Search

In this PhD thesis, we address the challenge of integrating Brain-Computer Interfaces (BCI) and music technologies on the specific application of music source separation, which is the task of isolating individual sound sources that are mixed in the audio recording of a musical piece. This problem has been investigated for decades, but never considering BCI as a possible way to guide and inform separation systems. Specifically, we explored how the neural activity characterized by electroencephalographic signals (EEG) reflects information about the attended instrument and how we can use it to inform a source separation system.First, we studied the problem of EEG-based auditory attention decoding of a target instrument in polyphonic music, showing that the EEG tracks musically relevant features which are highly correlated with the time-frequency representation of the attended source and only weakly correlated with the unattended one. Second, we leveraged this ``contrast'' to inform an unsupervised source separation model based on a novel non-negative matrix factorisation (NMF) variant, named contrastive-NMF (C-NMF) and automatically separate the attended source.Unsupervised NMF represents a powerful approach in such applications with no or limited amounts of training data as when neural recording is involved. Indeed, the available music-related EEG datasets are still costly and time-consuming to acquire, precluding the possibility of tackling the problem with fully supervised deep learning approaches. Thus, in the last part of the thesis, we explored alternative learning strategies to alleviate this problem. Specifically, we propose to adapt a state-of-the-art music source separation model to a specific mixture using the time activations of the sources derived from the user's neural activity. This paradigm can be referred to as one-shot adaptation, as it acts on the target song instance only.We conducted an extensive evaluation of both the proposed system on the MAD-EEG dataset which was specifically assembled for this study obtaining encouraging results, especially in difficult cases where non-informed models struggle., Dans cette thèse, nous abordons le défi de l'utilisation d'interfaces cerveau-machine (ICM) sur l'application spécifique de la séparation de sources musicales qui vise à isoler les instruments individuels qui sont mélangés dans un enregistrement de musique. Ce problème a été étudié pendant des décennies, mais sans jamais considérer les ICM comme un moyen possible de guider et d'informer les systèmes de séparation. Plus précisément, nous avons étudié comment l'activité neuronale caractérisée par des signaux électroencéphalographiques (EEG) reflète des informations sur la source à laquelle on porte son attention et comment nous pouvons l'utiliser pour informer un système de séparation de sources.Tout d'abord, nous avons étudié le problème du décodage par l'EEG de l'attention auditive d'un instrument spécifique dans une pièce musicale polyphonique, en montrant que l'EEG suit les caractéristiques musicales pertinentes qui sont fortement corrélées avec la représentation temps-fréquence de la source à laquelle on porte l'attention et seulement faiblement corrélées avec les autres. Ensuite, nous avons exploité ce "contraste" pour informer un modèle de séparation de sources non supervisé basé sur une nouvelle variante de factorisation en matrices positives (NMF), appelée contrastive-NMF (C-NMF) et séparer automatiquement la source à laquelle on porte l'attention.La NMF non supervisée est une approche efficace dans de telles applications ne disposant pas ou peu de données d'apprentissage, comme c'est le cas dans des scénarios nécessitant des enregistrements EEG. En effet, les jeux de données EEG liés à la musique disponibles sont coûteux et longs à acquérir, ce qui exclut la possibilité d'aborder le problème par des approches d'apprentissage profond entièrement supervisées. Dans la dernière partie de la thèse, nous avons exploré des stratégies d'apprentissage alternatives. Plus précisément, nous avons étudié la possibilité d'adapter un modèle de séparation de sources de l'état de l'art à un mélange spécifique en utilisant les activations temporelles de sources dérivées de l'activité neuronale de l'utilisateur au moment du test. Cette approche peut être considérée comme étant " à adaptation unitaire" (one-shot), car l'adaptation agit uniquement sur une instance de chanson.Nous avons évalué les approches proposées sur les jeu de données MAD-EEG qui a été spécifiquement assemblé pour cette étude, obtenant des résultats encourageants, en particulier dans les cas difficiles où les modèles non informés sont mis à mal.

The diagnosis of consciousness in brain-injured patients has a major impact on diagnostic, therapeutic and prognostic management. Thus, the physician tries to determine what state of consciousness or level of consciousness the patient is in, but does not have access to what the patient is thinking, the conscious content. The study of the state of consciousness and the conscious content in these patients is therefore a major challenge. This thesis aims to attempt to disentangle the content and state of consciousness in brain-damaged patients. In a first part, we propose a simple graphical representation (a 2d space) that can be useful to contrast the characteristics of brain activity in two dimensions according to conscious content and conscious state. In a second part, We prospectively explored the prognostic value of an EEG signature of consciousness -the global effect, a marker of conscious content- for predicting the prognosis of conscious state in a large cohort of patients with brain injury. In a third part, we explore the relationship between states/content of consciousness and the embodied paradigm to cognition. We show that fluctuations in heart rate depend on top-down attention to the stimulus, predict memory performance and may represent a simple measure to assess conscious state and conscious content in patients., Le diagnostic de conscience chez les patients neurolésés a un impact majeur sur la prise en charge diagnostique, thérapeutique et pronostique. Ainsi, le médecin essaie de déterminer dans quel état de conscience ou niveau de conscience se trouve le patient, mais n'accède pas à ce que le patient pense, au contenu de sa conscience. L'étude de l'état de conscience et du contenu conscient chez ces patients est donc un défi majeur. Cette thèse vise à tenter de démêler le contenu et l'état de conscience chez les patients atteints de lésions cérébrales. Dans une première partie, nous proposons une représentation graphique simple (un espace 2d) qui peut être utile pour contraster les caractéristiques de l'activité cérébrale dans deux dimensions selon le contenu conscient et l'état conscient. Dans une deuxième partie, nous avons exploré de manière prospective la valeur pronostique d'une signature EEG de conscience -l'effet global, marqueur de contenu conscient- pour prédire le pronostic d'état de conscience dans une large cohorte de patients avec une lésion cérébrale. Dans une troisième partie, nous explorons la relation entre les états/contenu de conscience et un paradigme incarné de la cognition. Nous montrons que les fluctuations de la fréquence cardiaque dépendent de l'attention top-down portée au stimulus et peuvent représenter une mesure simple pour évaluer l'état et le contenu de la conscience chez les patients.

The most prominent type of artifact contaminating electroencephalogram (EEG)signals are the eyeblink (EB) artifacts, which could potentially lead tomisinterpretation of the EEG signal. Online detection and removal of eyeblink artifacts from EEG signals are essential in applications such a Brain-Computer Interfaces (BCI), neurofeedback and epilepsy diagnosis. In this thesis, algorithms that combine unsupervised eyeblink artifact detection (eADA) with enhanced Empirical Mode Decomposition (FastEMD) and Canonical Correlation Analysis (CCA) are proposed,i.e. FastEMD-CCA2 and FastCCA, to automatically identify eyeblink artifacts andremove them in an online setting. FastEMD-CCA2 and FastCCA have outperformedone of the existing state-of-the-art methods, FORCe. The average artifact removalaccuracy, sensitivity, specificity and error rate of FastEMD-CCA2 is 97.9%, 97.65%,99.22%, and 2.1% respectively, validated on a Hitachi dataset with 60 EEG signals,consisting more than 5600 eyeblink artifacts. FastCCA achieved an average of99.47%, 99.44%, 99.74% and 0.53% artifact removal accuracy, sensitivity, specificityand error rate respectively, validated on the Hitachi dataset too. FastEMD-CCA2 andFastCCA algorithms are developed and implemented in the C++ programming language to investigate the processing speed these algorithms could achieve in adifferent medium. Analysis has shown that FastEMD-CCA2 and FastCCA took about10.7 and 12.7 milliseconds respectively, on average to clean a 1-second length of EEG segment. This makes them a feasible solution for applications requiring onlineremoval of eyeblink artifacts from EEG signals., Parmi les artefacts contaminant les signaux d'électroencéphalogramme (EEG) les plus importants sont les clins d’oeuil qui pourraient potentiellement conduire à une mauvaise interprétation du signal EEG. La détection et la suppression en ligne des artefacts de clignement des yeux des signaux EEG sont essentielles dans des applications telles que les interfaces cerveau ordinateur (BCI), le neurofeedback et la surveillance de l’épilepsie. Dans cette thèse, des algorithmes qui combinent la détection non supervisée des artefacts de clignement des yeux (eADA) avec une décomposition en mode empirique améliorée (FastEMD) et une analyse de corrélation canonique (CCA) sont proposés, sous le nom de FastEMD-CCA2 et FastCCA, pour identifier automatiquement les artefacts de clignement des yeux et les supprimer en temps réel. Les algorithmes FastEMD-CCA2 et FastCCA sont comparés à la méthode FORCe. La précision, la sensibilité, la spécificité et le taux d’erreur moyens de suppression des artefacts de FastEMD-CCA2 sont respectivement de 97,9%, 97,65%, 99,22% et 2,1%, validés sur un ensemble de données Hitachi. Pour ces mêmes critères nous obtenons avec FastCCA une moyenne de 99,47%, 99,44%, 99,74% et 0,53% validés également sur l’ensemble de données Hitachi. Les algorithms FastEMD-CCA2 et FastCCA sont développés et mis en œuvre dans le langage de programmation C ++ pour étudier la vitesse de traitement qu’ils pourraient atteindre en cas d’implémentation embarquée. L’analyse a montré que FastEMD-CCA2 et FastCCA ont pris respectivement environ 10,7 et 12,7 millisecondes, en moyenne, pour traiter un segment d’EEG de 1 seconde. Cela en fait une solution realisable pour les applications nécessitant la suppression en temps réel des clins d’oeuil dans les signaux EEG.

International audience; Objectives - Assessment of current practice and the need for tele-transmission and remote interpretation of EEG in France. Transmission of EEG to a distant center could be a promising solution to the problem of decreasing availability of neurophysiologists for EEG interpretation, in order to provide equity within health care services in France. This practice should logically follow the legal framework of telemedicine and the recommendations that were recently edited by the Société de neurophysiologie clinique de langue française (SNCLF) and the Ligue française contre l'épilepsie (LCFE). Methods - A national survey was designed and performed under the auspices of the SNCLF. Results - This survey reveals that there is an important gap between the official recommendations and the "reality on the ground". These local organizations were mainly established through the impulse of individual initiatives, rarely driven by health regulatory authorities and sometimes far from legal frameworks. For the majority, they result from a need to improve medical care, especially in pediatrics and neonatology, and to ensure continuity of care. When present, tele-transmission of EEG is often only partially satisfactory, since many technical procedures have to be improved. Conversely, the lack of tele-transmission of EEG would penalize medical care for some patients. Conclusions - The survey shows both the wealth of local initiatives and the fragility of most existing networks, emphasizing the need for better cooperation between regulatory authorities and health care professionals to establish or improve the transmission of EEG in France.

International audience; Electroencephalography allows the functional analysis of electrical brain cortical activity and is the gold standard for analyzing electrophysiological processes involved in epilepsy but also in several other dysfunctions of the central nervous system. Morphological imaging yields complementary data, yet it cannot replace the essential functional analysis tool that is EEG. Furthermore, EEG has the great advantage of being non-invasive, easy to perform and allows repeat testing when follow-up is necessary, even at the patient's bedside. Faced with advances in knowledge, techniques and indications, the Société de neurophysiologie clinique de langue française (SNCLF) and the Ligue française contre l'épilepsie (LFCE) found it necessary to provide an update on EEG recommendations. This article will review the methodology applied to this work, refine the various topics detailed in the following chapters. It will go over the summary of recommendations for each of these chapters and highlight proposals for writing an EEG report. Some questions could not be answered by review of the literature; in such cases, in addition to the guidelines the working and reading groups provided their expert opinion.

International audience; EEG recordings can be sent for remote interpretation. This article aims to define the tele-EEG procedures and technical guidelines. Tele-EEG is a complete medical act that needs to be carried out with the same quality requirements as a local one in terms of indications, formulation of the medical request and medical interpretation. It adheres to the same quality requirements for its human resources and materials. It must be part of a medical organization (technical and medical network) and follow all rules and guidelines of good medical practices. The financial model of this organization must include costs related to performing the EEG recording, operating and maintenance of the tele-EEG network and medical fees of the physician interpreting the EEG recording. Implementing this organization must be detailed in a convention between all parties involved: physicians, management of the healthcare structure, and the company providing the tele-EEG service. This convention will set rules for network operation and finance, and also the continuous training of all staff members. The tele-EEG system must respect all rules for safety and confidentiality, and ensure the traceability and storing of all requests and reports. Under these conditions, tele-EEG can optimize the use of human resources and competencies in its zone of utilization and enhance the organization of care management.

L'interface cerveau-ordinateur (ICO) décode les signaux électriques du cerveau requise par l’électroencéphalographie et transforme ces signaux en commande pour contrôler un appareil ou un logiciel. Un nombre limité de tâches mentales ont été détectés et classifier par différents groupes de recherche. D’autres types de contrôle, par exemple l’exécution d'un mouvement du pied, réel ou imaginaire, peut modifier les ondes cérébrales du cortex moteur. Nous avons utilisé un ICO pour déterminer si nous pouvions faire une classification entre la navigation de type marche avant et arrière, en temps réel et en temps différé, en utilisant différentes méthodes. Dix personnes en bonne santé ont participé à l’expérience sur les ICO dans un tunnel virtuel. L’expérience fut a était divisé en deux séances (48 min chaque). Chaque séance comprenait 320 essais. On a demandé au sujets d’imaginer un déplacement avant ou arrière dans le tunnel virtuel de façon aléatoire d’après une commande écrite sur l'écran. Les essais ont été menés avec feedback. Trois électrodes ont été montées sur le scalp, vis-à-vis du cortex moteur. Durant la 1re séance, la classification des deux taches (navigation avant et arrière) a été réalisée par les méthodes de puissance de bande, de représentation temporel-fréquence, des modèles autorégressifs et des rapports d’asymétrie du rythme β avec classificateurs d’analyse discriminante linéaire et SVM. Les seuils ont été calculés en temps différé pour former des signaux de contrôle qui ont été utilisés en temps réel durant la 2e séance afin d’initier, par les ondes cérébrales de l'utilisateur, le déplacement du tunnel virtuel dans le sens demandé. Après 96 min d'entrainement, la méthode « online biofeedback » de la puissance de bande a atteint une précision de classification moyenne de 76 %, et la classification en temps différé avec les rapports d’asymétrie et puissance de bande, a atteint une précision de classification d’environ 80 %., A Brain-Computer Interface (BCI) decodes the brain signals representing a desire to do something, and transforms those signals into a control command. However, only a limited number of mental tasks have been previously detected and classified. Performing a real or imaginary navigation movement can similarly change the brainwaves over the motor cortex. We used an ERS-BCI to see if we can classify between movements in forward and backward direction offline and then online using different methods. Ten healthy people participated in BCI experiments comprised two-sessions (48 min each) in a virtual environment tunnel. Each session consisted of 320 trials where subjects were asked to imagine themselves moving in the tunnel in a forward or backward motion after a randomly presented (forward versus backward) command on the screen. Three EEG electrodes were mounted bilaterally on the scalp over the motor cortex. Trials were conducted with feedback. In session 1, Band Power method, Time-frequency representation, Autoregressive models and asymmetry ratio were used in the β rhythm range with a Linear-Discriminant-analysis classifier and a Support Vector Machine classifier to discriminate between the two mental tasks. Thresholds for both tasks were computed offline and then used to form control signals that were used online in session 2 to trigger the virtual tunnel to move in the direction requested by the user's brain signals. After 96 min of training, the online band-power biofeedback training achieved an average classification precision of 76 %, whereas the offline classification with asymmetrical ratio and band-power achieved an average classification precision of 80%.

A brain-computer interface (BCI) is a system that enables control of devices or communication with other persons, only through cerebral activity, without using muscles. The main application for BCIs is assistive technology for disabled persons. Examples for devices that can be controlled by BCIs are artificial limbs, spelling devices, or environment control systems. BCI research has seen renewed interest in recent years, and it has been convincingly shown that communication via a BCI is in principle feasible. However, present day systems still have shortcomings that prevent their widespread application. In part, these shortcomings are caused by limitations in the functionality of the pattern recognition algorithms used for discriminating brain signals in BCIs. Moreover, BCIs are often tested exclusively with able-bodied persons instead of conducting tests with the target user group, namely disabled persons. The goal of this thesis is to extend the functionality of pattern recognition algorithms for BCI systems and to move towards systems that are helpful for disabled users. We discuss extensions of linear discriminant analysis (LDA), which is a simple but efficient method for pattern recognition. In particular, a framework from Bayesian machine learning, the so-called evidence framework, is applied to LDA. An algorithm is obtained that learns classifiers quickly, robustly, and fully automatically. An extension of this algorithm allows to automatically reduce the number of sensors needed for acquisition of brain signals. More specifically, the algorithm allows to perform electrode selection. The algorithm for electrode selection is based on a concept known as automatic relevance determination (ARD) in Bayesian machine learning. The last part of the algorithmic development in this thesis concerns methods for computing accurate estimates of class probabilities in LDA-like classifiers. These probabilities are used to build a BCI that dynamically adapts the amount of acquired data, so that a preset, approximate bound on the probability of misclassifications is not exceeded. To test the algorithms described in this thesis, a BCI specifically tailored for disabled persons is introduced. The system uses electroencephalogram (EEG) signals and is based on the P300 evoked potential. Datasets recorded from five disabled and four able-bodied subjects are used to show that the Bayesian version of LDA outperforms plain LDA in terms of classification accuracy. Also, the impact of different static electrode configurations on classification accuracy is tested. In addition, experiments with the same datasets demonstrate that the algorithm for electrode selection is computationally efficient, yields physiologically plausible results, and improves classification accuracy over static electrode configurations. The classification accuracy is further improved by dynamically adapting the amount of acquired data. Besides the datasets recorded from disabled and able-bodied subjects, benchmark datasets from BCI competitions are used to show that the algorithms discussed in this thesis are competitive with state-of-the-art electroencephalogram (EEG) classification algorithms. While the experiments in this thesis are uniquely performed with P300 datasets, the presented algorithms might also be useful for other types of BCI systems based on the EEG. This is the case because functionalities such as robust and automatic computation of classifiers, electrode selection, and estimation of class probabilities are useful in many BCI systems. Seen from a more general point of view, many applications that rely on the classification of cerebral activity could possibly benefit from the methods developed in this thesis. Among the potential applications are interrogative polygraphy ("lie detection") and clinical applications, for example coma outcome prognosis and depth of anesthesia monitoring.