Search

The glaucoma is an eye disease that causes blindness when it progresses in an advanced stage. Early glaucoma diagnosis is essential to prevent the vision loss. However, early detection is not covered due to the lack of ophthalmologists and the limited accessibility to retinal image capture devices. In this paper, we present an automated method for glaucoma screening dedicated for Smartphone Captured Fundus Images (SCFIs). The implementation of the method into a smartphone associated to an optical lens for retina capturing leads to a mobile aided screening system for glaucoma. The challenge consists in insuring higher performance detection despite the moderate quality of SCFIs, with a reduced execution time to be adequate for the clinical use. The main idea consists in deducing glaucoma based on the vessel displacement inside the Optic Disk (OD), where the vessel tree remains sufficiently modeled on SCFIs. Within this objective, our major contribution consists in proposing: (1) a robust processing for locating vessel centroids in order to adequately model the vessel distribution, and (2) a feature vector that relevantly reflect two main glaucoma biomarkers in terms of vessel displacement. Furthermore, all processing steps are carefully chosen based on lower complexity, to be suitable for fast clinical screening. A first evaluation of our method is performed using the two public DRISHTI-DB and DRIONS-DB databases, where 99% and 95% accuracy, 96.77% and 97,5% specificity and 100% and 95% sensitivity are respectively achieved. Thereafter, the method is evaluated using two fundus image databases respectively captured through a smartphone and retinograph for the same persons. We achieve 100% accuracy using both databases which assesses the robustness of our method. In addition, the detection is performed on 0.027 and 0.029 second when executed respectively on the Samsung-M51 on the Samsung-A70 smartphone devices. Our proposed smartphone app provides a cost-effective and widely accessible mobile platform for early screening of glaucoma in remote clinics or areas with limited access to fundus cameras and ophthalmologists.

International audience; Objective: SARS-CoV-2 is a worldwide health emergency with unrecognized clinical features. This paper aims to review the most recent medical Imaging techniques used for the diagnosis of SARS-CoV-2 and their potential contributions to attenuate the pandemic. Recent researches, including Artificial Intelligence tools, will be described. Methods We review the main clinical features of SARS-CoV-2 revealed by different medical imaging techniques. First, we present the clinical findings of each technique. Then, we describe several artificial intelligence approaches introduced for the SARS-CoV-2 diagnosis. Results CT is the most accurate diagnostic modality of SARS-CoV-2. Additionally, ground-glass opacities and consolidation are the most common signs of SARS-CoV-2 in CT images. However, other findings such as reticular pattern, and crazy paving could be observed. We also found that pleural effusion and pneumothorax features are less common in SARS-CoV-2. According to the literature, the B lines artifacts and pleural line irregularities are the common signs of SARS-CoV-2 in ultrasound images. We have also stated the different studies, focusing on artificial intelligence tools, to evaluate the SARS-CoV-2 severity. We found that most of the reported works based on deep learning focused on the detection of SARS-CoV-2 from medical images while the challenge for the radiologists is how to differentiate between SARS-CoV-2 and other viral infections with the same clinical features. Conclusion The identification of SARS-CoV-2 manifestations on medical images is a key step in radiological workflow for the diagnosis of the virus and could be useful for researchers working on computer-aided diagnosis of pulmonary infections.

Deep learning, the fastest growing segment of Artificial Neural Network (ANN), has led to the emergence of many machine learning applications and their implementation across multiple platforms such as CPUs, GPUs and reconfigurable hardware (Field-Programmable Gate Arrays or FPGAs). However, inspired by the structure and function of ANNs, large-scale deep learning topologies require a considerable amount of parallel processing, memory resources, high throughput and significant processing power. Consequently, in the context of real time hardware systems, it is crucial to find the right trade-off between performance, energy efficiency, fast development, and cost. Although limited in size and resources, several approaches have showed that FPGAs provide a good starting point for the development of future deep learning implementation architectures. Through this paper, we briefly review recent work related to the implementation of deep learning algorithms in FPGAs. We will analyze and compare the design requirements and features of existing topologies to finally propose development strategies and implementation architectures for better use of FPGA-based deep learning topologies. In this context, we will examine the frameworks used in these studies, which will allow testing a lot of topologies to finally arrive at the best implementation alternatives in terms of performance and energy efficiency.

Due to their ability to reduce the size of high-dimensional input data, self-organizing maps (SOMs) can be employed as data quantizers. The widely used software implementations of SOM enjoy flexibility and adaptability, usually to the detriment of performances, which limits their use in real time applications. On the contrary, the hardware counterparts of SOMs exploit the inherent parallelism of hardware to boost the overall performances, but generally lack adaptability without considerable design efforts. To benefit from both, the flexibility of software and performances of hardware SOM implementations, unconventional design approaches of SOMs should be used. In this work, a scalable and adaptable hardware implementation of a SOM network is presented. The proposed architecture allows to dynamically extend the SOM operation from a smaller to a larger map only by (re-)configuring the parameters of each neuron. The gained scalability is obtained by decoupling the computation layer composed of neurons, from the communication one, used to provide data exchange mechanisms between neurons. The proposed SOM architecture is also validated through simulation on variable-sized SOM networks applied to image compression.

A novel approach is proposed for medical analysis and clinical decision support of the Electrocardiogram (ECG) signals based on the deterministic finite automata (DFA) with the addition of some requirements. This paper proves regular grammar is effective in the extraction of QRS complex and interpretation of ECG signals. The DFA will be used to represent a normalized QRS complex as a sequence of negative and positive peaks. A QRS is considered as a set of adjacent peaks that satisfy certain criteria of standard deviation and duration. The proposed method is applied on several kinds of ECG signals collected from the standard MIT-BIH arrhythmia database. Several metrics are calculated including QRS durations, RR distances and peak amplitudes. Furthermore, σRR and σQRS metrics were added to quantify RR distances regularity and QRS durations, respectively. Regular grammar with the addition of some requirements and deterministic automata proved functional for both biomedical signals and ECG signal diagnosis. The suggested method provided a sensitivity rate of 99.74% and the positive predictivity rate of 99.86%. The algorithm was compared to other works in the literature and the quality performance detection was compared with several algorithms tested and validated on the MIT-BIH database. A head-to-head comparison in terms of sensitivity and CPU runtime was provided with the wavelet method.

Context Cardiovascular disease remains the leading cause of death affecting the adult population. It involves disorders of myocardial perfusion which can lead to the disturbance in myocardial function and/or myocardial infarction. Finding an efficient method for the deformation analysis in the left ventricle of the heart (LV) is one of the major concerns in cardiac imaging. Objective In literature, most models provide global representation of the LV surface, that cannot specify the affected area. The present work is aimed at quantifying the local deformations in the LV by developing accurate computational approaches in order to refine the diagnosis and specify the ischemic territory. Method The proposed approach advocates the application of three different methods: (1) the numerical quantification of regional deformation based on SPHARM shape descriptors, (2) volume evolution, and (3) the Hotelling metric computation by use of the relative curvature at the vertices of the triangulation. In order to progress to a regional analysis, we carried out a division into 17 regions according to the AHA standard of LV object. It is worth noting that our two first methods are based on the application of the cloud closing process by generating new points on the cutting planes. Results The different approaches were tested on 44 patients. The obtained results show its effectiveness and validity by giving the degree and the location of the pathological deformation in the LV. We compared the obtained results with diagnostic offered by qualified nuclear medicine physicians. Also, we compared the obtained results of each methods in order to define the more robust approach.

Series of InxGa1-xAs/GaAs structures with indium vapor composition ranging from 13 to 100%, denoted samples A, B, C and D, were grown by metalorganic vapor phase epitaxy (MOVPE) at 450 °C and in situ monitored by spectral reflectance (SR). In order to contribute to the enhancement of crystal quality and to understand growth kinetic of InxGa1-xAs/GaAs structures, the dependence of structural and morphological properties on indium composition x was studied. Basing on high resolution x-ray diffraction (HRXRD) measurements, solid indium compositions x of samples A, B, C and D were determined. Also, the evolution of structural quality (dislocations density, grain size, etc.) as a function of indium composition x was quantified. Besides, morphological properties (hatching and islands formations, densities, sizes and uniformities, RMS surface roughness, etc.) and growth process (growth anisotropy, etc.) versus indium composition x were examined using atomic force microscopy (AFM) analysis. Also, reflectance three-dimensional plot as function of time and wavelength was recorded to quantify the evolution of reflectivity in the wavelength range from 400 to 1000 nm and to determine some growth parameters such as growth rates and thicknesses of InxGa1-xAs samples. A good correlation between experimental results issued from different characterizations tools was obtained.

GaAs/InxGa1 − xAs/GaAs strained and partially relaxed structures were grown by metalorganic vapor phase epitaxy and in situ monitored by laser reflectometry (LR). Two structures were formed by a single InxGa1 − xAs layer, and the third comprises three superposed InxGa1 − xAs layers having increasing indium contents. LR plots as function of time were recorded to extract growth rates and thicknesses of active and cap layers. In order to study the strain effect on structural and optical properties of these heterostructures, high resolution X-ray diffraction (HRXRD) and photoreflectance (PR) measurements were performed. HRXRD curves are developed to calculate strain tensor components, indium composition, and thicknesses of strained and partially relaxed layers. Besides, valence-band splitting and band-gap energy shift were measured by best fitting PR spectra at 300 K. Experimental energy values determined as a function of indium composition and relaxation rate were compared to those obtained by the elastic strain theory. For single and superposed InxGa1 − xAs active layers, a good correlation between experimental results and theoretical predictions was obtained.

Optical properties of InxGa1−xAs films grown on GaAs substrates by metalorganic vapor phase epitaxy were investigated. Spectral reflectance (SR) and photoreflectance (PR) at room temperature and near-infrared photoluminescence (PL) at 10 K were performed. SR signals in the range of 200–1700 nm provided the x-dependence of the critical point energies E1, E1 + Δ1 and E2. Furthermore, band-gap and spin-orbit splitting energies, as well as their broadening parameters were determined from PR spectra and studied as function of In composition ranging from 0 to 0.37. On the other hand, the origins of luminescence bands observed in PL spectra were revealed. A redshift of 16 meV/%In in the band-to-band transition was obtained. All results issued from different characterizations tools are correlated and compared to the literature.

In x Ga 1− x As/GaAs pseudomorphic structures were grown by metalorganic vapor phase epitaxy. Reciprocal space mapping were recorded in the vicinity of (0 0 4) and (1 1 5) nodes using high resolution X-ray diffraction (HRXRD) in order to determine strain tensor components, indium compositions and thicknesses of alloys. Near-infrared photoluminescence (PL) was performed at 10 K. The impact of strain on PL response was revealed by peak energy positions and line width. In addition, valence-band splitting (VBS) and the shift of the heavy-hole were measured. Besides, photoreflectance (PR) at room temperature was useful to establish experimentally the dependence of VBS and band energy shifts ( E 0 and E 0 + ∆ 0 ) on elastic strain due to lattice mismatches. Other parameters such as the internal electric-field and the electro-optical energy were determined from Franz–Keldysh oscillations analysis. Good correlation between the results obtained from all investigated techniques and theoretical predictions was confirmed.

Resume L’objectif de notre travail est de proposer un outil d’analyse automatique et d’aide a la decision base sur les reseaux de neurones artificiels (RNA). La premiere difficulte consiste dans le choix de la representation des signaux physiologiques et en particulier de l’electroencephalogramme (EEG). Une fois la representation adoptee, l’etape suivante est la conception du reseau de neurones optimal determine par un processus d’apprentissage et de validation sur des donnees issues d’un ensemble d’enregistrements de sommeil. Nous avons etudie plusieurs configurations classiques de RNA qui ont donnes des resultats variants de 62 a 71 % pour enfin proposer une nouvelle configuration hierarchique qui donne un taux de 74 % de bonne classification pour six stades. Ces resultats nous incitent a approfondir l’etude de cette problematique aux niveaux representation et conception des RNA pour ameliorer les performances de notre outil.

For texture analysis, several features such as co-occurrence matrices, Gabor filters and the wavelet transform are used. Recently, fractal geometry appeared to be an effective feature to analyze texture. But it is often restricted to 2D images, while 3D information can be very important especially in medical image processing. Moreover applications are limited to the use of fractal dimension. This study focuses on the benefits of fractal geometry in a classification method based on volumic texture analysis. The proposed methods make use of fractal and multifractal features for a 3D texture analysis of a voxel neighborhood. They are validated with synthetic data before being applied on real images. Their efficiencies are proved by comparison to some other texture features in supervised classification processes (AdaBoost and support vector machine classifiers). The results showed that features based on fractal geometry (by combining fractal and multifractal features) contributed to new texture characterization. Information on new features was useful and complementary for a classification method. This study suggests that fractal geometry can provide a new useful information in 3D texture analysis, especially in medical imaging.

A new approach for uniform parameterization and rotation invariant global description of 3D triangular surface meshes of objects with a spherical topology is presented. It consists of two steps. First, an initial mapping based on a heat conduction model is carried out and an optimization of the initial parameterization in a constrained optimization procedure is applied. Second, a rotation invariant 3D shape description is achieved using the abstract harmonic analysis and the shift theorem. Our approach is validated on both synthetic data and real data obtained from myocardial scintigraphy imaging technique as an example.