Search

Due to the sensitive nature of diabetes-related data, preventing them from being easily shared between studies, and the wide discrepancies in their data processing pipeline, progress in the field of glucose prediction is hard to assess. To address this issue, we introduce GLYFE (GLYcemia Forecasting Evaluation), a benchmark of machine learning-based glucose predictive models. We present the accuracy and clinical acceptability of nine different models coming from the literature, from standard autoregressive to more complex neural network-based models. These results are obtained on two different datasets, namely UVA/Padova Type 1 Diabetes Metabolic Simulator (T1DMS) and Ohio Type-1 Diabetes Mellitus (OhioT1DM), featuring artificial and real type 1 diabetic patients respectively. By providing extensive details about the data flow as well as by providing the whole source code of the benchmarking process, we ensure the reproducibility of the results and the usability of the benchmark by the community. Those results serve as a basis of comparison for future studies. In a field where data are hard to obtain, and where the comparison of results from different studies is often irrelevant, GLYFE gives the opportunity of gathering researchers around a standardized common environment.

Despite recent advances of deep neural networks in hand vein identification, the existing solutions assume the availability of a large and rich set of training image samples. These solutions, therefore, still lack the capability to extract robust and discriminative hand-vein features from a single training image sample. To overcome this problem, we propose a single-sample-per-person (SSPP) palm-vein identification approach, where only a single sample per class is enrolled in the gallery set for training. Our approach, named MSMDGAN + CNN, consists of a multi-scale and multi-direction generative adversarial network (MSMDGAN) for data augmentation and a convolutional neural network (CNN) for palm-vein identification. First, a novel data augmentation approach, MSMDGAN, is developed to learn the internal distribution of patches in a single image. The proposed MSMDGAN consists of multiple fully convolutional GANs, each of which is responsible for learning the patch distribution within an image at a different scale and at a different direction. Second, given the resulting augmented data by MSMDGAN, we design a CNN for single sample palm-vein recognition. The experimental results on two public hand-vein databases demonstrate that MSMDGAN is able to generate realistic and diverse samples, which, in turn, improves the stability of the CNN. In terms of accuracy, MSMDGAN + CNN outperforms other representative approaches and achieves state-of-the-art recognition results.

An optoelectronic Reservoir Computer is proposed and implemented numerically and on physical hardware for early-stage Alzheimer’s disease detection for the first time. The approach yields classification accuracy of 85% surpassing state-of-the-art performances.

Photoplethysmography (PPG) is a non-invasive and cost-efficient optical technique used to assess blood volume variation inside the micro-circulation. PPG technology is widely used in a variety of clinical and non-clinical devices in order to investigate the cardiovascular system. One example of clinical PPG device is the pulse oxymeter, while non-clinical PPG devices include smartphones and smartwatches. Such a wide diffusion of PPG devices generates plenty of different PPG signals that differ from each other. In fact, intrinsic device characteristics strongly influence PPG waveform. In this paper we investigate transfer learning approaches on a Covolutional Neural Network based quality assessment method in order to generalize our model across different PPG devices. Our results show that our model is able to classify accurately signal quality over different PPG datasets while requiring a small amount of data for fine-tuning.Clinical relevance- A precise detection and extraction of high quality PPG segments could enhance significantly the reliability of the medical analysis based on the signal.

We present a vision-based activity recognition system for centrally connected humanoid robots. The robots interact with several human participants who have varying behavioral styles and inter-activity-variability. A cloud server provides and updates the recognition model in all robots. The server continuously fetches the new activity videos recorded by the robots. It also fetches corresponding results and ground-truths provided by the human interacting with the robot. A decision on when to retrain the recognition model is made by an evolving performance-based logic. In the current article, we present the aforementioned adaptive recognition system with special emphasis on the partitioning logic employed for the division of new videos in training, cross-validation, and test groups of the next retraining instance. The distinct operating logic is based on class-wise recognition inaccuracies of the existing model. We compare this approach to a probabilistic partitioning approach in which the videos are partitioned with no performance considerations.

The adoption of deep learning in healthcare is hindered by their “black box” nature. In this paper, we explore the RETAIN architecture for the task of glucose forecasting for diabetic people. By using a two-level attention mechanism, the recurrent-neural-network-based RETAIN model is interpretable. We evaluate the RETAIN model on the type-2 IDIAB and the type-1 OhioT1DM datasets by comparing its statistical and clinical performances against two deep models and three models based on decision trees. We show that the RETAIN model offers a very good compromise between accuracy and interpretability, being almost as accurate as the LSTM and FCN models while remaining interpretable. We show the usefulness of its interpretable nature by analyzing the contribution of each variable to the final prediction. It revealed that signal values older than 1[Formula: see text]h are not used by the RETAIN model for 30[Formula: see text]min ahead of time prediction of glucose. Also, we show how the RETAIN model changes its behavior upon the arrival of an event such as carbohydrate intakes or insulin infusions. In particular, it showed that the patient’s state before the event is particularly important for the prediction. Overall the RETAIN model, thanks to its interpretability, seems to be a very promising model for regression or classification tasks in healthcare.

Palm-vein recognition has received increasing researchers’ attention in recent years. However, palm-vein recognition faces various challenges in practical applications, one of which is the lack of robustness against image quality degradation, resulting in reduction of the verification accuracy. To address this problem, this paper proposes an end-to-end convolutional neural network to automatically extract vein network features, thus without resorting to any hand-crafted features. Firstly, we label the palm-vein pixels based on several handcraft-based segmentation methods and reconstruct a training set accordingly. Secondly, an end-to-end vein segmentation model is proposed based on a generative adversarial network. After training, this model outputs a map where each value is the probability that the corresponding pixel belongs to a vein pattern. The resulting map is then subject to binarization by thresholding and stored in a binary image, used subsequently for verification matching. The experimental results on the public CASIA palm-vein dataset demonstrate the effectiveness of our proposed method.

Standard objective functions used during the training of neural-network-based predictive models do not consider clinical criteria, leading to models that are not necessarily clinically acceptable. In this study, we look at this problem from the perspective of the forecasting of future glucose values for diabetic people. In this study, we propose the coherent mean squared glycemic error (gcMSE) loss function. It penalizes the model during its training not only of the prediction errors, but also on the predicted variation errors which is important in glucose prediction. Moreover, it makes possible to adjust the weighting of the different areas in the error space to better focus on dangerous regions. In order to use the loss function in practice, we propose an algorithm that progressively improves the clinical acceptability of the model, so that we can achieve the best tradeoff possible between accuracy and given clinical criteria. We evaluate the approaches using two diabetes datasets, one having type-1 patients and the other type-2 patients. The results show that using the gcMSE loss function, instead of a standard MSE loss function, improves the clinical acceptability of the models. In particular, the improvements are significant in the hypoglycemia region. We also show that this increased clinical acceptability comes at the cost of a decrease in the average accuracy of the model. Finally, we show that this tradeoff between accuracy and clinical acceptability can be successfully addressed with the proposed algorithm. For given clinical criteria, the algorithm can find the optimal solution that maximizes the accuracy while at the same meeting the criteria.

Progress in the biomedical field through the use of deep learning is hindered by the lack of interpretability of the models. In this paper, we study the RETAIN architecture for the forecasting of future glucose values for diabetic people. Thanks to its two-level attention mechanism, the RETAIN model is interpretable while remaining as efficient as standard neural networks.

International audience; Intrusion detection is a critical Cyber Security subject. Different Machine Learning (ML) approaches have been proposed for Intrusion Detection Systems (IDS). However, their application to real-life scenarios remains challenging due to high data dimensionality. Representation learning (RL) allows discriminative feature representation in a low dimensionality space. The application of this technique in IDS requires more investigation. This paper examines and discusses the contribution of Siamese network based representation learning in improving the IDS performance. Extensive experimental results under different evaluation scenarios show different improvement rates depending on the scenario.

International audience; Nowadays, there is a huge amount of Historical Arabic Documents (HAD) in the national libraries and archives around the world. Analyzing this type of data manually is a difficult and costly task. Thus, an automatic process is required to exploit these documents more rapidly. Processing historical documents is a recent research subject that has seen a remarkable growth in the last years. Processing Historical Arabic Documents is a particularly challenging problem. First, due to complicated nature of Arabic script compared to other scripts and second because the documents are ancient. This paper focuses on this difficult problem and provides a comprehensive survey of existing research work. First, we describe in detail the challenges making the automatic processing of Historical Arabic Documents a difficult task. Second, we classify this task into four applications of automatic processing of HAD: i) Analyze the document to extract the main text ii) Identify the writer of the document iii) Recognize some words or parts of the document in a reference dataset andiv) Retrieve and extract specific data from the document. For each application, existing approaches are surveyed and qualitatively described. Finally, we focus on available datasets and describe how they can be used in each application.

International audience; Communication-enabled devices routinely carried by individuals have become pervasive, opening unprecedented opportunities for collecting digital metadata about the mobility of large populations. In this paper, we propose a novel methodology for the estimation of people density at metropolitan scales, using subscriber presence metadata collected by a mobile operator. Our approach suits the estimation of static population densities, i.e., of the distribution of dwelling units per urban area contained in traditional censuses. More importantly, it enables the estimation of dynamic population densities, i.e., the time-varying distributions of people in a conurbation. By leveraging substantial real-world mobile network metadata and ground-truth information, we demonstrate that the accuracy of our solution is superior to that granted by state-of-the-art methods in practical heterogeneous urban scenarios.

International audience; The recognition of human actions based on three-dimensional depth data has become a very active research field in computer vision. In this paper, we study the fusion at the feature and decision levels for depth data captured by a Kinect camera to improve action recognition. More precisely, from each depth video sequence, we compute Depth Motion Maps (DMM) from three projection views: front, side and top. Then shape and texture features are extracted from the obtained DMMs. These features are based essentially on Histogram of Oriented Gradients (HOG) and Local Binary Patterns (LBP) descriptors. We propose to use two fusion levels. The first is a feature fusion level and is based on the concatenation of HOG and LBP descriptors. The second, a score fusion level, based on the naive-Bayes combination approach, aggregates the scores of three classifiers: a collaborative representation classifier, a sparse representation classifier and a kernel based extreme learning machine classifier. The experimental results conducted on two public datasets, Kinect v2 and UTD-MHAD, show that our approach achieves a high recognition accuracy and outperforms several existing methods.

International audience; Speech analysis can be used for healthcare tasks such as pathology detection. Conventionally, a speech-language pathologist is specialized to detect anomalies from speech. Speech disorders result from a variety of causes such as brain injury, stroke, hearing loss, developmental delay or emotion alteration. Content of the speech is often not of interest for pathology detection, but characteristics are. In the literature of automatic pathology detection by speech analysis, physiological pathologies such as nodule and cancer are taken into account along with neurodegenerative brain disorders such as Parkinson's disease, Alzheimer's disease and mild cognitive impairment. As the problem of pathology detection from speech has become a vast research area, comprehensive reviews are needed by researchers to contribute novel approaches. In this study, a literature survey on pathology detection is provided including data types, features, classification methods and accuracy rates

L'objectif de cet article est d'analyser l'ecriture manuscrite en ligne, caracterisee par une intervention primordiale des muscles et du cerveau, afin de detecter des pathologies neurologiques comme l'Alzheimer et le Parkinson, qui sont liees a l'organisation de l'ensemble des facultes mentales. Dans ce cadre, une acquisition de donnees est realisee au sein du service neurologie au CHU Hassan II de Fes. Elle est faite sur une tablette graphique WACOM permettant de recuperer les donnees spatiotemporelles de l'ecriture manuscrite. La problematique de la detection de pathologies necessite prealablement l'etude des parametres descriptifs de l'ecriture manuscrite permettant la caracterisation des sujets controles et des personnes atteintes. Cet article decrit principalement la phase d'acquisition de donnees en cours au CHU de Fes, ainsi que l'analyse preliminaire de parametres spatiotemporels de l'ecriture realisee sur ces donnees.

International audience; As elderly care is getting more and more important, monitoring of activity of daily living (ADL) has become an active research topic. Both robotic and pervasive computing domains, through smart homes, are creating opportunities to move forward in ADL field. Multiple techniques were proposed to identify activities, each with their features, advantages and limits. However, it is a very challenging issue and none of the existing methods provides robust results, in particular in real daily living scenarios. This is particularly true for vision-based approaches used by robots. In this paper, we propose to refine a robot's visual activity recognition process by relying on smart home sensors. We assert that the consideration of further sensors and the knowledge about the target user together with the semantic by means of an ontology and a reasoning layer in the recognition process, has improved the existing works results. We experimented through multiple activity recognition scenarios with and without refinement to assess the relevance of such a combination. Although our tests reveal positive results, they also point out limits and challenges that we discuss in this paper

This paper provides an extended comparison of two temporal models for gesture recognition, namely Hybrid Neural Network-Hidden Markov Models (NN-HMM) and Recurrent Neural Networks (RNN) which have lately claimed the state-the-art performances. Experiments were conducted on both models in the same body of work, with similar representation learning capacity and comparable computational costs. For both solutions, we have integrated recent contributions to the model architectures and training techniques. We show that, for this task, Hybrid NN-HMM models remain competitive with Recurrent Neural Networks in a standard setting. For both models, we analyze the influence of the training objective function on the final evaluation metric. We further tested the influence of temporal convolution to improve context modeling, a technique which was recently reported to improve the accuracy of gesture recognition.

International audience; Finger-vein verification is a highly secure biometric authentication that has been widely investigated over the last years. One of its challenges, however, is the possible degradation of image quality, that results in spurious and missing vein patterns, which increases the verification error. Despite recent advances in finger-vein quality assessment, the proposed solutions are limited as they depend on human expertise and domain knowledge to extract handcrafted features for assessing quality. We have proposed, recently, the first Deep Neural Network (DNN) framework for assessing finger-vein quality, that does not require manual labeling of high and low quality images, as is the case for state of the art methods, but infers such annotations automatically based on an objective indicator, the biometric verification decision. This framework has significantly outperformed the existing methods, whether the input image is in grayscale or is binary. Motivated by these performances, we propose, in this work, a representation learning of finger vein image quality, where a DNN takes as input conjointly the grayscale and binary versions of the input image to predict vein quality. Our model allows to learn the joint representation from grayscale and binary images, for quality assessment. The experimental results, obtained on a large public dataset, demonstrates that our proposed method accurately identifies high and low quality images, and outperforms other techniques in terms of equal error rate (EER) minimization, including our previous DNN models, based either on grayscale or binary input.

International audience; Video-based recognition of activities of daily living (ADLs) is being used in ambient assisted living systems in order to support independent living of old people. In this work, we propose a new multimodal ADL recognition method by modeling the correlation between motion and object information. We encode motion using dense interest point trajectories which are robust to occlusion and speed variability. We formulate the learning problem using a two-layer SVM hidden conditional random field (HCRF) recognition model that is particularly relevant for multimodal sequence recognition. This hierarchical classifier opti-mally combines the discriminative power of SVM and the long-range feature dependencies modeling by the HCRF

Introduction L’ecriture est un processus cognitif qui resulte de l’interaction entre plusieurs facteurs du developpement cognitifs, linguistiques, et psychomoteurs. De ce fait, la deterioration de celle-ci est un signe de dysfonctionnement dans l’une des spheres cognitives. Objectifs Notre objectif est de developper une technique innovante, permettant d’aider les professionnels de sante a depister une deterioration cognitive a un stade precoce lors des maladies neurodegeneratives a l’aide de l’analyse de l’ecriture manuscrite. Methodes L’ecriture manuscrite est capturee sur tablette graphique (WACOM) et est analysee « en ligne » comme une sequence de signaux acquis (position, pression, vitesse et inclinaison du stylo) chez des patients marocains atteints de maladie de Parkinson, maladie d’Alzheimer et MCI ; et sera comparer a des volontaires normaux. Resultats Nous avons effectue une premiere analyse des resultats de 34 parkinsoniens compares a 34 controles apparies selon l’âge et le niveau d’etudes. Les resultats revelent des differences significatives entre le groupe temoin et le groupe Parkinson, mais aussi dans le groupe Parkinson entre des patients de bas niveau et haut niveau socioculturel. Les parkinsoniens ont une vitesse et une acceleration plus faible, avec des hesitations plus elevees que les SC. Discussion Nous retrouvons des parametres discriminants de l’ecriture manuscrite permettant d’identifier des clusters de personnes ayant un vieillissement normal et d’autres souffrants de pathologies neurodegeneratives en fonction de leurs niveau socioculturel. Conclusion L’etude a permis dans un premier temps de determiner un ensemble caracterisant les malades de Parkinson a un stade precoce par rapport aux sujets sains.

International audience; This paper presents a multi-shot person re-identification system from video sequences based on Interest Points (SURFs) matching. Our objective is to improve the Interest Points (IPs) matching using low resolution images in terms of re-identification accuracy and running time. First, we propose a new method of SURF matching via Local Sparse Representation (LSR). Each SURF in the test video sequence is expressed as a sparse representation of a subset of SURFs in the reference dataset. Our approach consists of searching the latter subset from the reference IPs that are located on a similar spatial neighborhood to the query IP. Second, it investigates whether IPs filtering can decrease the re-identification running time. An ensemble of binary classifiers are evaluated. Our approach is assessed on the large dataset PRID-2011 and shown to outperform favorably with current state of the art

International audience; Age characterization through handwriting is an important research field with several potential applications. It can, for instance, characterize normal aging process on one hand and detect significant handwriting degradation possibly related to early pathological states. In this work, we propose a novel approach to characterize age and gender from online handwriting styles. Contrary to previous works on handwriting style characterization, our contribution consists of a two-layer clustering scheme. At the first layer, we perform a writerindependent clustering on handwritten words, described by global features. At the second layer, we perform a clustering that considers style variation at the previous level for each writer, to provide a measure of his/her handwriting stability across words. We investigated different clustering algorithms and their effectiveness for each layer. The handwriting style patterns inferred by our novel technique show interesting correlations between handwriting, age and gender.

We describe an approach to conjointly locate and recognize a street name within a street line. The system developed is based on a probabilistic framework that naturally integrates various knowledge sources to emit a final decision. At the handwriting signal level, hidden Markov models are extensively used to provide the needed matching scores. Several optimization techniques are employed to speed up the processing time. Experiments carried out on large data sets of street line images, automatically extracted from real French mail envelope images, show very promising results.

International audience; This paper presents a comparative experimental study of the multidimensional indexing methods based on the approximation approach. We are particularly interested in the LSH family, which provides efficient index structures and solves the dimensionality curse problem. The goal is to understand the performance gain and the behavior of this family of methods on large-scale databases. E2LSH is compared to the KRA+-Blocks and the sequential scan methods. Two criteria are used in evaluating the E2LSH performances, namely average precision and CPU time using a database of one million image descriptors

International audience; This paper concerns the implementation of our online signature verification system on a mobile device. Verification involves confirming or denying a person's claimed identity. Our system is based on a Hidden Markov Model and outputs two complementary scores: the first one is related to the likelihood given by the HMM of the claimed identity; the second one is related to the segmentation given by such an HMM on the input signature. A claimed identity is confirmed when the arithmetic mean of the two scores obtained on such an input signature is higher than a threshold. Also, a personal normalization of the local parameters of the signature is carried out to make the system robust to changes of platforms. A patent was submitted with special emphasis on the latter claim. This system is implemented on a mobile platform PDA Qtek 2020 ARM 400 MHz. An acquisition interface is developed allowing an enrollment step of a person by acquisition of 5 of his/her signatures, and a verification step of a given signature of a registered person. Enrolment speed depends on the complexity of the signature, while verification is performed in real time. Performance assessment of our system, carried out on two databases acquired on a PDA, shows a degradation of system performance on mobile platform compared to a fixed platform. In order to improve the performance in the case of mobility, we propose a strategy for enhancing the quality of the reference signatures at the enrolment phase.

Locality Sensitive Hashing (LSH) is one of the most promising techniques for solving nearest Neighbours search problem in high dimensional space. Euclidean LSH is the most popular variation of LSH that has been successfully applied in many multimedia applications. However, the Euclidean LSH presents limitations that affect search performances. The main limitation of the Euclidean LSH is the large memory consumption. In order to achieve a good accuracy, a large number of hash tables is required. This paper propose a new hashing algorithm to overcome the storage space problem, while keeping a good accuracy and better query time. The Experimental results on a real large scale dataset show the interest of our approach, Locality Sensitive Hashing (LSH) est l'une des techniques les plus prometteuses pour la résolution des problèmes de la recherche des plus proches voisins dans l'espace de grande dimension. Euclidien Exact LSH (E2LSH) est la variante la plus populaire du LSH qui a été appliquée avec succès dans de nombreuses applications multimédia. Toutefois, l'E2LSH présente des limitations qui affectent les performances de recherche. La principale limitation de l'E2LSH est l'espace mémoire important utilisé. Afin de parvenir à une bonne qualité de recherche, un grand nombre de tables de hachage est nécessaire. Ce papier propose un nouvel algorithme de hachage pour remédier au problème d'espace de stockage, tout en conservant la bonne qualité de recherche et un meilleur temps de calcul. Les résultats expérimentaux obtenus sur une base de données réelle à grand échelle montrent l'intérêt de notre approche

International audience; In this paper, we propose a novel human activity recognition approach based on STIPs' trajectories as local descriptors of video sequences. This representation compares favorably with state of art feature extraction methods. In addition, we investigate the use of SVM/HCRF combination for temporal sequence modeling, where SVM is applied locally on short video segments to produce probability scores, the latter being considered as the input vectors to HCRF. This method constitutes a new contribution to the state of the art on activity recognition task. The obtained results demonstrate that our method is efficient and compares favorably with state of the art methods on human activity recognition

International audience; Understanding how handwriting (HW) style evolves as people get older may be key for assessing the health status of elder people. It can help, for instance, distinguishing HW change due to a normal aging process from change triggered by the early manifestation of a neurodegenerative pathology. We present, in this paper, an approach, based on a 2-layer clustering scheme that allows uncovering the main styles of online HW acquired on a digitized tablet, with a special emphasis on elder HW styles. The 1st level separates HW words into writer-independent clusters according to raw spatial-dynamic HW information, such as slant, curvature, speed, acceleration and jerk. The 2nd level operates at the writer level by converting the set of words of each writer into a Bag of 1st Layer Clusters, that is augmented by a multidimensional description of his/her writing stability across words. This 2nd layer representation is input to another clustering algorithm that generates categories of writer styles along with their age distributions. We have carried out extensive experiments on a large public online HW database, augmented by HW samples acquired at Broca hospital in Paris from people mostly between 60 and 85 years old. Unlike previous works claiming that there is only one pattern of HW change with age, our study reveals basically three major HW styles associated with elder people, among which one is specific to elders while the two others are shared by other age groups

National audience; Nous présentons dans ce papier, un système de vision, intégré dans le robot Nao pour la reconnaissance d'activités humaines réalisées par une personne dans des conditions réelles. Contrairement à une caméra fixe, le robot considéré peut filmer les activités à des endroits différents, ce qui implique qu'il doit opérer dans des conditions plus complexes, relativement aux fonds de scènes, aux conditions d'éclairage, aux angles de vues, et aux distances du robot par rapport à la personne. Le système développé est fondé sur l'extraction de trajectoires de points d'intérêt spatio-temporels uniformément et densément échantillonnés sur la scène, et l'encodage de ceux-ci par des histogrammes de gradients sur des volumes définis par leurs trajectoires. La représentation ainsi obtenue est ensuite transformée en un " sac de mots " (Bag of Words) qui est ensuite classifié par un SVM (Support Vector Machines). Ce système a été implémenté dans le robot Nao et testé dans des conditions réelles en considérant onze activités humaines. Les performances obtenues sont satisfaisantes dans l'ensemble, spécialement au regard des contraintes d'implémentation et de ressources de calcul

International audience; This paper presents an alternative to the mainstream approach of STIP-based SVM recognition for human recognition. First, it studies whether or not whole silhouette representation by Histogram-of-Oriented-Gradients (HOG) or Histogram-of-Optical-Flow (HOF) descriptors is more discriminated when compared to sparse spatio-temporal interest points (STIPs). Second, it investigates whether explicitly modeling the temporal order of features using continuous HMMs outperforms the standard Bag-of-Words (BoW) representation that overlooks such an order. When both whole silhouette representation and temporal order modeling are combined, a significant improvement is shown on the Weizmann database over STIP-based SVM recognition.