSrt on Bland–Altman plots when the RVEF or RVESV increase (Figure 1b,h).

The CSrt images were diagnostic in all examinations, but the overall image quality score was, as expected, significantly lower with this technique. This can be explained by the lower edge definition provided by CS, which resulted in a slightly blurry aspect of the images. This should be addressed by using a two-shot variant of the evaluated compressed-sensing cine, which would provide an improved edge sharpness and would preserve the important scan time reduction [25]. Nevertheless, this two-shot variant has not yet been evaluated for the right ventricular functional parameters and should be the subject of further study. However, a lower RV artifact score was found with CSrt due to the reduction of artifacts—which were mostly related to mis-triggering—achieved with this real-time acquisition technique. In addition, the performance of CSrt for RV wall-motion disorder depiction was very high, as there was no diagnostic loss in comparison with the reference images, and only mild tricuspid regurgitations were not depicted with CSrt cine (Figure 2; Video S1 (Supplementary Materials)). These findings are in line with a recently published study evaluating the same real-time CS cine sequences for both LV and RV assessment in a non-selected adult cohort [18].

The CSrt sequence consisted of a single-breath-hold cine acquisition; however, some patients could not fully achieve the required apnea due to their clinical condition. They were not excluded from the study, as our aim was to be as representative as possible of our CHD population that we encounter in daily practice. Despite these free-breathing ends of acquisition, no major artifacts (CMR RV quality score > 7/10) were noticed in the CSrt images, which all had diagnostic quality. These findings strongly suggest the possibility of the free-breathing acquisition of CSrt cine, which is particularly relevant for pediatric or end-stage CHD patients. Although the aim of this study was not to evaluate free-breathing imaging protocol, free-breathing CSrt has been demonstrated to be a reliable alternative that allows faster acquisition than sequences based on registration of multiple acquisitions and motion-correction algorithms [9,26]. The acceleration provided by CSrt may allow one to either (a) shorten breath-holding duration by splitting the stacks of cine slices to reduce the number of cine loops acquired per breath hold, especially for patients with shortness of breath, or (b) to shorten the overall examination duration, as suggested in the present study (average: 22 s (CSrt) versus 7 min and 22 s (SSFPref) for 13 cine slices) to improve the clinical workflow and tolerance in children or to spare examination time in order to acquire additional sequences. Indeed, a comprehensive study of hemodynamic patterns is a key point in the initial work-up of CHD or in repair follow-up. Four-dimensional (4D) flow is a promising but time-consuming technique that may take advantage of the thusly spared time [27,28]. Depending on the sequence design, this technique may provide both qualitative and quantitative assessments of flow patterns and ventricular volumes [29]. As extended examination durations are an obstacle for the routine use of 4D flow, compressedsensing 4D flow prototypes are also being developed [30–32].

Another interesting point is the decrease in the mis-triggering artifacts observed with CSrt (Figure 3; Video S2 (Supplementary Materials)). Even though it was not the purpose of our study and would require further dedicated studies, this finding suggests that CSrt might have an important part to play for functional or WMD evaluations in patients with irregular heart rates [33].

### *Limitations*

The minimum age in our population was 7 years, and further studies would be needed for validation in younger children. We also have to report that despite the drastic decrease in acquisition time, the data reconstruction process was more time consuming than with SSFPref, as 2 min were necessary in order to visualize the whole cine stack in spite of a graphics processing unit upgrade.

Regarding the blurry aspect of images that we observed with CS, it must be said that the CSrt sequence was designed in order to reduce the acquisition time as much as possible. In a different way, some authors have successfully tested CS cine to improve the spatial or temporal resolution with quite similar or even moderately shorter acquisition times than those for reference SSFPref, or even to achieve three-dimensional cine acquisitions [34–36]. Our scan time was, however, strongly reduced, as low as 22.4 ± 6.2 (SD) s versus about 6 to 10 min for segmented multi-breath-hold SSFPref, which can be very useful for patient comfort and workflow.

Although it is in line with the current literature, the bias observed between the two sequences in RVEF measurement (∼10%) is clinically relevant and must be taken into consideration in CHD follow-up [37,38]. The reason for such a bias may lie in the edge sharpness impairment induced by CSrt as compared to SSFPref [18]. Indeed, the partial Fourier and the interpolation performed to provide a constant cardiac frame rate for postprocessing induced a smoother and blurrier endocardial delineation than conventional cine. This limitation should be responsible for the increased bias in segmentation between the two techniques and may be solved by a multi-shot approach to CS acceleration [25].

### **5. Conclusions**

Compressed-sensing real-time cine imaging enables the assessment RV function and volumes in patients with CHD while providing a significant reduction in examination duration and allowing an improvement in time efficiency and patient care.

**Supplementary Materials:** The following are available online at https://www.mdpi.com/article/ 10.3390/jcm10091930/s1, Video S1: Four-chamber cine slice in a 31-year-old male patient referred for transposition of great arteries after a Senning repair (same patient as Figure 2); Video S2: Fourchamber cine slice in a 27-year-old female patient referred for tetralogy of Fallot follow-up.

**Author Contributions:** B.L.: data collection, interpretation and analysis, drafting of the manuscript, critical revision for important intellectual content; J.P.: study conception and design, data collection, interpretation and analysis, drafting of the manuscript, critical revision for important intellectual content; A.C.: study conception, critical revision for important intellectual content; H.F.: data collection, interpretation and analysis, drafting of the manuscript; M.S.: study conception and design, critical revision for important intellectual content; C.F.: study conception and design, critical revision for important intellectual content; V.S.: data collection and interpretation, critical revision for important

intellectual content; A.S.: data collection and interpretation, critical revision for important intellectual content; C.V.G.: data collection and interpretation, critical revision for important intellectual content; J.H.: data collection and interpretation, critical revision for important intellectual content; E.C.: study conception, critical revision for important intellectual content; S.T.: study conception and design, critical revision for important intellectual content; D.M.: study conception, critical revision for important intellectual content; F.P.: study conception and design, data collection, interpretation and analysis, drafting of the manuscript, critical revision for important intellectual content. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Institutional Review Board Statement:** The study was approved by the research ethics committee of Lille University Hospital.

**Informed Consent Statement:** Informed consent was obtained from all subjects involved in the study.

**Data Availability Statement:** The data presented in this study are available on reasonable request from the corresponding author, subject to approval by the research ethics committee of Lille University Hospital.

**Conflicts of Interest:** B.L., J.P., A.C., H.F., V.S., A.S., C.V.G., J.H., E.C., D.M., and F.P. have no competing interests. They are employed by an institution engaged in a contractual collaboration with Siemens Healthcare. M.S., C.F., and S.T. are employees of Siemens Healthcare GmbH.

### **References**


### *Article* **60-S Retrogated Compressed Sensing 2D Cine of the Heart: Sharper Borders and Accurate Quantification**

**Benjamin Longère 1,\*, Christos V. Gkizas 2, Augustin Coisne 1, Lucas Grenier 2, Valentina Silvestri 2, Julien Pagniez 2, Arianna Simeone 2, Justin Hennicaux 2, Michaela Schmidt 3, Christoph Forman 3, Solenn Toupin 4, David Montaigne <sup>1</sup> and François Pontana <sup>1</sup>**


**Abstract:** Background and objective: Real-time compressed sensing cine (CSrt) provides reliable quantification for both ventricles but may alter image quality. The aim of this study was to assess image quality and the accuracy of left (LV) and right ventricular (RV) volumes, ejection fraction and mass quantifications based on a retrogated segmented compressed sensing 2D cine sequence (CSrg). Methods: Thirty patients were enrolled. Each patient underwent the reference retrogated segmented steady-state free precession cine sequence (SSFPref), the real-time CSrt cine and the segmented retrogated prototype CSrg sequence providing the same slices. Functional parameters quantification and image quality rating were performed on SSFPref and CSrg images sets. The edge sharpness, which is an estimate of the edge spread function, was assessed for the three sequences. Results: The mean scan time was: SSFPref = 485.4 ± 83.3 (SD) s (95% CI: 454.3–516.5) and CSrg = 58.3 ± 15.1 (SD) s (95% CI: 53.7–64.2) (*p* < 0.0001). CSrg subjective image quality score (median: 4; range: 2–4) was higher than the one provided by CSrt (median: 3; range: 2–4; *p* = 0.0008) and not different from SSFPref overall quality score (median: 4; range: 2–4; *p* = 0.31). CSrg provided similar LV and RV functional parameters to those assessed with SSFPref (*p* > 0.05). Edge sharpness was significantly better with CSrg (0.083 <sup>±</sup> 0.013 (SD) pixel<sup>−</sup>1; 95% CI: 0.078–0.087) than with CSrt (0.070 <sup>±</sup> 0.011 (SD) pixel<sup>−</sup>1; 95% CI: 0.066–0.074; *<sup>p</sup>* = 0.0004) and not different from the reference technique (0.075 <sup>±</sup> 0.016 (SD) pixel−1; 95% CI: 0.069–0.081; *p* = 0.0516). Conclusions: CSrg cine provides in one minute an accurate quantification of LV and RV functional parameters without compromising subjective and objective image quality.

**Keywords:** cardiac; heart; magnetic resonance; CMR; compressed sensing; fast imaging; function; retrospective; retrogating; image quality

### **1. Introduction**

Cardiac magnetic resonance (CMR) is the reference standard method for quantification of volumes, ejection fraction (EF) and mass of left (LV) and right ventricles (RV) [1–3]. Reliable volumes assessment is required since EF has a strong prognostic value regarding clinical outcomes and survival [4–6]. However, besides steady-state free precession cine images essential for quantification, phase contrast angiography, gadolinium enhanced imaging, and additional sequences may be recommended depending on heart conditions,

**Citation:** Longère, B.; Gkizas, C.V.; Coisne, A.; Grenier, L.; Silvestri, V.; Pagniez, J.; Simeone, A.; Hennicaux, J.; Schmidt, M.; Forman, C.; et al. 60-S Retrogated Compressed Sensing 2D Cine of the Heart: Sharper Borders and Accurate Quantification. *J. Clin. Med.* **2021**, *10*, 2417. https://doi.org/ 10.3390/jcm10112417

Academic Editor: Mickaël Ohana

Received: 16 March 2021 Accepted: 26 May 2021 Published: 29 May 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

leading to an extended scan time, which may be difficult to handle for patients suffering from cardiac-related shortness of breath, since multiple breath-holdings are required for the acquisition [7].

To reduce acquisition time, compressed sensing (CS) has recently been applied to magnetic resonance imaging, especially CMR [8]. Based on Candès et al.'s work on signal recovery from incomplete sampling, Donoho proposed the CS acquisition [9,10]. Since the signal compression required for archiving and transfer implies the deletion of most acquired data, the principle of CS is to acquire only the pieces of information that would be preserved after this compression, sparing the time necessary for the acquisition of the data that would finally be purged during this process. Three prerequisites are mandatory for CS [11]. First, contrary to most medical images, the signal must be compressible. This means a sparsifying transform is required for most transformed coefficients to be insignificant, making the transformed image compressible. Second, the undersampling must be incoherent to provide noise-like overfolding and avoid ambiguity. Finally, nonlinear iterative reconstructions enforcing image consistency with measured signal and transformed image sparsity are needed. In the case of cine imaging, the redundancy in the cardiac cycle provides a strong spatiotemporal correlation, which can be exploited for additional acceleration [12–14]. Undersampling is not only performed in plane but also in the temporal domain by using variable sampling density maps for successive frames.

The first-generation real-time CS sequence (CSrt) has already been widely evaluated regarding volumes, EF and mass for both ventricles, at 1.5 and 3-Tesla [15–25]. These studies used 80 iterations to reconstruct CSrt data, this setting being supposed to provide the best compromise between image quality and reconstruction time [17]. Nevertheless, computing times over one minute per slice were reported, making such a setting not compatible with clinical use [26,27]. Vermersch et al. evaluated the same sequence halving the number of iterations and demonstrated a similar agreement for the quantification of volumes, EF and mass quantification than previously reported in the literature, extending the use of CSrt in clinical practice [22]. To provide an acceleration factor compatible with single-breath-hold real-time imaging, additional sensitivity encoding imaging (SENSE) and partial Fourier are implemented, the latter being a potential cause of edge sharpness impairment [28]. Another limitation is the limited temporal resolution of 49 ms which may provide variable numbers of frames per slice in the case of heart rate variability. To allow post-processing, a 25-frame per slice interpolation is performed. Besides the border blurring induced by this normalization, end-systolic volumes may be overestimated in the case of a fast heart rate [29,30].

A new generation of retrogated CS sequence (CSrg) has been released. It features segmented acquisition to improve temporal resolution and partial Fourier switch-off to improve edge sharpness. The purpose of this study was to assess the image quality and the ventricular functional parameters in comparison with CSrt and the reference retrogated segmented steady-state free precession cine sequence (SSFPref).

### **2. Materials and Methods**

### *2.1. Study Population*

From March to April 2019, 30 consecutive adult patients referred for rest CMR were included. Exclusion criteria were grown-up congenital heart disease work-up or follow-up, underaged patients, patients suffering from arrhythmia for whom the use of prospective ECG gating was necessary, MRI contraindications and patient refusal. Patients gave informed consent, and the protocol was approved by our Institutional Ethics Committee.

### *2.2. Imaging Protocol*

CMR studies were performed on a 1.5-T scanner (MAGNETOM Aera, Siemens Healthcare, Erlangen, Germany). Every patient underwent three series of cine images: first the reference retrogated segmented multi-breath-hold SSFP sequence (SSFPref); then the CS-accelerated SSFP real-time sequence (CSrt) acquired in two breath-holds, and finally

the retrogated segmented SSFP prototype with CS-fashioned acceleration requiring three breath-holds (CSrg). One LV 2-chamber, one RV 2-chamber, one 4-chamber slice and a LV short-axis stack covering the entire ventricles were acquired with the three above-cited sequences, providing identical slice position, thickness, and number. Imaging parameters for the three sequences are summarized in Table 1.

**Table 1.** Imaging parameters of the reference steady-state free precession cine, real-time compressed sensing cine and segmented retrogated compressed sensing cine.


Data are expressed as mean ± standard deviation in the absence of any indication. <sup>a</sup> Constant value. Abbreviations: SSFPref, reference steady-state free precession cine; CSrt, real-time compressed sensing cine; CSrg, segmented retrogated compressed sensing cine.

### *2.3. Cine Images Quality Assessment*

A 3-step evaluation was performed for the three sequences.

First, a subjective overall image quality was assessed using a 4-point Likert scale (1: non diagnostic, 2: fair, 3: good, 4: excellent).

Then acquisition quality was evaluated using a standardized score based on the "*LVfunction cine SSFP*" section of the criteria established from the European CMR registry, evaluating the artifact detection [15] (p. 3). This score was modified, removing its four last items since their score was systematically null, accordingly to our center practice (Table 2). This score increased with acquisition impairment.

**Table 2.** "*LV-Function cine SSFP*" section of the standardized objective quality criteria score based on the European CMR registry. Adapted from [15] (p. 3).



The four last items were nulled since acquisitions were repeated when orientation was not appropriated (item 9 = 0); all acquisitions were performed using a 8-mm thickness (item 10 score = 0) and a 2-mm gap (item 11 score = 0), horizontal and vertical long axis views were systematically acquired (item 12 score = 0). Consequently, italic criteria were not applied, and only bold criteria were used for objective quality assessment in our study, providing a maximum score of 10 points. The more artifacts there were, the higher the score was. Abbreviations: LV, left ventricle; SSFP, steady-state free precession.

> Finally, the edge sharpness (ε) between myocardium and LV blood pool was measured on end-diastole 4-chamber view (Figure 1). This assessment was performed using a MATLAB (version R2015a, The MathWorks, Natick, MA, USA) homemade script. An intensity profile line was drawn perpendicularly to the mid-cavity interventricular septum border with the LV blood pool at end-diastole [31–33]. The ε value was a spatial frequency (pixel<sup>−</sup>1) calculated as the inverse of the distance separating the two points corresponding to 20% and 80% of the difference between the minimum and maximum intensities along this line.

**Figure 1.** Edge sharpness measurement. An orthogonal profile line is drawn across the border between the interventricular myocardium and the left ventricular blood pool (blue line) on a 4-chamber view at end-diastole provided by (**a**) the reference segmented SSFP sequence, (**b**) the real-time compressed sensing sequence and (**c**) the retrogated compressed sensing prototype, providing intensity profiles (blue curves) along the line for (**d**) reference, (**e**) real-time and (**f**) prototype sequences. The edge sharpness was calculated as the inverse of the distance *d* (in pixels) from the positions corresponding to 20% and 80% (red stars) of the difference between the maximum and minimum signal intensities (black crosses) along the profile line and was expressed in pixel−1. Abbreviations: SSFPref, reference steady-state free precession; CSrt, real-time compressed sensing; CSrg, retrogated compressed sensing; ε, edge sharpness; Ymax, maximum signal intensity; Ymin, minimum signal intensity; d, distance along the intensity (pixels).

### *2.4. Functional Evaluation*

Assessment of end-diastolic volumes (EDV), end-systolic volumes (ESV), stroke volumes (SV) and EF were performed for both ventricles as well as LV mass (LVM). These parameters were measured on short-axis stacks with semi-automated segmentation with manual correction of the LV endocardium and epicardium while manual segmentation of the RV endocardium was necessary, using dedicated 4D analysis software (Cardiac MR analysis workflow, Syngo.via VB30A, Siemens Healthcare, Erlangen, Germany). Fourchamber, LV and RV 2-chamber slices were used to define mitral and tricuspid valve planes to ensure optimal assessment of ventricular bases.

### *2.5. Conditions of Image Analysis*

The three datasets were independently analyzed by a 4-year experience CMR radiologist (LG). After anonymization and randomization performed for each sequence, each dataset was analyzed separately. Each analysis session, which evaluated all the images of the same sequence, was separated from the previous one by one month. A second radiologist with 11 years of experience in CMR (BL) segmented both ventricles on CSrg images for interrater variability assessment.

### *2.6. Statistics Analysis*

Categorical data were presented as numbers (percentage) and continuous variables as mean ± standard deviation (SD) (95% confidence interval (CI)) in the case of normal distribution or median (range: minimum–maximum) in other cases. Variable normality was assessed using the D'Agostino–Pearson test.

Paired Wilcoxon signed-rank test was used to compare subjective image qualities and acquisition qualities between CSrg and SSFPref or CSrt. An analysis of variance was performed to compare edge sharpness and acquisition times of the three sequences. CSrg and SSFPref mean functional parameters were compared using a Student's *t* test, with linear regression and Bland–Altman analysis to assess the agreements between both methods. Inter and intra-observer variabilities were assessed using intra-class coefficient correlation. Significance of the test was defined by values of *p* < 0.05.

As for valvular regurgitations and WMD visualization, a receiver operating characteristic (ROC) curve was used. Statistical analysis was performed using dedicated commercially available software (MedCalc 14.8.1.0, MedCalc Software, Ostend, Belgium).

### **3. Results**

### *3.1. Population Description*

Demographics data are listed in Table 3. The 30 patients (22 men, 8 women; mean age: 48.0 ± 21.0 (SD) years; 95% CI: 40.2–55.9 years) were referred for: heart valve disease (*n* = 7/30; 23.3%), ischemic cardiopathy (*n* = 5/30; 16.7%); dilated cardiomyopathy (*n* = 5/30; 16.7%), myocarditis (*n* = 5/30; 16.7%), left ventricular hypertrophy (*n* = 5/30; 16.7%) and infiltrative cardiomyopathy (*n* = 3/30; 10%). All patients could fully complete CSrt and CSrg breath-holdings.


**Table 3.** Study population characteristics.

Abbreviations: SD, standard deviation; 95% CI, 95% confidence interval.

### *3.2. Scan Time and Image Quality*

SSFPref mean scan time was 485.4 ± 83.3 (SD) s (95% CI: 454.3–516.5 s) while CSrt scan time was 23.9 ± 5.2 (SD) s (95% CI: 21.9–25.8 s) and CSrg scan time was 58.3 ± 15.1 (SD) s (95% CI: 53.7–64.2 s) (*p* < 0.0001). Compared to SSFPref, the mean acceleration factor provided by CSrg was 8.7 ± 2.6 (SD) (95% CI: 4.5–15.7). A mean number of 15.0 ± 1.2 (SD) slices (95% CI: 14.6–15.4 slices) was acquired with each technique. For CSrg reconstruction of the full dataset, the mean time was 82.9 ± 23.4 (SD) s (95% CI: 77.4–110.1 s).

All images were rated as diagnostic. CSrg subjective quality score (median: 4; range: 2–4) was higher than the one provided by CSrt (median: 3; range: 2–4; *p* = 0.0008). SSFPref overall quality score (median: 4; range: 2–4) was not different from CSrg score (*p* = 0.31).

Regarding the acquisition quality based on the EuroCMR registry, the CSrg sequence (median: 0; range: 0–3) was not different from either the SSFPref acquisition (median: 0; range: 0–3; *p* = 0.38) or the CSrt cine (median: 0; range: 0–3; *p* = 0.83).

The CSrg demonstrated a significantly better edge sharpness than CSrt (εCSrg = 0.083 ± 0.013 (SD) pixel−<sup>1</sup> (95% CI: 0.078–0.087 pixel−1) versus <sup>ε</sup>CSrt = 0.070 ± 0.011 (SD) pixel−<sup>1</sup> (95% CI: 0.066–0.074 pixel<sup>−</sup>1); *p* = 0.0004). Moreover, no significant difference was demonstrated between CSrg and SSFPref (εSSFPref = 0.075 ± 0.016 (SD) pixel−<sup>1</sup> (95% CI: 0.069–0.081 pixel−1); *p* = 0.0516) (Figure 2; Figure 3; Video S1; Video S2 (Supplementary Materials)).

(**a**) (**b**) (**c**)

**Figure 2.** Midventricular short-axis cine slice acquired with the three cine sequences in a 37-year-old patient referred for myocarditis suspicion. (**a**) Reference steady-state free precession cine: Likert scale = 4/4; EuroCMR score = 0/10; ε = 0.082 pixel<sup>−</sup>1; LVEF = 59%, LVEDV = 135 mL, LVM = 83 g, RVEF = 63%, RVEDV = 159 mL; (**b**) Real-time compressed sensing cine: Likert scale = 4/3; EuroCMR score = 0/10; ε = 0.056 pixel−1; LVEF = 60%, LVEDV = 133 mL, LVM = 81 g, RVEF = 61%, RVEDV = 153 mL; (**c**) Retrogated compressed sensing cine: Likert scale = 4/4; EuroCMR score = 0/10; ε = 0.081 pixel<sup>−</sup>1; LVEF = 58%, LVEDV = 133 mL, LVM = 86 g, RVEF = 62%, RVEDV = 164 mL. Abbreviations: ε, edge sharpness; LVEF, left ventricular ejection fraction; LVEDV, left ventricular end-diastolic volume; LVM, left ventricular myocardial mass; RVEF, right ventricular ejection fraction; RVEDV, right ventricular end-diastolic volume.

**Figure 3.** Edge sharpness comparison assessed at end-diastole. (**a**) There was no significant difference regarding ε between CSrg and SSFPref (*p* = 0.0516) (**b**) but CSrg significantly improved ε while compared with CSrt (*p* = 0.0004). Abbreviations: CSrg, retrogated compressed sensing; CSrt, real-time compressed sensing; SSFPref, reference steady-state free precession; ε, edge sharpness.

### *3.3. Volumes, Functions and Mass Quantification*

Good agreements were yielded by Bland–Altman and linear regression analyses for both LV (Figure 4) and RV (Figure 5) assessments. No significant difference was demonstrated regarding LVM, LV and RV volumes (EDV, ESV, SV) and EF (*p* > 0.05) (Table 4). Intrarater variability was excellent, demonstrating intraclass correlation coefficients (ICC) greater than 0.99 for both ventricles, as were interrater variabilities for LV (ICC ≥ 0.97) and RV (ICC ≥ 0.96).

**Figure 4.** *Cont*.

**Figure 4.** Bland–Altman plots and linear regression trendlines applied to left ventricular functional parameters quantifications. Left column: Bland–Altman plots for (**a**) LVEF, (**c**) LVEDV, (**e**) LVESV, (**g**) LVSV and (**i**) LVM. Solid blue lines are the mean differences between parameters measured with SSFPref and CSrg sequences and dashed red lines are the 95% limits of agreement. Right column: linear regression trend lines for (**b**) LVEF, (**d**) LVEDV, (**f**) LVESV, (**h**) LVSV and (**j**) LVM. Abbreviations: SSFPref, reference steady-state free precession; CSrg, retrogated compressed sensing; SD, standard deviation; LVEF, left ventricular ejection fraction; LVEDV, left ventricular end-diastolic volume; LVESV, left ventricular end-systolic volume; LVSV, left ventricular stroke volume; LVM, left ventricular mass.

**Figure 5.** Bland–Altman plots and linear regression trendlines applied to right ventricular functional parameters quantifications. Left column: Bland-Altman plots for (**a**) RVEF, (**c**) RVEDV, (**e**) RVESV, and (**g**) RVSV. Solid blue lines are the mean differences between parameters measured with SSFPref and CSrg sequences and dashed red lines are the 95% limits of agreement. Right column: linear regression trend lines for (**b**) RVEF, (**d**) RVEDV, (**f**) RVESV, and (**h**) RVSV. Abbreviations: SSFPref, reference steady-state free precession; CSrg, retrogated compressed sensing; SD, standard deviation; RVEF, right ventricular ejection fraction; RVEDV, right ventricular end-diastolic volume; RVESV, right ventricular end-systolic volume; RVSV, right ventricular stroke volume.


**Table 4.** Functional parameters segmented on both the reference steady-state free-precession and retrogated compressedsensing cine.

ICC assessed the inter and intrarater agreements for ventricular segmentations. The significance of Student's *t* test is defined by values of *p* < 0.05. Abbreviations: SSFPref, reference steady-state free-precession cine; CSrg, retrogated compressed-sensing cine; SD, standard deviation; 95%CI, 95% confidence interval; LV, left ventricular; RV, right ventricular; EF, ejection fraction; EDV, end-diastolic volume; ESV, end-systolic volume; SV, stroke volume; LVM, left ventricular mass; ICC, intraclass correlation coefficient; Inter, interrater; Intra, intrarater.

### **4. Discussion**

Our clinical study performed a comprehensive evaluation of a retrogated CS sequence in daily practice. Results are in line with preliminary CSrg tests on eight healthy volunteers reported by Forman et al., regarding LVEF, LVEDV and LVESV, which were similar to SSFPref quantification [34]. However, the study population was small and LVM, RV volumes and EF were not assessed. In the present study, no significant difference was demonstrated regarding LVM, LV and RV volumes (EDV, ESV and SV) and EF.

First-generation 49-ms temporal resolution CSrt sequence provided LVEDV underestimation and LVM overestimation, which were clinically insignificant or smaller than intra or interrater variabilities [17,22,25,26,35]. Moreover, LVEDV underestimation was also reported with other acceleration techniques such as radial gradient-echo or k-space parallel imaging [36,37]. These differences were not depicted with CSrg. This might be explained by the improved edge sharpness facilitating segmentation, retrospective ECG gating allowing acquisition of the last phases of the cardiac cycle and the better temporal resolution (37 ms) [38]. This observation is in line with previous studies suggesting an optimal temporal resolution for accurate steady-state free precession quantification below 45 ms [29]. Since quantification is a major CMR point of interest, the absence of significant difference regarding each LV and RV parameter seems promising for clinical implementation [1]. Moreover, the high intra and interrater reproducibility allows the use of CSrg for chronic status follow-up such as heart failure, anthracycline induced cardiotoxicity and other cardiomyopathies [39–41].

The overall image quality alteration using CSrt has already been assessed in a previous publication [22]. The higher score provided by CSrg confirmed the image quality improvement. Indeed, CSrt interpolation was responsible for smoothed images, which required physicians to get used to this rendering. The absence of difference between acquisitions qualities (EuroCMR quality score) was expected. Indeed, no difference had been demonstrated regarding the first-generation CS sequence. The evaluated prototype features partial Fourier switch-off and segmented acquisition, which were not supposed to generate more artifacts. Regarding edge sharpness, the absence of interpolation and

acquisition of more data improves boarder delineation. Not only was εCSrg better than εCSrg but it was similar to εSSFPref. Edge sharpness was chosen as a metric for intrinsic image quality assessment due to its simple and reproducible implementation. Moreover, edge sharpness is an estimate of the edge spread function whose derivative is a line-spread function [33]. Fast Fourier transform of the latter gives the task-based modulation transfer function (MTFTask); edge sharpness can be considered as a reasonable approximation of MTFTask, which is widely used to evaluate the spatial frequency response of an imaging system, even using iterative reconstructions [42].

Predictably, CSrg scan time was approximatively twice as long as the first-generation CS sequence since acquisitions were segmented on two heart beats in contrast to real-time imaging. However, as compared to SSFPref, the mean CSrg scan time of about 60 s still provides an 8.7-fold acceleration factor. The study protocol divided CSrg acquisition in 3 stacks to shorten breath-holdings. Nevertheless, this setting is adjustable, such as the number of heart beats required for one slice to reduce apneas or increase spatiotemporal resolution. We chose to use CS to improve scan time, with workflow fluency being a major point of concern. However, it is possible to take advantage of this acceleration to improve spatial resolution, or maybe more interestingly, temporal resolution. Indeed, in the field of feature tracking in CMR, a resolution at least over 30 frames per cycle is recommended for accurate strain assessment, which is more time-consuming when using SSFPref [43]. Such settings may facilitate further feature tracking studies, though the impact of CS based reconstructions should be evaluated regarding strain analysis reliability.

The image quality improvement provided by the prototype CSrg as compared to the first-generation real-time CSrt should facilitate the spread of CS use in daily practice. However, the 2-shot acquisition is responsible for the loss of the real-time acquisition. Futures generations of CS cine sequences, maintaining the CSrt real-time feature and the CSrg image quality, should be oriented towards the application of additional motion correction algorithms to provide free-breathing acquisition with preserved image quality.

### *Limitations*

The size of the population was limited and cannot represent the whole variety of cardiac conditions encountered in daily practice. Nevertheless, the main objective of this study was to assess the image quality recovery as compared to SSFPref and CSrt. To facilitate the standardized assessment of edge sharpness, we chose to exclude congenital heart disease anatomy. Among the cardiac conditions assessed by CMR, dilated cardiomyopathy is a frequent pathology, usually responsible for shortness of breath, in which the LV wall may be thinned. The decreased wall thickness may be challenging for endocardium and epicardium delineation, thus impairing LV mass assessment. The impact of LV wall thickness for this assessment, for instance in dilated cardiomyopathy, could not be evaluated because of an insufficient subgroup. Nevertheless, we assume that the edge sharpness provided by CSrg should help to distinguish the endocardium from the epicardium as compared to the first-generation CSrt cine that provided blurrier borders. Although a high acceleration factor was demonstrated using CSrg, the improvement of CMR tolerance can only be assumed since this parameter was not evaluated. However, we suppose that the lower number of breath-holdings required by CSrg should help dyspneic or claustrophobic patients undergo CMR examinations. This acceleration might be overestimated and must be interpreted cautiously since the two 1.6 × 1.6 mm2 in-plane resolution CS sequences were compared with a 1.3 × 1.3 mm2 in-plane resolution reference sequence. However, Miller et al. demonstrated that the maximal accuracy for functional parameters quantification using SSFP sequence was reached between 1 and 2-mm in-plane spatial resolution [29]. Even though further acquisitions could be performed during CSrg reconstruction during this study (SSFPref images were acquired first and available), up to 2 min were necessary for images to be reconstructed and displayed to set the next sequences orientation in case of exclusive use of CSrg. Wall motion abnormalities were not evaluated, but the increased temporal resolution provided by segmented acquisition and the absence of interpolation

should not impair their visualization on CSrg since the first generation of compressed sensing cine was demonstrated to be reliable for this evaluation (Video S3 (Supplementary Materials)) [44]. The heart rate of the evaluated patients ranged from 54 to 101 beats per minute (R–R intervals: 594 ms to 1111 ms). Since CSrg is a 2-shot sequence and its temporal resolution is constant (37 ms), the amount of data acquired during the acquisition must vary and may have an impact on the quality of the reconstructed cine images and consequently on functional assessment. The impact of heart rate on image quality was not evaluated in this study due to the limited size of the study population. Finally, only sinus rhythm patients were enrolled in this study. Even though a real-time acquisition is more robust than a segmented acquisition versus arrhythmia, the need of only two heart beats per slice should be more arrhythmia-proof than the conventional 8-heart-beat SSFPref. Further comparison with other acceleration techniques such as generalized autocalibrating partial parallel acquisition would be interesting since the later requires more heart beats for identical temporal and spatial resolutions [45].

### **5. Conclusions**

CSrg allows reliable quantification of LV and RV volumes, EF and mass providing similar objective and subjective image quality to SSFPref. Performed in clinical conditions, CSrg is promising in terms of workflow improvement and image quality recovering in comparison with the first-generation real-time CSrt.

**Supplementary Materials:** The following are available online at https://www.mdpi.com/article/ 10.3390/jcm10112417/s1, Video S1: Short-axis cine stacks in a 45-year-old woman referred for myocarditis follow-up; Video S2: Four-chamber cine slice in the same patient as Video S1; Video S3, short-axis mid-cavity cine slice in a 68-year-old patient referred for viability assessment after myocardial infarction, demonstrating inferolateral akinesia.

**Author Contributions:** B.L.: study conception and design, data collection, interpretation and analysis, drafting of the manuscript, critical revision for important intellectual content; C.V.G.: data collection and interpretation, critical revision for important intellectual content; A.C.: study conception, critical revision for important intellectual content; L.G.: data collection, interpretation and analysis, drafting of the manuscript; V.S.: data collection and interpretation, critical revision for important intellectual content; J.P.: data collection and interpretation, critical revision for important intellectual content; A.S.: data collection and interpretation, critical revision for important intellectual content; J.H.: data collection and interpretation, critical revision for important intellectual content; M.S.: study conception and design, critical revision for important intellectual content; C.F.: study conception and design, critical revision for important intellectual content; S.T.: study conception and design, critical revision for important intellectual content; D.M.: study conception, critical revision for important intellectual content; F.P.: study conception and design, data collection, interpretation and analysis, drafting of the manuscript, critical revision for important intellectual content. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Institutional Review Board Statement:** The study was approved by the research ethics committee of Lille University Hospital.

**Informed Consent Statement:** Informed consent was obtained from all subjects involved in the study.

**Data Availability Statement:** The data presented in this study are available on reasonable request from the corresponding author, subject to approval by the research ethics committee of Lille University Hospital.

**Conflicts of Interest:** B.L.; C.V.G; A.C.; L.G.; V.S.; J.P.; A.S.; J.H.; D.M.; F.P. have no competing interest. They are employed by an institution engaged in a contractual collaboration with Siemens Healthcare. M.S.; C.F.; S.T. are employees of Siemens Healthcare GmbH.

### **References**


### *Article* **Long-Term Imaging Follow-Up in DIPNECH: Multicenter Experience**

**Cécile Chung 1,2, Sébastien Bommart 3,4, Sylvain Marchand-Adam 5, Mathieu Lederlin 6, Ludovic Fournel 2,7, Marie-Christine Charpentier 8, Lionel Groussin 2,9, Marie Wislez 2,10,11, Marie-Pierre Revel 1,2 and Guillaume Chassagnon 1,2,\***


**Abstract:** Diffuse pulmonary neuroendocrine cell hyperplasia (DIPNECH) is a rare pre-invasive disease whose pathophysiology remains unclear. We aimed to assess long-term evolution in imaging of DIPNECH, in order to propose follow-up recommendations. Patients with histologically confirmed DIPNECH from four centers, evaluated between 2001 and 2020, were enrolled if they had at least two available chest computed tomography (CT) exams performed at least 24 months apart. CT exams were analyzed for the presence and the evolution of DIPNECH-related CT findings. Twenty-seven patients, mostly of female gender (*n* = 25/27; 93%) were included. Longitudinal follow-up over a median 63-month duration (IQR: 31–80 months) demonstrated an increase in the size of lung nodules in 19 patients (19/27, 70%) and the occurrence of metastatic spread in three patients (3/27, 11%). The metastatic spread was limited to mediastinal lymph nodes in one patient, whereas the other two patients had both lymph node and distant metastases. The mean time interval between baseline CT scan and metastatic spread was 70 months (14, 74 and 123 months). Therefore, long-term annual imaging follow-up of DIPNECH might be appropriate to encompass the heterogeneous longitudinal behavior of this disease.

**Keywords:** carcinoid tumor; neoplasm metastasis; lymphatic metastasis; multidetector computed tomography; neuroendocrine cells

### **1. Introduction**

Diffuse pulmonary neuroendocrine hyperplasia (DIPNECH) is presumed to be a rare pulmonary disorder characterized by a diffuse proliferation of pulmonary neuroendocrine cells of the airway mucosa. The proliferation may be limited to the mucosa or locally cross the basement membrane to form tumorlets or develop into carcinoid tumors [1]. Additionally, DIPNECH may result in small airway obstruction due to the combination

**Citation:** Chung, C.; Bommart, S.; Marchand-Adam, S.; Lederlin, M.; Fournel, L.; Charpentier, M.-C.; Groussin, L.; Wislez, M.; Revel, M.-P.; Chassagnon, G. Long-Term Imaging Follow-Up in DIPNECH: Multicenter Experience. *J. Clin. Med.* **2021**, *10*, 2950. https://doi.org/10.3390/ jcm10132950

Academic Editor: Mickaël Ohana

Received: 31 May 2021 Accepted: 29 June 2021 Published: 30 June 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

of luminal protrusion of the neuroendocrine cells and constrictive bronchiolitis related to peribronchiolar fibrosis induced by the secretion of amines and peptides [2]. DIPNECH has been recognized as a separate entity in 1992 [3] and is distinct from reactive pulmonary neuroendocrine cells hyperplasia that can be observed in several conditions, including chronic lung disease, cigarette smoking and living at high altitude [3]. In a series of 1090 patients who received resection for primary lung carcinomas, Ruffini et al. found DIPNECH in 3 patients out of 55 with lung carcinoid tumor having a prevalence of 5.4% in this subgroup [4]. While being recognized as a pre-invasive condition for lung carcinoid tumors by the World Health Organization (WHO) [1] and being increasingly reported, DIPNECH pathophysiology remains poorly understood.

The first uncertainty regarding this disease concerns its exact definition, and to date, there is no consensus on diagnostic criteria. The WHO definition [1] is strictly based on histopathology and encompasses patients with different clinical presentations. Indeed, about one-half of patients present with pulmonary symptoms and functional impairment, whereas the other half of patients remain asymptomatic, according to the literature [5,6]. At the same time, some asymptomatic patients may present the computed tomography (CT) abnormalities usually encountered in DIPNECH (e.g., mosaic perfusion, multiple nodules and micronodules), others only present neuroendocrine hyperplasia on histopathologic examinations, and no CT anomalies. Some authors have proposed to differentiate between i/ DIPNECH syndrome in case of symptomatic diffuse pulmonary neuroendocrine cell proliferation on histology, ii/ DIPNECH in case of asymptomatic diffuse pulmonary neuroendocrine cell proliferation on histology with suggestive radiological features, and iii/ pulmonary neuroendocrine cell hyperplasia (PNECH) in case of asymptomatic diffuse pulmonary neuroendocrine cell proliferation on histology without compatible radiological findings [7]. The lack of symptoms in DIPNECH patients with signs of small airway disease on chest CT is related to the fact that small airways represent the "silent zone" of the lung where the disease might accumulate for many years with very little effect [8,9]. Chest CT and functional magnetic resonance imaging are reported to have better sensitivity than spirometry for early detection of small airway disease, as reported in cystic fibrosis children [10,11].

Uncertainties regarding DIPNECH also include the pathophysiological mechanisms and the appropriate treatment as well as long-term evolution and risk factors for poor outcome. Although DIPNECH is generally considered to be an indolent disease, some patients develop respiratory failure leading to lung transplantation [6,12] or metastatic spread [13]. Given the scarcity of longitudinal data available, there are currently no specific guidelines for DIPNECH imaging follow-up.

Therefore, the aim of this study was to assess long-term evolution in imaging in a cohort of patients with histologically proven DIPNECH, which could serve as a basis for follow-up recommendations.

### **2. Materials and Methods**

### *2.1. Study Population*

This retrospective multicenter study was approved by our Institutional Review Board (AAA-2019-08012) which waived the need for patients' consent.

Clinical, pathological and radiological records from 4 University hospitals were reviewed for patients with a histologically confirmed diagnosis of DIPNECH over a 19-year period, from 2001 to 2020. The diagnosis of DIPNECH encompassed DIPNECH and DIPNECH syndrome and required at least one of the following two criteria:


Exclusion criteria were the unavailability of two chest CT scans performed at least 24 months apart and the association with an unrelated active cancer.

### *2.2. Image Analysis*

The baseline and the latest chest CT exams were retrospectively analyzed. In the case of surgical resection between these 2 CT examinations, the CT exam performed just before surgery was also analyzed. All CT exams were read in consensus by a junior radiologist (CC) and a chest radiologist (GC) with 1 and 7-year experience in chest imaging, respectively.

Baseline CT exams were analyzed for the presence of the seven different CT features reported in DIPNECH [14], namely: mosaic perfusion, pulmonary nodules, subpleural atelectasis, mucoid impactions, bronchial thickening, bronchiectasis and pulmonary cysts. Regarding the pulmonary nodules, micronodules/nodules suggestive of tumorlets (<5 mm in diameter) and nodules suggestive of carcinoid tumors (≥5 mm in diameter) were counted, the size of the largest nodule was measured and the presence of calcifications, as well as central/peripheral location, were assessed. A nodule was considered to be centrally located if it was connected to a segmental or more proximal bronchus. Side-by-side comparison between baseline and follow-up CT exams was also performed to determine changes in mosaic perfusion extent, changes in size or number of pulmonary nodules and occurrence of lymphatic or metastatic spread.

### *2.3. Clinical Data*

Patient charts were reviewed for demographic characteristics, symptoms, smoking history and pulmonary function tests (PFTs) at the time of the baseline CT scan. The diagnosis circumstances were investigated and classified into 3 categories: incidental finding, respiratory symptom or CT staging of unrelated neoplasm. The date and type of surgery, as well as histopathological results, were also retrieved.

In cases where another surgery was performed after the diagnosis of DIPNECH had already been established, the CT examination performed prior to treatment and clinical information were both reviewed.

Additionally, the latest follow-up PFTs were collected to assess the evolution of obstruction based on forced expiratory volume in 1 second (FEV1). A decrease of at least 10% of FEV1 absolute predicted value was considered significant.

### *2.4. Statistical Analysis*

Statistical analysis was performed using R software (version 3.6, R Foundation, Vienna, Austria). Qualitative variables were expressed as percentages, whereas quantitative variables were expressed as the median and interquartile range (IQR). Patient characteristics and baseline CT findings were compared between patients with progressive disease requiring additional intervention and the remaining patients. Comparisons were performed using the Fisher exact test for qualitative variables and the Wilcoxon–Mann–Whitney test for quantitative variables. *p* values less than 0.05 were regarded as being significant.

### **3. Results**

### *3.1. Patients' Characteristics at Baseline*

A total of 66 patients considered to have DIPNECH were identified in the databases of four university hospitals. Of these, the 27 patients who had at least two years of CT follow-up and met the criteria for DIPNECH diagnosis were included (Figure 1, flow chart). Patients were mostly of female gender (*n* = 25/27; 93%) and had a median age of 63 years (IQR: 59–72 years; range: 42–88 years) at the time of baseline CT scan (Table 1).

**Figure 1.** Flow chart.



Note: Quantitative data are presented as median [interquartile range] and qualitative data are presented as proportion (percentage). \* at time of the baseline computed tomography scan. FEV1: forced expiratory volume in 1 second, PFT: pulmonary function test.

> Neuroendocrine proliferation was histologically confirmed on surgical resection specimen in 25 patients (93%), CT-guided transthoracic biopsy in one patient (4%) and ultrasound-guided biopsy of liver metastasis in one patient (4%). All patients in this series had at least one histologically confirmed lung carcinoid tumor. Atypical carcinoids were observed in six patients (*n* = 6/27; 22%), whereas only typical carcinoids were found in the remaining 21 patients with the limit that not all nodules were resected.

> For the 24 patients for whom the information about circumstances of diagnosis was available, the majority were diagnosed because of thoracic symptoms (*n* = 15/24; 62%), whereas for the other nine patients, nodules were found at the initial staging of an unrelated neoplasm (*n* = 5/24; 21%) or incidentally (*n* = 4/24; 17%).

> The majority of patients for whom the information was available were symptomatic (*n* = 18/22; 82%) with cough, dyspnea or other respiratory symptoms in 12 (*n* = 12/22; 55%), 10 (*n* = 10/22; 45%) and 8 patients (*n* = 8/22; 36%), respectively. Other respiratory symptoms consisted in recurrent bronchitis (*n* = 3/8; 38%) and unrelated obstructive sleep apnea (*n* = 3/8; 38%). The median delay from symptoms onset to the first CT scan was 56 months (IQR: 27–90 months; range: 0–274 months).

> Most patients were non-smokers (*n* = 11/20; 55%). Baseline PFTs were available for 18 patients, showing an obstructive syndrome in 9 of them (*n* = 9/18; 50%). The median FEV1 was 82 percent of the predicted value (IQR: 72–89%; range: 26–110%).

### *3.2. Imaging Findings at Baseline*

All but two available baseline CT scans (25/27; 93%) were performed prior to any surgery. All patients had pulmonary nodules associated with mosaic perfusion (*n* = 27/27, 100%) (Table 2).



Note: Quantitative data are presented as median (interquartile range) and qualitative data are presented as proportion (percentage). \* at time of the baseline computed tomography scan. CT = computed tomography.

> A total of 20 patients (*n* = 20/27, 74%) had 10 or more nodules measuring less than 5 mm, which are conventionally considered tumorlets, while 3 patients (*n* = 3/27, 11%) had 10 or more nodules measuring at least 5 mm, which are conventionally considered as carcinoid tumors. The remaining 7 patients had less than 10 nodules, with a median number of tumorlets and carcinoid tumors of 3 (IQR = 1–3, range: 0–7) and 2 (IQR = 1–4, range: 0–8), respectively. Only three patients had three or fewer lung nodules (tumorlets + carcinoids) on baseline chest CT, but these patients also had diffuse mosaic perfusion related to the diffuse neuroendocrine cell hyperplasia. In all three cases, the diagnosis has been confirmed by surgical biopsy showing in addition to the carcinoid tumors, multiple tumorlets (some being beyond CT resolution) and diffuse neuroendocrine cell hyperplasia. The median size of the largest nodule was 9 mm (IQR = 8–13 mm, range: 2–32 mm). Centrally located nodules, suggesting central carcinoid tumors, and calcifications were found in only three patients (*n* = 3/27, 11%).

> Other features of airway involvement included band-like subpleural atelectasis in 67% of patients (*n* = 18/27), mucoid impaction in 37% (*n* = 10/27), mild bronchial thickening in 33% (*n* = 9/27) and mild cylindrical bronchiectasis in 19% (*n* = 5/27). In one patient (*n* = 1/27, 4%), the chest CT scan revealed the presence of pulmonary cysts, but pathologic evaluation of these cysts was not available. Baseline CT exams revealed no evidence of lymphatic or extrapulmonary dissemination.

### *3.3. Disease Evolution and Treatment*

The median duration of radiological follow-up was 63 months (IQR: 31–80 months; range: 24–170 months). During this period, lung nodules increased in 19 patients (*n* = 19/27; 70%) of whom 17 (17/27, 63%) presented an increased number and 18 (18/27, 67%) a size increase, whereas nodules remained stable in the remaining eight patients (*n* = 8/27; 30%). Despite nodule increase in the majority of patients, only two patients had an additional local treatment. In the first patient, DIPNECH had been diagnosed 15 years (176 months) earlier on a surgical biopsy. The patient had Cushing's syndrome and a stable 18 mm lung nodule in the right lower lobe, which was percutaneously biopsied. The pathological study demonstrated an atypical lung carcinoid showing ACTH antibody staining and the patient was treated by stereotactic body radiotherapy. Following treatment, Cushing's disappeared but relapsed six months later. In the second patient, DIPNECH had been diagnosed 62 months earlier after wedge resections in the right upper, middle and lower lobes showed three atypical lung carcinoids along with tumorlets and pulmonary neuroendocrine cell hyperplasia. The patient was re-operated on 25 months later, for the resection of a newly appeared 15 mm nodule, which turned out to be a granuloma.

With regard to CT signs of small airway involvement, mosaic perfusion and bronchial dilatations remained stable in the majority of patients (*n* = 22/27; 81% and *n* = 26/27; 96%, respectively).

Lymph node enlargement suggestive of lymphatic spread was observed and confirmed in three patients (*n* = 3/27, 11%). All had at least one atypical lung carcinoid (*n* = 3/3; 100%).

Distant metastases occurred in two patients (*n* = 2/27, 7%), all of whom also had lymphatic dissemination. Distant and lymphatic metastases were synchronous in one patient, while in another patient, the nodal spread had preceded distant metastases by 25 months (Figure 2). Both patients developed liver metastasis and then pleural carcinomatosis. In one patient, the lung carcinoid was also metastatic to the bone.

Pathological confirmation of the metastatic stage was obtained for all patients and showed features of atypical carcinoid for all.

Therefore, three patients (*n* = 3/27, 11%) developed lymphatic spread with or without distant metastases, in a mean time interval time of 67 months following the baseline CT scan (14, 63 and 123 months). The frequency of histologically confirmed atypical carcinoids was significantly higher in patients who developed lymphatic or distant metastases (100% vs. 13%; *p* = 0.005). The other patient characteristics were not significantly different when comparing these two patient categories (*p* > 0.05) (Tables 1 and 2).

Seven patients had baseline and follow-up PFTs with a median follow-up interval of 51 months (IQR: 38–79 months; range: 24–97 months). Functional worsening was observed in four patients (*n* = 4/7; 57%) with a median FEV1 decrease of −14.5% in absolute predicted value (−35%, −15%, −14% and −13%), whereas FEV1 remained stable or improved in the remaining three patients (−4%, −3% and +13% in absolute predicted value).

**Figure 2.** DIPNECH with metastatic course over an 8 year period in a 60-year-old woman. (**a**) Maximum intensity projection (MIP) reformation of baseline chest CT scan shows several nodules (arrow and arrowheads), including a 12 mm nodule in the left upper lobe (arrow). (**b**) Minimum intensity projection (MinIP) reformation reveals the characteristic association to diffuse mosaic perfusion. (**c**) There was no mediastinal lymph node enlargement. (**d**) Histopathological analysis of the 12 mm nodule shows carcinoid tumor with necrosis spots (arrows) consistent with an atypical form. (**e**,**f**) Analysis of distant parenchyma revealed bronchial cell proliferation (arrows) on hematoxylin and eosin stain (E), which corresponded to neuroendocrine cells hyperplasia on chromogranin A stain (F). (**g**) Five years later, contrast-enhanced CT shows the development of mediastinal lymphadenopathy leading to the new surgery. (**h**) Two years later, Gadolinium-enhanced T1 weighted axial image demonstrates liver metastasis leading to the start of chemotherapy. (**i**) Contrast-enhanced CT scan performed 6 months later shows disease progression with the occurrence of pleural metastasis (arrow). Nodal and liver metastasis were histologically confirmed and showed features of atypical lung carcinoid.

### **4. Discussion**

In this multicenter study, we found that DIPNECH was a progressive disease in the majority of patients, over a median follow-up time of 63 months, with an increase in the size and/or number of lung nodules. Despite this evolution, the disease course remained indolent in all but three patients (11%) who developed lymphatic spread, followed by a distant metastatic spread in two patients.

To the best of our knowledge, this multicenter study is the first to focus on long-term CT follow-up in DIPNECH, and our cohort is one of the largest reported to date [5,6,15–20]. As in other reported series, we found a strong female predominance with 93% of women in our cohort compared to 79 to 100% in the literature [6,15]. Most patients (82%) were symptomatic (DIPNECH syndrome) and all had mosaic perfusion on chest CT scan, a sign of small airway disease. The median 56-month delay between the onset of symptoms and DIPNECH diagnosis is in phase with previous reports. Indeed, many patients are misdiagnosed as having late-onset asthma. However, the association of multiple lung nodules and mosaic perfusion observed in all our patients should be considered as highly suggestive of DIPNECH in middle-aged women. This association is part of the diagnostic criteria proposed by Carr et al., allowing DIPNECH to be diagnosed without histopathological confirmation [15].

Unlike some other studies, mosaic perfusion was present in all patients of our series (100% vs. 17–100%) [2,15,21]. Mosaic perfusion, which is due to vasoconstriction in areas of small airway obstruction and blood flow redistribution to normal areas [22], may be subtle on native CT images and standard lung windowing. Its detection is enhanced by the use of minimum intensity projection (minIP) reformation with narrowed window width and optimized center level settings [14,23].

Disease evolution in DIPNECH may be related to either small airway disease worsening or malignant progression. Several studies have reported an increase of pulmonary symptoms or a decline in pulmonary function in 7 to 55% of patients [5,6,15]; for some patients, functional worsening leads to end-stage respiratory failure, lung transplantation or death. Gorshtein et al. reported that 55% of patients had both functionally and radiologically progressive disease, the other remaining stable [5]. We observed increased mosaic perfusion in 19% of our patients and functional worsening in 57% of patients who had PFTs re-evaluation. By contrast, the majority of patients presented an increase in pulmonary nodules. Only two series have reported imaging follow-up data in DIPNECH patients and found a progression of lung lesions in 49 and 55% of cases but over shorter median follow durations (19.5 and 56 months vs. 63 months) [5,24]. While none of our patients were metastatic at the time of diagnosis, 11% developed a metastatic nodal or extranodal spread during the disease course. Only a few cases of metastatic spread have been reported to date in DIPNECH, the prevalence widely ranging from 0 to 27% [5,17]. The reported cases mostly consist of lymphatic spread, with only three previously reported cases of extranodal metastasis in the eye and bone [17], the adrenal gland [5] and the liver [13]. Similar to previous reports, we found metastatic disease to occur in patients with proven atypical carcinoid tumors, which was the only characteristic significantly associated with a metastatic evolution. The presence of atypical carcinoid is reported to be rare in DIPNECH, with a prevalence ranging from 0 to 27% [5,6,19]. However, in the setting of multiple bilateral nodules, not every nodule can be pathologically evaluated, and thus, the proportion of patients with atypical histology may be underestimated. Similarly, it is not possible to exclude that some typical carcinoids could metastasize, since lymphatic spread has been reported in 5% of resected typical lung carcinoid [25].

In view of the long delay, up to 123 months, between baseline CT and the occurrence of metastatic spread, long-term CT surveillance should be performed in DIPNECH. According to the current recommendations of the European Neuroendocrine Tumor Society (ENETS) on pulmonary carcinoids [26], CT scan should be performed at 3 and 6 months after treatment and then every 12 months for the first 2 years in the setting of typical pulmonary carcinoid. Then long-term annual chest X-ray and CT every 3 years are recommended. For atypical pulmonary carcinoids, closer monitoring is recommended with CT imaging 3 months post-surgery and then every 6 months for 5 years. After 5 years, an annual CT should be performed. In the setting of DIPNECH, atypical pulmonary carcinoids are presumed to be rare, and due to the diffuse nature of the disease and the advanced patient age, resection of all lesions is usually not feasible. Furthermore, an increasing number of patients are being diagnosed with DIPNECH without surgical resection, potentially leaving

in place atypical carcinoids. Therefore, close long-term monitoring seems necessary in these patients. Based on our experience and previous reports, we propose contrast-enhanced CT of the chest, abdomen and pelvis at 3 months, 6 months and 12 months after baseline CT, and then long-term annual CT (Figure 3). In the case of confirmed atypical pulmonary carcinoid, the ENETS follow-up recommendations should be applied.

**Figure 3.** Proposal of follow-up recommendations for patients with DIPNECH.

Our study has several limitations; first, because of its retrospective design and the lack of available recommendations for DIPNECH follow-up, there was no standardization of imaging follow-up, and several patients could not be included due to the lack of 2-year follow-up. Others were not included because histological confirmation was lacking. We chose to propose recommendations based only on histologically proven follow-up cases. Additionally, the contribution of the 68Ga DOTATOC PET-CT scanner could not be evaluated because only a few patients underwent PET/CT in this multicenter cohort study. Finally, the limited size of our cohort made it impossible to identify significant differences other than the presence of atypical carcinoid between patients with and without metastatic spread.

### **5. Conclusions**

In conclusion, the majority of DIPNECH patients present long-term disease progression on imaging and 11% develop metastatic spread at 63-month median follow-up. Long-term imaging follow-up is therefore required. Further prospective studies are needed to evaluate these follow-up recommendations.

**Author Contributions:** Conceptualization, C.C., L.F., M.-P.R. and G.C.; data curation, C.C., S.B., S.M.-A., M.L., L.F., M.-C.C., L.G. and M.W.; writing—original draft preparation, C.C. and G.C.; writing—review and editing, S.B., S.M.-A., M.L., L.F., M.-C.C., L.G., M.W. and M.-P.R.; supervision, G.C. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Institutional Review Board Statement:** The study was conducted according to the guidelines of the Declaration of Helsinki, and approved by the Institutional Review Board of Cochin Hospital (protocol code AAA-2019-08012 and approved on 16 April 2019).

**Informed Consent Statement:** Patient consent was waived due to the retrospective design of the study.

**Data Availability Statement:** Not applicable.

**Conflicts of Interest:** The authors declare no conflict of interest.

### **References**


### *Article* **Interstitial Lung Abnormalities Detected by CT in Asbestos-Exposed Subjects Are More Likely Associated to Age**

**François Laurent 1,2,3,\*, Ilyes Benlala 1,2,3, Gael Dournes 1,2,3, Celine Gramond 4, Isabelle Thaon 5, Bénédicte Clin 6,7, Patrick Brochard 1,8, Antoine Gislard 9,10, Pascal Andujar 11,12,13,14, Soizick Chammings 14, Justine Gallet 4, Aude Lacourt 4, Fleur Delva 4, Christophe Paris 15,16, Gilbert Ferretti 17,18,19 and Jean-Claude Pairon 11,12,13,14**


**Abstract:** Objective: the aim of this study was to evaluate the association between interstitial lung abnormalities, asbestos exposure and age in a population of retired workers previously occupationally exposed to asbestos. Methods: previously occupationally exposed former workers to asbestos eligible for a survey conducted between 2003 and 2005 in four regions of France, underwent chest CT examinations and pulmonary function testing. Industrial hygienists evaluated asbestos exposure and calculated for each subject a cumulative exposure index (CEI) to asbestos. Smoking status information was also collected in this second round of screening. Expert radiologists performed blinded independent double reading of chest CT-scans and classified interstitial lung abnormalities into: no abnormality, minor interstitial findings, interstitial findings inconsistent with UIP, possible or definite UIP. In addition, emphysema was assessed visually (none, minor: emphysema <25%, moderate: between 25 and 50% and severe: >50% of the lung). Logistic regression models adjusted for age and smoking were used to assess the relationship between interstitial lung abnormalities

**Citation:** Laurent, F.; Benlala, I.; Dournes, G.; Gramond, C.; Thaon, I.; Clin, B.; Brochard, P.; Gislard, A.; Andujar, P.; Chammings's, S.; et al. Interstitial Lung Abnormalities Detected by CT in Asbestos-Exposed Subjects Are More Likely Associated to Age. *J. Clin. Med.* **2021**, *10*, 3130. https://doi.org/10.3390/jcm10143130

Academic Editor: Mickaël Ohana

Received: 26 May 2021 Accepted: 12 July 2021 Published: 15 July 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

and occupational asbestos exposure. Results: the study population consisted of 2157 male subjects. Interstitial lung abnormalities were present in 365 (16.7%) and emphysema in 444 (20.4%). Significant positive association was found between definite or possible UIP pattern and age (OR adjusted =1.08 (95% CI: 1.02–1.13)). No association was found between interstitial abnormalities and CEI or the level of asbestos exposure. Conclusion: presence of interstitial abnormalities at HRCT was associated to aging but not to cumulative exposure index in this cohort of former workers previously occupationally exposed to asbestos.

**Keywords:** asbestos-exposition; HRCT; asbestosis

### **1. Introduction**

The development and severity of asbestosis is related to intensity of exposure to asbestos and time since first exposure [1]. The surveillance of the former exposed workers is justified by financial compensations and because of the elevated risk of bronchial cancer. Therefore, and because of the long latency of asbestosis, health surveillance should be prolonged after the exposure. High resolution computed tomography (HRCT) is able to detect asbestos-induced pulmonary changes much earlier than chest x-ray and is useful for early diagnosis of asbestosis [2–4]. However, today, the prevalence of asbestosis is lower than in studies of past decades [5,6]. In addition, the pattern of asbestosis has changed and mild fibrosis has been reported [6,7]. Interstitial lung abnormalities (ILA) are defined as early interstitial changes in nondependent areas of the lung, and has been validated as evidence of subclinical interstitial lung disease (ILD) [8]. Subjects with subclinical ILD exhibit more respiratory symptoms, physiologic decline and higher mortality [9,10]. However, in a population of middle-aged to elderly subjects, ILA have been associated with a mild form of pulmonary fibrosis in smokers and non-smokers [8,10] and has been reported to increase with age [11–14]. Moreover, longitudinal studies have shown progression of imaging patterns of ILA [9] and that have been related to mortality [15]. Indeed, a recent study showed that the prevalence of HRCT patterns of usual interstitial pneumonia (UIP) and chronic interstitial pneumonia were 0.3% and 3.8%, respectively, in a smokers' cohort with 25% of progression in those who underwent a 3 year follow-up CT scan [13]. Therefore, the relationships between age, asbestos-exposure and smoking status need to be clarified.

The present study was designed to evaluate the association between interstitial abnormalities, asbestos exposure and age in a population of former workers previously occupationally exposed to asbestos. The main objective of this study was to determine whether ILA identified in a population of former asbestos-exposed workers were due to asbestosis or also to other causes of interstitial lung disease such age and/or smoking.

### **2. Material and Methods**

### *2.1. Study Population*

A first round of a screening program for asbestos-related diseases was held between October 2003 and December 2005 in four regions of France. Retired workers exposed to asbestos during their working life and without already compensated asbestos-related disease were eligible in this surveillance program. As previously described, volunteers were invited to participate using different ways and constituted the Asbestos-Related Diseases Cohort (ARDCO). They could beneficiate of a free medical check-up including chest CT-scan and pulmonary function tests [1,16–18]. Subjects having CT-scan sent to the regional coordinating centers constituted the Asbestos Post-Exposure Survey (APEXS) population. All male subjects of the APEXS population were included in the present study. A second round of screening was organized 5 years later and this study is based on the second round. Indeed, digitalized thin-section CT examinations of the second round of the survey showed better image quality than CT exams of the first round where a number of digitalized examinations were missing.

The study was approved by the hospital ethics committee (CPPRB Paris-Cochin n 1946 (2002), CPP Ile De France III n 1946/11/02-02 (2010)). All participants received information about the study and gave their written informed consent.

### *2.2. Asbestos Exposure and Smoking Status*

All subjects completed a standardized questionnaire, having different parts: all successive jobs (date of beginning and end, location of the employer, main activity of the firm, precise job of the subject), indication of having ever worked in a list of specific jobs (known for high probability of exposure to asbestos) in the construction industry with indication of date of beginning and end of these jobs; 9 specific questions on tasks entailing exposure to asbestos with indication of date of beginning and end of each task; and finally free text to invite the subject to add any precision he would prefer to add to his previous answers. Therefore, industrial hygienists could evaluate asbestos exposure based on of the complete working life history of each subject. For each job considered exposed to asbestos, the duration (expressed in years) and dates of exposure were determined. The maximum level of exposure was defined as the highest exposure occurring during the entire work history. The following weighting factors were decided for the intensity level of exposure, based on the knowledge of the different situations of exposure: low (passive exposure): 0.01; low intermediate: 0.1; high intermediate: 1; high: 10. A cumulative exposure index (CEI) to asbestos was then calculated for each subject over his working life. It was calculated by summing the exposures calculated for each exposed job (duration x weighting factor). There were no atmospheric measurements in this cohort, and detailed information on the frequency of exposure (percentage of the working time) was frequently lacking. Therefore, the CEI was expressed in exposure units x years rather than fibers/mL x years. The elapsed time between the beginning of the first job considered to be exposed to asbestos and the date of CT scan was calculated as the time since first exposure (TSFE) [16–18].

The questionnaire also collected information on smoking status, allowing classification of subjects into three categories: current smokers, ex-smokers (those who had stopped smoking for at least one year) and never-smokers (having smoked less than 100 cigarettes over the working life).

### *2.3. CT Scanning*

A specific protocol was established by a group of chest radiologists as previously described [1]. Four experts trained in the interpretation of asbestos-related CT abnormalities performed a double-blinded independent evaluation of all CT examinations. Triple evaluation was performed in the case of disagreement between the first two readings. These evaluations were performed in a blinded fashion to the subject's cumulative exposure to asbestos and to the report of the initial reading made by the radiologists who performed the CT examination. Interstitial or pleural abnormalities were registered using a standardized form according to the Fleischner Society glossary of terms [19].

The reader was asked to classify the patient into one of four categories for parenchymal findings: no abnormality, minor interstitial findings, interstitial findings inconsistent with UIP, possible or definite UIP. The second category was based on the description of abnormal interstitial findings in aging people [11] and the 2 last categories based on the consensus ATS/ERS criteria for the diagnosis of idiopathic pulmonary fibrosis (IPF) [20]. Severity of emphysema was assessed visually on a 4-level scale (none, minor: emphysema occupying less than 25%, moderate: between 25 and 50% and severe: more than 50% of the whole pulmonary volume).

### *2.4. Statistical Analysis*

The relationship between interstitial lung abnormalities and occupational asbestos exposure was estimated using logistic regression models adjusted for age and smoking (never smokers, ex-smokers and current smokers). Occupational asbestos exposure was characterized by the CEI, time since first exposure and the maximum level of exposure. Linearity of quantitative variables was examined using fractional polynomial [21]. Supplementary analyses were conducted to study the association between severity emphysema at CT and occupational asbestos exposure using the same analysis strategy as described above. Statistical analyses were carried out using SAS software version 9.4 (SAS Institute, Inc, Cary, NC, USA) and R software version 3.4.2.

### **3. Results**

*3.1. Subjects' Demographic Data, Smoking Data, Asbestos Exposure and Frequency of Interstitial Abnormalities at CT*

At the second round of the screening program proposed to the APEXS population, 2268 subjects have been explored by CT with a double readings and triple reading if discordances. After exclusion of 92 women and 19 subjects with missing data for smoking status, the study population consisted of 2157 male subjects (Figure 1). Pleural plaques were present in 559 (25.7%) subjects, interstitial abnormalities in 365 (16.7%), emphysema in 444 (20.4%) (Table 1).

**Figure 1.** Study Flow-Chart. APEXS: Asbestos Post-Exposure Survey; ARDCO: Asbestos-Related Diseases Cohort; CT = computed tomography.

**Table 1.** Frequency of pleural plaques, interstitial abnormalities and emphysema as detected at CT (*n* = 2157).




UIP: usual interstitial pneumonia.

*3.2. Association between Possible and Definite CT Patterns of UIP and Age, Smoking Status and the Level of Exposure to Asbestos*

The frequency of interstitial findings according to demographic data, smoking status and asbestos exposure is shown Table 2.

A significant positive association was found between UIP pattern or possible UIP pattern and age (OR crude = 1.08 (95% CI: 1.03–1.13) and OR adjusted = 1.08 (95% CI: 1.02–1.13)) for each additional year (Table 3).

In addition, UIP pattern or possible UIP pattern were significantly associated to smoking status (OR crude for ex-smoker = 2.13 (95% CI: 1.03–4.39)) (Table 3).

**Table 2.** Frequency of interstitial abnormalities detected at CT according to demographic characteristics, smoking status and asbestos exposure data.


CEI: cumulative exposure index to asbestos.


**Table 3.** Association between the possible or definite UIP pattern at CT, asbestos exposure and age unadjusted and adjusted on age, occupational asbestos exposure and smoking status.

CEI: cumulative exposure index to asbestos. Bold values indicate statistical significance.

There was no significant association between interstitial lung abnormalities and asbestos exposure assessed either by CEI or by maximum level of exposure (Table 3). We have also combined the two intermediate categories of maximum level of exposure (intermediate low + intermediate high) and the results were not significant (data not shown).

When patients with either definite UIP pattern or possible UIP pattern and inconsistent with UIP pattern were grouped, this association with age remained significant (OR crude = 1.09 (95% CI: 1.06–1.12) and OR adjusted = 1.09 (1.05–1.13)) (Supplementary Table S1). On the other hand, the association with asbestos exposure remained nonsignificant (Supplementary Table S1).

### *3.3. Association between CT Severity of Emphysema and Age, Smoking Status and the Level of Exposure to Asbestos*

The frequency of emphysema findings according to demographic data, smoking status and asbestos exposure is shown in Table 4. No association was found between emphysema and CEI to asbestos, but an association was observed with the maximum level of exposure to asbestos (OR crude = 2.05 (95% CI: 1.06–3.99) and OR adjusted = 2.27 (95% CI: 1.08–4.80) for subjects with "high-level" of asbestos exposure). As expected, a significant association was found between emphysema and smoking status (Supplementary Table S2).

**Table 4.** Frequency of emphysema at CT according to age, smoking status and asbestos exposure data.



**Table 4.** *Cont.*

CEI: cumulative exposure index to asbestos.

### **4. Discussion**

This study has shown the association between interstitial abnormalities and age after adjustment on smoking status and asbestos exposure in a population of retired workers previously occupationally exposed to asbestos. This result is consistent with the CT lung cancer screening study by Vehmas et al. [22] among asbestos-exposed workers, where a positive correlation was found between interstitial lung abnormalities at CT and aging after adjustment to smoking status, asbestos exposure and body mass index.

The strengths of our study are the large number of subjects, individual estimation of cumulative occupational exposure to asbestos, HRCT acquisition and accurate analysis of CT by experts and categorization of interstitial abnormalities according to the ATS/ERS consensus criteria [20]. However, our analysis predates international consensus guideline for the diagnosis of IPF [23]. Nevertheless, definitions of UIP patterns are quite similar.

In a population of middle-aged to elderly subjects without occupational exposure, the presence of discrete lung parenchymal abnormalities has been reported to increase with age [11,24]. Jin et al. reported a positive association of fibrotic ILA at HRCT and age [25]. In this case, 60% of healthy subjects aged 75 or older have shown a basal reticular pattern whereas in those younger than 55 years old, no interstitial abnormalities have been reported [11]. This may be due to elastin degradation with aging, which leads to alveolar collapse or to mild interstitial fibrosis. In addition, interlobular septae thickening has been found more commonly in older subjects. However, subpleural lines may disappear at imaging in prone position in some healthy individuals. A study by Gamsu et al. [3] showed that CT findings of asbestosis are neither perfectly sensitive nor specific for asbestosis. In another study by Copley et al., the results of CT evaluation of 74 patients with asbestosis were compared to those of 212 patients with idiopathic pulmonary fibrosis showing that HRCT patterns of asbestosis are closely akin to the UIP pattern [26]. HRCT is today an essential component of the diagnostic pathway in interstitial lung disease. In order to clarify the management of patient with IPF, ATS/ERS guidelines have been published to assess the probability of the disease according to patterns [20].

Asbestosis, however, cannot be distinguished by HRCT from a possible or definite UIP pattern [26]. The first round of screening of this cohort has been reported [1] and conversely to the agreement between trained expert readers for the detection of pleural plaques that was good to excellent, the agreement between trained expert readers for the detection of interstitial abnormalities has been reported to be fair to good [27]. In an attempt to circumvent the subjects with possible mild or moderate asbestosis, we separated those subjects from those with an interstitial pattern inconsistent with UIP, using the criteria defined by ATS/ERS consensus paper [20]. Patterns with upper lobe or middle lung

interstitial abnormalities predominance, predominance of ground-glass opacities, nodular pattern, peribronchial pattern, cysts and/or air trapping or mosaic attenuation in three or more lobes were classified globally in the group inconsistent with UIP pattern [20]. Interestingly in these subjects, interstitial abnormalities were not associated with exposure but with age, even when adjusted on smoking status. However, the non-association between interstitial lung abnormalities and asbestos exposure evaluated using time related markers (i.e., duration of exposure and time since the first exposure) could be related to an over-adjustment with age. Nevertheless, the association between interstitial abnormalities with typical or atypical UIP pattern and age has been reported [15,25,26]. Therefore, these subjects may have a preclinical interstitial lung disease rather than asbestosis.

In our study, the association of UIP or possible UIP pattern at HRCT and smoking in univariate analysis is in line with literature showing that tobacco consumption is a risk factor for development of fibrotic lung abnormalities [28]. This finding was not significant in current smokers which could be explained by the low number of current smokers at the time of the study (*n* = 126). In the study by Jin et al., the association between ILA at HRCT and smoking status was found not significant [25]. However, fibrotic ILAs were differentiated into subtypes (i.e., patients with ground glass opacities: *n* = 12, reticulations: *n* = 9 and honeycombing: *n* = 9) which decreased the strength of the statistical analysis. Nonetheless, in the multivariate model when adjusted for occupational asbestos exposure and age, association with cigarette smoking was found not significant.

The relationships between emphysema and asbestos exposure remain unclear. A positive association was found between emphysema and the maximum level of asbestos exposure but not with the CEI to asbestos. Indeed, CEI is considered as a more precise parameter taking into account the total asbestos exposure along the working carrier, which could be more relevant than the maximum level of exposure that could possibly intermingles other associated exposures. In addition, in a population of heavily exposed people to asbestos, Huuskonen et al. [29] have reported an association between emphysema findings and asbestos exposure, after adjustment for age and smoking. They have differentiated, however, emphysema subtypes and used a more detailed score than we did. The causative role of asbestos on emphysema remains to be determined.

Our study had several limitations. Exclusion of already compensated subjects before the first round of this survey may introduce some selection bias. Moreover, this was a voluntary based participation survey thus, motivations of subjects to not participate in the second round were unknown CT images evaluated in this study were acquired at the second round of screening, where all the examinations were stored on digital support for expert analysis in contrast with first round screening examinations that were already reported [1]. Regarding the asbestos exposure evaluation, no atmospheric measurements was performed in this cohort and detailed information on the frequency of exposure was lacking. Therefore, the CEI was expressed in exposure units x years rather than in fibers/mL x years. However, the semi-quantitative analysis of exposure with ordinal classes allowed us to evaluate the association between increasing exposure levels and interstitial lung abnormalities. The low prevalence of asbestosis in this population is in agreement with recent data in the asbestosis epidemiology [5]. However, since this cohort is based on voluntary participation and subjects who have been already compensated for asbestos-related occupational disease before entering the survey were not included, it is likely than a substantial fraction of subjects with overt asbestosis were not included in the study. There was no histological proof of asbestosis in our population. However, the diagnosis is currently made on the basis of HRCT and an appropriate history of asbestos exposure. Indeed, as a consensus, no surgical lung biopsy is needed in these patients. In addition, we have no control group, but ethical considerations prevent the use of radiation exposure due to CT in non-asbestos-exposed subjects.

We have reported that the presence of interstitial abnormalities at HRCT was not associated to the level of exposure in a population of asbestos-exposed subjects, but to aging. This should raise the issue of ILA as an early stage of IPF with a need of an adequate surveillance.

**Supplementary Materials:** The following are available online at https://www.mdpi.com/article/10 .3390/jcm10143130/s1. Table S1: Association between the presence of UIP pattern (definite, possible UIP pattern) and interstitial pattern inconsistent with UIP pattern at CT and asbestos exposure, age, and smoking status. Table S2: Association between emphysema at CT and asbestos exposure. Table S3: Most represented occupations (ISCO 68) in the general population (*n* = 2157) and according interstitial abnormalities detected at CT. Table S4: Most represented industries (ISIC Rev 2) in all subjects and according interstitial abnormalities detected at CT.

**Author Contributions:** Conceptualization, F.L., C.P., G.F. and J.-C.P.; methodology, C.G., A.L., J.G., J.-C.P.; validation, F.L., I.B., G.D. and J.-C.P.; formal analysis, F.L., I.B., G.D., C.G., J.G., A.L. and J.-C.P.; investigation, F.L., I.B., G.D., I.T., B.C., P.B., A.G., P.A., S.C., C.P., G.F. and J.-C.P.; resources, F.L., I.B., G.D., I.T., B.C., P.B., A.G., P.A., S.C., F.D., C.P., G.F. and J.-C.P.; data curation, C.G., J.G., A.L.; writing—original draft preparation, F.L. and I.B.; writing—review and editing, F.L., I.B., G.D., C.G., I.T., B.C., P.B., A.G., P.A., S.C., J.G., A.L., F.D., C.P., G.F. and J.-C.P.; visualization, C.G.; supervision, J.-C.P.; project administration, S.C.; funding acquisition, F.L., I.T., B.C., P.B., A.G., P.A., S.C., C.P., G.F. and J.-C.P. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by French National Health Insurance (Occupational Risk Prevention Department), French Ministry of Labour and Social Relations, French Agency for Food, Environmental and Occupational Health and Safety (ANSES grant 07-CRD-51, EST 2006/1/43, EST 2009/68).

**Institutional Review Board Statement:** The study was approved by the hospital ethics committee (CPPRB Paris-Cochin n◦1946 (2002), CPP Ile De France III n◦1946/11/02-02 (2010)).

**Informed Consent Statement:** All participants received information about the study and gave their written informed consent.

**Data Availability Statement:** The data presented in this study are available on request from the corresponding author. The data are not publicly available due to privacy.

**Conflicts of Interest:** The authors declare no conflict of interest.

### **References**


### *Article* **Compressed Sensing Real-Time Cine Reduces CMR Arrhythmia-Related Artifacts**

**Benjamin Longère 1,\*, Paul-Edouard Allard 2, Christos V Gkizas 2, Augustin Coisne 1, Justin Hennicaux 2, Arianna Simeone 2, Michaela Schmidt 3, Christoph Forman 3, Solenn Toupin 4, David Montaigne <sup>1</sup> and François Pontana <sup>1</sup>**


**Abstract:** Background and objective: Cardiac magnetic resonance (CMR) is a key tool for cardiac work-up. However, arrhythmia can be responsible for arrhythmia-related artifacts (ARA) and increased scan time using segmented sequences. The aim of this study is to evaluate the effect of cardiac arrhythmia on image quality in a comparison of a compressed sensing real-time (CSrt) cine sequence with the reference prospectively gated segmented balanced steady-state free precession (Cineref) technique regarding ARA. Methods: A total of 71 consecutive adult patients (41 males; mean age = 59.5 ± 20.1 years (95% CI: 54.7–64.2 years)) referred for CMR examination with concomitant irregular heart rate (defined by an RR interval coefficient of variation >10%) during scanning were prospectively enrolled. For each patient, two cine sequences were systematically acquired: first, the reference prospectively triggered multi-breath-hold Cineref sequence including a short-axis stack, one four-chamber slice, and a couple of two-chamber slices; second, an additional single breath-hold CSrt sequence providing the same slices as the reference technique. Two radiologists independently assessed ARA and image quality (overall, acquisition, and edge sharpness) for both techniques. Results: The mean heart rate was 71.8 ± 19.0 (SD) beat per minute (bpm) (95% CI: 67.4–76.3 bpm) and its coefficient of variation was 25.0 ± 9.4 (SD) % (95% CI: 22.8–27.2%). Acquisition was significantly faster with CSrt than with Cineref (Cineref: 556.7 ± 145.4 (SD) s (95% CI: 496.7–616.7 s); CSrt: 23.9 ± 7.9 (SD) s (95% CI: 20.6–27.1 s); *p* < 0.0001). A total of 599 pairs of cine slices were evaluated (median: 8 (range: 6–14) slices per patient). The mean proportion of ARA-impaired slices per patient was 85.9 ± 22.7 (SD) % using Cineref, but this was figure was zero using CSrt (*p* < 0.0001). The European CMR registry artifact score was lower with CSrt (median: 1 (range: 0–5)) than with Cineref (median: 3 (range: 0–3); *p* < 0.0001). Subjective image quality was higher in CSrt than in Cineref (median: 3 (range: 1–3) versus 2 (range: 1–4), respectively; *p* < 0.0001). In line, edge sharpness was higher on CSrt cine than on Cineref images (0.054 <sup>±</sup> 0.016 pixel−<sup>1</sup> (95% CI: 0.050–0.057 pixel−1) versus 0.042 <sup>±</sup> 0.022 pixel−<sup>1</sup> (95% CI: 0.037–0.047 pixel−1), respectively; *p* = 0.0001). Conclusion: Compressed sensing real-time cine drastically reduces arrhythmia-related artifacts and thus improves cine image quality in patients with arrhythmia.

**Keywords:** cardiac; heart; magnetic resonance; CMR; compressed sensing; real-time; fast imaging; arrhythmia; artifact

**Citation:** Longère, B.; Allard, P.-E.; Gkizas, C.V.; Coisne, A.; Hennicaux, J.; Simeone, A.; Schmidt, M.; Forman, C.; Toupin, S.; Montaigne, D.; et al. Compressed Sensing Real-Time Cine Reduces CMR Arrhythmia-Related Artifacts. *J. Clin. Med.* **2021**, *10*, 3274. https://doi.org/10.3390/jcm10153274

Academic Editor: Mickaël Ohana

Received: 22 June 2021 Accepted: 21 July 2021 Published: 24 July 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

### **1. Introduction**

Cardiac magnetic resonance (CMR) is a major imaging modality for the assessment of left and right ventricular volumes and mass [1–3]. Moreover, it provides effective morphologic and kinetic assessment, including of the right ventricle which is not easily evaluated with ultrasounds due to its retrosternal location [4]. Multi-breath-hold segmented balanced steady state free precession (bSSFP) sequences are considered superior to gradient-echo imaging since they provide better endocardium delineation and reproducibility in a shorter scan time [5]. Retrospective electrocardiogram (ECG) gating requires the heart rate (HR) to be a regular periodic phenomenon as pieces of data are continuously acquired on multiple cardiac cycles, time-labelled and merged for the reconstruction of a whole cine slice, which is a weighted representation of successive heartbeats. It allows adapting the length of the acquisition window to the duration of the heartbeat during the continuous acquisition. This enables capturing of the complete cardiac cycle in segmented acquisitions. Typically, k-space interpolation or filtering is applied to retrospectively gate the acquired data to a reference heartbeat [6,7]. In the case of arrhythmia, artifacts occur since reconstruction is performed using incoming data from different frames of the cardiac cycle. Arrhythmia rejection algorithms can be applied with retrospective gating but may end in exceedingly long breath-holds. These arrhythmia-related artifacts (ARA) may be limited using prospectively triggered sequences by setting the acquisition window shorter than the briefest measured RR interval (time laps between two consecutive R peaks) [8]. However, this requires decreasing the number of k-space lines acquired per cardiac frame in order to preserve the widely accepted temporal resolution of 20 phases per cardiac cycle and misses to display the diastolic phases [9,10]. As a result of these adjustments, longer breath-holds and scan time are observed while the last phases of the cardiac cycle are not sampled.

Decreasing the amount of measured data is a simple way to reduce acquisition time. In recent years, compressed sensing was established as a powerful method to drastically reduce scan time [11–14]. This is achieved by highly undersampling k-space with a random sampling pattern. After Fourier transform, these acquired data result in noise-like, incoherent artifacts. These artifacts are compensated for in the final image with a non-linear iterative reconstruction exploiting the fact that medical images have a sparse representation. In combination with parallel imaging, acceleration rates can be achieved with CS that enable real-time cardiac cine imaging based on a balanced bSSFP readout with spatiotemporal resolution in a similar range to the reference (Cineref) acquisitions [15].

Various CMR studies have evaluated real-time CS cine sequences in 1.5 and 3 Tesla magnetic resonance scanners showing promising results for the assessment of left and right ventricles, including in patients with atrial fibrillation [16–21]. However, image quality was not specifically assessed in patients with irregular HR. Based on the assumption that real-time CS cine (CSrt) could reduce ARA, our study aimed at evaluating its image quality as compared to the reference multi-breath-hold segmented bSSFP cine (Cineref) in patients suffering from cardiac arrhythmia.

### **2. Materials and Methods**

### *2.1. Study Population*

From January 2019 to December 2019, 71 adult patients referred to our cardiovascular radiology department for CMR with concomitant arrhythmia during scanning were enrolled. Irregular HR was defined when the coefficient of variation of RR intervals (CVRR) was greater than 10% while scanning. The CVRR was calculated as the ratio of the standard deviation to the mean of RR intervals' durations which were obtained from digital imaging and communications in medicine (DICOM) fields. Patients under 18 years old, grown-up congenital heart disease, stress CMR, patients undergoing ECG retrogated CMR and patients with sinus rhythm were excluded. A graphic illustration of the study design is provided in Figure S1 (Supplementary Materials). The protocol was approved by our institutional ethics committee and patients gave informed consent. The study was approved

by the French National Agency for the Safety of Medicines and Health Devices (ANSM; ID-RCB: 2017-A00852-51).

### *2.2. Imaging Protocol*

CMR studies were performed on a 1.5 T scanner (MAGNETOM Aera, Siemens Healthcare, Erlangen, Germany). Every patient underwent two series of cine images: first, the reference prospectively triggered and segmented multi-breath-hold Cineref sequence; second, the prototype single-breath-hold real-time single-shot CSrt cine sequence. Both acquisitions included one left ventricular (LV) and one right ventricular (RV) two-chamber slice, one four-chamber slice and a LV short-axis stack covering both ventricles with an 8 mm slice thickness and a 2 mm gap. Regarding the prospectively gated Cineref sequence, 20 phases of the cardiac cycle were acquired and the number of views per frame was set to reach this sampling rate. In single-shot CSrt cine imaging, the data acquisition was performed in a single heartbeat. The acquisition was triggered by the R peak on the ECG. With adaptive triggering, the acquisition was stopped with the next R peak, which allowed capturing the complete cardiac cycle. However, for multi-slice acquisition it could lead to a variation in the number of cardiac phases acquired between different slices as the temporal resolution was fixed. Temporal interpolation was applied to generate an additional dataset with a fixed number of cardiac phases (*n* = 20). This dataset was used to quantify cardiac function using a dedicated post-processing software that required a fixed number of cardiac phases. To evaluate the CSrt sequence in clinical conditions, 40 iterations were used to perform image recovery to maintain an acceptable reconstruction time. An additional phase contrast imaging (PCI) flow sequence was acquired on the aortic root. Segmented Cineref and CSrt cine sequences parameters are available in Table 1.

**Table 1.** Imaging parameters of the reference prospectively triggered steady-state free-precession cine imaging and real-time compressed sensing cine imaging.


Data are expressed as mean ± standard deviation in the absence of any indication. <sup>a</sup> Constant value. <sup>b</sup> Interpolation was performed to provide a constant frame rate of 20 cardiac phases per cycle for post-processing. <sup>c</sup> The number of views per frame was set according to the shorter RR interval in order to acquire 20 cardiac phases. Prospective triggering allows data sampling during a fixed acquisition window after each R peak while adaptative triggering allow data sampling until the next R peak occurs. Abbreviations: Cineref, reference segmented cine; CSrt, real-time compressed sensing cine; ECG, electrocardiogram; *n*, data represented as numbers; NA, not applicable.

### *2.3. Cine Images Quality Assessment*

Image quality was evaluated in both groups using four indicators. First, the subjective overall image quality was evaluated using a subjective 4-point Likert scale (1: non diagnostic; 2: poor; 3: good; 4: excellent). Secondly, an objective image quality assessment was carried out based on standardized criteria adapted from the European CMR registry

"LV-Function cine SSFP" section (referred to below as "EuroCMR score") [22] (p. 3). Higher scores referred to more frequent artifact occurrence (Table 2).

**Table 2.** "LV-Function cine SSFP" section of the standardized objective quality criteria score based on the European CMR registry. Adapted from [22] (p. 3).


Every acquisition using both sequences marked a null score concerning the four last items. Indeed, acquisitions were repeated every time slice orientation was not appropriated (item 9 = 0); all acquisitions (Cineref and CSrt) were performed using the same slice thickness and gap which were 8 mm (item 10 = 0) and 2 mm (item 11 = 0), respectively, and both horizontal and vertical long-axis slices were systematically acquired (item 12 = 0). Criteria in italics were not applied, and only bold criteria were used for objective quality assessment in our study, providing a maximum score of 10 points. The more artifacts there were, the higher the score was. Abbreviations: LV, left ventricle; SSFP, steady-state free precession; CMR, cardiac magnetic resonance; Cineref, reference segmented cine; CSrt, real-time compressed sensing cine; ARA, arrhythmia-related artifacts.

> Third, the proportion of short-axis slices affected by ARA in each stack of both sequences was calculated, referred to as ARA rate. ARA were defined as a blurring of all or a part of the LV wall borders [22].

> Finally, the edge sharpness (ε) of the boundary between myocardium and blood pool, which is the spatial frequency (in pixel−1) reflecting the spatial resolution, was measured on paired Cineref and CSrt four-chamber slices at end-diastole, accordingly to the literature [23,24]. Additional measurement at end-systole was performed. The edge spread function (ESF), which is the response of the imaging system to a high contrast boundary, was measured on MATLAB (version R2015a, The MathWorks, Natick, MA, USA), by drawing a signal profile line perpendicularly across the edge between the interventricular septum and the LV blood pool (Figure 1a,b) [25]. Then, ε was calculated as the reciprocal of the distance separating the points corresponding to 20% and 80% of the difference between local minimum and maximum signal intensities (Figure 1c,d).

**Figure 1.** Example of edge sharpness assessment at end-diastole for a 56-year-old male patient suffering from atrial fibrillation. The same four-chamber view at end-diastole is acquired with (**a**) the Cineref sequence and (**b**) the CSrt sequence. An orthogonal profile line was drawn at mid-cavity across the border between the septal myocardium and the left ventricular blood pool (blue line) on a four-chamber view. It provided intensity profiles (blue curves) along the line for (**c**) Cineref and (**d**) CSrt cine. The edge sharpness was the inverse of the distance *d* (in pixels) between the positions corresponding to 20% and 80% (red stars) of the difference between the maximum and minimum signal intensities (blue crosses). The edge sharpness was expressed in pixel−1. This measurement was performed at end-diastole and end-systole for both sequences. Note that the peaks (arrows heads) added to the Cineref signal profile curve (**c**) correspond to the doubling of the interventricular septum border (arrows) on the cine view (**a**). The same assessment was performed on both sequences at end-diastole and end-systole for the 71 enrolled patients. Abbreviations: Cineref, reference segmented cine; CSrt, real-time compressed sensing cine; εCineref, edge sharpness measured on Cineref sequence; εCSrt, edge sharpness measured on CSrt cine; I, signal intensity; Imin, minimal signal intensity; Imax, maximal signal intensity; d, distance along the profile line.

### *2.4. Conditions of Image Analysis*

Images from both sequences were anonymized before transfer to a clinical workstation (Sygno.via VB30A, Siemens Healthcare, Erlangen, Germany). A radiologist with 4 years of experience (PEA) performed the image quality assessment according to the above-cited indicators. Image sets were randomly evaluated in each group. The same observer (PEA) first performed the quality assessment of the reference Cineref images and at least one month later evaluated the CSrt images. For each patient, arrhythmia was quantified by calculating the CVRR. An additional assessment was performed by a radiologist with 8 years of experience (BL) from 30 randomly selected patients to evaluate the interobserver

agreement and performed the same assessment regarding subjective quality, EuroCMR score and ARA rates. In the case of mismatch between the two readers, a radiologist with 15 years of experience (FP) performed the quality assessment with the two others to reach consensual scores which were used instead of those set by the first and less experienced observer. Mismatches were defined by discrepancies greater than or equal to 2 points regarding subjective quality score and EuroCMR score, or by a 20% difference in ARA rates. The edge sharpness assessment was automated and was not evaluated for interobserver agreement. Finally, semi-automated segmentation of LV endocardium and epicardium, and manual segmentation RV endocardium were performed on the same workstation with both cine sequences for each patient. LV stroke volume was also measured on PCI sequence.

### *2.5. Statistics Analysis*

Categorical data were represented as numbers (percentages), continuous variables as mean ± standard deviation (SD) (95% confidence interval (CI)) in case of normal distribution and median (range: minimum–maximum) in other cases. Sequences were compared using the Wilcoxon signed-rank test regarding the overall subjective quality score and the modified EuroCMR score. Paired Student's *t*-test was used for ARA rates, edge sharpness comparisons, and ventricular functional parameters comparison. An analysis of variance (ANOVA) was used to compare LV stroke volumes assessed by cine segmentation and PCI flow sequence. Intraclass correlation coefficient and kappa test were applied to assess the interobserver agreement [26]. Values of *p* < 0.05 were considered statistically significant. Statistical analysis was performed using MedCalc software (version 14.8.1.0, MedCalc Software, Ostend, Belgium).

### **3. Results**

### *3.1. Population Description*

The mean age of the population was 59.5 ± 20.1 (SD) years (95% CI: 54.7–64.2 years) with a male predominance (*n* = 41/71; 57.7%, women: *n* = 30/71; 42.3%). Patients were referred for initial work-up or follow-up of coronary artery disease (*n* = 17; 23.9%), heart rhythm disorder (*n* = 14; 19.7%), dilated cardiomyopathy (*n* = 11; 15.5%), infiltrative cardiomyopathy (*n* = 8; 11.3%), heart valve disease (*n* = 7; 9.9%), myocarditis (*n* = 6; 8.5%), hypertrophic cardiomyopathy (*n* = 5; 7.0%), and heart failure (*n* = 3; 4.2%). The mean HR was 71.8 ± 19.0 beats per minute (bpm) (95% CI: 67.4–76.3 bpm) and 38.0% of the patients (*n* = 27/71) demonstrated a mean HR above 75 bpm, meaning the 49 ms temporal resolution of the CSrt cine provided less than 16 frames of the cardiac cycle per slice. The mean CVRR was 25.0 ± 9.4% (95% CI: 22.8–27.2%). Arrhythmia was caused by atrial fibrillation (*n* = 42/71; 59.2%), ventricular hyperexcitability (*n* = 17/71; 23.9%), and conduction disorders (*n* = 12/71; 16.9%). Demographic data are summarized in Table 3. Biventricular functional assessment of the population is reported in Table 4.



Abbreviations: CVRR, coefficient of variation of RR interval; bpm, beat per minute; SD, standard deviation; 95% CI, 95% confidence interval.


**Table 4.** Biventricular functional assessment of the study population.

Data are presented as mean ± SD (95% CI). The significance of statistic tests is defined by values of *<sup>p</sup>* < 0.05. <sup>a</sup> Student's *<sup>t</sup>*-test; <sup>b</sup> Analysis of variance. Abbreviations: Cineref, reference segmented cine; CSrt, real-time compressed sensing cine; PCI, phase contrast imaging sequence; SD, standard deviation; 95% CI, 95% confidence interval; LV, left ventricular; RV, right ventricular; EF, ejection fraction; EDV, end-diastolic volume; ESV, end-systolic volume; SV, stroke volume; LVM, left ventricular mass.

### *3.2. Cine Acquisitions*

A total of 599 short-axis cine slices were acquired with each sequence. A median number of 8 (range: 6–14) cine slices was acquired twice for each patient, depending on cardiac morphology. To acquire the same slices, CSrt was significantly faster than Cineref (Cineref: 556.7 ± 145.4 (SD) s (95% CI: 496.7–616.7 s); CSrt: 23.9 ± 7.9 (SD) s (95% CI: 20.6–27.1 s); *p* < 0.0001).

### *3.3. Objective European CMR Standardized Criteria-Based Quality Score*

The EuroCMR score for the CSrt cine (median: 1 (range: 0–5)) was significantly better than for the Cineref sequence (median: 3 (range: 0–3); *p* < 0.0001) (Table 5) (Figure 2; Video S1 (Supplementary Materials)). Interobserver agreements were 0.94 and 0.89 regarding Cineref and CSrt, respectively. No mismatch was encountered between the readers.

**Table 5.** Objective image quality with EuroCMR criteria scores: comparison between Cineref and CSrt image sets.


The significance of Wilcoxon signed-rank test is defined by values of *p* < 0.05. Red values represent patients for whom CSrt score was equivalent to or better than that of Cineref for *n* = 68/71 patients (95.8%). Abbreviations: CMR, cardiac magnetic resonance; Cineref, reference segmented cine; CSrt, real-time compressed sensing cine; EuroCMR, European CMR registry.

**Figure 2.** Examples of comparisons between Cineref and CSrt sequences in three patients suffering from arrhythmia. Mid-cavity short-axis views acquired with (**a**,**c**,**e**) Cineref and (**b**,**d**,**f**) CSrt. The three patients were (**a**,**b**) a 74-year-old man suffering from atrial fibrillation, (**c**,**d**) a 37-year-old woman screened for a genetically proven arrhythmogenic right ventricular cardiomyopathy, (**e**,**f**) a 63-year-old woman scanned for a second-degree atrioventricular block. The image quality assessment demonstrated: (**a**) Likert scale = 1/4, EuroCMR score = 3/10, εCineref = 0.051 pixel<sup>−</sup>1; (**b**) Likert scale = 3/4, EuroCMR score = 0/10, εCSrt = 0.067 pixel−1; (**c**) Likert scale = 1/4, EuroCMR score = 3/10, εCineref = 0.015 pixel−1; (**d**) Likert scale = 3/4, EuroCMR score = 1/10, εCSrt = 0.050 pixel<sup>−</sup>1; (**e**) Likert scale = 1/4, EuroCMR score = 3/10, εCineref = 0.023 pixel−1; (**f**) Likert scale = 3/4, EuroCMR score = 0/10, εCSrt = 0.035 pixel−1. Abbreviations: Cineref, reference segmented cine; CSrt, real-time compressed sensing cine; εCineref, edge sharpness measured on Cineref; εCSrt, edge sharpness measured on CSrt cine; EuroCMR, European cardiac magnetic resonance registry.

### *3.4. Subjective Overall Quality Score*

The subjective quality score was significantly better (*p* < 0.0001) for the CSrt sequence with a median score of 3 (range: 1–3). A 0.85 interobserver agreement was reached. The Cineref sequence provided a median score of 2 (range: 1–4; *p* < 0.0001) with an intraclass coefficient of 0.82 (Table 6). No mismatch was encountered between the readers. The Cineref sequence provided 23 non-diagnostic acquisitions compromising functional and morphological assessments versus only 10 stacks with the CSrt cine (Figure 2; Video S1 (Supplementary Materials)).

**Table 6.** Subjective overall image quality scores: comparison between Cineref and CSrt image sets.


The significance of Wilcoxon signed-rank test is defined by values of *p* < 0.05. Red values represent patients for whom CSrt score was equivalent to or better than that of Cineref for *n* = 64/71 patients (90.1%). Abbreviations: Cineref, reference segmented cine; CSrt, real-time compressed sensing cine.

### *3.5. Arrhythmia-Related Artifacts Rate*

ARA using Cineref sequence were assessed on *n* = 514/599 (85.8%) cine slices from *n* = 70/71 (98.6%) patients, with a 0.90 interobserver agreement. One mismatch was encountered between the readers (PEA: *n* = 8/14, 57.1%; BL: *n* = 12/14, 85.7%; FP: *n* = 12/14, 85.7%). The mean proportion of impaired slices per patient was in 85.9 ± 22.7 (SD) %. No ARA could be depicted using the CSrt sequence.

### *3.6. Edge Sharpness*

The CSrt sequence provided a higher edge sharpness coefficient at end-diastole (εCSrt = 0.051 ± 0.016 pixel−<sup>1</sup> (95% CI: 0.048–0.055 pixel−1)) than the Cineref (εCineref = 0.040 ± 0.018 pixel−<sup>1</sup> (95% CI: 0.036–0.044 pixel<sup>−</sup>1)) (*p* = 0.0001). A similar finding was observed at end-systole (εCSrt = 0.054 ± 0.016 pixel−<sup>1</sup> (95% CI: 0.050–0.057 pixel−1); <sup>ε</sup>Cineref = 0.042 ± 0.022 pixel−<sup>1</sup> (95% CI: 0.037–0.047 pixel<sup>−</sup>1); *p* = 0.0001).

### **4. Discussion**

This prospective monocentric study based on a 71-patient cohort is, to our knowledge, the widest and most comprehensive study to evaluate the CSrt sequence in patients with irregular HR. Previous studies in non-selected patients confirmed that real-time CS cine imaging is a reliable alternative to segmented multi-breath-hold SSFP for the assessment of both ventricles' volumes and function in addition to reducing acquisition time [16–20]. Our study demonstrates a dramatic drop in ARA and a significant improvement of subjective and objective image quality with CSrt in patients suffering from heart rhythm disorders. However, no CSrt set was rated as excellent because of the smooth boundaries rendered by the interpolation process which are mandatory for post-processing. Indeed, since the temporal resolution of the CS sequence is fixed, a variable number of frames will be acquired from one cycle to another in the case of arrhythmia. For segmentation to be achieved, post-processing tools require all cine slices to display the same number of frames per cycle. Consequently, a standardization is performed to display 20 frames per cycle on all slices.

We previously showed on a small sub-group of 25 patients suffering from arrhythmia that CSrt and Cineref sequences allowed similar image quality [20]. However, the present study does not only suggest equivalent scores but significantly better objective and subjective image quality scores with the CSrt cine. Of note, this sequence still provided non-null EuroCMR scores since the slices were identically located on both sequences; accordingly, most of the wrap-around or metallic artifacts were reproduced on the CSrt acquisition.

Our results are in line with the previous study by Goebel et al. on 20 patients with atrial fibrillation [21]. This study focused on a subjective semi-quantitative 4-point quality score and the evaluation of the variation of the myocardial signal intensity which is the reciprocal of the signal-to-noise ratio (SNR). However, this last parameter, or its reciprocal, is considered as hardly suitable for non-linear iterative reconstructions. Moreover, CS is built to suppress pieces of the image signal, while SNR is suited for fully sampled data [27]. Besides our study being specifically designed to evaluate the image quality using additional quantitative and objective metrics, we also performed a clinically integrated evaluation in a larger population.

The higher edge sharpness of CSrt images reflected the faster signal variation along a distance and a better delineation of the image boundaries. We evaluated the edge sharpness both at end-diastole, when the myocardium is supposed to be relatively still, and at end-systole. This metric, regardless of the non-linearity of the reconstruction process, is more suitable than the SNR or its reciprocal to evaluate image quality [27]. The ESF and its inverse value ε measure the imaging system ability to restitute high contrast transition in images. This parameter, the derivative (the line spread function) and its Fourier transform (the so-called task-based modulation transfer function or task transfer function) are currently considered for image quality assessment in the field of non-linear image reconstructions [24,25,28].

Regarding volumetric evaluation in patients suffering from arrhythmia, CSrt provided significantly lower LV end-diastolic volume (−3.6 ± 7.2 mL) than measured on Cineref, which was already observed in previous studies [16,20]. As for RV, there was a significant underestimation of all evaluated functional parameters. Nevertheless, these variations compared to the conventional Cineref should be considered with caution in a population with arrhythmia. Indeed, irregular heartbeats induce variable ventricular preloads and contractions, making the real reference values impossible to determine. Moreover, since CSrt demonstrated a better image quality than Cineref that was impaired by ARA, one may consider the segmentation to be more reliable on CSrt.

Besides the image quality improvement, the single breath-hold CSrt sequence allowed a dramatic reduction in scan time. Not only was acquisition faster, but the ARA reduction avoided repeating the acquisition of non-diagnostic slices [20]. The workflow improvement being a major issue in the field of CMR, this real-time sequence is very promising and may improve cost-effectiveness [29].

### *Limitations*

Although the overall subjective image quality was improved with CSrt cine, 10 stacks were still considered as non-diagnostic. Indeed, iterative reconstructions occasionally failed or were not completely achieved on this prototype sequence. Nevertheless, such failures are now rare since the release of the final version of the sequence. Moreover, ECG-related issues occurred when R peaks were occasionally missed, which made the system consider two consecutive heartbeats as one single cycle. The corresponding slices then display a double heart cycle which could not be used for post-processing and was ranked as non-diagnostic. Special attention should be paid to skin preparation before ECG electrode placement.

Other fast real-time sequences, such as radial acquisition, have previously been reported [30,31]. Our study does not compare CSrt to other types of real-time sequences. To our knowledge, no such evaluation has been published and further study would be required for comparison.

A methodological limitation of our study is the impossibility to perform blinded evaluation of the sequences since CSrt cine displayed smoother boundaries than Cineref sequence. Consequently, observers could recognize the type of sequence they were evaluating. However, paired CSrt and Cineref stacks from the same patients were separated, randomized, and assessed during different sessions.

Regarding the sampling of heart cycles for the assessment of ventricular volumes and mass, the fixed temporal resolution leads to variable sampling rates from shorter cycler to longer ones. As a consequence, in the case of HR faster than 60 bpm, the recommended 20 frames per cycle could not be acquired [9,10]. However, in the field of analog-to-digital signal conversion, a 16-time oversampling is reputed sufficient for the signal restitution to be accurate, corresponding to a 75 bpm HR [32]. High CVRR in HR may be encountered, and some slices may display undersampled heart cycles. In our study, 38% of the patients demonstrated a mean HR above 75 bpm for whom the undersampling of the cardiac cycle should be considered cautiously during interpretation, especially for volume segmentations. Nevertheless, it must be balanced by the reduction of ARA provided by CSrt.

### **5. Conclusions**

In addition to reducing acquisition time, CSrt sequence drastically reduces arrhythmiarelated artifacts and improves image quality in patients with irregular heart rate. This rapid imaging technique allows practitioners, in daily practice, to improve quality, workflow and accessibility of CMR for patients with challenging cardiac conditions.

**Supplementary Materials:** The following are available online at https://www.mdpi.com/article/10 .3390/jcm10153274/s1, Figure S1: Study design. Video S1: Viability assessment in a 77-year-old man with premature ventricular contractions caused by ischemic scars.

**Author Contributions:** B.L.: study conception and design, data collection, interpretation and analysis, drafting of the manuscript, critical revision for important intellectual content; P.-E.A.: data collection, interpretation and analysis, drafting of the manuscript; C.V.G.: data collection and interpretation, critical revision for important intellectual content; A.C.: study conception, critical revision for important intellectual content; J.H.: data collection and interpretation, critical revision for important intellectual content; A.S.: data collection and interpretation, critical revision for important intellectual content; M.S.: study conception and design, critical revision for important intellectual content; C.F.: study conception and design, critical revision for important intellectual content; S.T.: study conception and design, critical revision for important intellectual content; D.M.: study conception, critical revision for important intellectual content; F.P.: study conception and design, data collection, interpretation and analysis, drafting of the manuscript, critical revision for important intellectual content. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Institutional Review Board Statement:** The study was approved by the research ethics committee of Lille University Hospital.

**Informed Consent Statement:** Informed consent was obtained from all subjects involved in the study.

**Data Availability Statement:** The data presented in this study are available on reasonable request from the corresponding author, subject to approval by the research ethics committee of Lille University Hospital.

**Conflicts of Interest:** B.L.; P.-E.A.; C.V.G.; A.C.; J.H.; A.S.; D.M.; F.P. have no competing interest. They are employed by an institution engaged in a contractual collaboration with Siemens Healthineers. M.S.; C.F.; S.T. are employees of Siemens Healthineers.

### **References**


### *Article* **Impact of Morphotype on Image Quality and Diagnostic Performance of Ultra-Low-Dose Chest CT**

**Anne-Claire Ortlieb 1, Aissam Labani 2, François Severac 3, Mi-Young Jeung 2, Catherine Roy <sup>2</sup> and Mickaël Ohana 2,4,\***


**Abstract:** Objectives: The image quality of an Ultra-Low-Dose (ULD) chest CT depends on the patient's morphotype. We hypothesize that there is a threshold beyond which the diagnostic performance of a ULD chest CT is too degraded. This work assesses the influence of morphotype (Body Mass Index BMI, Maximum Transverse Chest Diameter MTCD and gender) on image quality and the diagnostic performance of a ULD chest CT. Methods: A total of 170 patients from three prior prospective monocentric studies were retrospectively included. Renewal of consent was waived by our IRB. All the patients underwent two consecutive unenhanced chest CT acquisitions with a full dose (120 kV, automated tube current modulation) and a ULD (135 kV, fixed tube current at 10 mA). Image noise, subjective image quality and diagnostic performance for nine predefined lung parenchyma lesions were assessed by two independent readers, and correlations with the patient's morphotype were sought. Results: The mean BMI was 26.6 ± 5.3; 20.6% of patients had a BMI > 30. There was a statistically significant negative correlation of the BMI with the image quality (ρ = −0.32; IC95% = (−0.468; −0.18)). The per-patient diagnostic performance of ULD was sensitivity, 77%; specificity, 99%; PPV, 94% and NPV, 65%. There was no statistically significant influence of the BMI, the MTCD nor the gender on the per-patient and per-lesion diagnostic performance of a ULD chest CT, apart from a significant negative correlation for the detection of emphysema. Conclusions: Despite a negative correlation between the BMI and the image quality of a ULD chest CT, we did not find a correlation between the BMI and the diagnostic performance of the examination, suggesting a possible use of the ULD protocol in obese patients.

**Keywords:** multidetector computed tomography; radiation dosage; lung; helical computed tomography

### **1. Introduction**

Computed Tomography (CT) accounts for the large majority of medical exposures to ionizing radiation [1]. In chest imaging, the radiation dose delivered varies significantly between clinical indications, machine and institution [2] and the dose-length product (DLP) is estimated at 347 mGy·cm (50th percentile) in a 2017 review of 159,909 US patients [3]. Significant susceptible organs such as the breasts, lungs and thyroid are included in the field-of-view; this justifies the reduction in the radiation dose delivered to the minimum needed for adequate diagnosis [4,5].

Chest CT is an appropriate candidate to a significant radiation dose reduction, due to an overall low attenuation of the lung parenchyma resulting in a high contrast. Substantial efforts have been made to optimize the dose delivered in chest imaging, particularly

**Citation:** Ortlieb, A.-C.; Labani, A.; Severac, F.; Jeung, M.-Y.; Roy, C.; Ohana, M. Impact of Morphotype on Image Quality and Diagnostic Performance of Ultra-Low-Dose Chest CT. *J. Clin. Med.* **2021**, *10*, 3284. https://doi.org/10.3390/jcm10153284

Academic Editor: Ernesto Di Cesare

Received: 16 June 2021 Accepted: 22 July 2021 Published: 26 July 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

through the implementation of iterative reconstruction [6] and deep learning reconstruction techniques [7]. It is now possible to achieve a diagnostic chest CT at the radiation dose of a posteroanterior and lateral chest radiograph series [8,9], and this has been proven efficient for various clinical scenarios [10], such as lung nodule detection and follow-up [11,12], pulmonary infection [13], CT-guided percutaneous biopsy [14], lymphangioleiomyomatosis [15], etc. However, the image quality of these Ultra-Low-Dose (ULD) chest CTs remains dependent on the patient's morphotype, and prior studies [11,16] seldomly included patients with a Body Mass Index (BMI) greater than 30 kg·m<sup>−</sup>2. Consequently, the diagnostic performance of a ULD chest CT is not well documented in obese patients [17].

We hypothesize that beyond a certain morphotype, defined by a threshold of BMI and/or a threshold of Maximum Transverse Chest Diameter (MTCD) and/or gender (given the different distribution of fat and breasts between women and men), the deterioration of the image quality becomes too important and leads to a decrease in the diagnostic performance.

The objective of this study was, therefore, to determine the influence of these morphological parameters (BMI, MTCD and gender) on the objective and subjective image quality of a ULD chest CT and on its diagnostic performance for the detection of nine predefined pulmonary parenchymal lesions.

### **2. Materials and Methods**

### *2.1. Population*

All 170 patients from 3 prior prospective studies performed in Strasbourg University Hospital and carried out on the same second-generation 320-row scanner (Aquillion One Vision Edition, Toshiba, Japan), were retrospectively included. The first study [18] included 55 patients from July 2013 to May 2014 with an occupational exposure to asbestos of at least 15 years, referred for screening of asbestos-related pleuro-pulmonary lesions. The second study [19] (April 2014 to September 2014) and the third study [20] (April 2015 to September 2015) included 51 and 64 patients, respectively, that were referred for a clinically indicated unenhanced chest CT.

The inclusion criterion common to the 3 studies was an age greater or equal to 35 years for men and 40 years for women. The common exclusion criteria were pregnancy, the inability to maintain apnea for more than 5 s, the inability to raise the arms above the head and the inability to give informed consent.

The renewal of consent was not required for this retrospective study by our hospital institutional review board. Written informed consent and approval from the local Ethics Committee had been obtained in the first instance for all patients, at the time of their initial prospective inclusion in one of the 3 trials.

Age, gender, weight, height, BMI and clinical indication for chest CT were recorded at the time of the examination. The Maximum Transverse Chest Diameter (MTCD) was measured by a radiologist (AC, with 4 years of experience in chest CT) on a mediastinal window, from lateral thoracic wall to lateral thoracic wall.

### *2.2. CT Acquisition*

Each patient underwent an unenhanced chest CT with two successive acquisitions: one "full dose" (FD) acquisition (120 kV, automated tube current modulation maxed at 700 mA) used as the Gold Standard and one "Ultra-Low-Dose" (ULD) acquisition (135 kV, 10 mA fixed) [21]. Acquisitions were made on a second-generation 320-row scanner (Aquillion One Vision Edition, Canon), with the parameters given in Table 1. All examinations were performed in successive end-inspiratory apneas, arms raised above the head. The 55 patients from the first study were acquired in prone position, to avoid gravitydependent posterior parenchymal abnormalities—as recommended when screening for asbestos-related diseases [18]. The 115 patients from the two following studies were acquired in supine position. The posterior–anterior and lateral topograms were used to

define the acquisition field, extending from the pulmonary apexes to the diaphragm; the acquisition length was kept identical for the two consecutive FD and ULD acquisitions.

**Table 1.** Acquisition parameters of study protocols.


The FD and ULD acquisitions were reconstructed in lung window (width = 1500 HU; center = −700 HU) with a hard kernel and a slice thickness of 1 mm, using the constructor's iterative reconstruction algorithm set at a standard level (Adaptative Iterative Dose Reduction using 3 Dimensional AIDR-3D, Canon).

The whole set of 340 parenchymal reconstructions were anonymized and randomized into two series of 170 FD and 170 ULD examinations, in a different random order for both.

### *2.3. Dosimetry*

Radiation doses were expressed in dose-length product (DLP). The effective radiation dose (ED) was calculated by multiplying the DLP by the chest-specific conversion factor of 0.014 mSv/mGy·cm [22].

### *2.4. Quantitative Image Quality*

Image noise was defined as the standard deviation (SD) of the attenuation of air within the tracheal lumen, using a circular region of interest (ROI) averaging at least 40 mm2, 1 cm above the carina, in the parenchymal reconstruction. It was measured three times by the same reader (AC), with size and position of the ROI kept constant between patients. The average of the three measurements was used as the final value in the statistical analysis.

### *2.5. Qualitative Image Quality*

The overall subjective image quality of ULD acquisitions was assessed independently by two radiologists that specialized in chest imaging (M.O., with 9 years of experience in chest CT and A.C., with 4 years of experience) using a 5-point Likert scale (Table 2), after a training session with joint evaluation of 15 cases.


**Table 2.** Subjective image quality for ULD chest CT.

### *2.6. Diagnostic Performance*

Each radiologist independently read the ULD then the FD lung parenchyma acquisitions, both in a different random order and with a delay of at least two weeks in between to limit the risk of a memorization bias.

All images were analyzed on a dedicated Workstation (Vitrea version 6.4, Vital Images), with use of multiplanar reconstructions and Maximum Intensity Projection. For each acquisition, the presence or absence of 9 predefined parenchymal lesions, based on the Fleischner Society lexicon [23] was rated in a binary fashion (i.e., presence or absence of the anomaly). The following lesions were assessed: solid nodule ≥ 5 mm, ground glass nodule, mass ≥ 3 cm, ground glass opacity, alveolar consolidation, emphysema, interlobular septal thickening, bronchiectasis, fibrosis.

Disagreements between readers were resolved in a subsequent consensual reading session. The consensual analysis of the FD chest CT served as the reference standard to which the consensual analysis of ULD CT was compared.

### *2.7. Statistical Analysis*

Quantitative variables were reported as average ± SD or median (1st quartile; 3rd quartile) according to their Gaussian distribution and assessed using the Shapiro–Wilk test. Categorical variables were reported as frequencies or percentages.

The inter-reader agreement for the subjective assessment of image quality was determined using Cohen's kappa and quadratic weighting to account for the degree of disagreement. The confidence intervals for Cohen's Kappa coefficients were calculated using the Bootstrap method. A kappa value of (0.0–0.20] was regarded as poor, (0.2–0.40] as fair, (0.41–0.60] as moderate, (0.61–0.80] as good and (0.81–1.00] as excellent.

Continuous variables were compared with a Student's *t*-test when the parametric conditions were met, and otherwise with a Wilcoxon test.

The Pearson χ<sup>2</sup> test or the Fisher exact test were used to compare the effectiveness of lesion detection (presence or absence) among the two groups, based on theoretical numbers.

Correlations between continuous variables were evaluated using the Spearman's correlation coefficient. Comparison between the different correlation coefficients were made using the Hotelling Williams test.

Diagnostic performance of the ULD chest CT for the detection of 9 pulmonary lesions was compared to the FD chest CT: the sensitivity (Se), specificity (Sp), positive predictive value (PPV), negative predictive value (NPV), accuracy and error rate were calculated using a confidence interval of 95%.

A *p*-value smaller than 0.05 was considered statistically significant.

All statistical analyses were performed using R software, version 3.4.3. R Core Team (2015). (R: *A language and environment for statistical computing*. R Foundation for Statistical Computing).

### **3. Results**

### *3.1. Population*

A total of 170 patients were retrospectively included: 129 men (75.9%) and 41 women (24.1%). Age was 62.7 ± 11 years old (range 35–88).

The clinical indications were screening of asbestos-related pleuro-pulmonary diseases (*n* = 55), follow-up of pulmonary nodules (*n* = 39), pulmonary infections (*n* = 23), follow-up of lung cancer (*n* = 13), follow-up of interstitial pulmonary diseases (*n* = 10), follow-up of chronic obstructive lung diseases (*n* = 4) and other situations (*n* = 26).

The mean BMI was 26.6 ± 5.3 kg·m−<sup>2</sup> (range 14.5–54.9), with a median BMI of 25.8 kg·m−<sup>2</sup> (IQ 23.1–29.3). Of the patients, 45.9% (*<sup>n</sup>* = 78) had a BMI less than or equal to 25 kg·m−2; 33.5% (*<sup>n</sup>* = 57) had a BMI between 25 and 30 kg·m−<sup>2</sup> and 20.6% (*<sup>n</sup>* = 35) had a BMI greater than or equal to 30 kg·m−2, of which eight patients (4.7% of the population) had a BMI greater than 35 kg·m<sup>−</sup>2.

The MTCD was 35.5 ± 3.4 cm (range 27–46).

### *3.2. Dosimetry*

The DLP for FD acquisitions was 252.4 ± 143.2 mGy·cm (range 84.2–868.5), with a median of 204.4 mGy·cm (IQ 152.9–323.2). The average ED was estimated at 3.46 mSv.

The DLP for ULD acquisitions was 16.4 ± 1.8 mGy·cm (12.1–20.8), with a median of 16.3 mGy·cm (IQ 15.4–17.7). The average ED was estimated at 0.23 mSv, representing a 93% decrease compared to the FD acquisition.

### *3.3. Quantitative Image Quality*

Image noise was 39.9 ± 11.3 for FD acquisitions and 53.3 ± 14.1 for ULD acquisitions, corresponding to a 33% increase.

There was no statistically significant correlation between BMI and noise in the ULD acquisitions: ρ = −0.12; 95% CI (−0.263–0.015); *p* = 0.11 (Figure 1).

**Figure 1.** Influence of BMI on noise for ULD-CT.

There was no statistically significant correlation between MTCD and noise in the ULD acquisitions: ρ = −0.14; 95% CI (−0.276–0.007); *p* = 0.07 (Figure 2).

**Figure 2.** Influence of MTCD on image noise (in SD) for ULD-CT.

### *3.4. Subjective Image Quality*

The inter-reader agreement for the overall subjective image quality score was good, with a Cohen's Kappa of 0.61 (95% CI [0.46–0.72]; *p* < 0.001). The average score assigned by the two readers for ULD image quality was ultimately 4.5 ± 0.6 with a median of 5 (4.5–5). The overall subjective image quality was almost always diagnostic (i.e., equal to or greater than three), except for two patients according to the junior reader (BMI at 44.1 and 37.8 kg·m−2) and one patient according to the senior reader (BMI at 44.1 kg·m−2). The distribution of the ratings is detailed in Table 3. An illustrative example is given in Figure 3.

**Table 3.** Distribution of ratings for subjective image quality.


**Figure 3.** Axial slices in parenchymal window illustrating the subjective quality of the ULD (**a1**,**a2**) and FD (**b1**,**b2**) acquisitions in patients with a BMI of 30 kg·m−<sup>2</sup> (1) and 21.5 kg·m−<sup>2</sup> (2): diagnostic-quality images (QI = 5/5) despite the different morphotype.

> The BMI was negatively and significantly correlated with the subjective image quality of the ULD acquisitions, with a correlation coefficient of ρ = −0.325; 95% CI (−0.462; −0.178); *p* < 0.001 (Figure 4).

**Figure 4.** Influence of the BMI on subjective image quality.

There was no statistically significant correlation between the MTCD and the subjective image quality of the ULD acquisitions: ρ = −0.0979; 95% CI (−0.253–0.055); *p* = 0.2041 (Figure 5).

**Figure 5.** Influence of the MTCD on subjective image quality.

### *3.5. Diagnostic Performance*

At least one predefined parenchymal lesion was found in 114 (67%) out of the 170 examinations, with a total of 243 lesions on the reference FD acquisition. There was a disagreement in the reading of the reference CT scan in 44 patients: non-agreement rate = 25.9%; 95% CI, (19.5–33.2].

### *3.6. Diagnostic Performance per Patient*

The overall diagnostic performance of the ULD chest CT (all lesions considered) were sensitivity, 77%; specificity, 99%; PPV, 94% and NPV, 65%.

The overall agreement rate for the interpretation of the ULD-CT scanner for all the lesions considered (disagreement = at least one difference out of nine lesions), was 68% (116/170) 95% CI [60.67–75.15].

There was no statistically significant correlation between BMI (*p* = 0.34), MTCD (*p* = 0.92) or gender (*p* = 0.53) and the per-patient diagnostic performance of the ULD-CT.

### *3.7. Diagnostic Performance per Lesion*

The diagnostic performances per lesion are summarized in Table 4.


**Table 4.** Diagnostic performance of ULD-CT.

There was a statistically significant negative influence of the BMI on the diagnostic performance of the ULD CT for the detection of emphysema (*p* < 0.001). This negative correlation was not statistically found for the MTCD nor the gender.

There was no statistically significant correlation between BMI, MTCD or gender and the diagnostic performances for the eight other parenchymal lesions, provided that some lesions had a low incidence (ground glass nodules, masses, alveolar consolidation, septal thickening and fibrosis).

### **4. Discussion**

Our study compiling 170 patients found an overall sensitivity of 77%, specificity of 99% and an agreement rate of 74.1% for a ULD chest CT when compared to the reference FD CT, using standard iterative reconstruction techniques. Despite the statistically significant negative correlation between image quality and BMI (ρ = −0.325; *p* < 0.001), we did not find a negative correlation between BMI and diagnostic performance (per patients or per lesions, except for emphysema). As a result, the deterioration in the image quality of a ULD CT associated with the increase in BMI does not translate into a significant decrease in diagnostic performance. Based on these findings, we can recommend the use of ULD CT, even when the BMI is greater than 30 kg·m<sup>−</sup>2, and likely up to 35 kg·m<sup>2</sup> based on the BMI distribution in our population.

In prior publications, most studies did not include patients with a BMI greater than 30 kg·m<sup>−</sup>2.

In his study, Lee [11] compared the ULD CT (80 kV-30 mA fixed) to a low-dose (LD) CT in 81 patients, the average BMI of the population was 23.6 ± 3.8 kg·m−<sup>2</sup> and only four patients had a BMI greater than 30 kg·m−2. They demonstrated a significant influence of

BMI on the subjective image quality of the ULD CT (ρ = −0.480; *p* < 0.001), with the images being of diagnostic-quality in more than 95% of cases when the BMI was lower than 25, compared to only 70% when the BMI was greater than 25.

In the publication of a study conducted by Kim [13] using ULD CT (120 kV-15 mA fixed) in patients with neutropenic fever, all of the 207 patients included had a BMI lower than 30 kg·m−<sup>2</sup> and only 14% of them had a BMI greater than 25 kg·m−2, among them four had a non-diagnostic image quality.

In Macri et al.'s study [16], using 100 kV and automated tube current modulation (60 mAs reference), a large majority of the patients included had a normal BMI (average BMI of 23.9 kg·m−<sup>2</sup> for men and 23.1 kg·m−<sup>2</sup> for women) and the image quality was excellent for 90% of the patients with a BMI up to 25, compared to only 76% of patients with a BMI greater than 25. It showed an excellent agreement between the ULD-CT results compared to LD-CT, except in two patients with a BMI at 26.8 and 27.4 kg·m<sup>−</sup>2.

Messerli et al. [12] studied the diagnostic performance of a ULD chest CT (100 kV, fixed tube current at 70 mAs, tin filtration) on the detection of pulmonary nodule compared to the reference FD CT. The population studied included 202 patients with a mean BMI of 26.2 ± 5.3 kg·m−<sup>2</sup> (range 15.9–49), and the results showed a significantly decreased subjective image quality for the ULD CT in patients with a BMI greater than 30 (*p* < 0.001), in line with our findings. However, the number of patients with a BMI greater than 30 was not specified in the article. The authors found a statistically significant negative influence of the BMI on the sensitivity of the ULD-CT for the detection of pulmonary nodules: the sensitivity for the detection of all nodules was 92.6% in patients with a BMI of 25, compared to 84.9% in patients with a BMI of 35, which differs with our results.

To summarize, previous studies seldomly included obese patients (BMI > 30 kg·m<sup>−</sup>2) and always reported a significant deterioration in image quality for high BMI. Only nodules were analyzed, and conclusions were in favor of a negative correlation between diagnostic performance and BMI. Our work, which includes a significantly higher number of obese patients (35 patients) nuance these observations by confirming a negative correlation between subjective image quality and BMI, yet without negatively impacting the diagnostic performance for the detection of nodules and seven other parenchymal lesions.

There was no statistically significant correlation between the MTCD and the subjective image quality, confirming BMI as the only metric that can predict the image quality expected from a ULD CT examination. This is all the more useful since BMI is easily obtained before the acquisition. BMI and MTCD had an unexpected—even though not significant tendency to lower the image noise. This positive influence could be explained by the high capacity of iterative reconstruction methods to significantly reduce noise. Similar results can be extracted from the literature [24,25], where images reconstructed with ASIR were associated with a greater noise reduction in high-weight patients.

The diagnostic performance of the ULD-CT was lower for emphysema and ground glass nodule compared to solid nodule and alveolar consolidation, which is in line with previous publications that have shown excellent diagnostic performance for spontaneously high contrast pulmonary lesions and lower performance for less attenuating lesions [11]. Unlike in the studies by Lee [11] and Macri [16], we chose to compare the diagnostic performance of the ULD CT to a reference "full dose" CT, which is the current Gold standard.

Several limitations of this work must be mentioned.

First, its retrospective nature—even if it is based on three prospective studies—and the relatively small number of severely obese patients (BMI > 35 kg·m−2)—even if this number is higher than in all the other publications on the subject.

Second, the population included in our study was mainly male, this is probably due to an inclusion bias for the 55 patients from the study that analyzed asbestos-related diseases and included only men. Nevertheless, the statistical analysis did not reveal any significant correlation by gender. Indeed, with breast attenuation being more significant in women, one could have imagined that at an equal BMI, the female sex could negatively influence the subjective image quality of the ULD CT.

Third, one third of the population was acquired in the prone position, which might have an impact on image quality.

Fourth, as a limitation inherent to all in vivo studies, reference was based only on the consensual reading of the FD CT by two readers, with a lack of any pathological confirmation. In the evaluation of subtle parenchymal lesions such as ground glass opacities, the analysis of multiple reviewers with different levels of expertise might have been more appropriate, as it could have led to a stronger consensus.

Finally, one should mention the monovendor nature of this study, carried out with a unique ULD acquisition protocol with a fixed tube current and a single iterative reconstruction algorithm. The prior publications listed used different iterative reconstruction software, and although our results are comparable to those studies, they cannot be strictly generalized to other constructors and other reconstruction techniques. Of note, the recently introduced Deep Learning Reconstruction might well further enhance the robustness of low dose acquisitions to high BMIs [26].

To conclude, our study suggests that despite a significant negative correlation of the BMI with the image quality of a ULD chest CT, its diagnostic performance is maintained. Consequently, the use of a ULD protocol when an unenhanced chest CT is needed could remain relevant in patients with a BMI greater than 30 kg·m<sup>−</sup>2.

**Author Contributions:** Conceptualization, M.O.; methodology, F.S.; validation, M.O. and C.R.; formal analysis, F.S. and A.-C.O.; investigation, A.-C.O., M.O. and A.L.; resources, A.L., C.R. and M.-Y.J.; data curation, A.-C.O. and M.O.; writing—original draft preparation, A.-C.O.; writing review and editing, M.O.; visualization, A.-C.O., A.L., F.S., M.-Y.J., C.R. and M.O.; supervision, C.R. and M.O.; project administration, C.R. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Institutional Review Board Statement:** The renewal of consent was not required for this retrospective study by our hospital institutional review board.

**Informed Consent Statement:** Written informed consent and approval from the local Ethics Committee had been obtained in the first instance for all patients, at the time of their initial prospective inclusion in one of the three trials.

**Conflicts of Interest:** The authors declare no conflict of interest.

### **Abbreviations**


#### **References**


### *Article* **Diagnostic Performance of Extracellular Volume Quantified by Dual-Layer Dual-Energy CT for Detection of Acute Myocarditis**

**Salim Aymeric Si-Mohamed 1,2,\*, Lauria Marie Restier 3,†, Arthur Branchu 2,†, Sara Boccalini 1,2, Anaelle Congi 3, Arthur Ziegler 2, Danka Tomasevic 4, Thomas Bochaton 4, Loic Boussel 1,2 and Philippe Charles Douek 1,2**


**Abstract:** Background: Myocardial extracellular volume (ECV) is a marker of the myocarditis inflammation burden and can be used for acute myocarditis diagnosis. Dual-energy computed tomography (DECT) enables its quantification with high concordance with cardiac magnetic resonance (CMR). Purpose: To investigate the diagnostic performance of myocardial ECV quantified on a cardiac dual-layer DECT in a population of patients with suspected myocarditis, in comparison to CMR. Methods: 78 patients were included in this retrospective monocenter study, 60 were diagnosed with acute myocarditis and 18 patients were considered as a control population, based on the 2009 Lake and Louise criteria. All subjects underwent a cardiac DECT in acute phase consisted in an arterial phase followed by a late iodine enhancement phase at 10 min after injection (1.2 mL/kg, iodinated contrast agent). ECV was calculated using the hematocrit level measured the day of DECT examinations. Non-parametric analyses have been used to test the differences between groups and the correlations between the variables. A ROC curve has been used to identify the optimal ECV cut-off discriminating value allowing the detection of acute myocarditis cases. A *p* value < 0.05 has been considered as significant. Results: The mean ECV was significantly higher (*p* < 0.001) for the myocarditis group compared to the control (34.18 ± 0.43 vs. 30.04 ± 0.53%). A cut-off value of ECV = 31.60% (ROC AUC = 0.835, *p* < 0.001) allows to discriminate the myocarditis with a sensitivity of 80% and a specificity of 78% (positive predictive value = 92.3%, negative predictive value = 53.8% and accuracy = 79.5%). Conclusion: Myocardial ECV enabled by DECT allows to diagnose the acute myocarditis with a cut-off at 31.60% for a sensitivity of 80% and specificity of 78%.

**Keywords:** dual-energy CT; iodine; diagnostic imaging; myocarditis; extra-cellular volume

### **1. Introduction**

Affecting approximately 1.8 million people worldwide in 2017 [1], myocarditis is a frequent inflammatory disease of the heart that can be caused by infectious agents, exposure to toxic substances and immune system activation [2,3]. The diagnosis includes clinical, laboratory, imaging, and histological parameters [1]; so over the years, different diagnostic tests have been developed to identify patients that have acute myocarditis. The endomyocardial biopsy has been the gold standard for a while until cardiac magnetic resonance imaging (CMR) and computed tomography (CT) got considered as non-invasive alternatives [4,5]. CMR emerged as a powerful non-invasive method for tissue characterization, including recognition and quantification of inflammation and replacement fibrosis

L.M.; Branchu, A.; Boccalini, S.; Congi, A.; Ziegler, A.; Tomasevic, D.; Bochaton, T.; Boussel, L.; Douek, P.C. Diagnostic Performance of Extracellular Volume Quantified by Dual-Layer Dual-Energy CT for Detection of Acute Myocarditis. *J. Clin. Med.* **2021**, *10*, 3286. https:// doi.org/10.3390/jcm10153286

**Citation:** Si-Mohamed, S.A.; Restier,

Academic Editor: Mickaël Ohana

Received: 30 June 2021 Accepted: 19 July 2021 Published: 26 July 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

in the setting of acute myocarditis [6,7]. Thanks to the Lake Louise Criteria, published in 2009, three markers of myocardial inflammation have been identified: hyperemia, tissue edema and necrosis/fibrosis [7]. Among the markers of the inflammatory burden, one of them stands out from the crowd: the extracellular volume (ECV).

Myocardial ECV increases related to myocardial fibroses, cardiac amyloid or edema [8]. CMR is the reference to measure ECV, but some previous studies showed that ECV can also be successfully determined with computed tomography, whether with single-energy computed tomography (SECT) or with dual-energy computed tomography (DECT), with a high correlation between ECV measurements derived from CT, histologic quantification, and CMR [9–14]. Hence, CT can effectively be considered as an interesting alternative to CMR. In addition, CT is more accessible and cheaper, while having a faster acquisition with a better spatial resolution, which makes it a good candidate for cardiac emergency imaging [15].

Therefore, we investigated the diagnostic performance of myocardial ECV quantified by cardiac DECT in a population of patients with suspected myocarditis, in comparison to CMR.

### **2. Materials and Methods**

### *2.1. Study Design*

This study is a monocenter retrospective study which has been conducted in the Cardiologic Hospital Louis Pradel, in Lyon (FRANCE) from May 2018 to May 2021.

### *2.2. Population*

The population was constituted of two groups of patients that were addressed for suspicion of acute myocarditis and underwent a cardiac dual-energy computed tomography and a CMR at the acute phase: patients with confirmed acute myocarditis (MG = myocarditis group) and patients without myocarditis patterns (CG = control group). The diagnosis was confirmed according to the Lake and Louise criteria on CMR. In order to validate the diagnosis of myocarditis, the patient had to present at least two of the Lake Louise Criteria (i.e., hyperemia, tissue edema and/or fibrosis/necrosis) [7]. Exclusion criteria for the myocarditis group in order to be comparable to the control group were: an underlying cardiomyopathy and/or left ventricular ejection fraction (LVEF) ≤40% and/or acute cardiac complications (heart failure, life-threatening arrythmias, death). Exclusion criteria for the control group were: an underlying cardiomyopathy and/or LVEF <50% according to the definition of heart failure [16].

### *2.3. Data Registration*

The following clinical, biological, functional and imaging parameters were recorded at admission: (1) age, sex, weight, size, systolic and diastolic blood pressure (SBP, DBP), heart rate, (2) values of high-sensitivity troponin (Tn), brain natriuretic peptide (BNP), creatinine; (3) left ventricular ejection fraction (LVEF) on trans-thoracic ultrasound at admission.

### *2.4. DECT Imaging Protocol*

### 2.4.1. Injection Protocol

The contrast material used was Iomeprol (400 mg/mL, Iomeron®; Bracco, Milan, Italy). All patients underwent a standard coronary CT angiography injection protocol consisting on a bolus injected at 3.5 mL/s into an 18 G catheter. The injection material was followed up by a saline rinse of 20 mL. The bolus volume was calculated according to the weight of the patient (1.2 mL/kg of contrast material).

### 2.4.2. Image Acquisition

The examinations have been performed on a dual-layer dual-energy CT (iQon; Phillips, Haifa, Israel) consisting of a first pass arterial phase and late phase 10 min after injection. Acquisition parameters were a retrospective gated ECG acquisition at 120 kVp, with a

cardiac care mas dose modulation (full dose 78%, half dose 40%), a pitch value at 0.20, and a rotation time at 0.27 s. Further technical details are provided in previous studies [17–19].

### 2.4.3. Reconstruction Protocol

Conventional and iodine density images were reconstructed from the late cardiac acquisition with a 1.5 mm slice thickness, a standard filter (Filter B) and a large field-of-view at 500 mm.

### 2.4.4. Image Analysis

Images were analyzed using a clinical workstation (Spectral Phillips Intellispace Portal Station, Phillips; Haifa, Israel). The myocardium was analyzed and manually segmented in 16 AHA segments using this software, on the iodine density images. We extracted the iodine concentrations in mg/mL for each 16 AHA segments. A circular region-of-interest of ~530 mm2 was drawn in the left cardiac cavity for measuring the iodine concentration in blood. The extracellular volume (ECV) was then calculated such as following:

$$\text{ECV} = 100 \times (1 - \text{Ht}) \times \frac{(\text{Iodine concentration in the myocardium})}{(\text{Iodine concentration in the blood})} \tag{1}$$

The iodine concentration was measured on iodine images in mg/mL; "Ht" is the hematocrit measured the day of the DECT acquisition. From these measurements, we analyzed the global myocardial ECV-per-patient.

### *2.5. Radiation Dose*

Dose-length-product and volume CT dose index for the late cardiac acquisition were recorded. Mean effective dose was calculated multiplying the dose-length-product by the chest k-factor of 0.014 [20].

### *2.6. Statistical Analyses*

Statistical analyses were performed with the SPSS software (IBM SPSS Statistics 19; 2018). The data are expressed as mean ± MSE (mean standard error) and median (minimum-maximum), accordingly to the normality tests.

For comparison of continuous variables between the two study groups, a non-parametric Mann–Whitney test has been used. A non-parametric Kolmogorov–Smirnov test has been used for the nominal variables.

For correlation purposes, Spearman correlation coefficients using a 95% confidence interval were calculated between biological (BNP, creatinine), functional markers (LVEF) and the global myocardial ECV tested.

A ROC curve has been used to identify the best discriminative cut-off value of ECV and to calculate the sensitivity, the specificity, the positive predictive value, the negative predictive value and the accuracy for the diagnosis of myocarditis, after ranking all the values and linking each value to the diagnosis of myocarditis.

### **3. Results**

### *3.1. Characteristics of the Population*

Initially, 107 patients were addressed to the hospital for suspicion of acute myocarditis. A total 73 patients have been diagnosed with acute myocarditis while 34 patients were rules out from myocarditis on CMR. Among them, 13 patients were excluded: 2 patients had a left ventricular ejection fraction ≤ 40%, 10 patients had life-threatening arrhythmia, and one patient was missing data. Finally, 60 patients have been included in the acute myocarditis group (Figure 1). Concerning the control group (CG), only 18 subjects have been included. Among the 16 patients who have not be retained for the CG, 2 of them had a left ventricular ejection fraction < 55%, 13 had cardiomyopathies and one patient had missing data (Figure 1). The baseline characteristics of the study population are summarized in Table 1.

**Figure 1.** Flow chart of the study population.



MG = myocarditis group; CG = control group; MSE = mean standard error; BMI = body mass index; LVEF = left ventricular ejection fraction; BNP = brain natriuretic peptide; *p* \* = significant differences (Mann–Whitney non-parametric test).

> The patients of the two groups have been paired by age and sex. The mean age was 32.9 ± 1.4 years for the (MG) and respectively 35.1 ± 3.6 years for the (CG). No significant difference for mean age or male proportion has been found between the two study groups. When compared, the troponins measured level was significantly higher (*p* < 0.001; 8630.3 ± 1585.9 vs. 822.5 ± 339.9 ng/L), and the LVEF was significantly lower (*p* = 0.006; 57.6 ± 1.0 vs. 64.2 ± 1.7%) for the acute myocarditis group compared to the control group.

### *3.2. Measurement of the Myocardial Inflammation*

Mean ECV of the myocarditis group was significantly higher compared to the one measured for the control group (*p* < 0.001, CI 95% (28.9–31.2)). Results are summarized in Table 2 and in Figure 2.


**Table 2.** Statistics of the ECV for both groups.

ECV = extracellular volume; MG = myocarditis group; CG = control group. Data are expressed in percentage. \* Significant differences MG vs.CG (Mann–Whitney non-parametric test): *p* < 0.001.

**Figure 2.** Box plots of the mean ECV for both groups. Median is represented by the line and the mean by the cross in the center of the box, upper and lower margins correspond to the 25th and the 75th percentiles, and outliers indicate the minimal and the maximal values. ECV = extracellular volume; CG = control group; MG = myocarditis group.

### *3.3. Correlation between ECV and the Different Parameters*

Concerning the acute myocarditis group, a positive significant correlation (*p* = 0.011) has been found between ECV and the troponins level (Pearson coefficient = 0.325). No significant correlation between ECV and LVEF has been found for this group.

On the other hand, significant correlations (respectively *p* = 0.015 and *p* = 0.048) have been found between ECV and the weight (Pearson coefficient = −0.563) and the BMI (Pearson coefficient = −0.472). No significant correlation has been found between ECV and the troponins level for this group. All the results are summarized in Table 3.


**Table 3.** Spearman correlation statistics between mean ECV of each group and other parameters.

ECV = extracellular volume; MG = myocarditis group; CG = control group; *n* = number of observations; *p* = significant difference coefficient; BMI = body mass index; LVEF = left ventricular ejection fraction; BNP = brain natriuretic peptide; \* = significant correlations.

### *3.4. Measurement of the ECV Cut-Off Value*

A ROC curve has been realized to estimate the best ECV cut-off value to finally determine which value can be used to discriminate acute myocarditis in a group of patients with suspected myocarditis. This curve is significantly representative of the ECV cuts-off values (*p* < 0.001) with an area under the curve of 0.835 (with a 95% CI of (0.748–0.922)). The results are represented in the Figures 3 and 4. The retained cut-off value of 31.60% in our study has shown a sensitivity of 80%, a specificity of 78%, a positive predictive value of 92.3%, a negative predictive value of 53.8% and accuracy of 79.5% for the discrimination of acute myocarditis from the control subjects.

**Figure 3.** ROC curve representing the different cuts-off of ECV.

**Figure 4.** Representative cases of patients with suspected acute myocarditis. Left panel (**A**–**D**). Case of a 42-year-old woman with elevated high-sensitivity troponins at 83 ng/L at admission. A late phase DECT was performed at admission. The global ECV was measured at 29.7%. MRI did not show any late gadolinium enhancement. Right panel (**E**–**H**). Case of a 27-year-old man with elevated high-sensitivity troponins at 8000 ng/L at admission. A late phase DECT was performed at admission. The global ECV was measured at 35.1%. MRI showed multiple late gadolinium enhancement of the sub-epicardial myocardial wall in favor of myocarditis.

### *3.5. Radiation Dose Analysis*

The mean ± SD volume CT dose index was 6.2 ± 1.7 mGy, the total dose length product was 123.2 ± 39.6 mGy.cm−1. As a result, the mean ± SD equivalent dose was calculated to be 1.7 ± 0.5 mSv.

#### **4. Discussion**

In the present study, we showed that myocardial ECV quantified by DECT is a biomarker of the myocarditis burden and that can be used for discriminating acute myocarditis in a population of patients with suspected myocarditis. These findings support the diagnostic value of ECV in the diagnostic work-up of a suspected acute myocarditis.

The novelty of the present study is the use of CT in a suspected population of myocarditis and as so the report of ECV values which can be the starting point for its implementation in clinical routine. In the same line, few studies have reported the ECV values of different cardiopathies such as heart failure, global cardiomyopathies, cardiac amyloidosis or in aortic stenosis patients [14,21,22]. Taken together, these studies are holding great promises for cardiac CT imaging because of its many advantages. Cardiac CT allows a 3D registration along the heart muscle in a short time acquisition with an excellent spatial resolution, and direct measurement of ECV in opposition with CMR which relies on measuring the effect of GBCAs on protons [8]. Because of its poor availability and its numerous contraindications, CT seems to be an encouraging and interesting alternative to CMR despite its irradiation [23]. However, among the different CT systems, it has to be noted the great advantage for DECT technology that allows the measurement of iodine content in a tissue without requiring a pre-injection examination such as done with single-energy CT, which reduces the burden radiation dose [22,24–26]. By decomposing the X-ray spectrum in two different energies spectra, DECT systems measure the photoelectric and Compton effects. Their recombination will allow to reconstruct quantitative images of the iodine distribution in the myocardium for ECV measurement [24]. Hence, DECT imaging is more prone to cardiac tissue characterization than single-energy CT which opens the door to evaluation of the myocardial ECV.

Our results demonstrated a significant increase of ECV in acute myocarditis patients, in accordance with previous studies using CMR that underlined an elevation of this biomarker in different cardiomyopathies, including myocarditis [15,27]. We observed

significant correlation between ECV and both troponins and BNP, which is coherent with the physiopathology of myocarditis. This is explained by the fact that troponins are a marker of tissue inflammation damages and BNP of a myocardial stress [7]. While for the control group, which presented a low elevation of troponins—no correlation was found. One hypothesis would be the non-flawless performances of current DECT systems for quantification of low iodine concentrations, which are reflecting low ECVs [17]. Finally, we observed a significant association between ECV and presence of myocarditis with an AUC of 83.5% (*p* < 0.0001). We have determined a cut-off value of ECV of 31.60% with high sensitivity of 80% and specificity of 78% for discriminating myocarditis. These performances are similar to different CMR studies [28–32] with pooled performances reported in a meta-analysis with a sensitivity of 76%, a specificity of 76%, a PPV of 72%, a NPV of 79% [33]. In addition, Nadjiri et al. have proposed an optimal ECV cut-off of 32.4% with a sensitivity of 93% and specificity of 74%, using CMR [28]. Altogether, the ECV values reported in the present study are in line with the data available with CMR. This is not a surprising finding, considering the high concordance between these modalities for ECV quantification as demonstrated by recent comparative studies which permits the use of this biomarker for multiple prospects [8,11,14,21]. Hence, we showed recently that ECV enabled by DECT allows a prediction of cardiac complications in acute myocarditis [34]. In this recent study, the cut-off suggested was of 39.5% which is highlighting a higher myocarditis inflammation burden that is consequently at risk. Finally, the present study is bringing one more contribution to the DECT for being a quick appropriate alternative candidate to CMR in cardiac emergency facilities [23,35].

The present study has limitations. The main limitation relies on the unperfect sensitivity of CMR using the 2009 CMR Lake Louise criteria which do not take into account ECV [7]. This limitation is mainly explained by the retrospective design of the study that started before the revised criteria in 2018 [6]. This bias probably increases ECV in the control group via false negative patients. In consequence differences between groups is probably underestimated and the bias non-differential. The absence of a fair comparison of ECV with CMR is also a limitation that points out the availability issue of CMR in acute settings reflecting a real-world practice. Yet a recent study has demonstrated similar ECV between the DECT system used in the present study and CMR, reinforcing our confidence in ECV derived from DECT scans [36]. Finally, the incremental diagnostic value of ECV by DECT to the late iodine enhancement presence has not yet been evaluated and should be performed in further studies.

As a conclusion, the evaluation of the myocardial ECV quantified by DECT in a population suspected of acute myocarditis demonstrated good diagnostic performances which allows us to consider ECV as a reliable DECT biomarker for the discrimination of myocarditis.

**Author Contributions:** Conceptualization, S.A.S.-M.; methodology, S.A.S.-M.; software, L.B.; validation, P.C.D., T.B.; formal analysis, A.B., A.C., S.A.S.-M., L.M.R., A.Z.; investigation, S.B., S.A.S.-M., P.C.D., L.B., D.T., A.Z.; resources, P.C.D.; data curation, P.C.D., S.A.S.-M.; writing—original draft preparation, S.A.S.-M., L.M.R., A.C., A.B.; writing—review and editing, S.A.S.-M.; supervision, S.A.S.-M., P.C.D., L.B. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Institutional Review Board Statement:** Ethical review and approval were waived for this study, due to the retrospective design of the study.

**Informed Consent Statement:** Written inform consent was waived by the local IRB (Hospices Civiles de Lyon) due to the retrospective design.

**Conflicts of Interest:** The authors declare no conflict of interest.

### **References**


MDPI St. Alban-Anlage 66 4052 Basel Switzerland Tel. +41 61 683 77 34 Fax +41 61 302 89 18 www.mdpi.com

*Journal of Clinical Medicine* Editorial Office E-mail: jcm@mdpi.com www.mdpi.com/journal/jcm

MDPI St. Alban-Anlage 66 4052 Basel Switzerland

Tel: +41 61 683 77 34 Fax: +41 61 302 89 18

www.mdpi.com

ISBN 978-3-0365-3372-8