Search

Purpose The purpose of this study was to assess the precision of four-dimensional (4D) phase-contrast magnetic resonance imaging (PCMRI) to measure mean flow and peak velocity (Vmax) in a pulsatile flow phantom and to test its sensitivity to spatial resolution and Venc. Material and methods The pulsatile flow phantom consisted of a straight tube connected to the systemic circulation of an experimental mock circulatory system. Four-dimensional-PCMR images were acquired using different spatial resolutions (minimum pixel size: 1.5 × 1.5 × 1.5 mm3) and velocity encoding sensitivities (up to three times Vmax). Mean flow and Vmax calculated from 4D-PCMRI were compared respectively to the reference phantom flow parameters and to Vmax obtained from two-dimensional (2D)-PCMRI. Results 4D-PCI measured mean flow with a precision of −0.04% to + 5.46%, but slightly underestimated Vmax when compared to 2D-PCMRI (differences ranging from −1.71% to −3.85%). 4D PCMRI mean flow measurement was influenced by spatial resolution (P Conclusion Using an experimental pulsatile flow model 4D-PCMRI is accurate to measure mean flow and Vmax with better results obtained with higher spatial resolution. We also show that Venc up to 3 times higher than Vmax may be used with no effect on these measurements.

International audience; Semi-automatic measurements are performed on 18 FDG PET-CT images to monitor the evolution of metastatic sites in the clinical follow-up of metastatic breast cancer patients. Apart from being time-consuming and prone to subjective approximation, semi-automatic tools cannot make the difference between cancerous regions and active organs, presenting a high 18 FDG uptake. In this work, we combine a deep learning-based approach with a superpixel segmentation method to segment the main active organs (brain, heart, bladder) from full-body PET images. In particular, we integrate a superpixel SLIC algorithm at different levels of a convolutional network. Results are compared with a deep learning segmentation network alone. The methods are cross-validated on full-body PET images of 36 patients and tested on the acquisitions of 24 patients from a different study center, in the context of the ongoing EPICURE seinmeta study. The similarity between the manually defined organ masks and the results is evaluated with the Dice score. Moreover, the amount of false positives is evaluated through the positive predictive value (PPV). According to the computed Dice scores, all approaches allow to accurately segment the target organs. However, the networks integrating superpixels are better suited to transfer knowledge across datasets acquired on multiple sites (domain adaptation) and are less likely to segment structures outside of the target organs, according to the PPV. Hence, combining deep learning with superpixels allows to segment organs presenting a high 18 FDG uptake on PET images without selecting cancerous lesion, and thus improves the precision of the semi-automatic tools monitoring the evolution of breast cancer metastasis. Clinical relevance-We demonstrate the utility of combining deep learning and superpixel segmentation methods to accurately find the contours of active organs from metastatic breast cancer images, to different dataset distributions.

International audience; 18 FDG PET/CT imaging is commonly used in diagnosis and follow-up of metastatic breast cancer, but its quantitative analysis is complicated by the number and location heterogeneity of metastatic lesions. Considering that bones are the most common location among metastatic sites, this work aims to compare different approaches to segment the bones and bone metastatic lesions in breast cancer. Two deep learning methods based on U-Net were developed and trained to segment either both bones and bone lesions or bone lesions alone on PET/CT images. These methods were cross-validated on 24 patients from the prospective EPICURE seinmeta metastatic breast cancer study and were evaluated using recall and precision to measure lesion detection, as well as the Dice score to assess bones and bone lesions segmentation accuracy. Results show that taking into account bone information in the training process allows to improve the precision of the lesions detection as well as the Dice score of the segmented lesions. Moreover, using the obtained bone and bone lesion masks, we were able to compute a PET bone index (PBI) inspired by the recognized Bone Scan Index (BSI). This automatically computed PBI globally agrees with the one calculated from ground truth delineations. Clinical relevance - We propose a completely automatic deep learning based method to detect and segment bones and bone lesions on 18 FDG PET/CT in the context of metastatic breast cancer. We also introduce an automatic PET bone index which could be incorporated in the monitoring and decision process.

Disentangling and recovering physical attributes, such as shape and material, from a few waveform examples is a challenging inverse problem in audio signal processing, with numerous applications in musical acoustics as well as structural engineering. We propose to address this problem via a combination of time--frequency analysis and supervised machine learning. We start by synthesizing a dataset of sounds using the functional transformation method. Then, we represent each percussive sound in terms of its time-invariant scattering transform coefficients and formulate the parametric estimation of the resonator as multidimensional regression with a deep convolutional neural network. We interpolate scattering coefficients over the surface of the drum as a surrogate for potentially missing data, and study the response of the neural network to interpolated samples. Lastly, we resynthesize drum sounds from scattering coefficients, therefore paving the way towards a deep generative model of drum sounds whose latent variables are physically interpretable., 11 pages, 7 figures. To appear in the Proceedings of Forum Acusticum, Lyon (France), December 2020

Speckle based imaging consists of forming a super-resolved reconstruction of an unknown sample from low-resolution images obtained under random inhomogeneous illuminations (speckles). In a blind context where the illuminations are unknown, we study the intrinsic capacity of speckle-based imagers to recover spatial frequencies outside the frequency support of the data, with minimal assumptions about the sample. We demonstrate that, under physically realistic conditions, the covariance of the data has a super-resolution power corresponding to the squared magnitude of the imager point spread function. This theoretical result is important for many practical imaging systems such as acoustic and electromagnetic tomographs, fluorescence and photoacoustic microscopes, or synthetic aperture radar imaging. A numerical validation is presented in the case of fluorescence microscopy., Accepted in IEEE Trans. CI (Final version submitted on Nov. 2017)

International audience; Context Multiple myeloma (MM) is a bone marrow cancer that accounts for 10% of all haematological malignancies. It was reported that full-body FDG PET imaging provides prognostic information for both baseline and therapeutic follow-up of MM patients (MM). Aims Predict Progression-Free Survival (PFS). Provide predictive features (Clinics and Radiomics). Contribution There is yet much to discover in the survival analysis of MM. However, the Random Survival Forest (RSF)[2] has demonstrated robustness but is not studied in the PET imaging and MM context.We developed a two-stage computer-assisted method based on PET imaging features towards assisting current diagnosis and treatment decisions for MM patients, with RSF and "Variable importance" (VIMP) [2]. Definitions Right censoring: When no event (death/relapse) has taken place at the end of the evaluation period. C-index: The concordance probability is the frequency of concordant pairs among all pairs of subjects. Error prediction = 1-C-index Survival curve: Survival rates of a specific population, over a period of time.

International audience; Contexte L'imagerie TEP-FDG corps entier procure des informations pronostiques dans le bilan initial et le suivi thérapeutique des patients atteints de myélome multiple (MM). Etat de l'art Les biomarqueurs par imagerie TEP ont été peu étudiés dans ce contexte et souvent en utilisant des méthodes de Cox et de Kaplan-Meier. Cependant, une approche de type "Random Survival Forest" (RSF), peut fournir un procédé plus robuste pour l'analyse de la survie et n'est pour le moment pas étudié dans le cadre du MM et de l'imagerie TEP. Proposition Nous proposons une méthode pour la prédiction de la progression dans le contexte du MM en utilisant des RSF avec la méthode de sélection par 'Variable importance' (VIMP) [Ishwaran et al., Annals of Applied Statistics, 2008]. Matériel et méthodes 1) Les données • Etude multicentrique et prospective, IMAJEM [Moreau et al. J Clin Oncol 2017] • 99 patients • Prise en compte des caractéristiques cliniques et d'imagerie. 2) La méthode Résultats 2) Caractéristiques prédictives Figure 4 : Diagramme des caractéristiques trouvées comme prédictives par la méthode VIMP (Le nombre optimal de features gardées est entre 5 et 7.) Résultats 1) Evaluation de la méthode Figure 2 : Erreur de prédiction sur 50 itérations, en fonction de la méthode. GB : gradient boosting machine with Cox, GBCI : gradient boosting machine for concordance index, Lasso : Lasso-cox, Our method : sélection VIMP plus RSF, Without : RSF sans sélection préalable. Figure 3 : A) Courbe de Kaplan Meier dans deux groupes de pronostic obtenues sur le set de test (en rose : Bon pronostic, en bleu : mauvais pronostic). Erreur : 0,30, p-value = 0,0058, valeur log-rank = 7,6, nombre de patients par groupe = 12. B) P-value calculée pour 50 itérations pour différentes méthode de sélection suivies d'une RSF. Our method : VIMP, VH : Variable-Hunting, MD : Minimal Depth, Without : sans sélection. Conclusions et perspectives La méthode proposée permet de décrire un modèle plus performant que les approches conventionnelles pour la prédiction au bilan initial de la survie sans progression des patients atteints de myléome multiple. Nous avons montré qu'il était possible de correctement séparer deux classes de patients (bon/mauvais pronostic) à partir de données indépendantes. Elle permet également de déterminer les paramètres clinico-biologiques et par imagerie les plus prédictifs de la progression. Par la suite, plus de patients et de paramètres vont être inclus avec notamment des caractéristiques radiomiques. Cette approche peut être généralisée à d'autres pathologies. Remerciements : Ce travail a été partiellement financé par SIRIC ILIAD et MILCOM Connect Talent.

International audience; —Multiple sclerosis (MS) is a major auto-immune disease that is the leading cause of non-traumatic impairment of the central nervous system (CNS) in young adults. Successful treatment of MS patients depends on accurate tools for both the MS diagnosis and the disability progression. In current and up-coming studies the authors aim to explore the capabilities of applying a commercial electromyographic and inertial sensor (MYO Armband by Thalmic Labs Inc.), coupled with a multichannel signal processing tool, to standard neurological examination of MS progression. In this pilot study we formulate a two-class " healthy control "-" having MS " classification problem. A dataset of electromyographic signals and inertial sensor measurements from 71 individuals (31 MS patients and 40 healthy controls) was acquired during standard neurological examination routine. Temporal and spectral features of the signals were extracted in order to train and validate a classification model. Finally, a Support Vector Machine classifier was obtained giving AUROC = 0.94, 95% CI = [0.88, 0.99]. We propose a set of signal descriptors that correlate with objective components of the neurological examination. The proposed signal acquisition and processing technique, being easy to integrate into the traditional neurological exam, may have high potential for aiding in quantifying MS progression.

International audience; Computer sound matching poses an inverse problem, namely, the identification of resynthesis parameters. Borrowing from the differentiable digital signal processing (DDSP) framework, we propose to automate its resolution by training a deep neural network. In this context, we aim to reach a compromise between the computational efficiency of parametric loss (P-loss) versus the psychoacoustical fidelity of spectral loss. Our approach, named ``perceptual--neural--physical'' (PNP), estimates the Riemannian metric which is associated to the composition between parametric synthesis and time--frequency scattering. By doing so, we locally linearize spectral loss and accelerate convergence. Furthermore, resorting to Tikhonov regularization improves the conditioning of the inverse problem. On an analysis--synthesis task for musical arpeggios, PNP training outperforms state-of-the-art methods P-loss (wav2shape) and STFT-based DDSP, as measured in terms of JTFS-based similarity between reference signal and reconstructed signal.; La transformation de sons par ordinateur pose un problème inverse d'identification des paramètres de resynthèse adéquats. Empruntant au formalisme du traitement du signal différentiable (DDSP), nous proposons d'automatiser sa résolution par entrainement d'un réseau de neurones profond. Dans ce contexte, nous visons un compromis entre l'efficacité computationnelle de la perte paramétrique et la fidélité psychoacoustique de la perte spectrale. Notre approche, baptisée perceptuelle-neuronalephysique (PNP), consiste à estimer la métrique riemannienne associée à la composition entre synthèse paramétrique et diffusion temps-fréquence (JTFS). Ce faisant, nous linéarisons localement la perte spectrale et accélérons la convergence. De plus, le recours à une régularisation de Tikhonov améliore le conditionnement du problème inverse. Par rapport à l'état de l'art (wav2shape et DDSP), et pour une tâche difficile d'analyse-synthèse d'arpège musical, l'entrainement via PNP rapproche le signal reconstruit du signal de référence, d'après une mesure de similarité de timbre fondée sur la JTFS.

International audience; Sound matching algorithms seek to approximate a target waveform by parametric audio synthesis. Deep neural networks have achieved promising results in matching sustained harmonic tones. However, the task is more challenging when targets are nonstationary and inharmonic, e.g., percussion. We attribute this problem to the inadequacy of loss function. On one hand, mean square error in the parametric domain, known as "P-loss", is simple and fast but fails to accommodate the differing perceptual significance of each parameter. On the other hand, mean square error in the spectrotemporal domain, known as "spectral loss", is perceptually motivated and serves in differentiable digital signal processing (DDSP). Yet, spectral loss is a poor predictor of pitch intervals and its gradient may be computationally expensive; hence a slow convergence. Against this conundrum, we present Perceptual-Neural-Physical loss (PNP). PNP is the optimal quadratic approximation of spectral loss while being as fast as P-loss during training. We instantiate PNP with physical modeling synthesis as decoder and joint time-frequency scattering transform (JTFS) as spectral representation. We demonstrate its potential on matching synthetic drum sounds in comparison with other loss functions.

International audience; Survival analysis of DLBCL patients requires the interpretation of PET images characterised by multiple small lesions. Current machine-learning approaches addressing similar problems consider as input the cropped image of a single lesion or the whole volume. In this paper, we incorporate the information of all lesions by modeling their joint survival analysis with a graph learning approach. We propose a compact graph representation of the segmented lesions enriched by radiomics features and edge weights. The representation is fed to a graph attention network to predict the 2-year Progression-Free Survival of a DLBCL patient, formalised as a graph classification problem. Experimental results on a clinical prospective database with 583 patients show that our method improves over three baseline fusion approaches.

International audience; Un cadre perceptual--neuronal--physique pour l'écoute artificielle

During an activity, knowing the mental workload (MWL) of the user allows to improve the Human-Machine Interactions (HMI). Indeed, the MWL has an impact on the individual and its interaction with the environment. Monitoring it is therefore a crucial issue. In this context, we have created the virtual game Back to Pizza which is based on the N-back task (commonly used for measuring MWL). In this more playful variant, users must carry out orders from customers of a pizza food truck. It is an interactive game that involves the audience of the IHM'23 conference, choosing several parameters like the number of ingredients. During this experience, the objective is to measure MWL in real time through an ElectroEncephaloGraph (EEG) and visual feedback on MWL level is given to the audience. With this demonstration, we propose to present a concept of a virtual interactive game that measures MWL in real time., Comment: IHM'23 - 34e Conf{\'e}rence Internationale Francophone sur l'Interaction Humain-Machine, AFIHM; Universit{\'e} de Technologie de Troyes, Apr 2023, Troyes, France

International audience; The "Holistic Evaluation of Audio Representations" (HEAR) is an emerging research program towards statistical models that can transfer to diverse machine listening tasks. The originality of HEAR is to conduct a fair, "apples-to-apples" comparison of many deep learning models over many datasets, resulting in multitask evaluation metrics that are readily interpretable by practitioners. On the flip side, this comparison incurs a neural architecture search: as such, it is not directly interpretable in terms of audio signal processing. In this paper, we propose a complementary viewpoint on the HEAR benchmark, which we name GEAR: Generative Evaluation of Audio Representations. The key idea behind GEAR is to generate a dataset of sounds with few independent factors of variability, analyze it with HEAR embeddings, and visualize it with an unsupervised manifold learning algorithm. Visual inspection reveals stark contrasts in the global structure of the nearest-neighbor graphs associated to logmelspec, Open-L 3 , BYOL, CREPE, wav2vec2, GURA, and YAMNet. Although GEAR currently lacks mathematical refinement, we intend it as a proof of concept to show the potential of parametric audio synthesis in general-purpose machine listening research.

Tools based on deep learning models have been created in recent years to aid radiologists in the diagnosis of breast cancer from mammograms. However, the datasets used to train these models may suffer from class imbalance, i.e., there are often fewer malignant samples than benign or healthy, which can bias the model towards the healthy class. In this study, we systematically evaluate several popular techniques to deal with this class imbalance, namely, class weighting, over-sampling, and under-sampling, as well as a synthetic lesion generation approach to increase the number of malignant samples. These techniques are applied when training on three diverse Full-Field Digital Mammography datasets, and tested on in-distribution and out-of-distribution samples. The experiments show that a greater imbalance is associated with a greater bias towards the majority class, which can be counteracted by any of the standard class imbalance techniques. On the other hand, these methods provide no benefit to model performance with respect to Area Under the Curve of the Recall Operating Characteristic (AUC-ROC), and indeed under-sampling leads to a reduction of 0.066 in AUC in the case of a 19:1 benign to malignant imbalance. Our synthetic lesion methodology leads to better performance in most cases, with increases of up to 0.07 in AUC on out-of-distribution test sets over the next best experiment.

International audience; The popularity of yttrium-90 ( 90 Y) PET is growing. However, due to the very low branching ratio of 90 Y ( 3.2×10−5 ), images reconstructed are characterized by a high noise level and positive bias in low-activity regions. To overcome this problem, some algorithms use penalized reconstructions, and others allow for negative values in the image. Recently, a post-processing method has also been proposed that removes the induced negative values while maintaining bias reduction. The work presented in this article aims to evaluate and compare these methods to guide the reader in the choice of the best-suited reconstruction algorithm and associated parameters. First, several algorithms were tested using experimental phantom data. Pareto fronts and Pareto sets from the multiobjective optimization formalism were used for the comparison. Next, a dosimetric study (using phantom and patient data) was conducted to assess the final impact of these algorithms. The lowest biases were reached by unconstrained algorithms. When compared to penalized algorithms, the latter allowed for noise reduction at a fixed level of bias, with the best results obtained using a penalty based on relative differences. A good compromise between noise and bias could be reached by combining penalty, early stopping of iterative algorithms, unconstrained algorithms, and post-processing.

Instrumental playing techniques such as vibratos, glissandos, and trills often denote musical expressivity, both in classical and folk contexts. However, most existing approaches to music similarity retrieval fail to describe timbre beyond the so-called "ordinary" technique, use instrument identity as a proxy for timbre quality, and do not allow for customization to the perceptual idiosyncrasies of a new subject. In this article, we ask 31 human subjects to organize 78 isolated notes into a set of timbre clusters. Analyzing their responses suggests that timbre perception operates within a more flexible taxonomy than those provided by instruments or playing techniques alone. In addition, we propose a machine listening model to recover the cluster graph of auditory similarities across instruments, mutes, and techniques. Our model relies on joint time--frequency scattering features to extract spectrotemporal modulations as acoustic features. Furthermore, it minimizes triplet loss in the cluster graph by means of the large-margin nearest neighbor (LMNN) metric learning algorithm. Over a dataset of 9346 isolated notes, we report a state-of-the-art average precision at rank five (AP@5) of $99.0\%\pm1$. An ablation study demonstrates that removing either the joint time--frequency scattering transform or the metric learning algorithm noticeably degrades performance., Comment: 32 pages, 5 figures. To appear in EURASIP Journal on Audio, Speech, and Music Processing (JASMP)

Yttrium-90 PET imaging is becoming increasingly popular. However, the probability that decay of a yttrium-90 nucleus will lead to the emission of a positron is only 3.2 × 10−5, and the reconstructed images are therefore characterised by a high level of noise, as well as a positive bias in low activity regions. To correct these problems, classical methods use penalised algorithms or allow negative values in the image. However, a study comparing and combining these different methods in the specific context of yttrium-90 was still missing at the beginning of this thesis. This thesis, therefore, aims to fill this gap. Unfortunately, the methods allowing negative values cannot be used directly in a dosimetric study. Therefore, this thesis starts by proposing a new method of post-processing the images, aiming to remove the negative values while keeping the average values as locally as possible. A complete multi-objective analysis of these different methods is then proposed. This thesis ends by laying the foundations of what could become an algorithm providing a set of adequate reconstruction hyper-parameters from sinograms alone.; La popularité de l'imagerie TEP à l'yttrium 90 va grandissante. Cependant, la probabilité qu'une désintégration d'un noyau d'yttrium 90 mène à l'émission d'un positon n'est que de 3,2 × 10−5, et les images reconstruites sont par conséquent caractérisées par un niveau de bruit élevé, ainsi que par un biais positif dans les régions de faible activité. Pour corriger ces problèmes, les méthodes classiques consistent à utiliser des algorithmes pénalisés, ou autorisant des valeurs négatives dans l'image. Cependant, une étude comparant et combinant ces différentes méthodes dans le contexte spécifique de l'yttrium 90 manquait encore à l'appel au début de cette thèse. Cette dernière vise donc à combler ce manque. Malheureusement, les méthodes autorisant les valeurs négatives ne peuvent pas être utilisées directement dans le cadre d'une étude dosimétrique, et cette thèse commence donc par proposer une nouvelle méthode de post-traitement des images, visant à en supprimer les valeurs négatives en en conservant les valeurs moyennes le plus localement possible. Une analyse complète multi-objectifs de ces différentes méthodes est ensuite proposée. Cette thèse se termine en posant les prémices de ce qui pourra devenir un algorithme permettant de proposer un jeu d'hyper-paramètres de reconstruction adéquats, à partir des seuls sinogrammes.

Yttrium-90 PET imaging is becoming increasingly popular. However, the probability that decay of a yttrium-90 nucleus will lead to the emission of a positron is only 3.2 × 10−5, and the reconstructed images are therefore characterised by a high level of noise, as well as a positive bias in low activity regions. To correct these problems, classical methods use penalised algorithms or allow negative values in the image. However, a study comparing and combining these different methods in the specific context of yttrium-90 was still missing at the beginning of this thesis. This thesis, therefore, aims to fill this gap. Unfortunately, the methods allowing negative values cannot be used directly in a dosimetric study. Therefore, this thesis starts by proposing a new method of post-processing the images, aiming to remove the negative values while keeping the average values as locally as possible. A complete multi-objective analysis of these different methods is then proposed. This thesis ends by laying the foundations of what could become an algorithm providing a set of adequate reconstruction hyper-parameters from sinograms alone., La popularité de l'imagerie TEP à l'yttrium 90 va grandissante. Cependant, la probabilité qu'une désintégration d'un noyau d'yttrium 90 mène à l'émission d'un positon n'est que de 3,2 × 10−5, et les images reconstruites sont par conséquent caractérisées par un niveau de bruit élevé, ainsi que par un biais positif dans les régions de faible activité. Pour corriger ces problèmes, les méthodes classiques consistent à utiliser des algorithmes pénalisés, ou autorisant des valeurs négatives dans l'image. Cependant, une étude comparant et combinant ces différentes méthodes dans le contexte spécifique de l'yttrium 90 manquait encore à l'appel au début de cette thèse. Cette dernière vise donc à combler ce manque. Malheureusement, les méthodes autorisant les valeurs négatives ne peuvent pas être utilisées directement dans le cadre d'une étude dosimétrique, et cette thèse commence donc par proposer une nouvelle méthode de post-traitement des images, visant à en supprimer les valeurs négatives en en conservant les valeurs moyennes le plus localement possible. Une analyse complète multi-objectifs de ces différentes méthodes est ensuite proposée. Cette thèse se termine en posant les prémices de ce qui pourra devenir un algorithme permettant de proposer un jeu d'hyper-paramètres de reconstruction adéquats, à partir des seuls sinogrammes.

In breast cancer screening, binary classification of mammograms is a common task aiming to determine whether a case is malignant or benign. A Computer-Aided Diagnosis (CADx) system based on a trainable classifier requires clean data and labels coming from a confirmed diagnosis. Unfortunately, such labels are not easy to obtain in clinical practice, since the histopathological reports of biopsy may not be available alongside mammograms, while normal cases may not have an explicit follow-up confirmation. Such ambiguities result either in reducing the number of samples eligible for training or in a label uncertainty that may decrease the performances. In this work, we maximize the number of samples for training relying on multi-task learning. We design a deep-neural-network-based classifier yielding multiple outputs in one forward pass. The predicted classes include binary malignancy, cancer probability estimation, breast density, and image laterality. Since few samples have all classes available and confirmed, we propose to introduce the uncertainty related to the classes as a per-sample weight during training. Such weighting prevents updating the network's parameters when training on uncertain or missing labels. We evaluate our approach on the public INBreast and private datasets, showing statistically significant improvements compared to baseline and independent state-of-the-art approaches. Moreover, we use mammograms from Susan G. Komen Tissue Bank for fine-tuning, further demonstrating the ability to improve the performances in our multi-task learning setup from raw clinical data. We achieved the binary classification performance of AUC = 80.46 on our private dataset and AUC = 85.23 on the INBreast dataset.

International audience; Real-time intramuscular electromyography (iEMG) decomposition, as an identification procedure of individual motor neuron (MN) discharge timings from a streaming iEMG recording, has the potential to be used in human-machine interfacing. However, for these applications, the decomposition accuracy and speed of current approaches need to be improved. Methods: In our previous work, a real-time decomposition algorithm based on a Hidden Markov Model of EMG, using GPUimplemented Bayesian filter to estimate the spike trains of motor units (MU) and their action potentials (MUAPs), was proposed. In this paper, a substantially extended version of this algorithm that boosts the accuracy while maintaining real-time implementation, is introduced. Specifically, multiple heuristics that aim at resolving the problems leading to performance degradation, are applied to the original model. In addition, the recursive maximum likelihood (RML) estimator previously used to estimate the statistical parameters of the spike trains, is replaced by a linear regression (LR) estimator, which is computationally more efficient, in order to ensure real-time decomposition with the new heuristics. Results: The algorithm was validated using twenty-one experimental iEMG signals acquired from the tibialis anterior muscle of five subjects by fine wire electrodes. All signals were decomposed in real time. The decomposition accuracy depended on the level of muscle activation and was >90% when less than 10 MUs were identified, substantially exceeding previous real-time results. Conclusion: Single channel iEMG signals can be very accurately decomposed in real time with the proposed algorithm. Significance: The proposed highly accurate algorithm for singlechannel iEMG decomposition has the potential of providing neural information on motor tasks for human interfacing.

Few-shot sound event detection is the task of detecting sound events, despite having only a few labelled examples of the class of interest. This framework is particularly useful in bioacoustics, where often there is a need to annotate very long recordings but the expert annotator time is limited. This paper presents an overview of the second edition of the few-shot bioacoustic sound event detection task included in the DCASE 2022 challenge. A detailed description of the task objectives, dataset, and baselines is presented, together with the main results obtained and characteristics of the submitted systems. This task received submissions from 15 different teams from which 13 scored higher than the baselines. The highest F-score was of 60% on the evaluation set, which leads to a huge improvement over last year's edition. Highly-performing methods made use of prototypical networks, transductive learning, and addressed the variable length of events from all target classes. Furthermore, by analysing results on each of the subsets we can identify the main difficulties that the systems face, and conclude that few-show bioacoustic sound event detection remains an open challenge., Comment: submitted to DCASE2022 workshop

An adequate classification of proximal femur fractures from X-ray images is crucial for the treatment choice and the patients' clinical outcome. We rely on the commonly used AO system, which describes a hierarchical knowledge tree classifying the images into types and subtypes according to the fracture's location and complexity. In this paper, we propose a method for the automatic classification of proximal femur fractures into 3 and 7 AO classes based on a Convolutional Neural Network (CNN). As it is known, CNNs need large and representative datasets with reliable labels, which are hard to collect for the application at hand. In this paper, we design a curriculum learning (CL) approach that improves over the basic CNNs performance under such conditions. Our novel formulation reunites three curriculum strategies: individually weighting training samples, reordering the training set, and sampling subsets of data. The core of these strategies is a scoring function ranking the training samples. We define two novel scoring functions: one from domain-specific prior knowledge and an original self-paced uncertainty score. We perform experiments on a clinical dataset of proximal femur radiographs. The curriculum improves proximal femur fracture classification up to the performance of experienced trauma surgeons. The best curriculum method reorders the training set based on prior knowledge resulting into a classification improvement of 15%. Using the publicly available MNIST dataset, we further discuss and demonstrate the benefits of our unified CL formulation for three controlled and challenging digit recognition scenarios: with limited amounts of data, under class-imbalance, and in the presence of label noise. The code of our work is available at: https://github.com/ameliajimenez/curriculum-learning-prior-uncertainty., Comment: Medical Image Analysis

International audience; This position paper advocates for digital sobriety in the design and usage of wireless acoustic sensors. As of today, these devices all rely on batteries, which are either recharged by a human operator or via solar panels. Yet, batteries contain chemical pollutants and have a shorter lifespan than electronic components: as such, they hinder the autonomy and sustainability of the Internet of Sounds at large. Against this problem, our radical answer is to avoid the use of batteries altogether; and instead, to harvest ambient energy in real time and store it in a supercapacitor allowing a few minutes of operation. We show the inherent limitations of battery-dependent technologies for acoustic sensing. Then, we describe how a lowcost Micro-Controller Unit (MCU) could serve for audio acquisition and feature extraction on the edge. In particular, we stress the advantage of storing intermediate computations in ferroelectric random-access memory (FeRAM), which is nonvolatile, fast, endurant and consumes little. As a proof of concept, we present a simple-minded detector of sine tones in background noise, which relies on a fixed-point implementation of the fast Fourier transform (FFT). We outline future directions towards bioacoustic event detection and urban acoustic monitoring without batteries nor wires. CCS CONCEPTS • Hardware → Sound-based input / output; Digital signal processing; Power estimation and optimization; Impact on the environment; • Computer systems organization → Real-time systems; • Networks → Sensor networks.

The integration of users' perception in the design process is and important challenge for the optimization of products. This study describes how design recommendations can be drawn, from a perceptual experiment with a panel of subjects using a multi-objective interactive genetic algorithm (IGA). The application concerns the bi-objective optimization of the unpleasantness and the detectability of sounds for electric vehicles (EV). After a description of the experimental protocol for the assessment of the detectability and the unpleasantness of EV sounds (listening test), a set of optimal sounds (Pareto efficient) is defined with an IGA experiment. The analysis of these sounds, based on a probabilistic analysis of the selection process, leads to the definition of design recommendations. A second listening test, involving recommended sounds but also other design proposals, allows an evaluation of the validity of the approach. Results show that the sounds recommended obtain interesting performance, in particular to improve the detectability of EV sounds.

International audience; Real-time intramuscular electromyography (iEMG) decomposition, as an identification procedure of individual motor neuron (MN) discharge timings from a streaming iEMG recording, has the potential to be used in human-machine interfacing. However, for these applications, the decomposition accuracy and speed of current approaches need to be improved. Methods: In our previous work, a real-time decomposition algorithm based on a Hidden Markov Model of EMG, using GPUimplemented Bayesian filter to estimate the spike trains of motor units (MU) and their action potentials (MUAPs), was proposed. In this paper, a substantially extended version of this algorithm that boosts the accuracy while maintaining real-time implementation, is introduced. Specifically, multiple heuristics that aim at resolving the problems leading to performance degradation, are applied to the original model. In addition, the recursive maximum likelihood (RML) estimator previously used to estimate the statistical parameters of the spike trains, is replaced by a linear regression (LR) estimator, which is computationally more efficient, in order to ensure real-time decomposition with the new heuristics. Results: The algorithm was validated using twenty-one experimental iEMG signals acquired from the tibialis anterior muscle of five subjects by fine wire electrodes. All signals were decomposed in real time. The decomposition accuracy depended on the level of muscle activation and was >90% when less than 10 MUs were identified, substantially exceeding previous real-time results. Conclusion: Single channel iEMG signals can be very accurately decomposed in real time with the proposed algorithm. Significance: The proposed highly accurate algorithm for singlechannel iEMG decomposition has the potential of providing neural information on motor tasks for human interfacing.

International audience; In this paper the problem of restoration of unsupervised nonnegative sparse signals is addressed in the Bayesian framework. We introduce a new probabilistic hierarchical prior, based on the Generalized Hyperbolic (GH) distribution, which explicitly accounts for sparsity. On the one hand, this new prior allows us to take into account the non-negativity. On the other hand, thanks to the decomposition of GH distributions as continuous Gaussian mean-variance mixture, a partially collapsed Gibbs sampler (PCGS) implementation is made possible, which is shown to be more efficient in terms of convergence time than the classical Gibbs sampler.

International audience; The impact of urban sound on human beings has often been studied from a negative point of view (noise pollution). In the two last decades, the interest of studying its positive impact has been revealed with the soundscape approach (resourcing spaces). The literature shows that the recognition of sources plays a great role in the way humans are affected by sound environments. There is thus a need for characterizing urban acoustic environments not only with sound pressure measurements but also with source-specific attributes such as their perceived time of presence, dominance or volume.This paper demonstrates, on a controlled dataset, that machine learning techniques based on state of the art neural architectures can predict the perceived time of presence of several sound sources at a sufficient accuracy. To validate this assertion, a corpus of simulated sound scenes is first designed. Perceptual attributes corresponding to those stimuli are gathered through a listening experiment. From the contributions of the individual sound sources available for the simulated corpus, a physical indicator approximating the perceived time of presence of sources is computed and used to train and evaluate a multi-label source detection model. This model predicts the presence of simultaneously active sources from fast third octave spectra, allowing the estimation of perceptual attributes such as pleasantness in urban sound environments at a sufficient degree of precision.

International audience; Orthogonal greedy algorithms are popular sparse signal reconstruction algorithms. Their principle is to select atoms one by one. A series of unconstrained least-squares subproblems of gradually increasing size is solved to compute the approximation coefficients, which is efficiently performed using a fast recursive update scheme. When dealing with non-negative sparse signal reconstruction, a series of non-negative least-squares subproblems have to be solved. Fast implementation becomes tricky since each subproblem does not have a closed-form solution anymore. Recently, non-negative extensions of the classical orthogonal matching pursuit and orthogonal least squares algorithms were proposed, using slow (i.e., non-recursive) or recursive but inexact implementations. In this paper, we revisit these algorithms in a unified way. We define a class of non-negative orthogonal algorithms and exhibit their structural properties. We propose a fast and exact implementation based on the active-set resolution of non-negative least-squares and exploiting warm start initializations. The algorithms are assessed in terms of accuracy and computational complexity for a sparse spike deconvolution problem. We also present an application to near-infrared spectra decomposition.

Electric Vehicles (EVs) are very quite at low speed, which can be hazardous for pedestrians. It is necessary to add warning sounds but this can represent an annoyance if they are poorly designed. On the other hand, they can be not enough detectable because of the masking effect due to the background noise. In this paper, we propose a method for the design of EV sounds that takes into account in the same time detectability and unpleasantness. It is based on user tests and implements Interactive Genetic Algorithms (IGA) for the optimization of the sounds. Synthesized EV sounds, based on additive synthesis and filtering, are proposed to a set of participants during a hearing test. An experimental protocol is proposed for the assessment of the detectability and the unpleasantness of the EV sounds. After the convergence of the method, sounds obtained with the IGA are compared to different sound design proposals. Results show that the quality of the sounds designed by the IGA method is significantly higher than the design proposals, validating the relevance of the approach.

With an aging population, sedentary lifestyles, an increase in strokes and other motor deficiencies, mobility is a global concern of paramount importance. To this problem, exoskeletons and orthoses are one possible technological solution.An orthosis is a multidisciplinary system that must adapt to the user's body to assist him in his motion. Due to its complex design, it involves physical, technological and control constraints on the user. The materials used, the shape, the information and power chain, the human morphology and the motion are all factors that influence the type and amplitude of the constraints. If these constraints are too important, the user will have to adapt and modify his gait consequently.To predict the physical constraints applied by a knee orthosis on a human being during a cyclic motion such as squatting or walking, a simulator has been developed. It enables the generation of optimal trajectories in accordance with a cost function and the computation of the joint torques required to achieve the motion. An experimental study was conducted on several subjects to understand how weight is distributed during a squat motion. Based on these results, a model for the forces distribution was proposed.; Avec le vieillissement de la population, la sédentarité, l’augmentation des accidents vasculaires cérébraux et autres déficiences motrices, la mobilité est un enjeu mondial primordial. À ce problème, les exosquelettes et les orthèses constituent une réponse technologique possible.Une orthèse est un système pluridisciplinaire qui doit s’adapter au corps de l’utilisateur pour l’assister dans son mouvement. De par sa nature complexe, elle impose à l’utilisateur des contraintes physiques, technologiques et liées à la commande. Les matériaux utilisés, la géométrie, la chaîne de puissance, la chaîne d’information, la morphologie de l’être humain et la nature du mouvement sont autant d’éléments qui influent sur la nature et l’amplitude des contraintes. Si celles-ci sont trop importantes, l’utilisateur va devoir s’adapter et modifier sa démarche en conséquence.Pour prévoir les contraintes physiques appliquées par une orthèse de genou sur un individu au cours d’un mouvement cyclique tel que le squat ou la marche, un simulateur a été développé. Il permet de générer des trajectoires optimales au sens d’une fonction de coût et de calculer les couples articulaires nécessaires à la réalisation du mouvement. Une étude expérimentale a été menée sur plusieurs sujets pour comprendre comment le poids se répartit au cours d’un mouvement de squat. Basé sur ces résultats, un modèle de répartition des efforts a été proposé.

Digital Breast Tomosynthesis (DBT) is an emerging imaging technique for breast cancer screening aiming to overcome certain limitations of traditional mammography, such as the superimposition of tissues. On the downside, DBT increases the radiologists’ workload as it generates stacks of high-resolution images, which are time-consuming to review and annotate. In this work, we propose a deep- multiple-instance-based method for DBT volume classification that relies on the local summarization of DBT slices (referred to as slabbing) and only requires volume-wise labels for training. Slabbing offers several advantages: i) it reduces the classifier’s computational complexity across the depth, letting it focus on the higher transversal resolution. Thanks to this strategy, we are the first to train a method at almost full-resolution (as high as 120 \(\times \) 2500 \(\times \) 2000); ii) it produces slabs that are closer to standard mammography, favoring an efficient transfer from classifiers trained on larger mammography databases; and iii) the slabs combined with a Multiple-Instance Learning (MIL) classifier result in localized information favoring interpretability. The proposed slabbing MIL approach is also novel for the automatic classification of DBTs. Moreover, we propose a trainable alternative to the handcrafted slabbing algorithms based on slice-wise attention that improves performance. We perform an experimental validation on a subset of the public BCS-DBT dataset and achieve an AUC of 0.73 with five-fold cross-validation. On a private multi-vendor dataset we obtain a similar AUC of 0.73, demonstrating an excellent performance consistency.

International audience; A specific half squat motion is analyzed as a planar movement. Experimental data about this motion are recorded with a motion capture device and two force plates. Joint torques are estimated with-a 3D Opensim model and a sagittal Matlab model. Torque variations from the two models are consistent even if the magnitudes of the corresponding variables differ. These results are attributable to the different anthropometric tables which the two models are based on. Another strategy developed here consists in estimating joint torques without the measured ground reaction forces. In that case, global vertical reaction force is well estimated. The use of the sagittal Matlab model is an efficient way to preliminarily analyze squat trajectories. The next step in this work is the study of the influence of the load represented by the weight of a knee prosthesis.

We work on the breast imaging malignancy segmentation task while focusing on the training process instead of network complexity. We designed a training process based on a modified U-Net, increasing the overall segmentation performances by using both, benign and malignant data for training. Our approach makes use of only a small amount of annotated data and relies on transfer learning from a self-supervised reconstruction task, and favors explainability.

Introduction Dans le contexte du traitement du cancer du sein metastatique, l’imagerie TEP/TDM constitue un outil diagnostique important. Cependant, la segmentation manuelle ou semi-automatique des metastases, etape preliminaire a une caracterisation plus precise de ces lesions, reste tres chronophage. Dans un premier pas vers l’automatisation de cette etape, nous proposons une detection automatique des metastases hepatiques basee sur une combinaison d’apprentissage machine classique et d’apprentissage profond. Materiel et methodes Dix-huit patients traites pour un cancer du sein metastatique ont passe une TEP/TDM. Les lesions metastatiques ont ete manuellement delinees par un medecin nucleaire. Parmi les 1200 lesions contourees, 67 etaient localisees dans le foie. Notre approche consiste dans un premier temps a segmenter le foie sur les images TDM pour delimiter la zone de recherche des lesions hepatiques. Nous avons entraine un algorithme d’apprentissage profond base sur le U-Net en utilisant une base de donnees TDM publique (LiTS, 130 patients) sur laquelle le foie et les lesions hepatiques ont ete contourees. Apres recalage rigide TEP/TDM, les masques binaires du foie issus de la premiere etape ont ete reechantillonnes dans le domaine des images TEP. Enfin, les lesions ont ete extraites des images TEP par l’utilisation d’un classifieur SVM prenant en compte les seuls pixels appartenant au foie dont le SUV est superieur a 2,5 et les labels correspondants (lesion ou non lesion). Une validation croisee a ete utilisee pour tenir compte du faible nombre de patients. Resultats Un Dice score de 0,95 a ete obtenu pour la segmentation du foie sur les images TDM. Les segmentations des lesions sur la TEP ont obtenu une sensibilite de 0,47, une valeur predictive positive (vpp) de 0,80 et une specificite de 0,99. Les resultats de la vpp et de la specificite demontrent la capacite du modele a correctement detecter les lesions malignes presentes, tandis que la sensibilite relativement faible traduit une legere sous-estimation de l’etendue des lesions. Conclusion Cette etude preliminaire a permis de mettre en avant une methode automatique pour detecter les lesions dans les images TEP en combinant l’information provenant de la TDM et de la TEP. A terme, cette brique pourra s’integrer dans les developpements permettant d’automatiser la segmentation de toutes les lesions metastatiques dans le cadre du cancer du sein.

National audience; HypothesisIn the clinical follow-up of metastatic breast cancer patients, semi-automatic measurements are performed on 18FDG PET/CT images to monitor the evolution of the main metastatic sites. Apart from being time-consuming and prone to subjective approximations, semi-automatic tools cannot make the difference between cancerous regions and active organs, presenting a high 18FDG uptake. In this work, we develop and compare fully automatic deep learning-based methods segmenting the main active organs (brain, heart, bladder), from full-body PET images.MethodsWe combine deep learning-based approaches with superpixels segmentation methods. In particular, we integrate a superpixel SLIC segmentation at different depths of a convolutional neural network, i.e. as input and within the optimization process. Superpixels reduce the resolution of the images, keeping sharp the boundaries of the larger target organs while the lesions, mostly smaller, are blurred. Results are compared with a deep learning segmentation network alone. The methods are cross-validated on full-body PET images of 36 patients from the ongoing EPICUREseinmeta study. The similarity between the manually defined ground truth masks of the organs and the results is evaluated with the Dice score. Moreover, these methods being preliminary to tumor segmentation, the precision of the networks is defined by monitoring the number of segmented voxels labelled as “active organ”, but belonging to a lesion.ResultsAlthough the methods present similar high Dice scores (0.96 ± 0.006), the ones using superpixels present a higher precision (on average 6, 16 and 27 selected voxels belonging to a tumor, for the CNN integrating superpixels in input, in optimization and not using them, respectively).ConclusionCombining deep learning with superpixels allows to segment organs presenting a high 18FDG uptake on PET images without selecting cancerous lesions. This improves the precision of the semi-automatic tools monitoring the evolution of breast cancer metastasis.

International audience; Multichannel intramuscular EMG (iEMG) recordings provide information on motor neuron behaviour, muscle fiber (MF) innervation geometry and, recently, have been proposed as means for establishing human-machine interfaces. Objective: in order to provide a reliable benchmark for computational methods applied to such recordings, we propose a simulation model for iEMG signals acquired by intramuscular multi-channel electrodes. Methods: we propose a number of modifications to the existing iEMG simulation methods, such as farthest point sampling for more uniform motor unit in-nervation centers distribution in the muscle cross-section, fiber-neuron assignment algorithm, motor neuron action potential propagation delay modelling and a linear model for multichannel recordings simulation. The proposed approach is also extended to gradually shifting (scanning) electrodes. Results: we provide representative applications of this model to the validation of methods for the estimation of motor unit territories, and for iEMG decomposition. Moreover, we extend this model to a full multichannel iEMG simulator using classical linear EMG modelling and existing approaches to the generation of motor neuron discharge sequences. Conclusions: the obtained simulation model provides physiologically accurate MUAPs across entire motor unit territories and for various electrode configurations. Significance: it can be used for the development and evaluation of mathematical methods for multichannel iEMG processing and analysis.

International audience; Driving blood glycaemia from hyperglycaemia to euglycaemia as fast as possible while avoiding hypoglycaemia is a major problem for decades for type-1 diabetes and is solved in this study. A control algorithm is designed that guaranties hypoglycaemia avoidance for the first time both from the theory of positive systems point of view and from the most pragmatic clinical practice. The solution consists of a state feedback control law that computes the required hyperglycaemia correction bolus in real-time to safely steer glycaemia to the target. A rigorous proof is given that shows that the control-law respects the positivity of the control and of the glucose concentration error: as a result, no hypoglycaemic episode occurs. The so-called hypo-free strategy control is tested with all the UVA/Padova T1DM simulator patients (i.e. ten adults, ten adolescents, and ten children) during a fasting-night scenario and in a hybrid closed-loop scenario including three meals. The theoretical results are assessed by the simulations on a large cohort of virtual patients and encourage clinical trials.

We study the fully convolutional neural networks in the context of malignancy detection for breast cancer screening. We work on a supervised segmentation task looking for an acceptable compromise between the precision of the network and the computational complexity., Comment: in Proceedings of iTWIST'20, Paper-ID: 30, Nantes, France, December, 2-4, 2020

International audience; Noise pollution reduction in the environment is a major challenge from a societal and health point of view. To implement strategies to improve sound environments, experts need information on existing noise. The first source of information is based on the elaboration of noise maps using software, but with limitations on the realism of the maps obtained, due to numerous calculation assumptions. The second is based on the use of measured data, in particular through professional measurement observatories, but in limited numbers for practical and financial reasons. More recently, numerous technical developments, such as the miniaturization of electronic components, the accessibility of low-cost computing processors and the improved performance of electric batteries, have opened up new prospects for the deployment of low-cost sensor networks for the assessment of sound environments. Over the past fifteen years, the literature has presented numerous experiments in this field, ranging from proof of concept to operational implementation. The purpose of this article is firstly to review the literature, and secondly, to identify the expected technical characteristics of the sensors to address the problem of noise pollution assessment. Lastly, the article will also put forward the challenges that are needed to respond to a massive deployment of low-cost noise sensors.

International audience; Objective: Real-time intramuscular electromyogra-phy (iEMG) decomposition, which is needed in biofeedback studies and interfacing applications, is a complex procedure that involves identifying the motor neuron spike trains from a streaming iEMG recording. Methods: We have previously proposed a sequential decomposition algorithm based on a Hidden Markov Model of EMG, which used Bayesian filter to estimate unknown parameters of motor unit (MU) spike trains, as well as their action potentials (MUAPs). Here, we present a modification of this original model in order to achieve a real-time performance of the algorithm as well as a parallel computation implementation of the algorithm on Graphics Processing Unit (GPU). Specifically, the Kalman filter previously used to estimate the MUAPs, is replaced by a least-mean-square filter. Additionally, we introduce a number of heuristics that help to omit the most improbable decomposition scenarios while searching for the best solution. Then, a GPU-implementation of the proposed algorithm is presented. Results: Simulated iEMG signals containing up to 10 active MUs, as well as five experimental fine-wire iEMG signals acquired from the tibialis anterior muscle, were decomposed in real time. The accuracy of decompositions depended on the level of muscle activation, but in all cases exceeded 85%. Conclusion: The proposed method and implementation provide an accurate, real-time interface with spinal motor neurons. Significance: The presented real time implementation of the decomposition algorithm substantially broadens the domain of its application.

We demonstrate the feasibility of a fully automatic computer-aided diagnosis (CAD) tool, based on deep learning, that localizes and classifies proximal femur fractures on X-ray images according to the AO classification. The proposed framework aims to improve patient treatment planning and provide support for the training of trauma surgeon residents. A database of 1347 clinical radiographic studies was collected. Radiologists and trauma surgeons annotated all fractures with bounding boxes, and provided a classification according to the AO standard. The proposed CAD tool for the classification of radiographs into types "A", "B" and "not-fractured", reaches a F1-score of 87% and AUC of 0.95, when classifying fractures versus not-fractured cases it improves up to 94% and 0.98. Prior localization of the fracture results in an improvement with respect to full image classification. 100% of the predicted centers of the region of interest are contained in the manually provided bounding boxes. The system retrieves on average 9 relevant images (from the same class) out of 10 cases. Our CAD scheme localizes, detects and further classifies proximal femur fractures achieving results comparable to expert-level and state-of-the-art performance. Our auxiliary localization model was highly accurate predicting the region of interest in the radiograph. We further investigated several strategies of verification for its adoption into the daily clinical routine. A sensitivity analysis of the size of the ROI and image retrieval as a clinical use case were presented., Comment: Accepted at IPCAI 2020 and IJCARS

Modern prosthetic control can be significantly enhanced due to the use of EMG decomposition. This technique permits to extract the activity of motor neurons that control the movement, thus giving a direct representation of neural command. This activity, being unaltered by factors non-related to motion, such as type and position of EMG electrode, is of great interest in prosthetic control. Existing real-time decomposition methods, however, provide activities of a very limited number of motor neurons (up to ten). This can be considered insufficient for intent inference. In this work, we present a probabilistic approach to intent inference that uses existing models of relations between the behavior of motor neurons and the movement. We compare our approach with a conventional one presented in the literature and show that it produces significantly better results when provided with a small number of decomposed motor neurons. To assess its performance in a fully controlled environment, we have developed a physiology-based simulation model of EMG and muscle contraction. Moreover, the analysis was also performed using experimental recordings of muscle contractions.; Le pilotage moderne de prothèse robotisée de bras peut être sensiblement amélioré par l'utilisation de la décomposition d'EMG. Cette technique permet d'extraire l'activité des motoneurones de la moelle épinière, une représentation directe de la commande neuronale. Cette activité, qui est insensible aux facteurs non-liés au mouvement, tels que le type ou la position d'électrode EMG, est essentielle pour le pilotage des prothèses. Cependant les méthodes de décomposition existantes ne fournissent que l'activité d'un nombre limité de motoneurones. Cette information peut être considérée insuffisante pour en inférer l'intention de l'utilisateur. Dans ce travail, nous présentons une approche probabiliste qui utilise les modèles existants de la relation entre les activités des motoneurones et le mouvement. Nous comparons cette approche à une approche plus conventionnelle et montrons qu'elle fournit de meilleurs résultats même quand elle est alimentée avec un nombre très bas de motoneurones décomposés. Pour évaluer sa performance dans un environnement contrôlé, nous avons développé un modèle physiologique de simulation d'EMG et de contraction de muscle. De plus, une analyse sur les signaux expérimentaux a été réalisée.

Modern prosthetic control can be significantly enhanced due to the use of EMG decomposition. This technique permits to extract the activity of motor neurons that control the movement, thus giving a direct representation of neural command. This activity, being unaltered by factors non-related to motion, such as type and position of EMG electrode, is of great interest in prosthetic control. Existing real-time decomposition methods, however, provide activities of a very limited number of motor neurons (up to ten). This can be considered insufficient for intent inference. In this work, we present a probabilistic approach to intent inference that uses existing models of relations between the behavior of motor neurons and the movement. We compare our approach with a conventional one presented in the literature and show that it produces significantly better results when provided with a small number of decomposed motor neurons. To assess its performance in a fully controlled environment, we have developed a physiology-based simulation model of EMG and muscle contraction. Moreover, the analysis was also performed using experimental recordings of muscle contractions.; Le pilotage moderne de prothèse robotisée de bras peut être sensiblement amélioré par l'utilisation de la décomposition d'EMG. Cette technique permet d'extraire l'activité des motoneurones de la moelle épinière, une représentation directe de la commande neuronale. Cette activité, qui est insensible aux facteurs non-liés au mouvement, tels que le type ou la position d'électrode EMG, est essentielle pour le pilotage des prothèses. Cependant les méthodes de décomposition existantes ne fournissent que l'activité d'un nombre limité de motoneurones. Cette information peut être considérée insuffisante pour en inférer l'intention de l'utilisateur. Dans ce travail, nous présentons une approche probabiliste qui utilise les modèles existants de la relation entre les activités des motoneurones et le mouvement. Nous comparons cette approche à une approche plus conventionnelle et montrons qu'elle fournit de meilleurs résultats même quand elle est alimentée avec un nombre très bas de motoneurones décomposés. Pour évaluer sa performance dans un environnement contrôlé, nous avons développé un modèle physiologique de simulation d'EMG et de contraction de muscle. De plus, une analyse sur les signaux expérimentaux a été réalisée.

International audience; This paper proposes to assess the relevance of new automated tools for electromyography (EMG) analysis, in order to differentiate neuropathic from myo-pathic patterns. The challenge is to define the diagnosis with only one iEMG signal per patient. Our proposed method uses the decomposition of the EMG signal to characterize motor unit action potentials (MUAPs). The decomposition of each iEMG signal is carried out with EMGLAB. For each signal, the decomposition provides a code which is used by the automated classification algorithms. We use here the linear Support Vector Machine (SVM) and the Bagging Trees methods. For the learning process we use several EMG signals and in different parts of the muscle. Only one recorded electromyography EMG signal per subject is used for the diagnostic test. We evaluate the k − f old cross-validation and the confusion matrix for both models. The accuracy is 77.3% for the SVM and 68.2% for the Bagging Trees. These are the first developments of this tool to make it useful for clinical practice.

We propose an efficient approach to estimate the uncertainty of deep-neural network classifiers based on the tradeoff of two measurements. The first based on subjective logic and the evidence of soft-max predictions and the second, based the Mahalanobis distance between new and training samples in the embedding space. These measurements require neither modifying, nor retraining, nor multiple testing of the models. We evaluate our methods on different classification tasks including breast cancer risk, breast density, and patch-wise tissue type and considering both an in-house database of 1600 mammographies, as well as on the public INBreast dataset. Throughout the experiments, we show the ability of our method to reject the most evident outliers, and to offer AUC gains of up to 10%, when keeping 60% of most certain samples.

International audience; Abstract : A repetitive movement over a long period of time, habituation to poor posture, handling heavy load can lead to musculoskeletal disorders. The most important thing is to prevent MSDs by automating repetitive tasks. But it is also necessary to help a patient with rehabilitation when an MDS has appeared. Regarding the human musculoskeletal system of movement analyses, joint torques developed by muscles are useful to prevent or treat MSDs. The knowledge of these torques is important to evaluate the possibilities of assisting a joint with an orthosis for instance. In this study, we analyze a specific half squat motion considered as a planar movement. Joint torques are estimated with two models-a 3D Opensim and a sagittal Matlab model. Torque variations from the two models are consistent even if the amplitudes differ. This result is attributable to the different anthropometric tables which the two models are based on. Another strategy developed here consists in estimating joint torques without the measured ground reaction forces. In that case, global vertical reaction force is well estimated. The use of the sagittal Matlab model is a simple and efficient way to preliminarily analyze squat trajectories.; Un mouvement répété sur une longue période, l'accoutumance à une mauvaise posture, la manipulation de charges lourdes peuvent entraîner des troubles musculosquelettiques (TMS). Il est important de pré-venir les TMS en automatisant les tâches répétitives. Mais il est également nécessaire d'aider un patient en phase de rééducation lorsqu'une TMS apparait. En ce qui concerne le système musculosquelettique humain, les couples articulaires développés par les muscles sont utiles pour prévenir ou traiter les TMS. La connaissance de ces couples est importante pour évaluer les possibilités d'assister une articulation avec une orthèse par exemple. Dans cette étude, nous analysons un mouvement de demi-squat spécifique considéré comme un mouvement planaire. Les couples articulaires sont estimés à l'aide de deux mo-dèles : un modèle 3D sur Opensim et un modèle sagittal sur Matlab. Les variations de couple issues des deux modèles sont cohérentes même si les amplitudes peuvent différer. Ce résultat s'explique par le fait que les modèles se basent sur des tables anthropométriques différentes. Une autre stratégie développée ici consiste à estimer les couples articulaires sans utiliser la mesure des forces de réaction du sol sur les pieds. Dans ce cas, la force de réaction verticale totale est bien estimée. L'utilisation du modèle sagittal Matlab est un moyen simple et efficace d'analyser préalablement les trajectoires de squat.

National audience; Neural activity leads to an increase in cerebral blood volume by two distinct vasodilatator mechanisms. To quantify neurovascular coupling degradations, we propose a non-linear parametric model. We perform the estimation of six parameters on a data set simulating variations in blood volume observed in mice. Such a parametric identification method is successfully validated with different noise levels. The next step will be to enhance the model in order to analyze human blood flow variations.; L'activité neuronale engendre une augmentation du volume sanguin cérébral induite par deux mécanismes vasodilatateurs distincts. Pour quantifier les dégradations subies par ce couplage neurovasculaire, nous proposons un modèle paramétrique non linéaire. Nous avons évalué l'estimation de six paramètres sur un jeu de données simulant des variations de volumes sanguins observées chez le rat. L'identification paramétrique avec différentes puissances de bruit paraît efficace et encourage à enrichir le modèle pour analyser des variations de débits sanguins chez l'homme.

International audience; Abstract : A repetitive movement over a long period of time, habituation to poor posture, handling heavy load can lead to musculoskeletal disorders. The most important thing is to prevent MSDs by automating repetitive tasks. But it is also necessary to help a patient with rehabilitation when an MDS has appeared. Regarding the human musculoskeletal system of movement analyses, joint torques developed by muscles are useful to prevent or treat MSDs. The knowledge of these torques is important to evaluate the possibilities of assisting a joint with an orthosis for instance. In this study, we analyze a specific half squat motion considered as a planar movement. Joint torques are estimated with two models-a 3D Opensim and a sagittal Matlab model. Torque variations from the two models are consistent even if the amplitudes differ. This result is attributable to the different anthropometric tables which the two models are based on. Another strategy developed here consists in estimating joint torques without the measured ground reaction forces. In that case, global vertical reaction force is well estimated. The use of the sagittal Matlab model is a simple and efficient way to preliminarily analyze squat trajectories.; Un mouvement répété sur une longue période, l'accoutumance à une mauvaise posture, la manipulation de charges lourdes peuvent entraîner des troubles musculosquelettiques (TMS). Il est important de pré-venir les TMS en automatisant les tâches répétitives. Mais il est également nécessaire d'aider un patient en phase de rééducation lorsqu'une TMS apparait. En ce qui concerne le système musculosquelettique humain, les couples articulaires développés par les muscles sont utiles pour prévenir ou traiter les TMS. La connaissance de ces couples est importante pour évaluer les possibilités d'assister une articulation avec une orthèse par exemple. Dans cette étude, nous analysons un mouvement de demi-squat spécifique considéré comme un mouvement planaire. Les couples articulaires sont estimés à l'aide de deux mo-dèles : un modèle 3D sur Opensim et un modèle sagittal sur Matlab. Les variations de couple issues des deux modèles sont cohérentes même si les amplitudes peuvent différer. Ce résultat s'explique par le fait que les modèles se basent sur des tables anthropométriques différentes. Une autre stratégie développée ici consiste à estimer les couples articulaires sans utiliser la mesure des forces de réaction du sol sur les pieds. Dans ce cas, la force de réaction verticale totale est bien estimée. L'utilisation du modèle sagittal Matlab est un moyen simple et efficace d'analyser préalablement les trajectoires de squat.