J. Imaging, Volume 8, Issue 2 (February 2022) – 35 articles

Space-borne infrared remote sensing specifically for the detection and characterization of fires has a long history in the DLR Institute of Optical Sensor Systems. In the year 2001, the first DLR experimental satellite, Bi-spectral Infrared Detection (BIRD), was launched after an intensive test period with cooled IR sensor systems on airborne systems. The main basis for the development of the FireBIRD mission with the two satellites, Technology Erprobungsträger No 1 (TET-1) and Bi-spectral-Infrared Optical System (BIROS), was the already space-proven sensor and satellite technology with successfully tested algorithms for fire detection and quantification in the form of the so-called fire radiation power (FRP). This paper summarizes the development principles for the IR sensor system of FireBIRD and the most critical design elements of the TET-1 and BIROS satellites, especially concerning the attitude control system—all very essential tools for high-resolution infrared fire monitoring. Key innovative tools necessary to increase the agility of small IR satellites are discussed. Full article

Percutaneous femoral arterial access is daily practice in a variety of medical specialties and enables physicians worldwide to perform endovascular interventions. The reported incidence of percutaneous femoral arterial access complications is 3–18% and often results from suboptimal puncture location due to insufficient visualization of the target vessel. The purpose of this proof-of-concept study was to evaluate the feasibility and the positional error of a mixed-reality (MR)-assisted puncture of the common femoral artery in a phantom model using a commercially available navigation system. In total, 15 MR-assisted punctures were performed. Cone-beam computed tomography angiography (CTA) was used following each puncture to allow quantification of positional error of needle placements in the axial and sagittal planes. Technical success was achieved in 14/15 cases (93.3%) with a median axial positional error of 1.0 mm (IQR 1.3) and a median sagittal positional error of 1.1 mm (IQR 1.6). The median duration of the registration process and needle insertion was 2 min (IQR 1.0). MR-assisted puncture of the common femoral artery is feasible with acceptable positional errors in a phantom model. Future studies should aim to measure and reduce the positional error resulting from MR registration. Full article

Background: Response to induction chemotherapy (IC) has been predicted in patients with sinonasal cancer using early delta radiomics obtained from T1- and T2-weighted images and apparent diffusion coefficient (ADC) maps, comparing results with early radiological evaluation by RECIST. Methods: Fifty patients were included in the study. For each image (at baseline and after the first IC cycle), 536 radiomic features were extracted as follows: semi-supervised principal component analysis components, explaining 97% of the variance, were used together with a support vector machine (SVM) to develop a radiomic signature. One signature was developed for each sequence (T1-, T2-weighted and ADC). A multiagent decision-making algorithm was used to merge multiple signatures into one score. Results: The area under the curve (AUC) for mono-modality signatures was 0.79 (CI: 0.65–0.88), 0.76 (CI: 0.62–0.87) and 0.93 (CI: 0.75–1) using T1-, T2-weighted and ADC images, respectively. The fuse signature improved the AUC when an ADC-based signature was added. Radiological prediction using RECIST criteria reached an accuracy of 0.78. Conclusions: These results suggest the importance of early delta radiomics and of ADC maps to predict the response to IC in sinonasal cancers. Full article

Bone and soft-tissue primary malignant tumors or sarcomas are a large, diverse group of mesenchymal-derived malignancies. They represent a model for intra- and intertumoral heterogeneities, making them particularly suitable for radiomics analyses. Radiomic features offer information on cancer phenotype as well as the tumor microenvironment which, combined with other pertinent data such as genomics and proteomics and correlated with outcomes data, can produce accurate, robust, evidence-based, clinical-decision support systems. Our purpose in this narrative review is to offer an overview of radiomics studies dealing with Magnetic Resonance Imaging (MRI)-based radiomics models of bone and soft-tissue sarcomas that could help distinguish different histotypes, low-grade from high-grade sarcomas, predict response to multimodality therapy, and thus better tailor patients’ treatments and finally improve their survivals. Although showing promising results, interobserver segmentation variability, feature reproducibility, and model validation are three main challenges of radiomics that need to be addressed in order to translate radiomics studies to clinical applications. These efforts, together with a better knowledge and application of the “Radiomics Quality Score” and Image Biomarker Standardization Initiative reporting guidelines, could improve the quality of sarcoma radiomics studies and facilitate radiomics towards clinical translation. Full article

The present paper proposes an implementation of a hybrid hardware–software system for the visual servoing of prosthetic arms. We focus on the most critical vision analysis part of the system. The prosthetic system comprises a glass-worn eye tracker and a video camera, and the task is to recognize the object to grasp. The lightweight architecture for gaze-driven object recognition has to be implemented as a wearable device with low power consumption (less than 5.6 W). The algorithmic chain comprises gaze fixations estimation and filtering, generation of candidates, and recognition, with two backbone convolutional neural networks (CNN). The time-consuming parts of the system, such as SIFT (Scale Invariant Feature Transform) detector and the backbone CNN feature extractor, are implemented in FPGA, and a new reduction layer is introduced in the object-recognition CNN to reduce the computational burden. The proposed implementation is compatible with the real-time control of the prosthetic arm. Full article

Computer vision plays an important role in the accurate foreground detection of medical images. Diagnosing diseases in their early stages has effective life-saving potential, and this is every physician’s goal. There is a positive relationship between improving image segmentation methods and precise diagnosis in medical images. This relation provides a profound indication for feature extraction in a segmented image, such that an accurate separation occurs between the foreground and the background. There are many thresholding-based segmentation methods found under the pure image processing approach. Minimum cross entropy thresholding (MCET) is one of the frequently used mean-based thresholding methods for medical image segmentation. In this paper, the aim was to boost the efficiency of MCET, based on heterogeneous mean filter approaches. The proposed model estimates an optimized mean by excluding the negative influence of noise, local outliers, and gray intensity levels; thus, obtaining new mean values for the MCET’s objective function. The proposed model was examined compared to the original and related methods, using three types of medical image dataset. It was able to show accurate results based on the performance measures, using the benchmark of unsupervised and supervised evaluation. Full article

Developing Field Programmable Gate Array (FPGA)-based applications is typically a slow and multi-skilled task. Research in tools to support application development has gradually reached a higher level. This paper describes an approach which aims to further raise the level at which an application developer works in developing FPGA-based implementations of image and video processing applications. The starting concept is a system of streamed soft coprocessors. We present a set of soft coprocessors which implement some of the key abstractions of Image Algebra. Our soft coprocessors are designed for easy chaining, and allow users to describe their application as a dataflow graph. A prototype implementation of a development environment, called SCoPeS, is presented. An application can be modified even during execution without requiring re-synthesis. The paper concludes with performance and resource utilization results for different implementations of a sample algorithm. We conclude that the soft coprocessor approach has the potential to deliver better performance than the soft processor approach, and can improve programmability over dedicated HDL cores for domain-specific applications while achieving competitive real time performance and utilization. Full article

Advances in computing and AI technology have promoted the development of connected health systems, indirectly influencing approaches to cataract treatment. In addition, thanks to the development of methods for cataract detection and grading using different imaging modalities, ophthalmologists can make diagnoses with significant objectivity. This paper aims to review the development and limitations of published methods for cataract detection and grading using different imaging modalities. Over the years, the proposed methods have shown significant improvement and reasonable effort towards automated cataract detection and grading systems that utilise various imaging modalities, such as optical coherence tomography (OCT), fundus, and slit-lamp images. However, more robust and fully automated cataract detection and grading systems are still needed. In addition, imaging modalities such as fundus, slit-lamps, and OCT images require medical equipment that is expensive and not portable. Therefore, the use of digital images from a smartphone as the future of cataract screening tools could be a practical and helpful solution for ophthalmologists, especially in rural areas with limited healthcare facilities. Full article

This paper presents a photometric stereo (PS) method based on the dichromatic reflectance model (DRM) using colour images. The proposed method estimates surface orientations for surfaces with non-Lambertian reflectance using diffuse-specular separation and contains two steps. The first step, referred to as diffuse-specular separation, initialises surface orientations in a specular invariant colour subspace and further separates the diffuse and specular components in the RGB space. In the second step, the surface orientations are refined by first initialising specular parameters via solving a log-linear regression problem owing to the separation and then fitting the DRM using Levenburg-Marquardt algorithm. Since reliable information from diffuse reflection free from specularities is adopted in the initialisations, the proposed method is robust and feasible with less observations. At pixels where dense non-Lambertian reflectances appear, signals from specularities are exploited to refine the surface orientations and the additionally acquired specular parameters are potentially valuable for more applications, such as digital relighting. The effectiveness of the newly proposed surface normal refinement step was evaluated and the accuracy in estimating surface orientations was enhanced around $30\%$ on average by including this step. The proposed method was also proven effective in an experiment using synthetic input images comprised of twenty-four different reflectances of dielectric materals. A comparison with nine other PS methods on five representative datasets further prove the validity of the proposed method. Full article

The Bologna Archaeological Museum, in cooperation with prestigious Italian universities, institutions, and independent scholars, recently began a vast investigation programme on a group of Egyptian coffins of Theban provenance dating to the first millennium BC, primarily the 25th–26th Dynasty (c. 746–525 BC). Herein, we present the results of the multidisciplinary investigation carried out on one of these coffins before its restoration intervention: the anthropoid wooden coffin of Un-Montu (Inv. MCABo EG1960). The integration of radiocarbon dating, wood species identification, and CT imaging enabled a deep understanding of the coffin’s wooden structure. In particular, we discuss the results of the tomographic investigation performed in situ. The use of a transportable X-ray facility largely reduced the risks associated with the transfer of the large object (1.80 cm tall) out of the museum without compromising image quality. Thanks to the 3D tomographic imaging, the coffin revealed the secrets of its construction technique, from the rational use of wood to the employment of canvas (incamottatura), from the use of dowels to the assembly procedure. Full article

Transfer learning from natural images is used in deep neural networks (DNNs) for medical image classification to achieve a computer-aided clinical diagnosis. Although the adversarial vulnerability of DNNs hinders practical applications owing to the high stakes of diagnosis, adversarial attacks are expected to be limited because training datasets (medical images), which are often required for adversarial attacks, are generally unavailable in terms of security and privacy preservation. Nevertheless, in this study, we demonstrated that adversarial attacks are also possible using natural images for medical DNN models with transfer learning, even if such medical images are unavailable; in particular, we showed that universal adversarial perturbations (UAPs) can also be generated from natural images. UAPs from natural images are useful for both non-targeted and targeted attacks. The performance of UAPs from natural images was significantly higher than that of random controls. The use of transfer learning causes a security hole, which decreases the reliability and safety of computer-based disease diagnosis. Model training from random initialization reduced the performance of UAPs from natural images; however, it did not completely avoid vulnerability to UAPs. The vulnerability of UAPs to natural images is expected to become a significant security threat. Full article

Holotomography is an extension of computed tomography where samples with low X-ray absorption can be investigated with higher contrast. In order to achieve this, the imaging system must yield an optical resolution of a few micrometers or less, which reduces the measurement area (field of view = FOV) to a few mm at most. If the sample size, however, exceeds the field of view (called local tomography or region of interest = ROI CT), filter problems arise during the CT reconstruction and phase retrieval in holotomography. In this paper, we will first investigate the practical impact of these filter problems and discuss approximate solutions. Secondly, we will investigate the effectiveness of a technique we call “multiscalar holotomography”, where, in addition to the ROI CT, a lower resolution non-ROI CT measurement is recorded. This is used to avoid the filter problems while simultaneously reconstructing a larger part of the sample, albeit with a lower resolution in the additional area. Full article

Shape-from-Shading represents the problem of computing the three-dimensional shape of a surface given a single gray-value image of it as input. In a recent paper, we showed that the introduction of an attenuation factor in the brightness equations related to various perspective Shape-from-Shading models allows us to ensure the well-posedness of the corresponding differential problems. Here, we propose a unified convergence result of a numerical scheme for several non-Lambertian reflectance models. This result is interesting since it can be easily extended to other non-Lambertian models in a unified and, therefore, powerful framework. Full article

Ischemic chronic cardiomyopathy (ICC) is still one of the most common cardiac diseases leading to the development of myocardial ischemia, infarction, or heart failure. The application of several imaging modalities can provide information regarding coronary anatomy, coronary artery disease, myocardial ischemia and tissue characterization. In particular, coronary computed tomography angiography (CCTA) can provide information regarding coronary plaque stenosis, its composition, and the possible evaluation of myocardial ischemia using fractional flow reserve CT or CT perfusion. Cardiac magnetic resonance (CMR) can be used to evaluate cardiac function as well as the presence of ischemia. In addition, CMR can be used to characterize the myocardial tissue of hibernated or infarcted myocardium. Echocardiography is the most widely used technique to achieve information regarding function and myocardial wall motion abnormalities during myocardial ischemia. Nuclear medicine can be used to evaluate perfusion in both qualitative and quantitative assessment. In this review we aim to provide an overview regarding the different noninvasive imaging techniques for the evaluation of ICC, providing information ranging from the anatomical assessment of coronary artery arteries to the assessment of ischemic myocardium and myocardial infarction. In particular this review is going to show the different noninvasive approaches based on the specific clinical history of patients with ICC. Full article

In this paper, we proposed a method for matching the color and glossiness of an object between different displays by using tone mapping. Since displays have their own characteristics, such as maximum luminance and gamma characteristics, the color and glossiness of an object when displayed differs from one display to another. The color can be corrected by conventional color matching methods, but the glossiness, which greatly changes the impression of an object, needs to be corrected. Our practical challenge was to use tone mapping to correct the high-luminance part, also referred to as the glossy part, which cannot be fully corrected by color matching. Therefore, we performed color matching and tone mapping using high dynamic range images, which can record a wider range of luminance information as input. In addition, we varied the parameters of the tone-mapping function and the threshold at which the function was applied to study the effect on the object’s appearance. We conducted a subjective evaluation experiment using the series category method on glossy-corrected images generated by applying various functions to each display. As a result, we found that the differences in glossiness between displays could be corrected by selecting the optimal function for each display. Full article

By aligning virtual augmentations with real objects, optical see-through head-mounted display (OST-HMD)-based augmented reality (AR) can enhance user-task performance. Our goal was to compare the perceptual accuracy of several visualization paradigms involving an adjacent monitor, or the Microsoft HoloLens 2 OST-HMD, in a targeted task, as well as to assess the feasibility of displaying imaging-derived virtual models aligned with the injured porcine heart. With 10 participants, we performed a user study to quantify and compare the accuracy, speed, and subjective workload of each paradigm in the completion of a point-and-trace task that simulated surgical targeting. To demonstrate the clinical potential of our system, we assessed its use for the visualization of magnetic resonance imaging (MRI)-based anatomical models, aligned with the surgically exposed heart in a motion-arrested open-chest porcine model. Using the HoloLens 2 with alignment of the ground truth target and our display calibration method, users were able to achieve submillimeter accuracy (0.98 mm) and required 1.42 min for calibration in the point-and-trace task. In the porcine study, we observed good spatial agreement between the MRI-models and target surgical site. The use of an OST-HMD led to improved perceptual accuracy and task-completion times in a simulated targeting task. Full article

Pore scale analysis of flow through porous media is of interest because it is essential for understanding internal erosion and piping, among other applications. Past studies have mainly focused on exploring macroscopic flow to infer microscopic phenomena. An innovative method is introduced in this study which permits visualization of interstitial fluid flow through the pores of a saturated synthetic transparent granular medium at the microscale. Several representative images of Ottawa sand were obtained using dynamic image analysis (DIA), for comparison with flow through perfect cylinders. Magnified transparent soil particles made of hydrogel were cast in 3D printed molds. Custom 3D printed jigs were employed for accurate positioning of the particles to ensure that particles have the same flow area within the soil. The pore fluid was embedded with silver-coated hollow microspheres that allowed for their florescence and tracking their movement within the model when illuminated by a laser light source. Images of the flow were captured from the model using a high-speed camera. This, along with particle image velocimetry (PIV) provided for the velocity and direction analysis of fluid flow movements within the pore space of a planar 2D model. Comparison of interstitial flow through homogeneous porosity-controlled Ottawa-shaped and cylindrical particles demonstrates that the magnitude of turbulence is related to particle roundness. Full article

During an Over-the-Board (OTB) chess event, all players are required to record their moves strictly by hand, and later the event organizers are required to digitize these sheets for official records. This is a very time-consuming process, and in this paper we present an alternate workflow of digitizing scoresheets using a BiLSTM network. Starting with a pretrained network for standard Latin handwriting recognition, we imposed chess-specific restrictions and trained with our Handwritten Chess Scoresheet (HCS) dataset. We developed two post-processing strategies utilizing the facts that we have two copies of each scoresheet (both players are required to write the entire game), and we can easily check if a move is valid. The autonomous post-processing requires no human interaction and achieves a Move Recognition Accuracy (MRA) around 95%. The semi-autonomous approach, which requires requesting user input on unsettling cases, increases the MRA to around 99% while interrupting only on 4% moves. This is a major extension of the very first handwritten chess move recognition work reported by us in September 2021, and we believe this has the potential to revolutionize the scoresheet digitization process for the thousands of chess events that happen every day. Full article

Model understanding is critical in many domains, particularly those involved in high-stakes decisions, e.g., medicine, criminal justice, and autonomous driving. Explainable AI (XAI) methods are essential for working with black-box models such as convolutional neural networks. This paper evaluates the traffic sign classifier of the Deep Neural Network (DNN) from the Programmable Systems for Intelligence in Automobiles (PRYSTINE) project for explainability. The results of explanations were further used for the CNN PRYSTINE classifier vague kernels’ compression. Then, the precision of the classifier was evaluated in different pruning scenarios. The proposed classifier performance methodology was realised by creating an original traffic sign and traffic light classification and explanation code. First, the status of the kernels of the network was evaluated for explainability. For this task, the post-hoc, local, meaningful perturbation-based forward explainable method was integrated into the model to evaluate each kernel status of the network. This method enabled distinguishing high- and low-impact kernels in the CNN. Second, the vague kernels of the classifier of the last layer before the fully connected layer were excluded by withdrawing them from the network. Third, the network’s precision was evaluated in different kernel compression levels. It is shown that by using the XAI approach for network kernel compression, the pruning of 5% of kernels leads to a 2% loss in traffic sign and traffic light classification precision. The proposed methodology is crucial where execution time and processing capacity prevail. Full article

The discrete shearlet transformation accurately represents the discontinuities and edges occurring in magnetic resonance imaging, providing an excellent option of a sparsifying transform. In the present paper, we examine the use of discrete shearlets over other sparsifying transforms in a low-rank plus sparse decomposition problem, denoted by $L+S$. The proposed algorithm is evaluated on simulated dynamic contrast enhanced (DCE) and small bowel data. For the small bowel, eight subjects were scanned; the sequence was run first on breath-holding and subsequently on free-breathing, without changing the anatomical position of the subject. The reconstruction performance of the proposed algorithm was evaluated against k-t FOCUSS. $L+S$ decomposition, using discrete shearlets as sparsifying transforms, successfully separated the low-rank (background and periodic motion) from the sparse component (enhancement or bowel motility) for both DCE and small bowel data. Motion estimated from low-rank of DCE data is closer to ground truth deformations than motion estimated from L and S. Motility metrics derived from the S component of free-breathing data were not significantly different from the ones from breath-holding data up to four-fold undersampling, indicating that bowel (rapid/random) motility is isolated in S. Our work strongly supports the use of discrete shearlets as a sparsifying transform in a $L+S$ decomposition for undersampled MR data. Full article

This paper shows that the X-ray analysis method known from the medical field, using a priori information, can provide a lot more information than the common analysis for high-speed experiments. Via spatial registration of known 3D shapes with the help of 2D X-ray images, it is possible to derive the spatial position and orientation of the examined parts. The method was demonstrated on the example of the sabot discard of a subcaliber projectile. The velocity of the examined object amounts up to 1600 m/s. As a priori information, the geometry of the experimental setup and the shape of the projectile and sabot parts were used. The setup includes four different positions or points in time to examine the behavior over time. It was possible to place the parts within a spatial accuracy of 0.85 mm (standard deviation), respectively 1.7 mm for 95% of the errors within this range. The error is mainly influenced by the accuracy of the experimental setup and the tagging of the feature points on the X-ray images. Full article

This paper proposes a reversible image processing method for color images that can independently improve saturation and enhance brightness contrast. Image processing techniques have been popularly used to obtain desired images. The existing techniques generally do not consider reversibility. Recently, many reversible image processing methods have been widely researched. Most of the previous studies have investigated reversible contrast enhancement for grayscale images based on data hiding techniques. When these techniques are simply applied to color images, hue distortion occurs. Several efficient methods have been studied for color images, but they could not guarantee complete reversibility. We previously proposed a new method that reversibly controls not only the brightness contrast, but also saturation. However, this method cannot fully control them independently. To tackle this issue, we extend our previous work without losing its advantages. The proposed method uses the HSV cone model, while our previous method uses the HSV cylinder model. The experimental results demonstrate that our method flexibly controls saturation and brightness contrast reversibly and independently. Full article

This paper presents two techniques in the matching and re-identification of multiple aerial target detections from multiple electro-optical devices: 2-dimensional and 3-dimensional kinematics-based matching. The main advantage of these methods over traditional image-based methods is that no prior image-based training is required; instead, relatively simpler graph matching algorithms are used. The first 2-dimensional method relies solely on the kinematic and geometric projections of the detected targets onto the images captured by the various cameras. Matching and re-identification across frames were performed using a series of correlation-based methods. This method is suitable for all targets with distinct motion observed by the camera. The second 3-dimensional method relies on the change in the size of detected targets to estimate motion in the focal axis by constructing an instantaneous direction vector in 3D space that is independent of camera pose. Matching and re-identification were achieved by directly comparing these vectors across frames under a global coordinate system. Such a method is suitable for targets in near to medium range where changes in detection sizes may be observed. While no overlapping field of view requirements were explicitly imposed, it is necessary for the aerial target to be detected in both cameras before matching can be carried out. Preliminary flight tests were conducted using 2–3 drones at varying ranges, and the effectiveness of these techniques was tested and compared. Using these proposed techniques, an MOTA score of more than 80% was achieved. Full article

Human coronaviruses (HCoV) are causative agents of mild to severe intestinal and respiratory infections in humans. In the last 15 years, we have witnessed the emergence of three zoonotic, highly pathogenic HCoVs. Thus, early and accurate detection of these viral pathogens is essential for preventing transmission and providing timely treatment and monitoring of drug resistance. Herein, we applied enhanced darkfield hyperspectral microscopy (EDHM), a novel non-invasive, label-free diagnostic tool, to rapidly and accurately identify two strains of HCoVs, i.e., OC43 and 229E. The EDHM technology allows collecting the optical image with spectral and spatial details in a single measurement without direct contact between the specimen and the sensor. Thus, it can directly map spectral signatures specific for a given viral strain in a complex biological milieu. Our study demonstrated distinct spectral patterns for HCoV-OC43 and HCoV-229E virions in the solution, serving as distinguishable parameters for their differentiation. Furthermore, spectral signatures obtained for both HCoV strains in the infected cells displayed a considerable peak wavelength shift compared to the uninfected cell, indicating that the EDHM is applicable to detect HCoV infection in mammalian cells. Full article

The fetus head circumference (HC) is a key biometric to monitor fetus growth during pregnancy, which is estimated from ultrasound (US) images. The standard approach to automatically measure the HC is to use a segmentation network to segment the skull, and then estimate the head contour length from the segmentation map via ellipse fitting, usually after post-processing. In this application, segmentation is just an intermediate step to the estimation of a parameter of interest. Another possibility is to estimate directly the HC with a regression network. Even if this type of segmentation-free approaches have been boosted with deep learning, it is not yet clear how well direct approach can compare to segmentation approaches, which are expected to be still more accurate. This observation motivates the present study, where we propose a fair, quantitative comparison of segmentation-based and segmentation-free (i.e., regression) approaches to estimate how far regression-based approaches stand from segmentation approaches. We experiment various convolutional neural networks (CNN) architectures and backbones for both segmentation and regression models and provide estimation results on the HC18 dataset, as well agreement analysis, to support our findings. We also investigate memory usage and computational efficiency to compare both types of approaches. The experimental results demonstrate that even if segmentation-based approaches deliver the most accurate results, regression CNN approaches are actually learning to find prominent features, leading to promising yet improvable HC estimation results. Full article

The proper inspection of a cracks pattern over time is a critical diagnosis step to provide a thorough knowledge of the health state of a structure. When monitoring cracks propagating on a planar surface, adopting a single-image-based approach is a more convenient (costly and logistically) solution compared to subjective operators-based solutions. Machine learning (ML)- based monitoring solutions offer the advantage of automation in crack detection; however, complex and time-consuming training must be carried out. This study presents a simple and automated ML-based crack monitoring approach implemented in open sources software that only requires a single image for training. The effectiveness of the approach is assessed conducting work in controlled and real case study sites. For both sites, the generated outputs are significant in terms of accuracy (~1 mm), repeatability (sub-mm) and precision (sub-pixel). The presented results highlight that the successful detection of cracks is achievable with only a straightforward ML-based training procedure conducted on only a single image of the multi-temporal sequence. Furthermore, the use of an innovative camera kit allowed exploiting automated acquisition and transmission fundamental for Internet of Things (IoTs) for structural health monitoring and to reduce user-based operations and increase safety. Full article

The TRIMAGE project aims to develop a brain-dedicated PET/MR/EEG (Positron Emission Tomography/Magnetic Resonance/Electroencephalogram) system that is able to perform simultaneous PET, MR and EEG acquisitions. The PET component consists of a full ring with 18 sectors. Each sector includes three square detector modules based on dual sstaggered LYSO:Ce matrices read out by SiPMs. Using Monte Carlo simulations and following NEMA (National Electrical Manufacturers Association) guidelines, image quality procedures have been applied to evaluate the performance of the PET component of the system. The performance are reported in terms of spatial resolution, uniformity, recovery coefficient, spill over ratio, noise equivalent count rate (NECR) and scatter fraction. The results show that the TRIMAGE system is at the top of the current brain PET technologies. Full article

This paper presents the evaluation of 36 convolutional neural network (CNN) models, which were trained on the same dataset (ImageNet). The aim of this research was to evaluate the performance of pre-trained models on the binary classification of images in a “real-world” application. The classification of wildlife images was the use case, in particular, those of the Eurasian lynx (lat. “Lynx lynx”), which were collected by camera traps in various locations in Croatia. The collected images varied greatly in terms of image quality, while the dataset itself was highly imbalanced in terms of the percentage of images that depicted lynxes. Full article