**An Automatic Unmixing Approach to Detect Tissue Chromophores from Multispectral Photoacoustic Imaging**

#### **Valeria Grasso 1,2, Joost Holthof <sup>1</sup> and Jithin Jose 1,\***


Received: 29 April 2020; Accepted: 4 June 2020; Published: 6 June 2020

**Abstract:** Multispectral photoacoustic imaging has been widely explored as an emerging tool to visualize and quantify tissue chromophores noninvasively. This modality can capture the spectral absorption signature of prominent tissue chromophores, such as oxygenated, deoxygenated hemoglobin, and other biomarkers in the tissue by using spectral unmixing methods. Currently, most of the reported image processing algorithms use standard unmixing procedures, which include user interaction in the form of providing the expected spectral signatures. For translational research with patients, these types of supervised spectral unmixing can be challenging, as the spectral signature of the tissues can differ with respect to the disease condition. Imaging exogenous contrast agents and accessing their biodistribution can also be problematic, as some of the contrast agents are susceptible to change in spectral properties after the tissue interaction. In this work, we investigated the feasibility of an unsupervised spectral unmixing algorithm to detect and extract the tissue chromophores without any a-priori knowledge and user interaction. The algorithm has been optimized for multispectral photoacoustic imaging in the spectral range of 680–900 nm. The performance of the algorithm has been tested on simulated data, tissue-mimicking phantom, and also on the detection of exogenous contrast agents after the intravenous injection in mice. Our finding shows that the proposed automatic, unsupervised spectral unmixing method has great potential to extract and quantify the tissue chromophores, and this can be used in any wavelength range of the multispectral photoacoustic images.

**Keywords:**photoacoustic; optoacoustic; spectralimaging; blind source separation; unsupervised unmixing

#### **1. Introduction**

The accurate detection and quantification of tissue chromophores is vital in molecular imaging, as it can facilitate the early detection, prediction, and monitor the disease conditions. In recent years, multispectral photoacoustic imaging has emerged as a noninvasive tool to visualize the tissue chromophores [1–4]. The underlying principle of photoacoustic (PA) imaging is based on the conversion of absorbed nanosecond laser pulses into acoustic waves that can be detected just as conventional ultrasound [5,6]. Based on this approach, PA images combine the peculiar optical absorption contrast of the tissue chromophores and the spatial resolution of ultrasound imaging (US). Being a hybrid imaging modality of ultrasound and optical, this multimodal imaging technology can provide anatomical, functional, and molecular information several centimeters deep in the tissues with a resolution up to tens of micrometers. The potential of PA imaging has been demonstrated in various preclinical applications, such as tumor progression and the prediction of tumor recurrence, therapy monitoring, imaging of vasculature, and the biodistribution of the contrast agents [7–13].

Apart from the preclinical applications, PA imaging is also used in clinical research. In addition to breast cancer monitoring [14,15] and sentinel lymph node imaging [16,17], the PA approach is also used to examine inflammatory bowel disease (IBD) [18] and the temporal arteries in patients with suspected giant cell arteritis (GCA) [19]. To expedite the clinical applications of PA imaging, recently, there has been a lot of focus on developing affordable light sources and the use of this technology in low-resource settings. Xia et al. demonstrated the feasibility of a Light-Emitting Diode (LED) based PA imaging system for the visualization of superficial vasculatures and needle guidance for minimally invasive procedures [20]. Zhu et al. used the LED-based approach to explore more clinical applications, such as diagnosing inflammatory arthritis and assessing peripheral microvascular function in patients [21–23].

Although there has been a lot of emphasis on PA hardware development, in the field of affordable settings, data analysis and reconstruction algorithms also play a crucial role in increasing the utility of the technology. Multiwavelength acquisition and spectral image processing is one of the commonly used techniques in PA. Since the optical absorption coefficient of the tissue chromophores varies over the spectrum, multispectral image processing approach can be applied to distinguish and characterize the molecules present in the tissues [24].

In general, the pixel intensity of the multispectral photoacoustic image is proportional to the absorption value of the respective tissue at a specific wavelength. In reality, due to the finite dimension of the pixel (partial volume effect) and the presence of instrumental noise, each spectrum can be a combination of different tissue chromophores. Therefore, it is a challenging task to unmix these signals spectrally and estimate their concentrations. The most common solution to detect the tissue chromophores from multispectral PA imaging is the supervised unmixing [25]. Although this technique yields acceptable results, it requires user interaction to provide the expected source spectral curves as an input to unmix the signals. For translational research with patients, these types of supervised spectral unmixing can be challenging, as the spectral signature of the tissues differs with respect to the disease condition. Imaging exogenous contrast agents and accessing their biodistribution can also be problematic, as some of the contrast agents are susceptible to change in spectral properties after the tissue interaction; thus, the algorithm can forfeit the sensitivity and specificity of imaging.

Hence, an unsupervised unmixing algorithm that can automatically detect the tissue chromophores, without any a-priori knowledge and user interaction, will be optimal, as this can facilitate and improve sensitivity and specificity. Generally, this class of algorithms is referred to as blind source separation (BSS) algorithms, as no a-priori information is required. The study reported by Glatz et al. [26] demonstrated the potential of these approaches to "blindly" extract the oxygenated and deoxygenated hemoglobin absorption spectra from the multispectral photoacoustic images. In the study, they evaluated different unsupervised algorithms, such as multivariate curve resolution analysis (MCR), principal component analysis (PCA), and independent component analysis (ICA) [27–31]. PCA yields an orthogonal transformation that decorrelates the variables. This approach relies upon the hypothesis that the source components are uncorrelated. On the other hand, ICA is based on a different assumption that the source components are maximally independent and non-Gaussian. Recently, Arabul et al. [32] used a similar approach to explore human carotid plaques, in which the ICA blind unmixing approach was constrained non-negatively. Although these approaches demonstrated the potential to detect the tissue chromophores, they suffer from limitations related to the interpretability of the mixed-sign values of their outcomes. Indeed, these aim to fit the training data well but often do not generalize the real and positive data sets.

In this paper, we investigated the possibility of using another blind source separation approach, which is based on non-negative matrix factorization (NNMF) [33]. The concept of NNMF has been widely used in a variety of applications, such as image recognition [34], text classification [35,36], and recommender systems [37]. This approach uses only the non-negative matrices to estimate the prominent components and their spatial distribution, from a linear mixture model. Montcuquet et al. [38] used this approach for in-vivo fluorescent imaging and demonstrated that the positivity condition enhances the convergence and, thus, improves the sensitivity of spectral unmixing [39].

Here, we examine the performance of non-negative matrix factorization (NNMF) to unmix tissue chromophores from multispectral PA images. The algorithm has been optimized to extract the tissue chromophores in the wavelength range of 680–900 nm. We tested the NNMF on synthetic data and on experimental data that mimic the blood vessels. Further, we validated the potential of the approach on an in-vivo study to detect and quantify the endogenous absorbers and exogenous contrast agent accumulation. To our knowledge, this is the first time the NNMF algorithm has been used for PA imaging.

#### **2. Non-Negative Matrix Factorization (NNMF)**

NNMF is a data decomposition approach, and it is based on the linear mixing model [40]. In this algorithm, the acquired mixed pixel spectra are differentiated into a collection of constituent spectra (called endmembers) and a set of fractional abundance maps. The endmembers represent the pure molecule absorption spectra present in the imaged sample, and these are extracted from the mixed pixel spectra. The maps of abundance at each pixel represent the percentage of each endmember present in that pixel.

Since the acquired spectral images are known, and the rest has to be estimated, the mixed data (multispectral PA images) can be arranged as a matrix *<sup>X</sup>* <sup>∈</sup> <sup>R</sup>*n*×*m*, where *<sup>n</sup>* represents the number of observations (pixels), and *m* corresponds to the number of variables per object (different wavelengths). In particular, the unmixing problem can be formulated as a matrix factorization:

$$X \approx WS \tag{1}$$

where *X* represents the mixed multispectral PA images, *W* the abundance maps, and *S* the source spectra. The dimensions of the matrices *W* and *S* are *n* × *k* and *k* × *m*, respectively, where *k* is the hyperparameter which represents the number of prominent components.

The NNMF constrained cost function of the optimization problem can be formulated as follows:

$$\mathbb{E}\left[\mathcal{W},\mathcal{S}\right] = \min\_{\mathcal{W},\mathcal{S}} \frac{1}{2} \parallel X - \mathcal{W}S \parallel\_{\mathcal{F}}^2 \tag{2}$$

$$\mathcal{W}\_{ij} \ge 0, \quad \mathcal{S}\_{ij} \ge 0 \tag{3}$$

where, in Equation (2), the defined cost function considers a Frobenius distance between the acquisition *X* and the model *WS* [33]. *W* and *S* are iteratively obtained until both matrices satisfy Equation (1), where the distance defined in the cost function (2) is constrained non-negatively (3). The NNMF learns a parts-based representation of the data, and the whole image is formed as a combination of additive components. The non-negativity constraint is computationally expensive to implement but it can lead to more interpretable data.

To solve the iterative optimization, the multiplicative update rules [41] can be used, and the steps can be defined as:

$$S\left(p+1\right) = S(p) \otimes \frac{\left(W^T \cdot X\right)}{\left(W^T W \cdot S\right)}\tag{4}$$

$$\mathcal{W}(p+1) = \mathcal{W}(p) \otimes \frac{\begin{pmatrix} \mathbf{X} \cdot \mathbf{S}^T \end{pmatrix}}{\left(\mathbf{W} \cdot \mathbf{S} \cdot \mathbf{S}^T\right)}\tag{5}$$

where *p* is the iteration step, and the operations of ⊗ and division in (4) and (5) are considered element by element. Each component is estimated only up to a multiplying scale factor. Hence, the factorization problem does not have a unique solution, where *WS* is a lower-rank approximation of *X.*

#### **3. Experimental Methods**

#### *3.1. Simulated Multispectral PA Images*

To evaluate the performance of the NNMF algorithm, simulation studies were performed by using a synthetic data set. The synthetic data set contains implemented photoacoustic spectral images within the wavelength range of 680–900 nm, with a step size of 5 nm. Each image at the respective wavelength contains 400 × 600 pixels. From the photoacoustic signal generation, it is evident that the PA signal is not only proportional to the absorption coefficient but also depends on the local fluence. The light fluence generally decreases with depth, and thus degrades the image uniformity, causing spatial fluence variations within the tissue. Consequently, the fluence compensation is significant for quantitative spectral imaging. Since the main focus of the simulation was to test the unmixing algorithm and its accuracy to detect the spectral signature of the prominent components, in this study, the local fluence was assumed to be constant.

Figure 1a depicts the schematic of the 2-dimensional (2-D) data set with six homogeneous inclusions. The inclusions mimic the cross section of blood vessels with different concentrations of oxyhemoglobin and deoxyhemoglobin. The inclusions 1, 2, and 3 contain oxyhemoglobin at 100%, 70%, and 30% of the intensity, respectively. Conversely, the inclusions 4, 5, and 6 include deoxyhemoglobin at 30%, 70%, and 100% of the intensity. Figure 1b shows the theoretical absorption spectra [42] expected from these inclusions, in addition to the background tissue absorption. To mimic the experimental conditions, a positive Gaussian distribution of noise (*mean* = 0.04; *std* = 0.1% ; *SNR* = 30 dB) was also added to the respective data set.

**Figure 1.** (**a**) Schematic representation of the simulated multispectral photoacoustic (PA) images with six inclusions, (**b**) ideal spectral curves of the inclusions.

#### *3.2. Experimental Set-Up and Tissue-Mimicking Vessel Phantom*

In addition to the simulation studies, experiments were performed by using Vevo LAZR-X photoacoustic image technology (FUJIFILM VisualSonics, Inc., Toronto, ON, Canada), as described elsewhere [43]. Vevo Phantom (FUJIFILM VisualSonics, Inc., Toronto, ON, Canada) containing two capillary tubes filled with Indocyanine Green (ICG, PULSION Verwaltungs, GmbH) and Methylene Blue (MB, Sigma-Aldrich), was used to mimic the blood vessels in the tissue. Transparent polyethylene (PE) tubes (SAI Infusion Technologies, Lake Villa, IL, USA), with an inner diameter of 15 μm and an outer diameter of 33 μm, were used. The tubes were positioned at a reciprocal distance of 6 mm and fixed at the same depth of 14 mm from the surface of the transducer. Demineralized water was used as a coupling medium, and the multispectral PA images were obtained in the wavelength range of 680–900 nm, with a step size of 5 nm. Figure 2a shows the schematic of the phantom and the experimental set-up. A 256-element linear array transducer with a central frequency of 21 MHz (MX250), including the integrated light delivery fibers from the sides of the transducer, was used

to acquire the PA images. The transducer was aligned perpendicular to the capillary tubes, and the cross-sectional image of the tubes was acquired throughout the wavelength range. 3-Dimensional (3-D) data sets were also collected by linearly translating the transducer with a stepper motor over the capillary tubes while capturing cross-sectional 2-D slices. Figure 2b shows the photoacoustic spectra measured from the capillary tubes by using the Spectro-Mode in the VevoLab software (FUJIFILM VisualSonics, Inc., Toronto, ON, Canada). In the measurement tool, the system allows the user to select the region of interest (ROI) and calculate the average intensity of the photoacoustic signal at different wavelengths. Although this is not quantitative and the values are in arbitrary units, it can provide the spectral absorption trend of the agents in the respective wavelength range.

**Figure 2.** (**a**) Schematic of the tissue-mimicking vessel phantom. (**b**) The PA absorbance spectral graph of the agents measured by using VevoLab.

#### *3.3. In-Vivo Study*

Further, in-vivo animal experiments were performed to evaluate the feasibility of the NNMF data analysis on multispectral PA Imaging. The animal experiments were performed at the FUJIFILM Sonosite/VisualSonics facility in Amsterdam. The animal protocols used in this work were evaluated and approved by the Animal Use and Ethics Committee (CEUA) of The Netherlands (Protocol AVD2450020173644). They are in accordance with FELASA guidelines and the National Law for Laboratory Animal Experimentation (Law No. 18.611). The experiments were performed by using the same apparatus (Vevo LAZR-X) used for the phantom studies. A CD-1 female mouse model (Envigo, Horst, the Netherlands) was used for the experiments. The animal was anesthetized with isoflurane and placed on the animal imaging platform of the Vevo LAZR-X system, where temperature, heart rate, and respiration rate were monitored in real time. During the experiments, anesthesia was maintained using a vaporized isoflurane (1 L/min of oxygen and 0.75% isoflurane) gas system. The animal was positioned in right lateral recumbency, and the transducer was aligned perpendicularly. The kidney–spleen area of the animal was imaged before and after the intravenous injection of ICG. To obtain a concentration of 800 μM, 25-mg vial of ICG (PULSION Verwaltungs, GmbH) was resuspended in sterile water. With the help of an infusion pump (flowrate of 15 μL/sec), 80 μL of ICG was injected into the tail vein, and multispectral PA images were acquired in the wavelength range of 680–900 nm.

#### **4. Results and Discussion**

#### *4.1. Simulated Multispectral PA Images*

Figure 3 shows the main component spectra (a) and the respective abundance maps (b, c, and d) extracted from the synthetic PA data set by using the NNMF algorithm. The obtained spectral graphs (Figure 3a) show that the prominent absorbers present within the inclusions consist of two different endmembers: oxyhemoglobin and deoxyhemoglobin. As expected from the synthetic data, the oxyhemoglobin is mainly distributed on the first row of inclusions, with decreasing intensity from left to right (Figure 3c). The deoxyhemoglobin is present in the second row of inclusions, with increasing intensity from left to right (Figure 3b). Figure 3d displays the spatial distribution of the detected third component (named as background in Figure 3a), which is principally present in the region around the inclusions.

**Figure 3.** (**a**) Source spectra of oxyhemoglobin, deoxyhemoglobin, and background extracted by the non-negative matrix factorization (NNMF) algorithm. Abundance maps of (**b**) deoxyhemoglobin, (**c**) oxyhemoglobin, and (**d**) background.

NNMF appears to be an accurate method, yielding an explicit unmixing of specific tissue biomarkers, such as oxyhemoglobin and deoxyhemoglobin, and extracting the respective spectral signatures. The calculated Pearson correlation coefficient between the ideal spectra of the oxy and deoxyhemoglobin, and the extracted spectra was equal to 1 (Table 1). This confirms that NNMF can provide encouraging results on extracting tissue chromophores from multispectral PA images.


**Table 1.** Correlation values between the extracted source components by using the NNMF unmixing approach, and the respective absorption spectral curves used as a reference.

Further, we investigated the performance of the NNMF approach to quantify the extracted prominent components. Figure 4a shows a graph with the quantitation of the source components (oxyhemoglobin, deoxyhemoglobin) per each circular inclusion of the synthetic data. Considering the synthetic data: inclusions 1, 2, and 3 present approximatively a decreasing content of oxyhemoglobin, and the regions 4, 5, and 6 present an increasing content of deoxyhemoglobin. In the graph, the variations are in the range of zero to one, and it is evident that these normalized intensities are matching the expected proportion of the components. Figure 4b shows the overlapped abundance maps of oxyhemoglobin and deoxyhemoglobin, and it also confirms the expected distribution of the prominent components.

**Figure 4.** (**a**) Quantitative evaluation of the prominent source components (oxyhemoglobin and deoxyhemoglobin), per each circular region of the synthetic phantom. (**b**) Overlapped abundance maps of oxyhemoglobin and deoxyhemoglobin.

#### *4.2. Tissue-Mimicking Vessel Phantom*

Figure 5 shows the prominent spectral curves (a) and the abundance maps of the dyes in the capillary tubes, obtained by using the NNMF algorithm. The abundance maps overlapped in Figure 5b correspond to the 2-D spatial distribution of the contrast agents (ICG in green and MB in blue). Figure 5c shows the unmixing of the capillary tubes in 3-D, where the total imaged range was 6.5 mm with a step size of 200 μm. The Pearson correlation coefficient was evaluated between the extracted spectral curves, by using the NNMF, and the spectra measured by using the Spectro-Mode in the VevoLab.

**Figure 5.** (**a**) Spectral absorption curves of the detected source components by NNMF, from 2-D spectral PA images of the tissue-mimicking vessel phantom. The overlapped abundance 2-D maps (**b**) and 3-D maps (**c**) of the detected source components: ICG and MB.

Table 1 reports the correlation coefficients measured for the ICG and methylene blue. The correlation value obtained for the methylene blue was 0.8344, and it was comparatively lower to the value obtained for the ICG, which was 0.9943. This could be due to the experimental conditions, as the tubes are located adjacently, into the phantom chamber. Hence, due to the short distance, the ICG that has an absorption peak at around 880 nm could influence the absorption spectrum of the MB. The graph reported in Figure 2b supports this assumption, as the MB spectrum shows an additional peak at around 880 nm. This may entail that measured spectral curves in Figure 2b are not the pure agent spectra. On the other hand, the spectra extracted by using NNMF are in accordance with the expected spectral signatures, and it shows promising unmixing performance.

#### *4.3. In-Vivo Study*

Figure 6a shows a high-resolution ultrasound (US) image of the kidney–spleen region and (b) is the PA image obtained at 880 nm. Figure 6c shows the oxygen saturation (*SO*2) map, obtained before the contrast agent injection. To obtain the oxygen saturation map, we have followed the algorithm reported by Needles et al. [5], where the pixel values of the image are in the range of 0% (lower oxygenation) to 100% (higher oxygenation).

**Figure 6.** Pre-injection conditions: (**a**) ultrasound (US) image of the kidney–spleen view; respectively (**b**) photoacoustic (PA) image obtained at 880nm, and (**c**) *SO*<sup>2</sup> map.

Figure 7 shows the post-ICG injection condition. Figure 7a displays the source absorption spectra extracted by the NNMF approach. As expected, in the post-injection condition, the NNMF extracted the typical absorption spectral curve of the ICG, in addition to the endogenous chromophores such as oxy and deoxyhemoglobin spectra. The extracted spectrum of the ICG appears slightly different than the vessel mimicking phantom. Although the peak absorption was at 880 nm, the ICG spectrum at the lower wavelengths was altered. This could be due to the ICG interaction with other chromophores within the tissues. Figure 7 also shows the abundance maps of deoxyhemoglobin (b), oxyhemoglobin (c), and ICG (d). In the ICG map, it is evident that the dye is mostly accumulated in the spleen region. This is in accordance with the biodistribution of the ICG, as the kinetic of the kidney is much faster than the spleen [44].

**Figure 7.** Post-injection conditions: (**a**) spectral signature of the endmembers obtained by using NNMF and abundance distribution maps of (**b**) deoxyhemoglobin, (**c**) oxyhemoglobin, and (**d**) ICG.

#### **5. Conclusions**

In summary, we investigated the spectral decomposition of various tissue biomarkers from multispectral PA images. In particular, we have explored the performance of an unsupervised spectral unmixing algorithm, NNMF, in the wavelength range of 680–900 nm. Considering the evidence obtained from the initial results, the NNMF can extract both endogenous and exogenous agents from the multispectral PA data. The unmixing results obtained from the simulation studies performed on synthetic data revealed a high correlation with the expected spectra, and also yielded the quantification of the chromophores. The experiment performed on the tissue-mimicking phantom also supported the results obtained on the synthetic data. Indeed, the NNMF showed promising unmixing performance, allowing the accurate detection of ICG and MB, by eliminating the spectral influence of the other dyes. The in-vivo experiments in the animal model showed the detection of a contrast agent signature, accounting for the spectral variations that may ensue due to tissue–dye interactions. The NNMF can also provide the maps of abundance distribution of contrast agents in different anatomical targets, facilitating in-vivo biodistribution and kinetics studies. Since the algorithm facilitates the automatic unmixing of the tissue chromophores, without any a-priori knowledge about the source components and user interactions, it is easy to adapt, and promising for data-driven studies in multispectral photoacoustic imaging.

In the current algorithm, we have used the time gain compensation (TGC) approach to overcome the PA signal attenuation through depth. Although the approach is not quantitative, it gives improved results. Recent studies demonstrate that taking into account the fluence variance [45,46] can overcome the quantification limits of photoacoustic imaging. Therefore, in future studies, we will investigate the corruption effects of the fluence variation, the finite size, and band-limited frequency response of the detectors, to consider the respective changes on the absorption spectra. Besides, we will expand the current wavelength range of 680–900 nm to the far infrared (FIR), as this may entail the detection of less prominent tissue chromophores, such as melanin, lipids, and collagens.

In conclusion, to the best of our knowledge, this would be the first time that NNMF was used for unmixing multispectral PA imaging. The obtained results confirmed that the NNMF algorithm automatically and accurately detects the component spectra. This proves that the imposed positivity constraints, to the source spectra and abundance distribution maps, are appropriate requirements to unmix tissue chromophores from multispectral PA images.

**Author Contributions:** Conceptualization, V.G. and J.J.; Methodology, V.G., J.H. and J.J.; Software, V.G.; Validation, V.G. and J.J.; Investigation, J.J.; Resources, J.J.; Data curation, V.G., J.H. and J.J.; Writing—original draft preparation, V.G.; Writing—review and editing, V.G. and J.J.; Visualization, V.G. and J.J.; Supervision, J.J.; Project administration, J.J.; Funding acquisition, J.J. All authors have read and agreed to the published version of the manuscript.

**Acknowledgments:** This publication is part of a project that has received funding from the European Union's Horizon 2020 research and innovation program under the Marie Skłodowska-Curie grant, agreement No 811226.

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

#### **References**


© 2020 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 (http://creativecommons.org/licenses/by/4.0/).

## *Article* **Multiangle Long-Axis Lateral Illumination Photoacoustic Imaging Using Linear Array Transducer**

**João H. Uliana <sup>1</sup> , Diego R. T. Sampaio <sup>1</sup> , Guilherme S. P. Fernandes 1, María S. Brassesco <sup>2</sup> , Marcello H. Nogueira-Barbosa 3, Antonio A. O. Carneiro <sup>1</sup> and Theo Z. Pavan 1,\***


**\*** Correspondence: theozp@usp.br

Received: 19 June 2020; Accepted: 19 July 2020; Published: 21 July 2020

**Abstract:** Photoacoustic imaging (PAI) combines optical contrast with ultrasound spatial resolution and can be obtained up to a depth of a few centimeters. Hand-held PAI systems using linear array usually operate in reflection mode using a dark-field illumination scheme, where the optical fiber output is attached to both sides of the elevation plane (short-axis) of the transducer. More recently, bright-field strategies where the optical illumination is coaxial with acoustic detection have been proposed to overcome some limitations of the standard dark-field approach. In this paper, a novel multiangle long-axis lateral illumination is proposed. Monte Carlo simulations were conducted to evaluate light delivery for three different illumination schemes: bright-field, standard dark-field, and long-axis lateral illumination. Long-axis lateral illumination showed remarkable improvement in light delivery for targets with a width smaller than the transducer lateral dimension. A prototype was developed to experimentally demonstrate the feasibility of the proposed approach. In this device, the fiber bundle terminal ends are attached to both sides of the transducer's long-axis and the illumination angle of each fiber bundle can be independently controlled. The final PA image is obtained by the coherent sum of subframes acquired using different angles. The prototype was experimentally evaluated by taking images from a phantom, a mouse abdomen, forearm, and index finger of a volunteer. The system provided light delivery enhancement taking advantage of the geometry of the target, achieving sufficient signal-to-noise ratio at clinically relevant depths.

**Keywords:** photoacoustic imaging; illumination scheme; in vivo; mouse; Monte Carlo; linear array

#### **1. Introduction**

Photoacoustic imaging (PAI) is a technique based on the photoacoustic (PA) effect, which consists of pressure waves generation due to the absorption of light [1–5]. Currently, laser-based PAI systems use short-duration laser pulses (i.e., ~10−<sup>9</sup> s) ensuring thermal and stress confinement. As pulsed-light propagates within the target material, its absorption increases the local temperature, causing a thermal-elastic expansion [6] and generating a pressure wave. Thus, PAI encodes the optical absorption information into pressure waves, therefore combining optical contrast with ultrasound spatial resolution [4].

PAI can provide physiological and anatomical information of tissues by accessing their optical, thermal, and mechanical proprieties [5,7]. Since PA signal magnitude is temperature-dependent, PAI has been used, for example, to map temperature variation within tissues during hyperthermia procedures [8–11]. Moreover, PA magnitude is proportional to the optical absorption of a chromophore; therefore, multi-wavelength PAI is capable of identifying structures with different optical absorption profiles [12,13]. In this context, a typical application of multispectral PAI is to estimate blood oxygen saturation (sO2) from the relative concentrations of oxyhemoglobin (HbO2) and deoxyhemoglobin (Hb) [14–19]. In addition, exogenous contrast agents, for example, nanoparticles and organic dyes, can be accessed to obtain molecular PAI and for drug delivery studies [20]. PAI is frequently combined with clinical ultrasound arrays. This approach allows simultaneously displaying conventional ultrasound images of different modalities (e.g., B-mode and Doppler) with PAI. Different preclinical and clinical, e.g., breast cancer [21] and joint arthritis [22,23], applications of PAI integrated with ultrasound scanners are under investigation.

Hand-held PAI systems usually operate in reflection mode (also known as epi-mode), where illumination and PA wave detection are arranged on the same side [24–28]. For linear array transducers, it is common to illuminate the tissue using a rectangular optical fiber output, which is attached to both sides of the elevation plane (short-axis) of the transducer, see Figure 1a (here this strategy will be referred to as standard dark-field illumination, following the terminology used in [29–32]). For this dark-field illumination scheme, when the transducer face is in contact with the skin, light is delivered obliquely and relies on light scattering within the target to illuminate the whole field-of-view (FOV) of the transducer. In addition, light absorption outside the imaging plane generates PA waves that can reach the ultrasound transducer. These signals are a source of clutter, which is also an important limiting factor to obtain PA images at deeper regions [33–35]. Different studies have investigated, for this standard dark-field illumination scheme, light delivery optimization to enhance PA image contrast and signal-to-noise ratio (SNR) by varying the distance between the fiber output and the transducer and incidence angle between the light beam and the imaging plane [30,33,36–38]. In [39], the authors verified, through Monte Carlo simulations, that the optimal illumination configuration depends on the optical properties of the tissues under investigation. They observed that thickness and optical scattering of skin play a major role for this optimization. Another possible strategy to optimize light delivery is to accommodate an optically transparent spacer between the transducer and target's surface to deliver light directly to the tissue underneath the transducer [25,27,28,32]. Improvements in light delivery could also be achieved by using a concave-shaped light catcher that redirects the light reflected by the skin surface back to the tissue, improving the PA signal magnitude at higher depths [40,41]. The aforementioned studies evaluated laser-based PAI systems. More recently, pulsed light-emitting diodes (LED)-based PAI technique has been proposed as an interesting and cost-efficient option [42,43]. LED-based PAI with linear array usually operates using a similar setup as shown in Figure 1a [42,44]. The study [44] suggested that the high divergence of LED illumination decreases the source direction dependency on PA signal compared to laser.

Since optimal light delivery to the tissue is essential to increase image depth and SNR, custom transducers, new materials, and new strategies have been developed to explore different illumination geometries to improve the quality of the PA image. For example, an ultrasonic transducer fabricated on a glass substrate has an improved transparency allowing the laser beam to propagate through the transducer's material with low absorption, resulting in overlapped optical excitation and acoustic detection [45]. An ultrasound transducer with a hollow central bore [46] or an optically transparent acoustic transducer [47] could be other options to provide reflection mode illumination. However, these approaches require an extensive redesign of the ultrasound probe and cannot be easily integrated into standard clinical scanners. In epi-mode PAI using standard linear array transducers, optical and acoustic fields can be coaxially arranged (see Figure 1b) by redirecting the laser beam using an optical/acoustic coupler [48] or by using a single or double acoustic reflector to redirect the acoustic waves [29,31,49]. Another strategy for coaxial illumination consists of a custom linear array transducer where the optical fiber outputs and piezoelectric elements are linearly and alternately arranged [50]. These studies [29,31,48–50] showed that this illumination strategy improved light delivery when

compared with the standard dark-field approach. In the present paper, this strategy will be referred to as bright-field illumination, following the terminology used in [30–32,49].

An alternative illumination approach, not yet investigated in the literature, would consist of attaching the fiber bundle terminal ends to both sides of the transducer's long-axis (from now on we will refer to this technique as long-axis lateral illumination), see Figure 1c. In the present paper, we propose a long-axis lateral illumination scheme as a new epi-mode PAI strategy, where the light is delivered within the imaging plane similarly to the coaxial arrangement. In the first part of the paper, Monte Carlo simulations of photon propagation were used to compare light delivery for different illumination strategies; i.e., standard dark-field, long-axis lateral, and bright-field illumination. A transparent spacer positioned between the transducer and the tissue surface was considered for all cases. Tissues with three different geometries were simulated; i.e., targets larger and smaller than the lateral dimension of the imaging plane simulating the human forearm and index finger, respectively and an intermediate situation simulating the cross section of a mouse torso where the abdomen was smaller than the image width and the lower limbs fitted the transducer FOV. The simulations demonstrate that the lateral illumination strategy can provide remarkably improved fluence distribution for targets smaller than the imaging plane.

**Figure 1.** (**a**) Standard dark-field illumination scheme to acquire photoacoustic (PA) image in reflection mode; a rectangular optical fiber terminal illuminates the surface of the target. (**b**) Bright-field illumination where the laser beam and the acoustic field are coaxially aligned. (**c**) Proposed long-axis lateral illumination architecture; the variation of light incidence angle provides wide illumination to the surface of the target.

In the second part of this paper, the development of a simple and easy way to construct a device for long-axis lateral illumination PAI is described. This device employed a nonexpensive commercially available bifurcated optical fiber bundle for light delivery, where no other optical components were required. Since the setup, as shown in Figure 1c, would irradiate only a limited area within FOV, the optical fiber bundle outputs were mounted on movable sockets arranged parallel to the imaging plane to provide multiangle long-axis lateral illumination. The final PA image, covering the full scan area, is then obtained by combining the PA sub-images acquired at different angles. This is a similar strategy as described in [51], where a narrow laser beam scanning approach was proposed for a combined real-time PA-ultrasound imaging system. Then the final PA image was the summation of the sub-images obtained at each laser beam scanning position.

Therefore, this paper presents a novel PAI light delivery where light and sound are coaxially illuminated. Different from other approaches with similar capability [29,31,48,49], the proposed technique does not require an acoustic/optical coupling device. This is an advantage because these coupling modules usually induce important phase distortion to the PA wavefront which can reduce

image quality [32]. The light delivery device was prototyped to provide freedom to independently choose the illumination angle, at each side of the transducer, allowing multiangle illumination planning. To show the feasibility of the multiangle long-axis lateral illumination PAI device, images taken from phantom, mouse abdomen, forearm, and index finger of a volunteer were analyzed.

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

#### *2.1. Monte Carlo Simulation of Illumination Schemes*

The spatial energy deposition may vary depending on the illumination scheme and target shape. To evaluate the performance of the illumination schemes depicted in Figure 1, light transport was simulated using the MCXLAB Matlab toolbox, which is a 3D voxel-based Monte Carlo model [52], for three different target geometries: (i) a cylindrical target shape simulating a situation similar to what was observed for the human index finger (lateral dimension smaller than transducer's width); (ii) geometry similar to the human forearm (lateral dimension larger than transducer's width); (iii) mouse torso as an intermediate case, i.e., part of the target was smaller (mouse abdomen), while the lower limbs of the animal was larger than the transducer's width.

The standard dark-field illumination scheme shown in Figure 1a, based on the setup described in [53], was composed of two optical fiber terminals (38 mm × 1.25 mm) with the same width as the ultrasound linear array used in the experiments of the present paper. Each terminal was positioned so that the light beam incident angle was 20◦ and the light beams overlapped at the upper surface of the target. For all three illumination schemes, an optically transparent spacer of 19.5 mm (i.e., water) was positioned between the transducer and the target. For the bright-field illumination scheme shown in Figure 1b, the laser beam was coaxial with acoustic detection. In this case, the illumination dimension hitting the target was 38 mm × 5 mm, which is in accordance with [48]. For the long-axis lateral illumination, Figure 1c, the fiber optic bundle terminals were circular in shape with 5 mm diameter. To illuminate the entire transducer FOV, the same multiangle illumination strategy used for the experiments (see next sections for a detailed description) were adopted in the simulations. We verified that at least 5 laser beam incident angles were necessary to ensure a complete illumination. In this case, all simulation parameters were the same as the experimental setup. A total of (5.0 <sup>×</sup> 106) photons were used to simulate each situation.

The volume dimension for all simulations was 89 mm × 60 mm × 30 mm with a voxel size of 0.25 mm. The volume consisted of two different materials, the background (water) and an inclusion (target) to simulate the tissue. The optical proprieties of the background were: absorption coefficient μ*bkg <sup>a</sup>* <sup>=</sup> 3.5640 <sup>×</sup> <sup>10</sup>−<sup>5</sup> mm<sup>−</sup>1, scattering coefficient <sup>μ</sup>*bkg <sup>s</sup>* = 1.0 mm<sup>−</sup>1, *gbkg* = 1, and η*bkg* = 1.37; where *g* denotes the anisotropic factor and η denotes the refraction index. For the target, the optical scattering and optical absorption coefficients were chosen for a generic tissue, following the equations [54]:

$$
\mu'\_{\ s} = a(\lambda/500nm)^{-b},
\tag{1}
$$

$$
\mu\_{\text{a}} = B O \mu\_{a}^{HbO\_{2}} + B(1 - O)\mu\_{a}^{Hb} + \mathcal{W}\mu\_{a}^{water} + F\_{a}\mu\_{a}^{liquid} \tag{2}
$$

where μ *<sup>s</sup>* denotes the reduced scattering coefficient, μ*<sup>a</sup>* denotes the absorption coefficient, *B* denotes the average blood volume fraction, *O* is the oxygen saturation of blood, *W* is the water content, and *Fa* is the fat content. Selecting a generic tissue composed of 15% of blood at 75% of oxygen saturation, 20% of water, and 10% of fat results in μ*tissue <sup>a</sup>* = 0.062 mm−<sup>1</sup> at 800 nm. The *a* and *b* values were chosen as the mean values estimated for soft tissues (*a* = 1.89 mm<sup>−</sup>1, *b* = 1.286) [54] resulting in μ *tissue <sup>s</sup>* = 1.033 mm−<sup>1</sup> at 800 nm. The anisotropic factor (*gtissue*) and the refraction index (η*tissue*) were 0.95 and 1.37, respectively.

#### *2.2. Device for Multiangle Long-Axis Lateral Illumination*

The PA system is composed by an Nd:YAG Laser (Brilliant B, Quantel Laser, Les Ulis, France) and an Optical Parametric Oscillator (MagicPRISM OPO, Opotek, Carlsbad, CA, USA) connected to a trifurcated optical fiber bundle (Oriel Glass Fiber Optic Bundle; numerical aperture 0.56; core diameter: 7.9 mm (common), 5.5 mm (legs); fiber length 36 in; Newport, Irvine, CA, USA). One terminal end of the optical fiber bundle was connected to the sensor of an energy meter (FieldMax II-TOP, Coherent, Santa Clara, CA, USA) providing the measurement of laser fluence in real-time. The other two terminals were used to illuminate the sample. PA and ultrasonic radiofrequency (RF) data were acquired using a commercial ultrasound system (SonixOP, Ultrasonix Medical Corp., Richmond, BC, Canada) connected to a parallel acquisition receiver module (SonixDAQ, Ultrasonix Medical Corp., Richmond, Canada), operating at a sampling frequency of 40 MHz.

The device for multiangle long-axis lateral illumination was developed using a new architecture that differs from most used configurations presented in previous studies [9,23–26,33–38]. The optical fiber bundle terminal ends were attached to movable sockets placed on the lateral sides of the transducer's long-axis (see Figure 1c), so that different focal illumination spots within the transducer FOV could be obtained by varying the incident laser beam angles (see Figures 2 and 3).

**Figure 2.** (**a**) Three-dimensional model and (**b**) prototype of the multiangle long-axis lateral illumination device attached to a linear array ultrasound transducer.

**Figure 3.** Depiction of the experimental setup used to acquire the multiangle long-axis lateral illumination PA images. The distance of 19.5 mm between the transducer and the phantom surface forces the focal illumination region to be at the phantom surface for θ*<sup>i</sup>* = 0◦.

The device consists of three main parts: the ultrasound transducer support, the motion transmission system, and two motors. The support and the motion transmission system were designed using the FreeCAD open-source parametric modeling software, see Figure 2a, and printed with Acrylonitrile Butadiene Styrene (ABS) plastic using a 3D printer (ZMorph 2.0 SX, ZMorph, Wroclaw, Poland), as shown in Figure 2b. The angle of the movable sockets is controlled by the servo motors (MG996R, Tower Pro, Shenzhen, China) connected to the motion transmission systems and controlled by an open-source microcontroller (Arduino UNO, Arduino, Turin, Italy).

A linear L14-5/38 ultrasound transducer (Ultrasonix Medical Corp., Richmond, Canada) with 128 piezoelectric elements and a nominal center frequency of 7.2 MHz, was positioned inside the support, which was then attached to the 3D linear stage (HSC-103, Sigmakoki, Tokyo, Japan). A LabVIEW virtual interface (National Instruments Corp., Austin, TX) was developed to control the position of the device as well as the illumination angles. The timing sequence of the synchronous RF data acquisition and multiangle illumination consists in acquiring a pair of PA and B-mode images for each laser pulse (laser repetition rate is 10 Hz). After *K* laser pulses (*K*·100 ms) the illumination angle is incremented, the process is then repeated for *n* illumination angles. A 3D volume is obtained by moving the transducer along the elevation axis with a 3-axes translational stage.

#### *2.3. Coherently Summing the PA Subframes*

The RF data were acquired intercalating the laser pulse with the pulse-echo transmission for obtaining a prebeamformed PA subframe and then a prebeamformed B-mode frame. The PA sub-image and B-mode image were generated with the delay and sum technique. In PA images, the RF signal*s*(*xi*, *t*) represents the pressure waves generated by the light absorbers and detected by the *i*-th transducer element. The PA wave time of flight from the absorber position to each element of the array is

$$\delta(\mathbf{x}, \mathbf{x}\_{i\star} \ y) = \sqrt{y^2 + \left(\mathbf{x} - \mathbf{x}\_i\right)^2} / c \tag{3}$$

where *x* and *y* are the lateral and axial position of the pressure wave source, respectively, while *xi* denotes the lateral distance of the *i*-th element of the array to the central element. The delay and sum technique for PA subframe reconstruction consists in applying a delay δ(*x*, *xi*, *y*) to the RF signal *s*(*xi*, *t*) detected by the elements of the transducer and adding coherently [51]

$$S(\mathbf{x}, \mathbf{y}) = \sum\_{\mathbf{x} = \mathbf{a}}^{\mathbf{x} + \alpha} s(\mathbf{x}\_{i\prime} \delta(\mathbf{x}, \mathbf{x}\_{i\prime} \ \mathbf{y})) \tag{4}$$

where α is the aperture of receive beamforming, i.e., the number of adjacent elements summed.

The coherent sum of reconstructed PA subframes acquired using each illumination angle, without other processing steps, is equivalent to a reconstructed PA image acquired using a wide illumination due to the linear behavior of the delay and sum operation (Huygens–Fresnel principle) [51]. Thus, PA signals reconstructed using the delay and sum technique can be coherently added to gather the contribution of each illumination angle as:

$$S\_{\mathbb{C}}(\mathbf{x}, \mathbf{y}) = \sum\_{\iota \neq 0}^{O\_n} S\_{\partial\_{\iota}}(\mathbf{x}, \mathbf{y}), \tag{5}$$

where *Sc*(*x*, *y*) is the reconstructed RF signal coherently summed (PA final image), *S*θ*<sup>i</sup>* (*x*, *y*) is the reconstructed RF signal acquired at the *i*-th illumination angle (PA subframe), θ*<sup>n</sup>* is the maximum illumination angle.

#### *2.4. Phantom Experiments: Evaluation of Multiangle Long-Axis Lateral Illumination PAI*

A cubic phantom with a homogeneous distribution of light absorbers (magnetic nanoparticles) was used to evaluate the multiangle illumination and the PA images. The phantom dimensions were 8.0 cm × 8.0 cm × 3.5 cm, and it was manufactured using a mixture of gelatin (Bloom 250, Gelita, Eberbach, Germany) and agar powder (RM026; Himedia Laboratories-LLC, Kennett Square, USA), diluted at dry-weight concentrations of 4% and 2% of water mass, respectively. Iron oxide nanoparticles (Fe3O4) with dimensions ranging from 20 nm to 30 nm (Nanostructured and Amorphous Materials Inc., Houston, TX, USA) in the concentration of 0.1% of water mass were added to act as light absorbers. Formaldehyde in a weight concentration of 0.5% of gelatin mass was added to increase stiffness and melting temperature. The phantom was manufactured according to the description in [55,56].

To avoid any coupling issues, the experiments were performed using targets immersed in water to guarantee that the gap between the ultrasound transducer and the target was completely filled by an optically transparent coupling medium. However, we believe it would be possible to acquire images using a matching layer. For example, this layer could be ultrasound imaging gel (see, for example, [57]) or a gel pad (see, for example, [58]). This is a topic of ongoing research and should appear in future publications.

The phantom was immersed in a water tank with its surface 19.5 mm from the transducer face. For this condition, the focal illumination region was at the phantom surface for θ*<sup>i</sup>* = 0 ◦ , considered as the smallest possible angle (see Figure 3). Then, for each illumination angle, two PA subframes were acquired and the angle was varied *n* times in steps of Δθ until the *n*-th angle was achieved

$$
\theta\_n = \theta\_{\min} + n\Delta\theta. \tag{6}
$$

The device moved across the elevation axis to obtain a volume (Table 1), resulting in a total of 360 frames. The PA images were acquired at 720 nm with an average fluence of 15 mJ/cm2 at the phantom surface. This wavelength was selected for the phantom experiment because one of the energy peaks of the laser is observed at 720 nm; in addition, iron oxide nanoparticles present higher optical absorption at lower wavelengths within the near infrared spectrum region [59]. For the in vivo experiments, 800 nm was selected because it is the isosbestic point of blood [60].


**Table 1.** Multiangle illumination and acquisition parameters.

The phantom was assumed to have a homogeneous distribution of light absorbers; the optical attenuation coefficient was calculated measuring the fluence of transmitted light through the layers of the phantom with different thickness. The estimated light attenuation of the phantom was μ*att phantom* <sup>=</sup> (0.133 <sup>±</sup> 0.011) mm<sup>−</sup>1. Therefore, an analysis of the light delivery was performed by evaluating the PA signal as a function of axial and lateral directions of an averaged PA image taken over the elevation axis. Since the magnetic nanoparticles at low concentration, which is the case of the present experiment, mainly absorb the light energy, optical scattering was considered negligible for this analysis [59].

We defined image depth as the axial distance between the position of a RF signal inside the target and the target surface, therefore not considering the distance between the transducer face and target surface. The average RF signal of the PA subframes at depths and in lateral direction were evaluated using distinct regions of interest (ROI). Based on the number of elements of the transducer (i.e., 128 elements), we defined a central ROI-1 within FOV, which included the five central elements (62–66) and extended from the phantom surface to the maximum depth, with dimensions 1.5 mm × 25.5 mm. Also, peripheral ROIs (ROI-2 and ROI-3) were defined including two sets of five elements positioned at opposite sides of transducer elements: 5–9 (ROI-2) and 119–123 (ROI-3). ROI-2 and ROI-3 had the same dimensions as ROI-1. In addition, the average PA signal magnitude, in the lateral direction from 0 mm to 2 mm of depth, was calculated for all elements (1–128).

Since the phantom had a homogenous distribution of light absorbers, the PA signal amplitude was related to the amount of light delivered. The quantitative analysis of illumination, in the ROI-1 region, as a function of illumination angle was performed using the mean square root of the RF signal amplitude (*ARMS*) calculated in each PA subframe:

$$A\_{RMS}(\theta) = \sqrt{\frac{\sum\_{j=x\_a}^{x\_b} \sum\_{i=y\_a}^{y\_b} \left(S\_{\theta}(\mathbf{x}\_{j\prime}, y\_i)\right)^2}{\left(y\_b - y\_a\right)\left(\mathbf{x}\_b - \mathbf{x}\_a\right)}}\tag{7}$$

where *xa*, *xb*, *ya*, and *yb* are the limits of the ROI-1.

The spatial light delivery information, in the central region of the final PA image, was estimated taking the mean axial position of the RF signal (*y* ) as a function of illumination angle in ROI-1

$$\overline{y}(\theta) = \frac{\sum\_{j=x\_a}^{x\_b} \sum\_{i=y\_a}^{y\_b} \left( \left| S\_{\mathcal{O}} \left( x\_{j,\prime} y\_i \right) \right| y\_i \right)}{\sum\_{j=x\_a}^{x\_b} \sum\_{i=y\_a}^{y\_b} \left( \left| S\_{\mathcal{O}} \left( x\_{j,\prime} y\_i \right) \right| \right)} \tag{8}$$

Thus, *ARMS* provides information about the mean amount of light delivered per illumination angle in the central region of the transducer while *y* provides spatial information about the mean axial position of generated pressure waves.

The final multiangle PA image SNR was calculated taking the envelope-detected image amplitude [61]:

$$\text{SNR} = \sum\_{i} \sum\_{j} \left[ S\_H(\mathbf{x}\_i, \mathbf{y}\_j) - \overline{S}\_{nH} \right] / \sigma\_{nH} \tag{9}$$

where *SH* is the Hilbert transform modulus of the RF signal (*SH* = |*H*{*SC*}|), *SnH* and σ*nH* are the average and standard deviation of the background noise in *SH*, respectively.

#### *2.5. In Vivo Experiments: Human (Finger and Forearm) and Animal (Balb*/*C Mouse)*

The index finger of the left hand and left anterior forearm of a human volunteer were photoacoustically imaged using multiangle PAI with illumination parameters according to Table 1. Figure 4a shows photographs of the index finger and forearm where the dashed lines indicate the position and orientation of the transducer. The volunteer immersed his hand and forearm in a water tank; the distance between the transducer face and the skin was chosen so that the laser beams were focused at the skin surface when illumination angle was minimum (θ*<sup>i</sup>* = 0◦). The experiments were performed using an average fluence of 9.0 mJ/cm2 at 800 nm, obtaining a total of 90 frames.

**Figure 4.** (**a**) Photographs of the index finger and forearm of the volunteer. The dashed lines indicate the position and orientation of the transducer. (**b**) Depiction of the experimental setup used to acquire in vivo PA images of Balb/C mouse.

A male Balb/C mouse, at the age of ten weeks, was anesthetized using vaporized isoflurane (1.0–1.5% isoflurane, Vetflurano, Virbac, São Paulo, Brazil). The animal was positioned in a ramp platform immersed in water with a controlled temperature of 36 ◦C, as shown in Figure 4b. Using the same wavelength and fluence of the human experiment, PA images of the animal abdomen were acquired.

The experiments involving humans and animals used controlled fluence lower than 20.0 mJ/cm2, considering the limit for short-pulse lasers at the skin, which is defined by the American National Standards Institute (ANSI). The animal procedures were approved by the Animal Ethical Committee of Ribeirão Preto Medical School, University of São Paulo (process No. 005/2017-1). The experiments with the volunteer were conducted according to the procedure approved by the Research Ethical Committee of Faculty of Philosophy, Science, and Letters of Ribeirão Preto, University of São Paulo (CAAE: 08860819.4.0000.5407).

#### **3. Results and Discussion**

#### *3.1. Comparison of Illumination Schemes for Di*ff*erent Target Shapes Using Monte Carlo simulation*

Monte Carlo simulations were conducted to analyze the influence of the target shape and illumination scheme on light delivery. Figure 5 shows the normalized fluence maps obtained for all illumination strategies. Figure 5a–c,e–g show the results for the cases where the target is larger and smaller than the image width, respectively. Figure 5i–k show an intermediate situation representing a mouse torso. In this case, the geometry was obtained by segmenting an experimental B-mode image which will be shown in the next sections. Figure 5a,e,i show the results obtained for multiangle long-axis lateral illumination, while in Figure 5b,f,j, the results for the bright-field illumination coaxial with acoustic detection are shown. Finally, Figure 5c,g,k show the results obtained using the standard dark-field illumination scheme. In these images, ROIs were used to compare the light fluence for different spatial locations (white square is ROI-A; black square is ROI-B and magenta square is ROI-C). Figure 5d,h,l show bar graphs comparing the average fluence estimated within ROIs A, B, and C.

In the central region, all illumination schemes presented similar relative fluence at a shallow depth (ROI-A); yet, to some extent it was consistently higher for the bright-field illumination independent of the target shape. For targets with a nonflat surface, the focus region of the two laser beams used in the standard dark-field arrangement can be partially outside of the material and imaging plane as can be seen on the left side of Figure 6. The light delivered outside of the imaging plane contributes less to the PA image generation and can be a clutter source [62]. In this case, both situations, bright-field and long-axis lateral illumination, have the advantage of delivering light inside the imaging plane even when the target's surface is not flat. For this reason, the average fluence within ROIs B and C, located at higher depths, was consistently lower for the standard dark-field illumination scenario. It is important to recall that the light beams used for the standard dark-field were overlapping at the tissue surface. Other studies have shown that light delivery can be increased at higher depths, by using a deeper located illumination focus [30,37,38,53]. However, this strategy can dramatically decrease light delivery at shallow depths. For example, the study [48] showed that by positioning the focus at 13.5 mm depth, the simulated fluence estimated for the standard dark-field optical illumination was considerably lower than what was estimated for the bright-field illumination scheme at depths lower than 10 mm.

For targets smaller than the images' lateral dimension, the long-axis lateral illumination can deliver light to the sides of the target, increasing the penetration of light inside the material, which is depicted at the right side of Figure 6 and can be observed in the Monte Carlo simulation results. Moreover, the long-axis lateral illumination redirects the light to the target while part of the bright-field and standard dark-field illumination schemes do not contribute to PA signal generation. The simulations show that the relative fluence obtained with multiangle long-axis lateral illumination was dramatically improved for the case of a cylindrical geometry with a diameter smaller than the width of the

ultrasound probe, which is a similar situation as the human finger as will be described in the next sections. Figure 5h shows that the average fluence for the long-axis lateral illumination scheme, measured within ROI-C, was four times higher than fluence delivered by the bright-field illumination and one order of magnitude higher compared to the dark-field illumination scheme.

For a target that combines parts smaller and parts larger than image width, as the mouse's torso, both long-axis lateral and the bright-field illumination schemes provided a relatively uniform light delivery to the entire target surface. On the other hand, the light delivered by the standard dark-field illumination was considerably higher at the top surface. For this situation, the long-axis lateral illumination provided a little increment of fluence within ROI-C compared to the bright-field illumination scheme.

The next two sections aim to evaluate the feasibility of generating PA images using the long-axis lateral illumination scheme. First, the device and the multiangle imaging strategy are evaluated with a phantom experiment; then the possibility of generating the PA images, in vivo, of targets with similar geometries adopted for the simulations are verified.

**Figure 5.** Normalized fluence maps obtained by Monte Carlo simulation for targets larger (**a**–**c**) and smaller (**e**–**g**) than the image width. An intermediate situation representing a mouse torso was also considered (**i**–**k**). All cases were simulated for the bright-field, standard dark-field, and long-axis lateral illumination schemes. Average fluence values were estimated within regions of interest (ROIs) A, B, and C for all cases (**d**,**h**,**l**).

**Figure 6.** Comparison between the standard dark-field illumination scheme and the multiangle long-axis lateral illumination. For targets larger than the image width with a nonflat surface, the long-axis lateral and the bright-field illumination schemes have the advantage of delivering light within the imaging plane. For targets smaller than the transducer width, the long-axis lateral illumination scheme can deliver light to the sides of the target.

#### *3.2. Analysis of Illumination Angles Contribution to the PA Image of the Phantom*

The homogeneous phantom with a flat surface is useful for the characterization of light delivery using different illumination angles. To evaluate the light delivered to the phantom, each PA subframe at θ*<sup>i</sup>* is represented as the average of the 19 PA subframes acquired at different positions of elevation axis (slice) using the same *i*-th illumination angle. The averaged PA subframes in Figure 7 shows the light propagation along depth. The blue arrows indicate the PA signal generated beyond the laser focal region for the illumination angles 0◦, 2◦, and 4◦. This observation can be understood as an advantage of providing illumination from the laterals of the transducer, therefore generating PA signals within FOV for regions not only at the focus. However, the amplitude of the PA signal is a function of the illumination angle, because the laser focus region moves towards higher depth, while the light path increases, reducing the fluence due to light attenuation.

PA signal magnitude increased for depths greater than 10 mm and illumination angles higher than 8◦, as it can be seen in Figure 8a. Although illumination along the peripheral areas of the transducer is mostly achieved by just one of the optical fiber outputs, the incident angle of the laser beam in this region decreased relative to the normal surface, delivering light at higher depths, as shown in Figure 8a,b. Moreover, the average PA signal showed the separation of laser beams along the lateral direction; see Figure 8c.

**Figure 7.** PA subframes of the homogeneous phantom for increasing illumination angles in the range 0◦–18◦. Each subframe is an average of the phantom's elevational dimension (i.e., 3.8 cm). Blue arrows indicate the generation of PA signals beyond the laser focal region.

**Figure 8.** Average PA signal as a function of the illumination angle, along the axial direction at (**a**) central (ROI-1) and (**b**) peripheral ROIs (ROI-2 and ROI-3); (**c**) average PA signal magnitude along lateral direction for depths ranging from 0 mm to 2 mm.

The analysis of the *ARMS* at the central region revealed a proportional decrease in the amount of light delivered for angles higher than 4◦, which is probably related to light attenuation within the phantom (Figure 9a). Besides, a peak of maximum *ARMS* could be observed for θ*<sup>i</sup>* = 4◦, showing that the maximum light delivery to the central area occurred when the laser focus region was completely inside the phantom, where the light was less attenuated (shallow depths). These results show the contribution of illumination using θ*<sup>i</sup>* < 4◦ was less significant for the image of the phantom. In addition, the mean depth of the PA signal increased as a function of the illumination angle (Figure 9b), which could be qualitatively inferred from the plots in Figure 8a.

The final PA image was obtained from the summation of the PA subframes at different illumination angles. For example, a PA image of a single image slice of the phantom is shown in Figure 10a. The average PA signal calculated at depths ranging from 0 mm to 25.5 mm for all elements of this PA image showed the contribution of all illumination angles; see Figure 10b. Figure 10c shows SNR as a function of depth demonstrating the multiangle long-axis illumination PA imaging could be used to perform studies at relevant imaging depths.

**Figure 9.** (**a**) Mean square root of PA signal and (**b**) mean depth of PA signal as a function of illumination angle.

**Figure 10.** (**a**) Final PA image of a single image slice of the phantom, (**b**) PA signal profile of this PA image shows the contribution of all illumination angles to generate PA signal at depths greater than 10 mm, and (**c**) SNR as a function of image depth.

#### *3.3. In Vivo PA Images*

Different in vivo experiments were conducted to evaluate the multiangle long-axis lateral illumination PAI. The first in vivo PA images were acquired from a human forearm. In this case, the shape of the surface was larger than the long-axis dimension of the transducer, providing similarities with the flat surface of the phantom and the simulation study. The anatomical structures of the human forearm such as palmaris longus tendon, subcutaneous blood vessels, and epithelial tissue could be identified in the B-mode image as well as in the PA image, as it can be seen in Figure 11a,b. SNR was calculated within two ROIs of 2.4 mm × 1.6 mm in the axial and lateral dimensions, respectively. For the in vivo evaluation, noise ROIs, with the same dimensions, were positioned within 5 mm distance from the structure under analysis. At a tissue depth of 7.3 mm, ROI overlaid the tendon of flexor digitorum superficialis [63] and provided SNR = 14 dB (green rectangle in Figure 11a), while another ROI overlaid a subcutaneous blood vessel and provided SNR = 25 dB at a 2.5 mm tissue depth (cyan rectangle in Figure 11a). Differences in the SNR values between deep and shallow regions were mostly due to light attenuation; the tendon (collagen) also presents an optical absorption coefficient lower than those for melanin or hemoglobin at 800 nm [64], which reduced its SNR on the PA image.

The second acquisition of PA images was obtained from the human index finger, which has a cylindrical-like shape that differs from the forearm or phantom's surface and has a maximum lateral extension of approximately 20 mm. However, the finger cross-section lateral dimension is smaller than the lateral FOV, allowing illumination from laterals to be more efficient. Consequently, the laser focus becomes deeper while the light path within the tissue is shortened, reducing the light attenuation. SNR was analyzed within ROIs of 2.0 mm × 3.0 mm in the axial and lateral dimensions, respectively. Those ROIs were placed over the location of the dorsal and palmar digital arteries at 3.4 mm (cyan rectangle in Figure 11c) and 10.3 mm tissue depths (green rectangle in Figure 11c), resulting in SNR of 22.5 dB and 22 dB, respectively.

Lastly, a challenging combination of both aforementioned surface shapes was observed when acquiring the PA image of the cross-section of the mouse abdomen. The abdomen had a lateral dimension smaller than the lateral length of the transducer, while the lower limbs fit FOV. In this case, illumination angles provided light delivery to the sides of the mouse abdomen and hit the surface of lower limbs obliquely. Furthermore, the mouse skin has an average thickness of 0.5 mm and optical absorption lower than human skin [65], increasing light penetration. In the B-mode image of the mouse, the bladder, femoral artery, and a branch of the abdominal aorta can be identified [66]. PA signals from the femoral artery were evaluated at 1.3 mm (cyan rectangle in Figure 11e) and the aortic branch at 10.5 mm of depth (green rectangle in Figure 11e) presenting SNR = 21.5 dB for the femoral artery and SNR = 17.5 dB for the aortic branch (ROIs of 2.0 mm × 2.8 mm).

For three among the most studied PA targets in biomedical applications, the multiangle long-axis lateral illumination was able to provide PA images with high SNR at a depth of 10.5 mm for target shapes with lateral extension smaller than the lateral length of transducer (mouse and index finger). In the human forearm, a tendon at depth of 7 mm generated PA signal with sufficient SNR for good

visualization of the structure. However, it should be mentioned that the in vivo PA images were acquired using a measured laser fluence of 9 mJ/cm2, which is much lower than the limit for short-pulse laser at the skin. Increasing the laser fluence to values close to the safety limit can improve SNR of PA images at greater depths.

**Figure 11.** In vivo multiangle long-axis lateral illumination PA and B-mode images. (**a**) PA image and (**b**) B-mode image of the human forearm, anatomical structures such as tendons and subcutaneous blood vessels can be identified; (**c**) PA image and (**d**) B-mode image of the human index finger, the cylindrical shape allows light delivery by the laterals promoting the visualization of the palmar digital artery at depth of 10.3 mm. (**e**) PA image and (**f**) B-mode image of the Balb/Cmouse abdomen, PA signal from an aortic branch at 10.5 mm of depth can be visualized.

Figure S1 shows the subframes acquired at each illumination angle for the in vivo experiments and plots with the corresponding SNR for ROIs positioned at the selected structures (vessels and tendon). The pronounced variation in the SNR values across the subframes acquired at different angles is evident. Clearly, SNR, at each ROI, observed for the final PA image is similar (only slightly higher) to that subframe with the highest SNR where the illumination area comprised the structure of interest. In the case of combining *N* PA images at the same illumination angle, it is expected that the SNR will be increased by <sup>√</sup> *N*, which would be higher than SNR obtained with the multiangle approach for a particular ROI. A more concentrated illumination in the proposed approach, when compared to the techniques illustrated in Figure 1a,b, can also help improve SNR at specific locations of the image. This concept has been also explored in other studies [51,53].

Multiangle long-axis lateral illumination could be a useful approach to improve the quality of PA images in preclinical studies with mice, since important anatomic structures are smaller than the FOV. Even for higher frequency linear arrays with a width smaller than the probe used here, murine tumor models still fit in this category [53]. In [53], the authors demonstrated that improving light delivery for this situation can greatly improve PA image quality. Since PA images taken from the human finger joint has shown great potential to evaluate inflammatory arthritis [22,23], it is a possible clinical application where the proposed technique could be beneficial. Future studies will include: (i) evaluating multi-wavelength PA images and fluence correction strategies [57] to monitor blood oxygen saturation and (ii) evaluate the feasibility of using a high divergent source like LED [44] to reduce the number of illumination angles.

Since the pulse repetition frequency of the laser (*LPRF*) is 10 Hz, the in vivo images of the present study were acquired at a frame rate of 1 frame per second. Each image was composed of two subframes per angle and five different angles were used. The maximum frame rate for this configuration is two frames per second if a single frame was acquired per angle. To acquire the full transducer FOV, a few subframes acquired at different angles are needed; therefore, the frame rate for the proposed technique will be lower than that obtained for the configurations shown in Figure 1a,b. This is a limitation of the proposed approach, especially for the cases where the illumination source operates at low PRF as the laser used in the present study (*LPRF* = 10 Hz) and monitor fast-changing dynamics is the goal. However, for lasers working at higher repetition rate (*LPRF* ~ 100 Hz), see, for example, [67]) this limitation can be minimized and the frame rate can be increased. The maximum *LPRF* supported by the setup depends of the angular velocity of servomotors (ω*s*), and the step angle (Δθ): *LPRF* = ω*s*/Δθ. The servomotors used in the setup can take 0.2 s to rotate the fibers output from 0◦ to 60◦, resulting in a maximum angular velocity of <sup>ω</sup>*<sup>s</sup>* = 5.2 *rad*·*s*−<sup>1</sup> when operating at 5 V. In this case, the setup could acquire PA subframes using = 75 Hz with Δθ = 4◦ and provide 15 frames per second when five illumination angles are used.

The present paper introduced and evaluated the feasibility of the multiangle long-axis lateral illumination to generate PA images, contributing to the development of new illumination strategies in PAI. An advantage of the setup presented here is that it could be additive to other existing illumination schemes. Light delivery by the laterals of the target could be used together with more conventionally used illumination schemes to improve light delivery to targets with lateral dimension smaller than the transducer's width. The concept of PA images acquired using multiangle illumination can also be applied to setups with similar design as described in [37,38], where the angle of incident light delivered by transducer's short-axis can be controlled.

#### **4. Conclusions**

This paper demonstrated the feasibility of using a novel multiangle long-axis lateral illumination PAI. Monte Carlo simulations compared light delivery to tissue for three different illumination schemes: bright-field, standard dark-field, and long-axis lateral illumination. Illumination schemes performance were evaluated for three preclinical and clinically relevant cases for PAI. The shape of the target influenced light delivery for all illumination schemes. Long-axis lateral illumination provided substantial improvement when targets smaller than the lateral width of the transducer were evaluated. The prototype developed to produce multiangle long-axis lateral illumination was evaluated with phantom and in vivo experiments. PA images of good quality were generated from mouse abdomen, forearm, and index finger of a volunteer. Based on the results presented here, a novel PAI system was proposed for preclinical and clinical research. In addition, long-axis lateral illumination could be used together with more conventional illumination schemes to improve light delivery in reflection mode PAI.

**Supplementary Materials:** The following are available online at http://www.mdpi.com/1424-8220/20/14/4052/s1, Figure S1: Subframes acquired at each illumination angle used to generate the in vivo PA images shown in Figure 11 of the main article for the (a) forearm, (c) index finger, and (e) mouse abdomen. SNR as a function of illumination angle for each subframe of selected structures located at the (b) forearm, (d) index finger, and (f) mouse abdomen.

**Author Contributions:** Conceptualization and design, J.H.U. and T.Z.P.; Data curation, J.H.U., D.R.T.S. and G.S.P.F.; Analysis and interpretation of data, J.H.U., D.R.T.S., G.S.P.F., M.S.B., M.H.N.-B. and A.A.O.C. and T.Z.P.; Funding acquisition, A.A.O.C. and T.Z.P.; Investigation, J.H.U and T.Z.P.; Methodology, J.H.U., D.R.T.S., G.S.P.F., M.S.B., M.H.N.-B., A.A.O.C. and T.Z.P.; Project administration, T.Z.P.; Resources, M.S.B., A.A.O.C. and T.Z.P.; Software, J.H.U., D.R.T.S. and G.S.P.F.; Supervision, T.Z.P.; Writing—original draft, J.H.U. and T.Z.P.; Writing—review & editing, J.H.U., D.R.T.S., G.S.P.F., M.S.B., M.H.N.-B. and A.A.O.C. and T.Z.P. All authors critically read and approved the final version of the manuscript.

**Funding:** This study was supported by Coordenação de Aperfeiçoamento de Pessoal de Nível Superior-Brasil (CAPES)-Finance Code 001, Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq) grant 436657/2018-0 and Fundação de Amparo à Pesquisa do Estado de São Paulo (FAPESP) grant 2016/22374-8, 2015/05684-0, 2017/14482-8, and 2013/18854-6.

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

#### **References**


© 2020 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 (http://creativecommons.org/licenses/by/4.0/).

## *Article* **Enhancement of Photoacoustic Signal Strength with Continuous Wave Optical Pre-Illumination: A Non-Invasive Technique**

**Anjali Thomas <sup>1</sup> , Souradip Paul 1, Joy Mitra <sup>2</sup> and Mayanglambam Suheshkumar Singh 1,\***


**Abstract:** Use of portable and affordable pulse light sources (light emitting diodes (LED) and laser diodes) for tissue illumination offers an opportunity to accelerate the clinical translation of photoacoustic imaging (PAI) technology. However, imaging depth in this case is limited because of low output (optical) power of these light sources. In this work, we developed a noninvasive technique for enhancing strength (amplitude) of photoacoustic (PA) signal. This is a photothermal-based technique in which a continuous wave (CW) optical beam, in addition to short-pulse (∼nsec) laser beam, is employed to irradiate and, thus, raise the temperature of sample material selectively over a pre-specified region of interest (we call the process as pre-illumination). The increase in temperature, in turn enhances the PA-signal strength. Experiments were conducted in methylene blue, which is one of the commonly used contrast agents in laboratory research studies, to validate change in temperature and subsequent enhancement of PA-signal strength for the following cases: (1) concentration or optical absorption coefficient of sample, (2) optical power of CW-optical beam, and (3) time duration of pre-illumination. A theoretical hypothesis, being validated by numerical simulation, is presented. To validate the proposed technique for clinical and/or pre-clinical applications (diagnosis and treatments of cancer, pressure ulcers, and minimally invasive procedures including vascular access and fetal surgery), experiments were conducted in tissue-mimicking Agar phantom and ex-vivo animal tissue (chicken breast). Results demonstrate that pre-illumination significantly enhances PA-signal strength (up to ∼70% (methylene blue), ∼48% (Agar phantom), and ∼40% (chicken tissue)). The proposed technique addresses one of the primary challenges in the clinical translation of LED-based PAI systems (more specifically, to obtain a detectable PA-signal from deep-seated tissue targets).

**Keywords:** photoacoustic imaging; signal enhancement; pre-illumination; photo-thermal effect; heat capacity

#### **1. Introduction**

Photoacoustic imaging (PAI) has been proven as a promising technology for nondestructive recovery of vital patho-physiological parameters (functional, structural, hemodynamics, mechanical, and molecular distribution [1–6] of biological tissues with a microscopic resolution at an unprecedented penetration depth (~cm) [7]. On the other hand, LED-based systems hold great potential in clinical translation because of their portability and affordability [8–10]. In LED-based PAI systems, the pulsed laser source is replaced by LEDs and, thus, the optical power of LEDs is insufficient to induce detectable photoacoustic (PA)-signal at higher penetration depths, which is the major drawback of the system [9]. Similar is true for the cases of laser diode based PAI systems [11]. In this context,

**Citation:** Thomas, A.; Paul, S.; Mitra, J.; Singh, M.S. Enhancement of Photoacoustic Signal Strength with Continuous Wave Optical Pre-Illumination: A Non-Invasive Technique. *Sensors* **2021**, *21*, 1190. https://doi.org/10.3390/s21041190

Academic Editors: Mithun Kuniyil Ajith Singh and Wenfeng Xia Received: 25 December 2020 Accepted: 23 January 2021 Published: 8 February 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/).

enhancement of the achievable PA-signal strength is a longstanding problem in low energy light source based PAI systems and temperature dependent PA-signal enhancement is one of the most focusing areas of research studies in the past few years [12–23]. Enhancement in signal strength enables not only to improve obtainable imaging depth but also to increase accuracy in the quantitative measurement of physiological parameters by improving signal-to-noise-ratio (SNR) [12]. In this regard, an injection-based technique—in which exogenous contrast agents or target specific biomarkers are introduced to tissue sample of interest externally through injection—has been proven successful and remains the only standard technique [13,24,25]. Unfortunately, the technique suffers from serious drawbacks: (1) limited studies of patho-physiological activities due to limited availability or selection of target-specific biomarkers [24,25], (2) bio-compatibility, (3) bio-toxicity, and (4) invasive in nature because of the introduction of exogenous dyes—which are foreign substances—to body as well as breakage of the intervening tissues including skin. To our best knowledge, experimental studies on noninvasive techniques for enhancement of the PA-signal strength (or amplitude) are limited to the articles reported in Refs. [12–23]. In all of these reported studies for enhancement of PA-signal strength, imaging sample was immersed inside a heating bath filled with water that controls the thermodynamic equilibrium temperature (with an exception of the study [20] that controls the temperature of the imaging target by connecting an electrically conducting wire to imaging target of interest). Thus, all of the reported (experimental) techniques are not suitable for the end applications being targeted to clinical settings. Refs. [21,26,27] reported studies on theoretical aspects of the generation of PA-waves and its contribution from thermodynamic equilibrium temperature (*T*) and temperature raise (Δ*T*) induced by transient optical illumination. In our present article, we report a unique photothermal-based technique for enhancing the achievable PA-signal strength. Experimentally, enhancement is facilitated by pre-illumination of the sample with a continuous wave (CW)-optical beam (in addition to irradiation of the sample with short-pulse laser (with pulse-width ~ nsec)), thereby raising (thermodynamic equilibrium) temperature (*T*) of the imaging sample selectively over a pre-specified region of interest. Shortly, enhancement of PA-signal strength is achieved by control of thermodynamic property of imaging sample over a particular region of interest (unlike controlling temperature of entire sample or background as it is the case for the above mentioned or reported techniques). In this way, our proposed technique is noninvasive (without introducing contrast agents) and nondestructive (without damaging tissues).

With a close similarity to our proposed technique, in PAI-guided phototherapy or photothermal therapy [28–30], a CW-optical beam is employed to irradiate target-specific contrast dyes (being introduced to tissue targets), and the subsequent temperature raise (Δ*T*) is imaged by PAI modality. In the study [16], a CW-laser beam was used for inducing interstitial tissue coagulation while heating of the imaging targets was facilitated by hot air blow from a heat gun or conducting heating. Introduction of photo-absorbing contrast agents—like gold nanoparticles, nano-shells, and plasmonic nanoparticles—in the target region facilitates enhancement in contrast and photoacoustic signal [31–33]. This is an invasive procedure. On the other hand, recovery of the temperature distribution (*T*( → *r* )) of tissue samples with PA-imaging modality has been studied for more than a decade of years or so [12,14–16,23,34]. However, experimental techniques (including CW-irradiation) for enhancement of PA-signal strength from endogenous signal contrasts of (imaging) tissue sample and its clinical applications are not yet reported in literature. In this article, we propose a theoretical (analytical) hypothesis that derives a mathematical relationship of explicit dependence of PA-signal strength on ambient temperature (*T*) of the given physical system (tissue sample, in our case). This proposed theoretical hypothesis is validated with numerical simulation studies. For the simulation studies, we adopted k-wave (MATLAB) platform [35–37], which is widely employed as a standard numerical simulation tool for photoacoustic imaging. The experimental results, with experiments being performed in tissue-mimicking Agar phantom, as well as chicken breast, sample, demonstrate that the proposed photothermal-based technique can be adapted for clinical applications. In other

words, the proposed technique was validated with experiments being conducted in tissuemimicking phantom (Agar, in our case) as it is done conventionally in (biomedical) research laboratory settings [3–5,12,15,18] while the feasibility of clinical translation of the proposed technique was validated with experiments being conducted in tissues (chicken breast).

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

#### *2.1. Theory*

In PA-imaging, a beam of short laser pulses (pulse width ~ nsec) is delivered to the sample surface so as to irradiate a particular target of interest. Thermoelastic expansion occurs [38] due to rapid heating and subsequent cooling of the irradiated sample material. This results in the generation of pressure waves in the sample, which are known as initial photoacoustic pressure waves. One may consider the photoacoustic effect as a thermodynamic process [38]. The initial PA-pressure can be expressed as [3,38]:

$$P\_0 = \frac{\beta}{\kappa} \frac{1}{\rho c\_V} \mu\_a \phi \;=\; \Gamma \mu\_a \phi. \tag{1}$$

where *Γ* = *<sup>β</sup> κ* 1 *<sup>ρ</sup>cV* is a dimensionless physical quantity, commonly referred as Grüeisen parameter, and it is a measure of thermoelastic efficiency of a given material. Here, in the case of PAI, *Γ* gives the measure of conversion efficiency from pulse optical energy to acoustic energy. We know that optical absorption coefficient (*μa*) is characterized by optical extinction coefficient (*ε*) that can be expressed as:

$$
\mu\_a = \varepsilon [\mathbb{C}]\_\prime \tag{2}
$$

where [*C*] is the concentration.

For a physical system or medium including solution of low concentration (methylene blue solution, as it is the case for our present study) and biological tissue, one can deduce Equation (1) as (a detailed derivation is provided in Appendix A):

$$P\_0 = \frac{\beta^{(water)}}{\kappa^{(water)}} \frac{1}{\rho^{(water)} c\_V^{(water)}} \varepsilon^{(methyl)} [\mathbb{C}]^{(methyl)} \phi \tag{3}$$

Equation (3) implies that photoacoustic wave generation is dictated by the thermodynamic properties of the surrounding (background) medium/fluid (water, in our study). It is similar to the studies [21–23,26,27,39] that demonstrated PA-wave generation is dictated by thermodynamic properties of the fluid in which optical absorbing targets of vanishingly small size (point source or nanoparticles) were immersed. In view of the arguments (discussed in Appendix A), the above Equation (3) holds true for the case of Agar gel phantom where superscripts, (water) and (methylene blue) are replaced by (agar gel) and (ink), respectively. We adopt mathematical representation corresponding to methylene blue solution.

The reported study [40] gives the variation of thermodynamic parameters with (ambient) thermodynamic equilibrium temperature (*T*) (as it is given in Table 1). From Table 1, one can conclude that, in the temperature range (∼<sup>20</sup> ◦<sup>C</sup> to 40 ◦C), *<sup>β</sup>*(*water*) changes linearly with *T,* which can be represented by a straight line (*β(water)* =a+bΔ*T*, where 'a' and 'b' are intercept and slope, respectively). This assumption is in agreement with Taylor expansion of *<sup>β</sup>* close to equilibrium temperature (*T*0) given *by <sup>β</sup>*(*T*0) <sup>→</sup> *<sup>β</sup>*(*T*<sup>0</sup> <sup>+</sup> <sup>Δ</sup>*T*) <sup>≈</sup> *<sup>β</sup>*(*T*0) <sup>+</sup> <sup>Δ</sup>*<sup>T</sup> <sup>d</sup><sup>β</sup> dT* = *<sup>β</sup>equil* <sup>+</sup> <sup>Δ</sup>*<sup>T</sup> <sup>d</sup><sup>β</sup> dT* , as it is done in Refs. [21,26,27], where Δ*T* is the differential change in temperature. *βequil* (≡ *β*(*T*0)) is the thermal expansion coefficient under thermodynamic equilibrium temperature (*T*0), which is again dependent on *T*. For water, thermal expansion coefficient (*β*) vanishes at *T* ~ 3.98 ◦C, i.e., *β*(*T*) = 0 for *T* ∼ 3.98 ◦C [13,17]). At this zero-crossing temperature, from Equation (3), one can conclude that PA-waves cannot be generated under any circumstances (including increasing optical absorption coeffi-

cient (*μa*) and/or optical fluence (*φ*)), which was validated by experimental studies in the past [21,26,41]. In our present study, we conducted experiments with *T* ∼ 20–30 ◦C, which is far above the zero-crossing temperature (∼4 ◦C) and the equilibrium term (*βequil*) cannot be neglected [21]. In this temperature range of interest (∼20–30 ◦C) for laboratory as well as clinical studies, from Table 1, we estimated 'b' to be ∼9.64 ◦C and relative change of *β* is found to be ∼46.63%. However, other thermodynamic parameters give negligible relative changes compared to *β*. Under these circumstances, we assume that *κ*(*water*), *ρ*(*water*), and *c* (*water*) *<sup>V</sup>* are independent of *<sup>T</sup>* or constant in comparison to the dependence of *<sup>β</sup>*(*water*) on *T* [13,42]. Specific heat capacity (*cV*) is independent of temperature (for soft tissue) over temperature range < 50 ◦C [16,43]. From Equation (3), we can obtain the explicit expression for *P*<sup>0</sup> at an arbitrarily chosen temperature (*T*) in the neighborhood of thermodynamic equilibrium temperature (*T*0), i.e., *P*<sup>0</sup> at *T*<sup>0</sup> → *T* = *T*<sup>0</sup> + Δ*T* can be expressed as:

$$P\_0(T\_0) \to P\_0(T = T\_0 + \Delta T) = P\_0(T\_0) \frac{[\mathbb{C}]^{(\text{metty})}}{a^{(\text{unit}\tau)}} \left( \left( \varepsilon^{(\text{metty})} \left( \frac{\partial \beta^{(\text{water})}}{\partial T} \right)\_{T\_0} \Delta T + \beta^{\text{water}} \left( \frac{\partial \varepsilon^{(\text{metty})}}{\partial T} \right)\_{T\_0} \Delta T \right) \phi \tag{4}$$

where *α*(*water*) (=*κ*(*water*)*ρ*(*water*)*c* (*water*) *<sup>V</sup>* ) and [*C*] (*methy*) are considered as independent of *T*. Equation (4) shows that the strength of initial PA-pressure waves (*P*0) is characterized by the ambient temperature (*T* = *T*<sup>0</sup> + Δ*T*) of the medium in addition to *μ<sup>a</sup>* and *φ*. In our present study, the differential raise in thermodynamic equilibrium temperature (Δ*T*) is facilitated by the pre-illumination of the sample with a CW-laser beam.

**Table 1.** Dependence of thermodynamic parameters on thermodynamic equilibrium temperature (*T*) for water.


Equation (4) gives an explicit representation of *P*<sup>0</sup> in integral form, into two separate thermal terms. The first term attributes to thermodynamic equilibrium temperature (*T*0), while the second term corresponds to transient changes in thermal expansion coefficient (*β*) and optical extinction coefficient (*ε*) that are resulted due to thermal perturbation in the physical (imaging) system by an external agency and its associated heating [44]. In our present study, we employed a CW-laser beam to raise the thermodynamic equilibrium temperature *T*<sup>0</sup> → *T*<sup>0</sup> + Δ*T* and, thus, the second term in the generation of PA-signals (in Equation (4)) while pulse laser beam, which is contributing to the enhancement from the two terms, is kept fixed. Equation (4) shows that enhancement of strength of initial PApressure waves, being generated in mechanical medium (by transient optical illumination), can be achieved (experimentally) by control of ambient temperature (a) directly through heating of the imaging sample or by thermal technique and (b) indirectly through photoillumination of the specimen raising incident optical power or photo-thermal technique. Direct method demands heating of the imaging specimen as a whole and, thus, it is not of much clinical importance (as it is discussed in Section 1). Indirect method facilitates to raise the temperature (*T*) of deep-seated tissue targets selectively over a pre-specified region of interest and, thus, this technique is of significant clinical impacts. Here, we focus on the indirect technique where we employed the CW-laser beam (for photo-illumination) in addition to the pulsed-laser beam that is adopted for transient illumination and subsequent generation of initial *P*0.

#### *2.2. Enhancement of PA-Signal Strength*

One can express relative enhancement in PA-signal strength, due to a raise in thermodynamic equilibrium temperature *T*0, as:

$$P\_0^{(cnh)} = \frac{P\_0(T = T\_0 + \Delta T) - P(T = T\_0)}{P\_0(T = T\_0)} \times 100 \text{ (in } \% \text{)}\tag{5}$$

#### *2.3. Numerical Simulation*

A circular target of radius ∼3 mm having contrast in *μ<sup>a</sup>* in comparison to that of the background is situated at the center of the background of area 21.6 × 21.6 mm2. This numerical sample represents the deep-seated inhomogeneous target illuminated with CW-laser (to raise the temperature) in the background tissue medium. A circular sensor (diameter ∼20 mm) of 145 identical detectors is collecting the PA-waves for reconstruction as it is done in our previous study [45]. Figure 1a presents a representative image of the distribution of initial PA-pressure waves (*P*0) at a temperature *T*<sup>0</sup> (say, room temperature), while Figure 1c presents the same at a temperature *T*- = *T*<sup>0</sup> + Δ*T* where Δ*T* is the temperature raise due to pre-illumination. The values of Δ*T* corresponding to different concentration [*C*], absorption coefficient (*μa*), and optical power are taken from experiment. Figure 1b,d present the reconstructed images corresponds to Figure 1a,c. Variation of PA-signal along the marked lines, which are indicated in Figure 1b,d, plotted in Figure 1e. We estimate the strength of initial PA-pressure waves, for the reconstructed pressure distribution, as the average value over full width at the 75% of the maximum of the line plots and it is found to be 130 Pa at room temperature (*T*0) and 193 Pa at a temperature *T*- = *T*<sup>0</sup> + Δ*T*. (Here, Δ*T* is the temperature raise corresponding to a concentration of ∼7 mM). From these values, PA-signal enhancement due to temperature raise is calculated as 51%. In similar ways, we study variation PA-signal strength with change in temperature due to the changes in the physical parameters of the targets, and the results are plotted in Section 3.

#### *2.4. Experimental Set-Up*

For experiments, we employed a home-built acoustic resolution photoacoustic microscopy (AR-PAM) imaging system in transmission configuration. Figure 2a depicts a schematic diagram of the experimental set-up. A train of short duration pulses of optical beam—beam diameter ~2.3 mm, power ~25 mW, pulse width ~6 nsecs, pulse repetition frequency ~100 Hz, and wavelength ~670 nm—from a tunable pulsed OPO laser source (SpitLight OPO Evo 150–532, Innolas Lasers, Krailling, Germany) was coupled by an optical fibre (diameter ~6 mm) so as to irradiate tissue sample of interest. A tightly focusing ultrasound transducer (V375-SU, Olympus, Shinjuku City, Tokyo, Japan)—focal spot size ~154 <sup>μ</sup>m (calculated using, *BD* <sup>=</sup> 1.02 *<sup>F</sup>*∗*<sup>v</sup> <sup>f</sup>* <sup>∗</sup>*<sup>D</sup>* [46], where *<sup>F</sup>* is focal length; *<sup>v</sup>* is speed of sound; *f* is operating frequency; *D* is the diameter of US transducer), operating frequency ~30 MHz, and focal length ~19.10 mm—was employed to pick-up transient light-induced PA-signals (*P*0) selectively from its narrow focal zone.

The detected PA-signals were amplified using a pulser-receiver (Part No.: 5073PR-40-P, Olympus, Shinjuku City, Tokyo, Japan) and, then, acquired using a data acquisition card (Part No.: 779745-02, NI PCI-5114, 250 MS/s, National Instruments, Austin, Texas, USA) being attached to a computer system. From time-resolved A-scan, depth-resolved data is obtained (as it was done in previous studies [3,5]). We obtained 2D or 3D data representative of PA-signals by raster scanning. We employed a high precession translation stage (Newmark NSC-G series, Newmark Systems Inc., Rancho Santa Margarita, CA, USA) for raster scanning of the transducer with step-size (~100 μm) while the optical pulse beam and the sample was kept fixed in a position. For a scanning area of ~4.8 × 3 mm2—that can be achieved by ~48 × 30 raster scanning steps—we can acquire an image at the frame rate ~0.69 Hz (~4 frames per min). In addition to pulse (nsec) laser beam, as it is shown in Figure 2a, the imaging sample is illuminated with a collimated optical beam (diameter ~3 mm) from a CW-laser source (Stradus-642-110, VORTRAN Laser Technology, Roseville,

CA, USA; wavelength, 642 nm) in such a way that the CW-beam and pulse-laser beam intersect each other at a pre-specified region of interest. Optical wavelengths of CW-illumination (for enhancement of PA-signal strength) and PA-excitation (to induce PA-signals) can be selected independently. CW-laser is continuously switched on during the experiments to maintain the temperature. The entire AR-PAM imaging system is controlled by a LabVIEWbased software. Figure 2b gives a typical 1D PA-signal acquired by our AR-PAM imaging system, while Figure 2c depicts a photograph of the sample holder (transparent cuvette). During experiments, the cuvette filled with sample material (say, methylene blue solution) is completely immersed inside an acoustic-coupling medium (water, in our case) for proper coupling of acoustic signals. We employed AMIRA software for the generation of 3D images from a sequence of 2D images, which are, in turn, reconstructed by using MATLAB.

**Figure 1.** Representative images of initial pressure distribution of the tissue-mimicking numerical sample with a circular target being embedded in the background and sensor array distributed over a circular ring (**a**,**c**). Numerical samples with circular targets at a temperature *T*<sup>0</sup> (**a**), and temperature *T*- = *T*<sup>0</sup> + Δ*T* (**c**). The corresponding reconstructed images (**b**) for target at a temperature *T*<sup>0</sup> and (**d**) for target at temperature *T* obtained using k-wave MATLAB toolbox. (**e**) Variation of PA-signal strength along the marked lines shown in (**b**,**d**).

**Figure 2.** Schematic diagram of the experimental set-up (**a**). A typical 1D PA-signal acquired by our acoustic resolution photoacoustic microscopy (AR-PAM) imaging system (**b**). (**c**) Photograph of transparent cuvette (inner thickness ~1 mm) that is employed as holder for imaging sample (methylene blue solution).

#### **3. Experimental Results and Discussion**

#### *3.1. Enhancement of PA-Signal from Methylene Blue*

Figure 3 gives the experimental results to compare PA-signal strengths (*P*0) with and without pre-illumination of CW-laser beam. Experiments were conducted with methylene blue solution (concentration of ∼7 mM using water as a solvent) as an imaging sample. Figure 3a,b present 3D image representatives of *P*<sup>0</sup> with and without pre-illumination, respectively, while Figure 3c,d give 2D images corresponding to a randomly selected plane (at *z* = 7). For quantitative analysis, the variation of *P*<sup>0</sup> along the marked lines, which are indicated in Figure 3c,d, are plotted in Figure 3e. In Figure 3c,d, yellow spot appearing in blue background corresponds to *P*<sup>0</sup> acquired from pulse laser-irradiated spot in sample while the background corresponding to un-illuminated region, which gives no PA-signal. The size and shape of the image, which is characterized by pulse-laser spot, are in agreement with that of cross-section (FWHM) of pulse-laser beam. From Figure 3c,d, it is observed that maximum *P*<sup>0</sup> achievable with pre-illumination is 0.60 V against 0.40 V (without pre-illumination). Average *P*<sup>0</sup> over full-width at 75% of maximum is estimated to be 0.50 V (for pre-illumination) against 0.35 V (without pre-illumination) and enhancement (*P*(*enh*) <sup>0</sup> ), corresponding to the plane *<sup>z</sup>* <sup>=</sup> 7, is found to be 46%. In the similar fashion, *<sup>P</sup>*(*enh*) 0 for the planes (corresponding to *z* = 6 and *z* = 8) were obtained as 40% and 38%, respectively. The average value of the measurements from the three consecutive planes (41%) is considered as the *P*(*enh*) <sup>0</sup> with pre-illumination for a particular power and concentration. The results demonstrate that pre-illumination with CW-optical beam provides a significant enhancement in *P*0, which is due to enhancement in temperature change following optical illumination of CW-beam (as it is given by Equation (5)). Similar experiments were conducted for empty cuvette and water (filled in cuvette) being employed as imaging sample. The experimental results, presented in supplementary figures (Figures S1 and S2), demonstrate that cuvette and water give no significant PA-signal for both of the two cases (with and without pre-illumination). These experimental results imply that *P*(*enh*) <sup>0</sup> is not contributed from sample holder (cuvette) and acoustic-coupling medium (water), or enhancement is contributed only from the sample (methylene blue, in our case).

**Figure 3.** 3D image representatives of *P*<sup>0</sup> without (**a**) and with (**b**) pre-illumination for methylene blue while 2D image corresponding to a randomly selected plane (at *z* = 7) ((**c**), without pre-illumination and (**d**), with pre-illumination). (**e**) Variation of *P*<sup>0</sup> along the marked lines depicted in (**c**,**d**). Arrowheads in both sides of scale bar (being included in (**a**,**c**)) indicate measure of the entire length of the image along *x*-axis. Arrowheads, marked in (**e**), indicate the points of observations for estimation of size of the illuminated spot.

From Figure 3d, the size of the illuminated spot was estimated—for pre-illumination as well as without pre-illumination—from two different aspects: (a) measuring the full width at half maximum (FWHM) using four cross-sectional profiles (or line plots) (as it is shown in Figure S3). The profile along the *x*-axis is given in Figure 3e and FWHM corresponding to without (blue color arrow) and with (red color arrow) pre-illumination are estimated to be ∼2.3 mm and ∼2.6 mm, respectively. The average value of the measured FWHM from all profiles is calculated as ∼2.3 mm (with) and ∼2.2 mm (without), respectively. This shows that, by our proposed pre-illumination technique, the effective resolution is reduced. (b) Measuring the change in object size across the baseline, i.e., the distance between two points indicated in Figure 3e by arrowheads (grey color and violet color (without preillumination); grey color and green color (with pre-illumination)). The width of the spot was found to be ∼4.3 mm (without pre-illumination) and ∼4.7 mm (with pre-illumination) and the increment in width is obtained as 9.5%. This implies that the obtainable spatial resolution of the imaging system is slightly reduced (∼9.5%) by the proposed optical pre-illumination technique. This may be because of thermal excitation and diffusion of heat (resulted from pre-illumination of CW-laser beam) [18,22], thereby enhancing strength of PA-signal from the immediate background of the (pulse-laser) illuminated spot. More elaborately, transient heat generation, which results from transient pulse-laser excitation, is localized [47] (under the physical condition of thermal and stress confinement) while (ambient) temperature raise (and its heat generation) induced by CW-optical beam illumination is diffused (i.e., not localized over the illuminated region perfectly). In this way, the effective size of the illuminated spot becomes wider due to pre-illumination with CW-optical beam.

#### 3.1.1. Variation of PA-Signal Enhancement with Variation in Different Parameters

Figure 4a presents the variation of *P*<sup>0</sup> with concentration ([*C*]) of sample for both of the two cases (with pre-illumination and without pre-illumination) where *P*<sup>0</sup> is measured as an average over full-width at 75% of maximum as it is done in the previous case (shown in Figure 3e). In the experiments, we employed CW-laser with power (*P*) ∼ 60 mW and time duration of pre-illumination (*t*) ~20 min. Figure 4b depicts the variation of *P*<sup>0</sup> with respect to *μ<sup>a</sup>* (which is estimated from [*C*]), where *P* = 60 mW and *t* = 20 min being employed as the experimental parameters. Variation in *P*(*enh*) <sup>0</sup> with pre-illumination, which is estimated using Equation (5), is also depicted. It is observed that *<sup>P</sup>*<sup>0</sup> and *<sup>P</sup>*(*enh*) 0 increases linearly with concentration. The observed linearity may be because of an increase in deposition of optical energy as well as mechanical coupling of constituent particles with an increase in [*C*]. Variation in *<sup>P</sup>*<sup>0</sup> and *<sup>P</sup>*(*enh*) <sup>0</sup> with CW-optical power used for preillumination is shown in Figure 4c, (where [*C*] = 7 mM and *t* = 20 min being employed as the experimental parameters), while *P*(*enh*) <sup>0</sup> with time interval (*t*) of pre-illumination of CW-optical beam is depicted in Figure 4d (where [*C*] = 7 mM, *P* = 60 mW). Without pre-illumination, the optical power of incident pulse-beam remains the same so that *P*<sup>0</sup> remains unchanged (which are depicted by blue-color dotted lines in Figure 4c,d). In this case, *<sup>P</sup>*<sup>0</sup> and *<sup>P</sup>*(*enh*) <sup>0</sup> increases non-linearly, which is followed by saturation at a particular time-point *t* ∼ 20 min. This is due to saturation in the deposition of optical energy and, thus, thermo-elastic expansion. Saturation in *P*(*enh*) <sup>0</sup> suggests that, for practical applications, selection of stability point of signal enhancement is demanded. In our validation study, with experiments being conducted in tissue-mimicking Agar phantom and chicken tissue, pre-illumination of the sample with *t* = 20 min was performed.

**Figure 4.** (**a**) Variation of measured *<sup>P</sup>*<sup>0</sup> and *<sup>P</sup>*(*enh*) <sup>0</sup> with concentration ([*C*]) of sample solution. In the experiments, we employed a continuous wave (CW)-laser of power (*P*) ∼ 60 mW and time duration of pre-illumination (*t*) <sup>∼</sup> <sup>20</sup> min). (**b**) Variation of measured *<sup>P</sup>*<sup>0</sup> and *<sup>P</sup>*(*enh*) <sup>0</sup> with optical absorption coefficient (where *<sup>P</sup>* <sup>=</sup> <sup>60</sup> mW and *<sup>t</sup>* <sup>=</sup> <sup>20</sup> min). (**c**) Variation of measured *<sup>P</sup>*<sup>0</sup> and *<sup>P</sup>*(*enh*) <sup>0</sup> with optical power of CW-optical beam (where [*C*] <sup>=</sup> <sup>7</sup> mM and *<sup>t</sup>* <sup>=</sup> <sup>20</sup> min), (**d**) Variation of *<sup>P</sup>*<sup>0</sup> and *<sup>P</sup>*(*enh*) <sup>0</sup> with time-interval of pre-illumination (where [*C*] = 7 mM, *P* = 60 mW). In all of the experiments, we employed pulsed laser beam of wavelength, *λ* ∼ 670 nm.

#### 3.1.2. Variation of Temperature Raise with Variation in Different Parameters

Figure 5 presents the experimental results for characterization studies of the variation of temperature-rise (Δ*T*) with various physical parameters of interest. A detailed description of the experimental set-up employed for this study is provided in supplementary (see Figure S4). Figure 5a depicts the variation of Δ*T* with respect to the time interval

of pre-illumination (*t*) of the CW-optical beam where concentration ([C]) of methylene blue is considered as an experimental parameter. Meanwhile, the variation of Δ*T* with *t*, where optical power of CW-optical beam is considered as an experimental parameter, is shown in Figure 5b. In these figures, it is observed that the temperature raise increases non-linearly and attains saturation at a certain point of the time interval of pre-illumination (∼20 min). Inset in Figure 5b shows a curve-fit, for 80 mW power of CW-laser, using Δ*T* = Δ*Tsat* <sup>1</sup> <sup>−</sup> *<sup>e</sup>*−*γ<sup>t</sup>* (where Δ*Tsat* is the maximum temperature raise attained at saturation and γ is a constant that characterizes the growth rate) and it is found that maximum temperature raise is 9.13 ◦<sup>C</sup> while growth rate, *<sup>γ</sup>* ∼ 0.3909 sec<sup>−</sup>1. From the figures, it is observed Δ*Tmax* and *γ* are dependent on [*C*] and optical power of incident CW-beam. Figure 5c shows the variation of temperature raise with [*C*], which are obtained from Figure 5a with *t* = 20 min. From Figure 5c, we obtained variation of Δ*T* with *μ<sup>a</sup>* and it is shown in Figure 5d. Similarly, the variation of Δ*T* with optical power of CW-beam at *t* = 20 min is given in Figure 5e. In the figures (Figure 5a–e), *P*(*enh*) <sup>0</sup> , which is presented in Figure 4, is also presented for comparison of Δ*T* and *P*(*enh*) <sup>0</sup> . This linearity in Δ*T*—which is due to increase and subsequent saturation in the absorption of optical energy—is in agreement with that for *P*(*enh*) <sup>0</sup> . From these results (Figures 4 and 5), one may conclude that *P*(*enh*) <sup>0</sup> is dominantly contributed from Δ*T* due to CW-optical pre-illumination in support of our theoretical hypothesis given by Equation (4). This is in agreement with the (simulation and experimental) study, reported by Mahmood et al. [20], that demonstrated *Γ* and, thus, *P*<sup>0</sup> are dependent on temperature. The time duration for achieving the saturation point of signal enhancement is characteristic of samples, and it can be considered as a preparatory requirement for imaging.

**Figure 5.** (**a**) Variation of temperature-raise (Δ*T*) with respect to time-interval (*t*) of pre-illumination of CW optical beam (with concentration as experimental parameter (for *P* = 60 mW)). (**b**) Variation of Δ*T* against *t* with optical power (*P*) of CW laser beam as an experimental parameter (for [*C*] = 7 mM), (**c**) Variation of Δ*T* against concentration (where *P* = 60 mW, *t* = 20 min were employed), (**d**) Variation of Δ*T* against optical absorption coefficient (where *P* = 60 mW and *t* = 20 min were employed), and (**e**) Variation of Δ*T* against optical power of CW beam for pre-illumination (where [*C*] = 60mW and *t* = 20 min were employed).

#### *3.2. Variation of Optical Extinction Coefficient of Methylene Blue with Temperature*

Figure 6 gives the experimental results for the characterization study of variation of optical extinction coefficient (*ε*) with temperature. For this study, we employed a UV-VIS as spectrum analyzer (PerkinElmer, Lambda 750) and *ε* is estimated from absorbance spectrum using *ε* = *Absorbance <sup>d</sup>* [*C*] where *d* is the (inner) thickness of sample holding cuvette (∼1 cm) where methylene blue solution is filled in. The sample is heated before introduction

to the UV-VIS spectrum analyzer. Temperature (*T*) is measured before and after the experiments, and its mean value is considered as the temperature measurement for the particular spectrum analysis experiment. Figure 6a depicts the variation of *ε* against wavelength (*λ*) while Figure 6b gives the variation of *ε* against *T*. From the figure, it is observed that *ε* varies linearly with *T* where intercept and slope of the linear curve are characteristics of *λ*. From this experimentally estimated *ε*, for a given *T*, we estimated the contribution of optical absorption to the enhancement of photoacoustic signal strength (given in Equations (4) and (5)). However, thermal expansion coefficient at any given temperature *T* was estimated using a linear fit of *β* with *T* given in Table 1. Using these measurements of *ε* and *β* (in Equations (4) and (5)), we estimated *P*(*enh*) <sup>0</sup> , which we consider as analytical calculation (in Figure 7), wherein thermal perturbation (Δ*T*) is adopted from the measurements corresponding to Figure 5.

**Figure 6.** Variation of extinction coefficient (**a**) with respect to optical wavelength (*λ*) (at various thermodynamic temperature (*T*)). (**b**) Variation of optical extinction coefficient (*ε*) against (*T*) for different *λ*. Inset in (**b**) presents variation of *ε* against *T* for *λ* = 670 nm, which is the optical wavelength of pulsed laser beam used for imaging (in our present study). In the experiments, we used methylene blue solution with [*C*] ∼ 2 μM.

#### *3.3. Comparison of Numerical Simulation and Experimental Results, and Validation of Theoretical Hypothesis*

Figure 7 presents the results of numerical simulation studies—for validation of our proposed theoretical hypothesis (presented in Section 2.1)—in comparison to the experimental results (given in Figure 4). In the meantime, *P*(*enh*) <sup>0</sup> , which is estimated theoretically or analytically using Equations (4) and (5), is also included. Variation of *P*(*enh*) <sup>0</sup> with respect to [*C*], *μa*, optical power of CW-beam (used for pre-illumination), and Δ*T* are shown in Figure 7a–d. We estimate the slopes for variation of *P*(*enh*) <sup>0</sup> with: (i) concentration as 6.24/mM (from experiments) against 5.52/mM (from analytical calculation) and 5.92/mM (from simulations), (ii) optical absorption coefficient as 0.103 cm (from experiments) against 0.091 cm (from analytical calculation) and 0.098 cm (from simulations), (iii) optical power of CW-beam as 0.439/mW (from experiments) against 0.406/mW (from analytical calculation) and 0.442/mW (from simulations), (iv) temperature raise (due to concentration or optical power) as 5.97 ◦C−<sup>1</sup> (from analytical calculation) and 6.37 ◦C−<sup>1</sup> (from simulations). In the figures, it is observed that slopes of the variation of *P*(*enh*) <sup>0</sup> with the physical parameters of interest, obtained from experiments and numerical simulations, are in good agreement. This demonstrates that the proposed hypothesis of the dependence of *<sup>P</sup>*<sup>0</sup> and *<sup>P</sup>*(*enh*) <sup>0</sup> on *T* (as it is given in Equations (4) and (5)) is validated. A deviation in absolute measurements of enhancement, as it is observed in Figure 7, may be due to lower in measurement of temperature of the imaging sample at a pre-specified region of interest—over which CW laser beam illuminates—resulting from tips of the (thermocouple) temperature sensor not being situated exactly at the point of illumination while attempting to prevent direct exposure of the tips to the incident light beam. To note, direct exposure of light to thermocouple tips

results in the heating of the tips, which gives inaccurate measurements of the temperature of the medium.

**Figure 7.** Comparison of the results obtained from experiments, theoretical hypothesis, and numerical simulation. (**a**) Variation of *<sup>P</sup>*(*enh*) <sup>0</sup> with concentration (we employed *P* = 60 mW and *t* = 20 min), (**b**) Variation of *<sup>P</sup>*(*enh*) <sup>0</sup> with optical absorption coefficient (*P* = 60 mW and *t* = 20 min were employed). (**c**) Variation of *<sup>P</sup>*(*enh*) <sup>0</sup> with optical power (*where* [*C*] = 7mM *and t* = 20 min). (**d**) Experimental measurements in *<sup>P</sup>*(*enh*) <sup>0</sup> with respect to temperature raise (Δ*T*), with Δ*T* being obtained from variation of optical power as well as concentration. The wavelength of the pulsed laser used for experiment is 670 nm.

#### *3.4. Validation of the Proposed Technique for Pre-Clinical/Clinical Studies*

To validate the proposed technique for enhancement of PA-signal strength, in preclinical and/or clinical studies, experiments were conducted in tissue-mimicking Agar-phantom as well as biological tissue (chicken breast collected from supermarket). Experimental results are depicted in Figure 8 (for Agar phantom) and Figure 9 (for chicken tissue). Figure 8a,b give 3D images representative of *P*<sup>0</sup> obtained without and with pre-illumination. Figure 8c,d depict 2D images (corresponding to *y* = 3 in 3D images). Line plots, showing the variation of *<sup>P</sup>*<sup>0</sup> and *<sup>P</sup>*(*enh*) <sup>0</sup> along marked lines indicated in Figure 8b,c, are depicted in Figure 8d. *P*(*enh*) <sup>0</sup> is found to be 48%, which is in agreement with <sup>∼</sup>1.5% ◦C−<sup>1</sup> increase in the Grüneisen parameter with temperature (for biological tissue) [8,20]. It is observed that the measured *P*<sup>0</sup> is significantly enhanced with pre-illumination of CW-laser beam. Figure 8f gives a photograph of Agar-sample where a target (1.2 × 0.5 × 2.2 mm3(*xyz*)) was embedded in a background phantom (24 × 1.5 × <sup>10</sup> mm3). Similar to reported studies [3,5], optical and elastic properties were tailored such that the target has only contrast in *μ<sup>a</sup>* with respect to background (0.02 mm−<sup>1</sup> (target) against 0.01 mm−<sup>1</sup> (background)). Figure 9a,b gives 3D images, while Figure 9c,d give 2D images, representative of *P*<sup>0</sup> obtained without and with pre-illumination. Line plots, showing the variation of *<sup>P</sup>*<sup>0</sup> and *<sup>P</sup>*(*enh*) <sup>0</sup> along the marked lines indicated in Figure 9b,c, are depicted in Figure 9d. *P*(*enh*) <sup>0</sup> is estimated and found to be ∼40%, which is a significant enhancement with pre-illumination of CW-laser beam. Figure 9f gives a photograph of the chicken breast tissue sample where a transparent (glass) tube of inner diameter (∼1 mm) is inserted through the chicken breast sample at depth ∼2.5 mm. The tube was filled with methylene blue, whereby mimicking deep-seated blood vessels in (chicken breast) tissue sample. In Figure 8, one observes that PA-representative image of the rectangular-shaped target appears to be circular (against appearing as a line in

Figures 8c and 9c). This is because of the pre-illumination of CW-laser beam, which is circular in cross-section. These results, with experiments being conducted in (chicken) tissue sample, demonstrate the practical applicability of the proposed photo-thermal technique to pre-clinical and/or clinical studies.

**Figure 8.** 3D images representative of *P*<sup>0</sup> obtained without (**a**) and with pre-illumination (**b**), while (**c**,**d**) give corresponding 2D images for plane (at *<sup>y</sup>* <sup>=</sup> 3). (**e**) Variation of *<sup>P</sup>*<sup>0</sup> and *<sup>P</sup>*(*enh*) <sup>0</sup> along marked lines indicated in (**c**,**d**). Photograph of tissue-mimicking Agar-sample (**f**) where marked line indicates region for raster scanning. Arrowheads in both sides of the scale bar (being included in the figure) indicate measure of entire length of the image along *x* -axis.

#### *3.5. Discussion*

The enhancement of PA-signal strength can be contributed from various factors (including change in specific heat capacity and isothermal compressibility), which are neglected in view of the argument described in Section 2.1. In our study, we assume that signal strength enhancement is contributed from a raise in thermodynamic equilibrium temperature (*T*) and, thus, thermal expansion coefficient (*β*) and optical extinction coefficient (*ε*) of the tissue target. Moreover, the CW-laser beam enhances (but does not induce itself) PA-signal strength induced by pulse-laser excitation. Even though the CW-laser illumination increases the optical energy being irradiated to sample, the total optical energy is restricted within this safety limit (∼ <sup>20</sup> mJ/cm2, FDA, Silver Spring, MD, USA [48]). Experiments were carried out to compare efficiency in the enhancement of PA-signal strength with CW-laser pre-illumination and pulse-laser excitation. Experimental results demonstrate that it is more efficient (46% (for CW-laser pre-illumination) against 98% (for pulsed-laser excitation)) to increase PA-signal strength with an increase of pulsed-laser energy in comparison to a similar amount of increase in CW-laser energy, which is in agreement with the reported study [27]. Even though pulse-laser is more efficient to enhance PA-signal strength, the price of pulse-laser increases drastically as the output pulse energy increase, which then leads to an increase in the overall cost of the PAI system. In this way, the proposed technique paves a cost-effective way to improve PA-signal strengths. Again, from our study, it is observed that a temperature raise due to illumination of CW-optical beam is ∼10.5 ◦C, which is very low in comparison to temperature increase as in the case of photo-therapy (where targeted temperature measures ∼40 ◦C [28,43,49]), which remains as

a standard clinical therapy. To the best of our knowledge, this pre-illumination technique may not impose any significant side effects in tissue (in in-vivo study and its applications).

**Figure 9.** 3D images representative of *P*<sup>0</sup> obtained without (**a**) and with pre-illumination (**b**), while (**c**,**d**) give corresponding 2D images for plane (at *<sup>y</sup>* <sup>=</sup> 3). (**e**) Variation of *<sup>P</sup>*<sup>0</sup> and *<sup>P</sup>*(*enh*) <sup>0</sup> along marked lines indicated in (**c**,**d**). Photograph of chicken breast tissue (**f**) in which transparent glass of inner diameter (∼1 mm) filled with methylene blue, thereby, mimicking deep-seated blood vessels is inserted at depth ∼2.5 mm (as it is observed from attached scale). Marked line indicates region for raster scanning. Arrow heads in both sides of scale bar (being included in the figure) indicate measure of entire length of the image along *x* -axis.

Refs. [14,15] demonstrated that with an increase in temperature ∼37 ◦C (from ∼33 ◦C to ∼70 ◦C), PA- signal strength is enhanced by ∼120% (from ∼5 mV to ∼11 mV) for soft tissue (chicken breast) and ∼116% (from ∼6 mV to ∼13 mV) (for porcine liver) [14,15]. For porcine muscle, enhancement in PA-signal strength is ∼42% corresponding to change in temperature ∼9 ◦C [28]. The enhancement of PA-signal strength we achieved in our study is relatively high in comparison to that of Refs. [14–16]. This may be because of different experimental conditions. In our study, a deep-seated target of interest is selectively illuminated by CW-laser beam and, thus, the target is selectively heated while in Ref. [14], the entire sample is immersed inside a heating bath, and the temperature is raised as a whole including background (that also generate PA-waves), i.e., the difference in the enhancement of PA-signal strength may be due to reduce in the obtainable SNR resulted from the increasing background signal [12]. In Ref. [21], an optical fluence ∼<sup>5</sup> mJ/cm2 was employed against ∼<sup>15</sup> mJ/cm<sup>2</sup> in our present study that gives a difference of 6 times in enhancement of signal strength.

One of the drawbacks of our proposed photo-thermal technique is that enhancement of PA-signal strength is at the cost of degradation in the obtainable spatial resolution. However, the degradation in spatial resolution is ~9%, which is much lower than the enhancement in PA-signal strength (~70% (methylene blue), ∼48% (Agar phantom)), and ~40% (chicken tissue)). Our study suggests that one can employ a CW-optical beam to

perform pre-illumination specifically over a certain point deep inside the tissue without disturbing the intervening tissue medium so as to further enhance the PA-signal strength from a specific deep-seated target reducing the noise from the background or surroundings. LED-based PAI is recently gaining popularity in a wide range of superficial imaging applications, and it holds greater potential in clinical translation because of its portability and affordability. The proposed technique is a promising method to image deep-seated tissues with LED-based systems and accelerates its clinical translation. Strengthening the intensity of boundary measured signals—for improving signal detectability—is a critical factor in imaging (in general) and PA-imaging (in particular). In addition, the obtainable signal contrast between targets (including contrast agents) and the background is more relevant to the improvement of imaging quality. Our proposed photo-thermal technique offers improvement in signal contrast and, thus, the image quality.

#### **4. Conclusions**

In conclusion, we demonstrate an optothermal-based technique for enhancement of obtainable PA-signal strength that is, experimentally, facilitated by pre-illumination with CW-optical beam (in addition to pulse-optical beam) in imaging sample. We propose a theoretical hypothesis and it is validated by numerical simulation studies being performed using k-wave toolbox. Experimental studies were conducted in tissue-mimicking Agar phantom and ex-vivo animal tissue (chicken breast) samples. Experimental results demonstrate that a significant enhancement (up to ∼70% (methylene blue), ∼48% (Agar phantom), and ∼40% (chicken tissue)) in the measured PA-signal strength can be achieved by our proposed photo-thermal technique. This unique (non-invasive and non-destructive) technique for enhancement of PA-signal strength will have a significant impact in PAimaging while improving not only achievable penetration depth but also SNR and, hence, accuracy in quantitative measurement of vital patho-physiological parameters. Particularly, this technique can address the pertaining challenge associated with weak PA-signal from deep-seated tissue regions, which is the major issue associated with the clinical translation of the LED-based PAI systems.

**Supplementary Materials:** The following are available online at https://www.mdpi.com/1424-822 0/21/4/1190/s1, Figure S1: 2D images representative, for empty cuvette as sample specimen for imaging, of PA-signal strength obtained without and with pre-illumination, Figure S2: 2D images representative, for water (filled in cuvette) as sample specimen for imaging, of PA-signal strength obtained without and with pre- illumination, Figure S3: 2D images (corresponding to Figure 3) and line plots along different direction for the calculation of FWHM, Figure S4: A schematic diagram of experimental set-up for the study of variation of temperature rise.

**Author Contributions:** Conceptualization, A.T., J.M. and M.S.S.; Data curation, A.T. and S.P.; Formal analysis, A.T. and J.M.; Funding acquisition, M.S.S.; Investigation, A.T. and S.P.; Methodology, A.T. and M.S.S.; Project administration, M.S.S.; Resources, S.P. and J.M.; Software, A.T.; Supervision, M.S.S.; Validation, A.T.; Visualization, A.T.; Writing—original draft, M.S.S.; Writing—review & editing, A.T., S.P., J.M. and M.S.S. All authors have read and agreed to the published version of the manuscript.

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

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** The data that supports the findings of the study are provided within the article and the supplementary material.

**Acknowledgments:** The authors want to acknowledge Rinsa S.R., BII Lab, IISER-TVM, for her technical support and suggestions.

**Conflicts of Interest:** The authors have no relevant financial interests in this article and no other potential conflicts of interest to disclose.

#### **Appendix A. Dependence of Initial Pressure Waves on Thermodynamic Parameters for Low Concentration Materials**

For a physical system or medium (including solution and biological tissue), that is constituted by various individual components, its mass density is given by *<sup>ρ</sup>* <sup>≈</sup> <sup>∑</sup>*<sup>N</sup> <sup>i</sup>*=<sup>1</sup> *mi* ∑*<sup>N</sup> <sup>i</sup>*=<sup>1</sup> *Vi* (where *mi* and *Vi* are mass and volume of *i th* component, and *N* is the total number of individual components). Under physical condition of low concentration, as it is the case of our present study (methylene blue solution with concentration of methylene (∼mM) being dissolved in water (as a solvent) and tissue-mimicking Agar phantom with concentration of India ink (∼3–5 <sup>μ</sup>l) in Agar gel (∼<sup>100</sup> gms) [3–5]), one can approximate <sup>∑</sup>*<sup>N</sup> <sup>i</sup>*=<sup>1</sup> *mi* <sup>≈</sup> *<sup>m</sup>*(*water*) and ∑*<sup>N</sup> <sup>i</sup>*=<sup>1</sup> *Vi* <sup>≈</sup> *<sup>V</sup>*(*water*) (for methylene blue solution), i.e., *<sup>ρ</sup>* <sup>≈</sup> *<sup>ρ</sup>*(*water*) (for methylene blue solution) and *<sup>ρ</sup>* <sup>≈</sup> *<sup>ρ</sup>*(*agar*) (for tissue-mimicking Agar phantom added with India ink as an optical absorber). Similarly, specific heat capacity can also be approximated as *cV* <sup>=</sup> <sup>∑</sup>*<sup>N</sup> <sup>i</sup>*=<sup>1</sup> *Qi* ∑*<sup>N</sup> <sup>i</sup>*=<sup>1</sup> *mi* <sup>≈</sup> *<sup>Q</sup>*(*water*) *<sup>m</sup>water*Δ*<sup>T</sup>* = *c* (*water*) *<sup>V</sup>* (where *Qi* is the heat absorbed by *i th* components of the medium). In the meantime, optical properties of the system is given by collective effects of the individual components as linear combination [50], i.e., *μ<sup>a</sup>* = ∑*<sup>N</sup> <sup>i</sup>*=<sup>1</sup> *μ<sup>i</sup> <sup>a</sup>* = ∑*<sup>N</sup> <sup>i</sup>*=<sup>1</sup> *ε<sup>i</sup> Ci* (where *ε<sup>i</sup>* and [*C<sup>i</sup>* ] are extinction (optical absorption) coefficient and concentration of *ith* component, respectively). For methylene blue solution, we can approximate that *<sup>μ</sup><sup>a</sup>* <sup>=</sup> *<sup>μ</sup>*(*water*) *<sup>a</sup>* <sup>+</sup> *<sup>μ</sup>*(*methy*) *<sup>a</sup>* <sup>=</sup> *<sup>μ</sup>*(*water*) *<sup>a</sup>* <sup>+</sup> *<sup>ε</sup>*(*methy*)[*C*(*methy*)] <sup>≈</sup> *<sup>μ</sup>*(*methy*) *<sup>a</sup>* (for *<sup>μ</sup>*(*water*) *<sup>a</sup> <sup>μ</sup>*(*methy*) *<sup>a</sup>* [37]). Similarly for Agar phantom, one can approximate *<sup>μ</sup><sup>a</sup>* <sup>=</sup> *<sup>μ</sup>*(*agar*) *<sup>a</sup>* <sup>+</sup> *<sup>μ</sup>*(*ink*) *<sup>a</sup>* <sup>=</sup> *<sup>ε</sup>*(*agar*)[*C*(*agar*)] + *<sup>ε</sup>*(*ink*)[*C*(*ink*)] <sup>≈</sup> *<sup>ε</sup>*(*ink*)[*C*(*ink*)] <sup>≈</sup> *<sup>μ</sup>*(*ink*) *<sup>a</sup>* (for *<sup>ε</sup>*(*agar*) *<sup>ε</sup>*(*ink*) [51]). Again, from Dalton's law of partial pressures, total pressure of a thermodynamic system is given by an algebraic sum of partial pressures imparted by the individual components, i.e., *P* = ∑*<sup>N</sup> <sup>i</sup>*=<sup>1</sup> *Pi* (where *Pi* is the partial pressure of *i th* component) and it can be approximated as *<sup>P</sup>* <sup>≈</sup> *<sup>P</sup>*(*water*) (for low concentration methylene solution) or *<sup>P</sup>* <sup>≈</sup> *<sup>P</sup>*(*agar*) (for Agar phantom). Further, under thermodynamic equilibrium condition, there exists a uniform distribution of temperature (*T*) at any arbitrarily chosen (spatial) point ( → *r* ) over the entire space of the thermodynamic system, i.e., *T* is independent of spatial position ( → *r* ) and, thus, measurements of *T* for all of the individual constituent components are the same. From the thermodynamic relations of *<sup>κ</sup>* and *<sup>β</sup>*, we can approximate *<sup>κ</sup>* <sup>=</sup> <sup>−</sup> <sup>1</sup> *V ∂V ∂P <sup>T</sup>* ≈ − <sup>1</sup> *V*(*water*) *∂V*(*water*) *∂P*(*water*) *<sup>T</sup>* <sup>=</sup> *<sup>κ</sup>*(*water*) and *<sup>β</sup>* <sup>=</sup> <sup>−</sup> <sup>1</sup> *V ∂V ∂T <sup>P</sup>* ≈ − <sup>1</sup> *V*(*water*) *∂V*(*water*) *∂T*(*water*) *<sup>P</sup>* <sup>=</sup> *<sup>β</sup>*(*water*) (for methylene blue solution) while *<sup>κ</sup>* <sup>=</sup> <sup>−</sup> <sup>1</sup> *V ∂V ∂P <sup>T</sup>* ≈ − <sup>1</sup> *V*(*water*) *∂V*(*agar*) *∂P*(*agar*) *<sup>T</sup>* <sup>=</sup> *<sup>κ</sup>*(*agar*) and *<sup>β</sup>* <sup>=</sup> <sup>−</sup> <sup>1</sup> *V ∂V ∂T <sup>P</sup>* ≈ <sup>−</sup> <sup>1</sup> *V*(*agar*) *∂V*(*agar*) *∂T*(*agar*) *<sup>P</sup>* <sup>=</sup> *<sup>β</sup>*(*agar*) (for Agar phantom). Now, for the case of low concentration, we can write Equation (A1) as:

$$P\_0 = \frac{\beta^{(water)}}{\kappa^{(water)}} \frac{1}{\rho^{(water)} c\_V^{(water)}} \mu\_a^{(methyl)} \phi \tag{A1}$$

#### **References**


## *Communication* **Tomographic Ultrasound and LED-Based Photoacoustic System for Preclinical Imaging**

#### **Kalloor Joseph Francis 1,\* , Richell Booijink 2, Ruchi Bansal <sup>2</sup> and Wiendelt Steenbergen 1,\***


Received: 17 April 2020; Accepted: 11 May 2020; Published: 14 May 2020

**Abstract:** Small animals are widely used as disease models in medical research. Noninvasive imaging modalities with functional capability play an important role in studying the disease state and treatment progress. Photoacoustics, being a noninvasive and functional modality, has the potential for small-animal imaging. However, the conventional photoacoustic tomographic systems use pulsed lasers, making it expensive, bulky, and require long acquisition time. In this work, we propose the use of photoacoustic and ultrasound tomographic imaging with LEDs as the light source and acoustic detection using a linear transducer array. We have demonstrated full-view tomographic imaging of a euthanized mouse and a potential application in liver fibrosis research.

**Keywords:** tomography; photoacoustic; ultrasound; small animal; liver; fibrosis

#### **1. Introduction**

Studying different diseases and developing new drugs in a controlled environment is vital in biomedical research. Small animals are widely used for this purpose, as they allow for controlled disease staging and evaluating the performance of drugs through histopathological validation [1]. Longitudinal monitoring of disease progression and treatment response to drugs can improve the outcome of preclinical studies and can reduce the number of laboratory animal deaths. Imaging modalities can be used for longitudinal monitoring of small animal models. However, there are limitations in using conventional imaging modalities such as MRI, CT, and ultrasound for small animal imaging [2]. Micro MRI is costly and has a slow data acquisition. Micro CT and PET, on the other hand, use ionizing radiation, which hinders longitudinal observations [2]. Ultrasound (US) is a noninvasive and real-time imaging modality but, being a structural imaging modality, in most cases it cannot quantify disease state. Photoacoustic (PA) imaging is a new modality that has functional and molecular capability while being noninvasive and real-time. Thus, PA is considered to be ideal for small animal imaging [3].

PA imaging utilizes pulsed light excitation to induce a temperature rise in optical absorbing structures inside the tissue resulting in thermoelastic expansion and acoustic wave generation. These acoustic waves are detected for imaging [4]. The advantage of PA imaging is that with optical excitation and acoustic detection it combines optical contrast at ultrasound resolution. Additionally, the use of ultrasound transducers enables us to combine PA imaging with conventional US imaging providing co-registered structural and functional imaging of the tissue. Several PA and US imaging systems were successfully demonstrated for small-animal whole-body imaging [3,5,6]. However, the use of a pulsed laser source in these systems not only makes them expensive but demands laser safe small animal labs, eye safety goggles, and additional manpower to operate the system. Therefore, for the wide use of PA imaging in small animal labs, there is a requirement for low cost, compact, safe to use tomographic systems which can be operated by a non-expert. Recent developments in LED-based PA imaging, being compact, low-cost, and safe to use, offer promising avenues to fill this gap [7]. LED-based handheld PA systems were used previously for small animal studies for imaging superficial structures such as tumors [8], wounds [9], and knee joints [7,10]. A limitation of the hand-held PA system using a linear transducer array is the limited view of the target tissue due to the directional sensitivity of the transducer. Additionally, with a small number of LED elements arranged on either side of the transducer, the imaging depth is shallow. We have recently developed a tomographic imaging configuration using a linear transducer array and four LED arrays, to overcome the limited view and to improve the imaging depth [11,12]. The system was originally developed for imaging finger joints for diagnosis and monitoring of rheumatoid arthritis [11].

In this study, we propose the application of our tomographic US and LED-based PA system for preclinical research. First, we demonstrate full-view tomographic imaging of the abdominal region of a mouse. We also compare the results with B-scan images obtained using a handheld probe. Further, we present a potential application of the system in liver fibrosis research. A large number of preclinical studies are currently being performed in small animals to develop antifibrotic therapies. However, the outcome of the preclinical study relies on endpoint histopathological analysis. A noninvasive imaging technique can provide longitudinal monitoring of animals and can improve the study outcome. We present the use of noninvasive and low-cost US and PA tomographic imaging system for liver imaging and compared the outcome with histology images.

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

An LED-based photoacoustic and ultrasound imaging system, AcousticX (Cyberdyne Inc., Japan), was used in this work. Four LED arrays having 576 elements (36 × 4 array) were used as the light source. We used LEDs having a wavelength of 850 nm, and each array has a pulse energy of 200 μJ with a pulse duration of 70 ns. We used a linear transducer array (128 elements) with a center frequency 7 MHz with 80% bandwidth for acoustic detection. A tomographic imaging configuration with four LED arrays and transducer scanning around the sample was used in this study.

PA imaging using a linear transducer array suffers from low image quality due to incomplete acoustic detection. To overcome the limited view problem in linear array-based imaging, we recently developed a tomographic imaging configuration [11]. The system consists of an imaging probe with a linear transducer array and four LED arrays as shown in Figure 1a. Two LED arrays were placed parallel to the long axis of the transducer, geometrically pointing towards the focus of the transducer (20 mm depth). The other two LED arrays were placed on either side of the transducer along its short axis at an angle of 105◦ with the transducer surface. These two LED arrays were also placed at an angle of 5◦ with the imaging plane to minimize the acoustic reflections being detected. In this way, the object to be imaged is illuminated from three sides. The probe is then scanned around the object for tomographic imaging. Based on a simulation study, we have estimated that 16 angular views with an angular step of 22.5◦ is sufficient to form a full-view tomographic image [11]. The illumination configuration was also developed to provide a significant overlapping region between two angular scans. Interested readers can refer to the work in [11,12] for more details. A 3D printed holder encompassing the transducer and LED arrays to form the imaging probe is shown in Figure 1b. For small animal imaging, the probe was mounted on a motorized linear and circular scanning system, installed on top of a water tank as shown in Figure 1c. A 3D printed small animal holder was used to keep the mouse in place for imaging (Figure 1c).

Male Balb/c mice (8 weeks old) were used in this study. All the animals used in the study received ad libitum normal chow diet and normal water, and were housed with 12h-light/12h-dark cycle. All the animal experiments were carried out strictly according to the ethical guidelines for the Care and Use of Laboratory Animals (Utrecht University, The Netherlands). A proof-of-concept study was performed whereby liver fibrosis was induced in one mouse by a single injection of 1ml/kg of carbon tetrachloride (CCl4), and the control animal received olive oil. Mice were euthanized and imaged immediately using the PA and US tomographic system. The hair with the outer skin layer of the animals was removed around the abdominal region before imaging. After the imaging experiments, the livers of the animals were collected and fixed with cryomatrix in isopentane. Fixed livers were sectioned into 5 μm sections and immunohistological analysis using collagen I (fibrosis marker) antibody was performed.

**Figure 1.** Small animal tomographic imaging system. (**a**) Schematic of the set-up with the imaging probe scanning around the mouse in a water tank. (**b**) Imaging probe with linear transducer array and four LED arrays in a 3D printed holder. (**c**) Photograph of the imaging set-up showing mouse holder, imaging probe and the scanning stages.

Two scanning techniques were used in this study, involving a combination of circular and linear scans. To obtain a complete tomographic image, a circular scanning of the entire 360◦ was performed. Combined PA and US imaging were performed at each angular step. Typically, a circular scan takes 42.5 secs. However, due to memory limitations of the system, the RF data was saved after each 180◦ scan and combined in the postprocessing stage. Multiple tomographic images of the abdominal area were obtained by translating the probe to a different location and repeating the circular scan. In the second experiment, the liver region was considered as the region of interest. Here, a limited view tomographic imaging was performed by spatial compounding of B-scan images from three angles (20◦ steps). In this case, linear scans were performed for the entire length (15 mm length) of the liver at each angular step. Each linear scan took 8.5 secs and the 20◦ circular scan took 2.5 secs. The US and PA B-scan images were formed by delay and sum algorithm. For speckle reduction and to improve the contrast of the US image, Bayesian mean filtering using an open source software was used [13]. To form tomographic images, the B-scan images from both US and PA imaging were rotated to the acquired angle and spatially compounded using a custom developed MATLAB program [14].

#### **3. Results and Discussion**

In Figure 2a–c, co-registered ultrasound and photoacoustic B-scan images of a mouse abdomen, obtained from three different angles, have been presented. The images demonstrate the limited view problem in small animal imaging using a linear transducer array. The limited view arises from the directional sensitivity of the transducer resulting in structures oriented away from the transducer being undetected. The tomographic images obtained from spatial compounding of the B-scan images are shown in Figure 2d–f. The ultrasound image shows the capability of the tomographic system to

image major abdominal organs such as the liver, kidney, spleen, stomach, and intestines. Additionally, anatomical structures such as skin and spinal cord are also imaged in this mode. Kidney and lobes of the liver can be seen as hypoechogenic regions, while the stomach, spleen, and intestine are visible as hyperechogenic regions. In PA images major blood vessels are visible beneath the skin and around the spinal cord. In addition to that, high PA signals were also observed from the liver, kidney, and spleen. For the imaging configuration used in this work, it was estimated that PA signal strength drops to 30% at a depth of 9.7 mm in soft tissue [11]. This finding also holds in this study, as most of the structures visible in PA images are within the 10 mm depth from the skin surface. Illumination from three sides and spatial compounding from several angles enable PA tomographic imaging to visualize most of the absorbing structures in the small animal. The abdominal diameter of Balb/c mice is mostly less than 30 mm and most of the organs of interest are within the 10 mm depth from the skin surface, making the system useful for small animal studies. However, there is indeed a region at the center where the signal to noise ratio is low, which can only be improved with better illumination methods. Another challenge in PA imaging is reflection artifacts from multiple structures inside the small animal body. There are several methods proposed to solve reflection artifacts in PA images [15]. Methods for improving the imaging depth and removing artifacts will be considered in the future. In this study, we have imaged a euthanized animal; however, imaging a live animal in this set-up is also possible by positioning the scanning stages such that the snout of the animal is above the water while the rest of the body immersed and/or using a breathing mask. Imaging speed is another factor that needs to be considered while imaging a live animal. The current scanning time is 42.5 secs, which is longer when compared with the state-of-the-art system using a full ring transducer [3]. However, it is possible to synchronize the breathing cycle of the animal with the acquisition and by using postprocessing to reduce image distortion due to motion while acquisition [16].

#### **Figure 2.** Tomographic photoacoustic and ultrasound imaging. (**a**–**c**) Co-registered ultrasound and photoacoustic B-Scan images of mouse abdomen, acquired from three different angles. Tomographic (**d**) ultrasound, (**e**) photoacoustic, and (**f**) co-registered ultrasound and photoacoustic image. Several organs in the abdominal region indicated by Sc-spinal cord, K-kidney, L-liver, Sp -spleen, St-stomach, and I-intestine. Dashed lines are used to mark the organs in the photoacoustic image.

In Figure 2, the liver is visible in both US and PA. Being a superficial organ, small animal liver imaging is one of the potential applications where the proposed system can be used. To demonstrate

174

this aspect, tomographic imaging of a control and a fibrotic mouse was performed. Figure 3a,h shows five tomographic slices of control and fibrotic mouse, respectively, with the liver in the top region of the image. Remarkable differences can be observed between control and fibrotic liver in both US and PA images. To analyze this difference, let us consider a pair of US and PA images of the liver from both control and fibrotic mouse shown in Figure 3b,c,e,f. The liver is marked with a green boundary. Two lobes of the liver are visible in the US images. In the US image, the difference in control and fibrotic liver is evident with the difference in echogenicity. Echogenicity can be characterized by the brightness level or the mean value of the region of interest. The mean pixel value was calculated with the liver as the region of interest. For the control animal, the calculated mean value was 0.13 ± 0.02, while it was 0.25 ± 0.03 for the fibrotic case. Additionally, heterogeneity computed using the variance was found to be 0.52 ± 0.11 for the control liver and 0.85 ± 0.10 for the fibrotic liver. These metrics were computed over the liver region with multiple tomographic slices stacked and normalized with the maximum value. The difference in structure evidenced in the US image is due to the hepatic nodularity developed due to scar formation in fibrotic liver [17]. In addition to the differences in US images, the PA images in Figure 3c,f show a difference in contrast between control and fibrotic liver. The contrast value computed for the liver region for the control animal was 0.48 ± 0.08 and 0.69 ± 0.08 for the fibrotic case. The primary reason for the high PA contrast in the fibrotic liver is due to the neovasculature in hypoxic liver cells [18]. Angiogenesis or neovascularization plays a crucial role in the progression of liver fibrosis and is also considered as an early indicator of fibrosis thus PA imaging can be used for monitoring the disease progression, while heterogeneity observed with US imaging reflects scar formation at a later stage of the disease [18]. The liver fibrosis was further confirmed using histopathological analysis. Figure 3d,g show histology images of control and fibrotic liver respectively with Collagen I immunostainings. The fibrotic liver (Figure 3g) shows an increased level of collagen bridges resulting from an accumulation of the extracellular matrix protein. These results are also in line with our previous small animal liver fibrosis study [19]. The main improvement we achieved in this work is the superior image quality from the tomographic system and the use of LED-based PA and US imaging. Further works will focus on including more animals in the study and developing a facility to image live animals.

The advantage of using an LED-based PA system is primarily the cost-effectiveness of the system. Nanosecond pulsed lasers and laser diodes are commonly used for tomographic imaging. The cost for a pulsed laser is approximately 70–200k\$, and for laser diodes, with drivers, the cost ranges from 10–25k\$. For LEDs with driving electronics, the cost is as lower as 10–15k\$ [20]. Another benefit is that LED arrays can be integrated within the imaging probe, hence the system can be compact and suitable for small animal labs. Additionally, there is no requirement for a laser-safe imaging facility. However, the limitations are the low pulse energy resulting in low imaging depth, to an extent, this can be addressed using LED arrays with a large number of elements and frame averaging. The longer pulse duration (30–100 ns) achieved with driving circuits for LEDs is sufficient for stress confinement in tomographic imaging applications where the expected resolution is 1mm. However, unlike lasers, wavelength tuning is not possible with LED-based illumination. LED arrays with elements having multiple wavelengths were tested for oxygen saturation imaging compromising pulse energy and imaging depth. Importantly, for different imaging applications, a wide choice of LEDs is available ranging from visible to near-infrared wavelengths (470–980nm). Custom-developed power LED arrays arranged around the small animal along with ring US transducer can be an ideal configuration for small animal imagers in the future.

**Figure 3.** Tomographic imaging of the liver. (**a**) Tomographic photoacoustic and ultrasound images of a control mouse showing the liver at a scanning step of 2 mm. (**b**) Ultrasound image of the liver and (**c**) corresponding photoacoustic image. The green line markers the liver region. (**d**) Histology image of the control liver stained with Collagen I (fibrosis marker). (**e**) Ultrasound image of a mouse liver with fibrosis (CCl4-treated) and (**f**) corresponding photoacoustic image. (**g**) Histology image of the fibrotic liver stained with collagen I (fibrosis marker). (**h**) Tomographic photoacoustic and ultrasound images of the fibrotic mouse showing liver at a scanning step of 2 mm.

#### **4. Conclusions**

In this work, a tomographic ultrasound and LED-based photoacoustic system was demonstrated for small animal imaging. Results show that the tomographic ultrasound imaging can be used to distinguish major organs in the abdominal region of a small animal. The tomographic photoacoustic imaging is capable of imaging an approximate depth of 10 mm from the surface of the skin, allowing imaging of blood vessels and major organs. Limited imaging depth and artifacts are some of the hurdles for photoacoustic imaging in this proposed configuration. The applicability of the proposed imaging system is demonstrated in small animal liver imaging in a fibrotic mouse model. An increase in photoacoustic contrast and ultrasound echogenicity was observed in the fibrotic liver compared to that of the control mouse. The proposed tomographic ultrasound and photoacoustic imaging system offer a cost-effective, compact, and eye-safe alternative for the expensive laser-based system for small animal research.

**Author Contributions:** K.J.F.- conceptualization, methodology, imaging experiments, data processing, analysis, draft preparation; R.B. (Richell Booijink) - histopathological analysis; R.B. (Ruchi Bansal) - conceptualization, small animal handling, review and editing, W.S. - supervision, project administration, funding acquisition, review and editing. All authors have read and agreed to the published version of the manuscript.

**Funding:** This work was funded by the National Centre for the Replacement Refinement and Reduction of Animals in Research (CRACKITRT-P1-3)

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

#### **References**


© 2020 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 (http://creativecommons.org/licenses/by/4.0/).
