Estimating gaze direction often suffers from the presence of distracting, non-gaze-related features within full-face images, which can hinder accurate prediction. In this article, we present GazeMTNet, a lightweight multi-task convolutional neural network that was created from scratch for the joint estimate of head posture and gaze direction. Our architecture makes use of the dependency to increase the accuracy of gaze estimation while minimizing computational complexity. Experiments on the EyeDiap and the MPIIGaze datasets show that GazeMTNet outperforms current techniques in terms of efficiency and accuracy, attaining remarkable performance, with mean angular errors of 4.29 and 3.20°, respectively. The 0.05 G FLOPs and 1.012 million parameters make our model ideal for embedded and real-time applications. These outcomes support the efficacy of multi-task learning when it comes to gaze estimation.

Diabetic macular edema (DME) and age-related macular degeneration (AMD) are major causes of vision impairment and blindness. While many classification applications for these diseases achieve high performance, they often overlook the crucial aspect of dataset and image quality, leading to potential erroneous predictions. This study emphasizes the importance of data quality in medical image classification, specifically for retinal imaging. We propose an Optical Coherence Tomography (OCT) image quality evaluation model using the pre-trained ARNIQA (leArning distoRtion maNifold for Image Quality Assessment) model to accurately identify retinal diseases autonomously. Our methodology includes a three-class classification system utilizing two Convolutional Neural Network (CNN) models, ResNet50 and Xception, applied to three datasets: the original dataset, a subset of high-quality images, and a subset of low-quality images. Using a Tunisian OCT dataset of 2887 images, we demonstrate the efficacy of our approach, achieving $100 \%$ accuracy with highquality images.

International audience

Diabetic macular edema (DME) and Age-Related Macular Degeneration (AMD) are two of the most common disorders that can cause blindness in a population and primarily cause retinal degradation. The application of multiple deep learning algorithms on Optical Coherence Tomography OCT) images to detect these disorders demonstrates excellent performance. However, because these algorithms include black box features, medical professionals are hesitant to fully trust the results. To address these challenges, we present a modified convolutional neural network based on the xception architecture for diagnosing DME and AMD using optical coherence tomography (OCT) images. To demonstrate the model’s transparency and trustworthiness, we used the Grad-CAM technique, which incorporates Explainable AI into the research and improves model interpretability. This technique assists medical specialists in demystifying deep learning algorithms and obtaining more information about the critical areas in OCT images used for prediction. The proposed model achieved an accuracy of 99.87%, a precision of 99.67%, and a recall of 98.29% on a dataset of 934 images.

This paper undertakes a thorough analysis of the existing landscape of datasets within the beekeeping field, with the primary aim of furnishing precision beekeeping researchers with a comprehensive overview. Through an assessment of accessibility, scope, and applicability, this study endeavors to identify gaps, discern trends, and pinpoint potential areas for future exploration. The overarching goal is to propel advancements in bee health, productivity, and conservation through data-driven strategies. Stressing the pivotal role of technology in augmenting dataset effectiveness, the paper underscores the necessity for collaborative endeavors to standardize data collection and sharing practices. Serving as a cornerstone resource, this review aims to equip researchers with the necessary insights to harness data for the improvement of beekeeping practices and the sustainability of bee populations.

In bioinformatics, analyzing large-scale gene expression data using machine learning models is a common practice. However, selecting relevant features from high-dimensional data poses a significant challenge. Feature selection techniques aim to address this challenge by reducing data size, selecting important features, and reducing classification time. While filter selection is a fast approach, the selected features may lack biological relevance. Wrapper techniques have shown efficacy in selecting relevant features, but they can be computationally intensive and time-consuming. Embedded techniques bridge this gap by selecting relevant features while maintaining computational efficiency and low cost. This study introduces a framework that utilizes the embedded feature selection based on the RidgeClassifier technique for selecting gene expression data (EGSRC) from the ADNI dataset. Hyperparameter tuning using Optuna is also employed to further improve the performance of the SVM classifier in classification learning. The study evaluates the performance of five different classifiers using various evaluation metrics through cross-validation. SVM achieved the highest scores across all metrics, recording 99.47% accuracy, 99.25% precision, 98.41% specificity, 100% recall, 99.62% F1-score, and 99.20% AUC. The proposed framework’s RC selector outperforms filter, embedded techniques, and state-of-the-art techniques, which demonstrates its robustness.

Retinal pathologies like choroidal neovascularization (CNV), drusen, and diabetic macular edema (DME) can give rise to microvascular alterations in the retina, ultimately resulting in vision impairment. The manual detection of these diseases poses a significant challenge and necessitates specialized medical expertise. To address this challenge, our study introduces novel deep learning methods for the detection of these ocular pathologies automatically and based on optical coherence tomography (OCT) scans. In our experimental setup, we utilized a dataset comprising 6000 OCT images sourced from the publicly available Kaggle dataset. Through comprehensive evaluations, our study revealed that the implementation of a bilinear convolutional neural network (B-CNN) yielded the highest classification score, surpassing the accuracy achieved by alternative models. Furthermore, when compared to other deep learning networks, our proposed approach showcased superior performance in the early diagnosis of these three ocular diseases.

Background: Single Photon Emission Computed Tomography (SPECT) scan is a convenient diagnostic technique for Parkinson's Disease (PD). Nevertheless, certain subjects who met the clinical diagnosis measurements for PD had normal SPECT images. They are referred to have Scans Without Evidence of Dopamine Deficiency (SWEDD). Indeed, SWEDD is a heterogeneous group of Healthy Control (HC) and early PD subjects. Method: In this research work, we developed a Stepwise Multiple Linear Regression (SMLR) method in the PD group. This process is an optimal regression model for identifying the key SPECT image-derived features, which influence clinical scores. We also investigate the interaction between these features in the three groups together (PD, SWEDD, and HC). Results: SMLR results showed that only the Specific Binding Ratio (SBR) of the Worst-Putamen and SBR of the Worst-Caudate are the factors that influence Unified Parkinson's Disease Rating Scale (UPDRS) and State Trait Anxiety Inventory (STAI) scores, respectively. In addition, two tests of Two-Way-ANOVA showed significant interactions between UPDRS and SBR of the Worst-Putamen, and between STAI and SBR of the Worst-Caudate of the three groups. Conclusion: From the SMLR results, we found that the increased UPDRS and STAI scores severity is highly dependent on the dopamine transporter density in the Worst-Putamen and the Worst-Caudate, respectively.

Colorectal cancer (CRC) is a leading cause of mortality worldwide. Microsatellite instability (MSI) detection is crucial for clinical decision-making in CRC, where people with varied therapeutic responses and prognoses are identified by their MSI status. However, the process of MSI detection is very sensitive, as it requires a lot of time, money as well as expertise from pathologists. Thus, we propose in this paper a deep learning approach for MSI detection from histopathological H&E stained whole slide images based on GAN-CNN. The achieved results are very promising and demonstrate the robustness of our approach. The proposed method provides a valuable second opinion to the pathologists and reduces their effort to get the most correct result.

Ultrasound (US) imaging is the leading diagnostic method for assessing the early‐stage thyroid nodule. However, the visual evaluation of nodules can be influenced by the subjectivity of radiologists' interpretations. Computer‐aided Diagnostic (CAD) systems can be useful in classifying these nodules according to their benign or malignant nature. The extraction of the characteristics, which relate in the author's case to the US of thyroid nodules, is essential in the differentiation of these nodules. The complex nature of images, however, generates a significant number of features, many of which are either redundant or irrelevant. This study presents a new CAD system that has been developed to categorise thyroid nodules. In this survey, 447 US images of thyroid nodules were retained. These images were used to extract features using statistical features extraction methods. A feature selection method based on the multi objective particle swarm optimisation algorithm was used to choose the most relevant and non‐redundant ones. Then, support vector machine (SVM) and random forests (RFs) were applied to classify these nodules. 10‐fold cross‐validation was used to assess the classification performance metrics. Their proposed CAD has reached a maximum accuracy of 94.28% for SVM; and 96.13% for RF using the contour‐based ROI.

Machine Learning (ML) is a subfield of Artificial Intelligence (AI) that is increasingly applied to several medical diagnosis tasks, including a wide range of diseases. Importantly, various ML models were developed to address the complexity of Parkinson's Disease (PD) diagnosis. PD is a neurodegenerative disease characterized by motor and non‐motor disorders where its syndromes affect the daily lives of patients. Several Computer Aided Diagnosis and Detection (CADD) systems based on hand‐crafted ML algorithms achieved promising results in distinguishing PD patients from Healthy Control (HC) subjects and other Parkinsonian syndrome categories using clinical data (e.g., speech and gait impairments) and medical imaging [e.g., Position Emission Tomography (PET) and Single Photon Emission Computed Tomography (SPECT)]. Despite the good performance of hand‐crafted ML algorithms, there is still a problem linked to the features' extraction and selection. In fact, Deep Learning DL has provided an ultimate solution for the features' extraction and selection related issue. An important number of studies on the diagnosis of PD using DL algorithms were developed recently. This study provides an overview of the application of hand‐crafted ML algorithms and DL techniques for PD diagnosis. It also introduces key concepts for understanding the application of ML methods to diagnose PD.

Lung cancer is the most frequent and lethal malignant tumor. Detection and characterization of pulmonary nodules in Computed Tomography images (CT) is a primordial task for lung cancer diagnosis at early stages. As juxta-pleural nodules are directly linked to the lung pleura, they contain an open contour which makes their extraction a challenging task. In this paper, we propose an automatic detection and classification method of a part-solid juxta-pleural nodule. The proposed method was tested on 18 CT scan images in axial acquisition and gave an accurate quantification of the solid component in a part-solid nodule.

Computer-Assisted Detection (CAD) systems for detecting Cerebral Aneurysms (CA) present an important influence in the prevention of Intracranial HSA (Subarachnoid Hemorrhage), their usefulness haves been reported. We propose in this paper, a CAD to detect CA in DSA angiographic images by combining different methods for robust features/interest points detector which are Maximally Stable Extremal Regions (MSER), Speed Up Robust Features (SURF) and Scale Invariant Feature Transform (SIFT). The results on the proposed CAD over the provided benchmark are very encouraging.

Ultrasound is one of the most used imaging techniques for assessing and evaluating thyroid lesions. Indeed, it shows a good performance in terms of discrimination between benign and malignant thyroid nodules. Diagnosis by ultrasound is, however, not as easy as it seems and depends strongly on the experience of the radiologists. To help physician and radiologists to better diagnose, many Computer-Assisted Diagnosis (CAD) systems have been developed. These systems are based on image processing techniques and on machine learning. They represent effective and useful tools allowing doctors to have a second opinion far from human subjectivity. Among the machine learning techniques, the Deep Learning has recently made rapid progress in interpreting medical imaging and has demonstrated an impressive efficiency. Various CAD systems treating ultrasound images of the thyroid have widely use it since then. This paper reviews the most recent research works on the CAD systems for analyzing ultrasound images of the thyroid to diagnose the benign or malignant nature of the thyroid nodules. The CAD systems studied in this paper are based on the Deep Learning. We present a brief description of the CAD systems based on the Deep Leaning. Specifically, we describe the data collection and the CNN implementation. We report also the results obtained in these studies and highlight the limitations of such studies. This literature review is aimed at researchers but also at physician who are interested in CAD tools in ultrasound images of the thyroid gland and can represent a state of the art for all those interested in the classification of medical imaging.

Intracranial Aneurysm (IA) is a weaknesses of the cerebral arteries. Their rupture causes heavy consequences. Therefore, accurately assessing the risk of IA rupture and intervening timely is critical for improving the survival rate and prognosis of patients. We present in this paper a new IA detection method based on Convolutional Neural Networks (CNN). All the IA are detected in 2D-DSA dataset by the proposed method with zero false negative.

In this paper, we presented a new approach for speckle noise removal based on the combination of the trilateral filter and stationary wavelet transform. The main contribution of this paper is the integration of a new robust weighting factor into the bilateral filter, to develop a new multi-scale version of the trilateral filter. The experimental results prove the performance and the effectiveness of the proposed filter in the speckle reduction and edge preservation.

Brain Tractography using Diffusion Tensor Imaging is defined as the study of the path of the white matter fibres. It's based on the fact that the diffusion of water molecules in the white matter depends on the orientation of the fibres. This emerging technique is used in clinic to improve diagnosis, guide therapy and provide prognostic indication. Important Applications of DTI is the Presurgical Planning which allows to localize with accuracy of gray matter and white matter areas at risk of being resected during lesion removal. In our country, this technique has been introduced recently and in a very limited number of Magnetic Resonance Imaging (MRI) sites. The sequences of Multi Direction Diffusion weighted are available as standard in most MR systems while the post-processing software is optional, and require expensive and powerful workstation. The First objective of this work is to propose an algorithm which allow fibre tracking starting from Diffusion Weighted Images, through Diffusion Tensor Data calculation till the reconstruction and extraction of white matter fibres. The second objective is related to the validation of the used algorithm in clinical applications.

The evaluation of Cardiac Magnetic Resonance (CMR) imaging exam is mainly based on the visual aspect. This visual evaluation depends on the level of expertise of the radiologist and it is characterized by variability within and between observers. The aim of this work is to propose a new method based on a mathematical model, “Fourier Transform” which calculates an amplitude parametric image. This image, calculated from the Cine MR images, allows the localization and quantification of abnormalities related to difference in contraction and their extent. The suggested amplitude image is likely to assist in the diagnosis through reducing the time taken by the radiologist to specify the abnormal contraction and by improving the accuracy of the examination. After testing this approach on patients (healthy and pathological), we have proven a good concordance between the results obtained by the parametric image and those collected from the routine examination.

Medical image degradation has a significant impact on image quality, and thus affects human interpretation and the accuracy of computer-assisted diagnostic techniques. Unfortunately, Ultrasound images are mainly degraded by an intrinsic noise called speckle. Therefore, despekle filtering is a critical preprocessing step in medical ultrasound images. In this paper we propose a new image denoising technique based on the combination of bilateral filter and stationary wavelet transform. The main contribution of this paper is in the use of a new neighborhood relationship to develop a new multiscale bilateral filter. Experimental results validated the effectiveness and the accuracy of the proposed filter in speckle noise reduction and edge preservation for medical ultrasound images.

The detection of cerebral aneurysms is of a paramount importance in the prevention of intracranial sub-arachnoid hemorrhage. We propose in this paper, a complete detection scheme, consisting of two phases, to detect blobs and aneurysms in cerebral 2D-DSA images. The first classification phase extracts cerebral vasculature by means of the fusion of multiple classifiers. The second detection phase involves detecting the aneurysms in the vascular tree from the segmented images using fuzzy Mathematical Morphology. The idea is to model the imprecision on the size and the shape of aneurysms using fuzzy logic. The experimental results on our test images, demonstrate the usefulness of the proposed method, which induces to a high sensitivity with a low number of false positives,and compares favorably to existing detection approaches.

The instrumentation and physiological patient factors bound to the patient in-vivo complicate the treatment of the images during a medical examination. Thus, they can contribute to the generation of artifacts in the resulting images. The artifacts degrade the quality of the images and can lead, in certain cases, to a bad diagnosis. For this reason, the appeal to the use of the simulation techniques allows to evaluate and improve the devices of acquisition and image processing. The models (called phantoms) are important tools to simulate human anatomy and physiology and to allow the evaluation of acquisition methods and image analysis. Thereby, the simulation offers a way of great importance to evaluate and improve medical techniques of acquisition devices, treatment and the reconstruction of images in-vitro.