Search

The objective of this progressive research is to develop a spatio-temporal source separation scheme using temporal priors with parametric uncertainties. The method has potential applications in single and multi-modal data analysis. It is shown that the proposed scheme can be jointly used with classical source separation techniques based on generalized eigenvalue decomposition (GEVD) and joint approximate diagonalization of eigenmatrices (JADE). As a case study, the proposed method is evaluated over noninvasive fetal electrocardiogram signals.

The notion of spatially distributed sources arises in many fields of applied signal processing. Following previous empirical findings in this area, in this work, we attempt to validate some of these findings within a simplified mathematical framework for a finite electrostatic distributed source within a homogeneous volume conductor. The model is very accurate for signals such as the electrocardiogram and electroencephalogram. The method is used to analyze second-order statistics of distributed sources. The findings of this research are expected to be helpful for better interpretation of PCA and ICA results, specifically for biomedical applications.

International audience; In this paper an Extended Kalman Filter (EKF) has been proposed for the filtering of noisy ECG signals. The method is based on a modified nonlinear dynamic model, previously introduced for the generation of synthetic ECG signals. An automatic parameter selection method has also been suggested, to adapt the model with a vast variety of normal and abnormal ECG signals. The results show that the EKF output is able to track the original ECG signal shape even in the most noisiest epochs of the ECG signal. The proposed method may serve as an efficient filtering procedure for applications such as the noninvasive extraction of fetal cardiac signals from maternal abdominal signals.

International audience; In this paper several applications of the Independent Component Analysis (ICA) algorithm, for the analysis of biomedical signal recordings have been investigated. One of these applications is the removal of EEG artifacts such as the EOG. It is shown that ICA may serve as a powerful tool, which could help the analysis of biomedical recordings, and give better insights about the underlying sources of some disorders. Another application of the proposed method is the detection of sleep disorders in patients suffering from sleep apnea. The ultimate goal of this approach is to develop an automatic noninvasive data acquisition system, for clinical applications.

International audience; Independent component analysis (ICA) and other blind source separation (BSS) methods are important processing tools for multi-channel processing of electroencephalographic data and have found numerous applications for brain-computer interfaces. A number of solutions to the BSS problem are achieved by approximate joint diagonalization (AJD) algorithms, thus the goodness of the solution depends on them. We present a new least-squares AJD algorithm with adaptive weighting on the separating vectors. We show that it has good properties while keeping the greatest generality among AJD algorithms; no constraint is imposed either on the input matrices or on the joint diagonalizer to be estimated. The new cost function allows interesting extensions that are now under consideration.

September

Objective: The instantaneous phase (IP) and instantaneous frequency (IF) of the electroencephalogram (EEG) are considered as notable complements for the EEG spectrum. The calculation of these parameters commonly includes narrow-band filtering, followed by the calculation of the signal's analytical form. The calculation of the IP and IF is highly susceptible to the filter parameters and background noise level, especially in low analytical signal amplitudes. The objective of this study is to propose a robust statistical framework for EEG IP/IF estimation and analysis. Approach: Herein, a Monte Carlo estimation scheme is proposed for the robust estimation of the EEG IP and IF. It is proposed that any EEG phase-related inference should be reported as an average with confidence intervals obtained by repeating the IP and IF estimation under infinitesimal variations (selected by an expert), in algorithmic parameters such as the filter's bandwidth, center frequency and background noise level. In the second part of the paper, a stochastic model consisting of the superposition of narrow-band foreground and background EEG is used to derive analytically probability density functions of the instantaneous envelope (IE) and IP of EEG signals, which justify the proposed Monte Carlo scheme. Main results: The instantaneous analytical envelope of the EEG, which has been empirically used in previous studies, is shown to have a fundamental impact on the accuracy of the EEG phase contents. It is rigorously shown that the IP/IF estimation quality highly depends on the IE and any phase/frequency interpretations in low IE are statistically unreliable and require a hypothesis test. Significance: The impact of the proposed method on previous studies, including time-domain phase synchrony, phase resetting, phase locking value and phase amplitude coupling are studied with examples. The findings of this research can set forth new standards for EEG phase/frequency estimation and analysis techniques.

Noninvasive extraction of fetal electrocardiogram (fECG) from multichannel maternal abdomen recordings is an emerging technology used for fetal cardiac diagnosis. The strongest interference for the fECG is the maternal ECG (mECG), which is not totally removed through conventional methods including blind source separation (BSS). In this work, we address the problem of offline maternal cardiac signal removal and introduce an online subspace denoising procedure for mECG cancellation. The proposed method is a general online denoising framework, which can be used for the extraction of the signal subspace from noisy multichannel observations in low signal-to-noise ratios, using suitable prior information of the signal or noise. The method is fairly generic and may also be useful for the separation of other signals and noise even in the cases that BSS assumptions are not satisfied. The performance of the proposed technique is evaluated on both real and synthetic data and has shown significant outperformance as compared with the state-of-the-art methods.

Congenital heart defects are among the most common birth defects and the leading cause of birth defect-related deaths. Most cardiac defects have some manifestation in the morphology of cardiac electrical signals, which are recorded by electrocardiography and are believed to contain much more information as compared with conventional sonographic methods. Therefore, the noninvasive study of fetal cardiac signals can provide an effective means of monitoring the well-being of the fetal heart and may be used for the early detection of cardiac abnormalities.In previous studies, various methods have been developed for the processing and extraction of fetal electrocardiogram (ECG) signals recorded from the maternal body surface. However, due to the low signal-to-noise ratio of these signals, the application of fetal electrocardiography has been limited to heartbeat analysis and invasive ECG recordings during labor.In this research, the objective is to improve the signal processing aspects of fetal cardiography and to provide better insights of this problem, by developing new techniques for the modeling and filtering of fetal ECG signals recorded from an array of electrodes placed on the maternal abdomen. The basic idea behind the developed methods is to use a priori information about cardiac signals, such as their pseudo-periodic structure, to improve the performance of the currently existing techniques and to design novel filtering techniques that are customized for cardiac signals. Due to the overlap of the fetal signals and interferences/noises in different domains, the methods that use the information in only one of these domains do not usually succeed in extracting the fetal ECG. Therefore, we design methods that use the information from various domains, in order to improve the quality of the extracted signals.Theoretically, the proposed methods are combinations of morphological models of the ECG, ad hoc Bayesian filtering techniques based on estimation theory, and special classes of spatial filters adapted from the blind and semi-blind source separation context. It is shown that due to the generality of the proposed methods, the same procedures are also applicable to multichannel adult ECG recordings and can be used in real-time cardiac monitoring systems.Moreover, the developed methods are based on the cardiac signal morphology without going into the details of volume conduction theory and the conductivities of the propagation media. Hence, the same methods are applicable to other cardiac monitoring modalities such as the magnetocardiogram (MCG), which are morphologically similar to the ECG. We specifically present a case study on the extraction of twin fetal MCG signals.We also present an advanced deflation technique, which is able to separate subspaces of desired signals from degenerate mixtures of signal and noise. This idea has found various applications in other contexts., Les malformations cardiaques congénitales sont parmi les malformations les plus communes à la naissance et la première cause de décès des nouveau-nés. La plupart des anomalies cardiaques sont visibles dans la morphologie des signaux électriques cardiaques, qui sont enregistrés par l'électrocardiographie qui semble contenir plus d'informations par rapport aux méthodes conventionnelles sonographiques. Par conséquent, l'étude non invasive des signaux cardiaques du foetus peut fournir un moyen efficace pour contrôler le bon fonctionnement du coeur du foetus et peut être utilisé pour la détection précoce des anomalies cardiaques.Dans les précédentes études, diverses méthodes ont été mises au point pour le traitement et l'extraction d'électrocardiogramme (ECG) du foetus, à partir des signaux enregistrés de la surface du corps de la mère. Toutefois, en raison du faible rapport signal/bruit de ces signaux, l'application d'électrocardiographie fétale a été limitée à l'analyse des battements cardiaques et à des enregistrements ECG invasifs pendant l'accouchement.Dans cette recherche, l'objectif est d'améliorer les méthodes de traitement du signal utilisées en cardiographie du foetus et d'apporter de nouvelles solutions à ce problème, en développant de nouvelles techniques de modélisation et de filtrage des signaux d'ECG du foetus enregistrés par un réseau d'électrodes placées sur le ventre maternel. L'idée de base derrière les méthodes développées, consiste à utiliser les informations a priori des signaux cardiaques, tels que leur pseudo-périodicité, afin d'améliorer les performances des méthodes existantes et de concevoir de nouvelles techniques de filtrage qui sont spécifiques aux signaux cardiaques. En raison du recouvrement des signaux du foetus avec les interférences/bruits dans différents domaines, les méthodes qui utilisent l'information dans un seul de ces domaines, ne réussissent pas à extraire les ECG foetaux. Par conséquent, nous proposons des méthodes de traitement qui utilisent les informations provenant de différents domaines, afin d'améliorer la qualité des signaux extraits.Théoriquement, les méthodes proposées sont des combinaisons de modèles morphologiques de l'ECG, de techniques de filtrage bayésienne ad hoc basées sur la théorie de l'estimation et de classes spéciales de filtres spatiaux issus du contexte de la séparation aveugle et semi-aveugle de sources. Il est montré que, en raison de la généralité des méthodes proposées, les mêmes procédures sont également applicables aux signaux ECG multicapteurs chez l'adulte, et peuvent être utilisées en temps réel dans les systèmes de surveillance cardiaque.En outre, les méthodes développées sont fondées sur la morphologie du signal cardiaque, sans prendre en compte les particularités de la théorie du volume conducteur et la propagation électromagnétique dans les milieux physiologiques. Par conséquent, les mêmes méthodes sont applicables à d'autres modalités de surveillance cardiaque, comme le magnétocardiogramme (MCG), qui sont morphologiquement similaire à l'ECG. En particulier, nous présentons une étude de cas sur l'extraction des signaux MCG de jumeaux.Nous présentons également une technique originale de déflation, qui vise à séparer les sous-espaces formés par les signaux d'intérêt dans des mélanges sous-déterminés. Cette idée s'avère très performante et débouche sur des applications diverses dans d'autres contextes.

International audience; In some recent works, model-based filtering approaches have been proved as effective methods for extracting ECG signals from single channel noisy recordings. The previously developed methods, use a highly realistic nonlinear ECG model for the construction of Bayesian filters. In this work, a multi-channel extension of the previous approach is developed, by using a three dimensional model of the cardiac dipole vector. The results have considerable improvement compared with the single channel approach. The method is hence believed to be applicable to low SNR multi-channel recordings.

International audience; In this paper the Extended Kalman Filter (EKF) has been used for the filtering of Electrocardiogram (ECG) signals. The method is based on a previously nonlinear dynamic model proposed for the generation of synthetic ECG signals. The results show that the EKF may be used as a powerful tool for the extraction of ECG signals from noisy measurements; which is the state of the art in applications such as the noninvasive extraction of fetal cardiac signals from maternal abdominal signals.