Nous vous rappelons que, afin de garantir l'accès de tous les inscrits aux salles de réunion, l'inscription aux réunions est gratuite mais obligatoire.
Inscriptions closes à cette réunion.
35 personnes membres du GdR ISIS, et 56 personnes non membres du GdR, sont inscrits à cette réunion.
Capacité de la salle : 100 personnes.
Titre : Signal and Image Processing in Neuroscience
Thème A - Méthodes et modèles en traitement du signal et de l'image
Date : 12/05/2022
Lieu : Sorbonne Université, 4 place Jussieu, bâtiment Esclangon (SCAI)
Nous organisons le jeudi 12 mai la journée intitulée "Signal and Image Processing in Neuroscience" dont le programme est donné ci-dessous. Cette journée a lieu en présentiel mais, pour les personnes qui souhaiteraient y assister en visio, nous proposons un lien zoom :
Sujet : Zoom meeting invitation - GDR ISIS - 12 Mai
Heure : 12 mai 2022 09:45 AM Paris
Participer à la réunion Zoom
https://univ-rennes1-fr.zoom.us/j/85201366761
ID de réunion : 852 0136 6761
Code secret : 998649
The modeling and analysis of brain activity involves very large and diverse scientific areas, from mathematics to psychology, passing through biology and engineering. This workshop focuses on recent advances in signal and image processing having applications in neuroscience. Different research directions will be explored such as estimating the brain sources activities using signals and images acquired at different time and space scales and different modalities, or inferring cerebral connectivity from these signals, elaborating and analyzing the dynamic graphs modeling the brain networks. Neuroscientific applications will also be presented, including medical diagnosis assistance for different pathologies (epilepsy, Alzheimer's disease,...) and the understanding of fundamental cognitive tasks.
09h50 - 10h00 : Accueil
10h00 - 10h50 : Fabrice Wendling (keynote 1)
10h50 - 11h20 : Alexandre Gramfort
11h20 - 11h50 : Mouloud Adel
11h50 - 12h10 : Majd Abazid
12h10 - 12h40 : Ahmad Karfoul
12h40 - 14h00 : pause déjeuner
14h00 - 14h50 : Tulay Adali (keynote 2)
14h50 - 15h20 : Gaëtan Frusque
15h20 - 15h50 : Christian Bénar
15h50 - 16h10 : Pascal Helson
16h10 - 16h40 : Radu Ranta
16h40 - 17h00 : Victor Delvigne
Fabrice Wendling (LTSI, UMR Inserm-UR1, Rennes)
Abstract: Brain connectivity estimated from electroencephalography (EEG) has received increasing attention over the past decade. Although considerable advances have been made both in the recording and in the analysis of EEG signals, a number of methodological questions are still open regarding the optimal way to process the data in order to identify brain networks from scalp EEG. In this presentation, we will analyze the impact of various factors that intervene in this identification, among which the combination between the algorithm used to solve the EEG inverse problem and the algorithm used to measure the functional connectivity. Examples of large-scale networks identified from High-Resolution EEG recordings will be described in the context of cognitive tasks or brain disorders.
Alexandre Gramfort (Inria, Université Paris-Saclay)
Abstract: Data augmentation is a key element of deep learning pipelines, as it informs the network during training about transformations of the input data that keep the label unchanged. Manually finding adequate augmentation methods and parameters for a given pipeline is however rapidly cumbersome. In particular, while intuition can guide this decision for images, the design and choice of augmentation policies remains unclear for more complex types of data, such as electroencephalography (EEG) signals. This talk with present gradient-based automatic data augmentation algorithms. Motivated by supervised learning applications using EEG signals for which good augmentation policies are mostly unknown, we propose a new differentiable relaxation of the problem. Results on supervised tasks using EEG such as sleep stage classification will be presented.
Mouloud Adel (Institut Fresnel, CNRS, Aix-Marseille Université)
Abstract: This talk deals with feature selection and data classification on Positrons Emission Tomography images for Alzheimer's disease computer-aided diagnosis. It will give an overview of the main steps on a computer-aided diagnosis system and will focus on data reduction based on feature selection and classification using machine learning techniques.
Majd Abazid (SAMOVAR, Télécom SudParis, Institut Polytechnique de Paris)
Abstract: The present work addresses brain network analysis considering different clinical severity stages of cognitive dysfunction, based on resting-state electroencephalography (EEG). This study addresses brain network analysis over different clinical severity stages of cognitive dysfunction. We exploit EEG data of subjective cognitive impairment (SCI) patients, mild cognitive impairment (MCI) patients and Alzheimer?s disease (AD) patients. We propose a new framework to study the topological networks, with a spatiotemporal entropy measure for estimating the connectivity.
Ahmad Karfoul (LTSI, UMR Inserm-UR1, Rennes)
Abstract: In this talk, an approach to track brain effective connectivity (causal relation between different neural systems) in the context of epilepsy is presented. It relies on optimally inferring the most dominant effective connectivity patterns underlying the observed depth EEG time series. In addition, this approach permits to capture the variability of these connectivity patterns over time, frequency and possibly over epileptic seizures through a constrained low rank canonical polyadic decomposition of a 3rd/4th order tensor built from the observed depth epileptic EEG signals. Results on both simulated and real data will be also presented and discussed.
Tulay Adali (MLSP Lab, University of Maryland, Baltimore County)
Abstract: In many fields today, such as neuroscience, remote sensing, computational social science, and physical sciences, multiple sets of data are readily available. Matrix and tensor factorizations enable joint analysis, i.e., fusion, of these multiple datasets such that they can fully interact and inform each other while also minimizing the assumptions placed on their inherent relationships. A key advantage of these methods is the direct interpretability of their results. This talk presents an overview of models based on independent component analysis, and its generalization to multiple datasets, independent vector analysis (IVA) with examples in fusion and analysis of neuroimaging data. Relationship of IVA to other methods such as multiset canonical correlation analysis is discussed, and a number of important directions of research are addressed, along with the challenges.
Gaëtan Frusque (ETH Zürich)
Abstract: L'épilepsie est l'une des maladies neurologiques les plus rependues dans le monde affectant environ 1% de la population. Une crise peut être vue comme un système complexe qui évolue dans le temps. On peut la représenter par un graphe dynamique grâce au calcul des connectivités fonctionnelles. Les graphes dynamiques ont récemment fait l'objet d'une attention considérable. Cependant, il n'existe pas de consensus sur les manières de les étudier. Dans cette présentation on propose une décomposition tensorielle parcimonieuse afin d'extraire les informations importantes des graphes dynamiques. L'enjeu applicatif est de découvrir les réseaux pathologies caractéristique de l'épilepsie de plusieurs patients.
Christian Bénar (INS, Inserm, Aix-Marseille Université)
Abstract: Une difficulté des mesures non-invasives en électrophysiologie (MEG, EEG) est la validation de leurs résultat. Les enregistrements d?EEG intracérébral effectués lors de l?investigation préchirurgicale de l?épilepsie constituent une opportunité extraordinaire pour confronter les résultats des analyses à une réalité terrain enregistrée directement dans le cerveau. Pour éviter les fluctuations entre des sessions différentes, il est utile d?enregistrer en même temps les signaux de surface (MEG, EEG) et de profondeur (EEG intracérébral). Dans cette présentation, je montrerai les défis et les possibilités des enregistrements de MEG-SEEG simultanés, ainsi que des applications en épilepsie et en cognition.
Pascal Helson (KTH Royal Institute of Technology, Stockholm, Sweden)
Abstract: In this study, we aim at getting more insights into neuronal activity changes distributed across the whole brain in Parkinson?s disease (PD). To do so, we performed a source level analysis on resting state magnetoencephalogram (MEG) from two groups: PD patients and healthy controls. After a spectral analysis, we quantified the aperiodic activity on a whole-brain scale by fitting a power law to the spectrum of MEG and then studied its relationship with age and UPDRS score. The aperiodic part of the spectrum, which is usually ignored, may be linked with Excitation-Inhibition (EI) ratio according to recent works. Thus, our results for the first time indicate that PD is associated with change in EI ratio across the whole brain.
Radu Ranta (CRAN, CNRS, Université de Lorraine)
Abstract: This presentation introduces a new sparse algorithm from the iterative regression family, derived from the classical OLS. The target application is the inverse problem of brain sources estimation. In order to ensure plausible physiological hypothesis (such as dipolar sources with fixed orientations), the regression is performed on a dictionary having matrix (not vector) elements, while the (vectorial) regression coefficients representing time varying amplitudes are constrained to unit rank. The performances of the proposed algorithm are compared with other classical solutions on simulated data.
Victor Delvigne (ISIA Lab, Faculty of Engineering, University of Mons)
Abstract: For some years now, deep learning (DL) has proven its supremacy in the context of modalities classification. However, one of the most widely used approaches during the model training is based on supervised learning: during the training phase, a large number of examples are shown to the model to teach it how to automatically extract the features that help to differentiate classes. Nevertheless, this approach deviates from the natural functioning of biological neural networks and presents some limitations. One solution to this problem is to consider self-supervised learning where the goal is to understand and learn to process the modality before classifying it. In this presentation, the application of this new approach for the automatic processing of medical signals (electroencephalogram) will be presented. The presented tasks are adapted to EEG by learning its temporal, spectral and spatial properties.
Date : 2022-05-12
Lieu : Sorbonne Université, 4 place Jussieu, bâtiment Esclangon (SCAI)
Thèmes scientifiques :
A - Méthodes et modèles en traitement de signal
B - Image et Vision
Inscriptions closes à cette réunion.
Accéder au compte-rendu de cette réunion.
(c) GdR IASIS - CNRS - 2024.