Jean-Marc Vesin

An efficient P300-based brain-computer interface for disabled subjects

A brain-computer interface (BCI) is a communication system that translates brain-activity into commands for a computer or other devices. In other words, a BCI allows users to act on their environment by using only brain-activity, without using peripheral nerves and muscles. In this paper, we present a BCI that achieves high classification accuracy and high bitrates for both disabled and able-bodied subjects. The system is based on the P300 evoked potential and is tested with five severely disabled and four able-bodied subjects. For four of the disabled subjects classification accuracies of 100% are obtained. The bitrates obtained for the disabled subjects range between 10 and 25bits/min. The effect of different electrode configurations and machine learning algorithms on classification accuracy is tested. Further factors that are possibly important for obtaining good classification accuracy in P300-based BCI systems for disabled subjects are discussed.

Brain-computer interface in multimedia communication

This article raises various issues in the design of an efficient BCI system in multimedia applications. The main focus is on one specific modality, namely an electroencephalography (EEG)-based BCI. In doing so, we provide an overview of the most recent progress achieved in this field, with an emphasis on signal processing aspects.

Significant Molecular and Systemic Adaptations after Repeated Sprint Training in Hypoxia

While intermittent hypoxic training (IHT) has been reported to evoke cellular responses via hypoxia inducible factors (HIFs) but without substantial performance benefits in endurance athletes, we hypothesized that repeated sprint training in hypoxia could enhance repeated sprint ability (RSA) performed in normoxia via improved glycolysis and O2 utilization. 40 trained subjects completed 8 cycling repeated sprint sessions in hypoxia (RSH, 3000 m) or normoxia (RSN, 485 m). Before (Pre-) and after (Post-) training, muscular levels of selected mRNAs were analyzed from resting muscle biopsies and RSA tested until exhaustion (10-s sprint, work-to-rest ratio 1∶2) with muscle perfusion assessed by near-infrared spectroscopy. From Pre- to Post-, the average power output of all sprints in RSA was increased (p<0.01) to the same extent (6% vs 7%, NS) in RSH and in RSN but the number of sprints to exhaustion was increased in RSH (9.4±4.8 vs. 13.0±6.2 sprints, p<0.01) but not in RSN (9.3±4.2 vs. 8.9±3.5). mRNA concentrations of HIF-1α (+55%), carbonic anhydrase III (+35%) and monocarboxylate transporter-4 (+20%) were augmented (p<0.05) whereas mitochondrial transcription factor A (−40%), peroxisome proliferator-activated receptor gamma coactivator 1α (−23%) and monocarboxylate transporter-1 (−36%) were decreased (p<0.01) in RSH only. Besides, the changes in total hemoglobin variations (Δ[tHb]) during sprints throughout RSA test increased to a greater extent (p<0.01) in RSH. Our findings show larger improvement in repeated sprint performance in RSH than in RSN with significant molecular adaptations and larger blood perfusion variations in active muscles.

Study of unipolar electrogram morphology in a computer model of atrial fibrillation.

Introduction: Electrograms exhibit a wide variety of morphologies during atrial fibrillation (AF). The basis of these time courses, however, is not completely understood. In this study, data from computer models were studied to relate features of the signals to the underlying dynamics and tissue substrate.

Study of atrial arrhythmias in a computer model based on magnetic resonance images of human atria.

The maintenance of multiple wavelets appears to be a consistent feature of atrial fibrillation (AF). In this paper, we investigate possible mechanisms of initiation and perpetuation of multiple wavelets in a computer model of AF. We developed a simplified model of human atria that uses an ionic-based membrane model and whose geometry is derived from a segmented magnetic resonance imaging data set. The three-dimensional surface has a realistic size and includes obstacles corresponding to the location of major vessels and valves, but it does not take into account anisotropy. The main advantage of this approach is its ability to simulate long duration arrhythmias (up to 40 s). Clinically relevant initiation protocols, such as single-site burst pacing, were used. The dynamics of simulated AF were investigated in models with different action potential durations and restitution properties, controlled by the conductance of the slow inward current in a modified Luo-Rudy model. The simulation studies show that (1) single-site burst pacing protocol can be used to induce wave breaks even in tissue with uniform membrane properties, (2) the restitution-based wave breaks in an atrial model with realistic size and conduction velocities are transient, and (3) a significant reduction in action potential duration (even with apparently flat restitution) increases the duration of AF. (c) 2002 American Institute of Physics.

Heart Rate Dynamics at the Onset of Ventricular Tachyarrhythmias as Retrieved From Implantable Cardioverter-Defibrillators in Patients With Coronary Artery Disease

BACKGROUND The recent availability of implantable cardioverter-defibrillators (ICDs) that record 1024 R-R intervals preceding a ventricular tachyarrhythmia (VTA) provides a unique opportunity to analyze heart rate variability (HRV) before the onset of VTA. METHODS AND RESULTS Fifty-eight post-myocardial infarction patients with an implanted ICD for recurrent VTA provided 2 sets of 98 heart rate recordings in sinus rhythm: (1) before a VTA and (2) during control conditions. Three subgroups were considered according to the antiarrhythmic (AA) drug regimen. A state of sympathoexcitation was suggested by the significant reduction in HRV before VTA onset compared with control conditions. beta-Blockers and dl-sotalol enhanced HRV in control recordings; nevertheless, HRV declined before VTA independent of AA drugs. A gradual increase in heart rate and decrease in sinus arrhythmia at VTA onset were specific findings of patients who received dl-sotalol. CONCLUSIONS The peculiar heart rate dynamics observed before VTA onset are suggestive of a state of sympathoexcitation that is independent of AA drugs.

Cancellation of Ventricular Activity in the ECG: Evaluation of Novel and Existing Methods

Due to the much higher amplitude of the electrical activity of the ventricles in the surface electrocardiogram (ECG), its cancellation is crucial for the analysis and characterization of atrial fibrillation. In this paper, two different methods are proposed for this cancellation. The first one is an average beat subtraction type of method. Two sets of templates are created: one set for the ventricular depolarization waves and one for the ventricular repolarization waves. Next, spatial optimization (rotation and amplitude scaling) is applied to the QRS templates. The second method is a single beat method that cancels the ventricular involvement in each cardiac cycle in an independent manner. The estimation and cancellation of the ventricular repolarization is based on the concept of dominant T and U waves. Subsequently, the atrial activities during the ventricular depolarization intervals are estimated by a weighted sum of sinusoids observed in the cleaned up segments. ECG signals generated by a biophysical model as well as clinical ECG signals are used to evaluate the performance of the proposed methods in comparison to two standard ABS-based methods

A Boosting Approach to P300 Detection with Application to Brain-Computer Interfaces

Gradient boosting is a machine learning method, that builds one strong classifier from many weak classifiers. In this work, an algorithm based on gradient boosting is presented, that detects event-related potentials in single electroencephalogram (EEG) trials. The algorithm is used to detect the P300 in the human EEG and to build a brain-computer interface (BCI), specifically a spelling device. Important features of the method described here are its high classification accuracy and its conceptual simplicity. The algorithm was tested with datasets recorded in our lab and one benchmark dataset from the BCI Competition 2003. The number of correctly inferred symbols with the P300 speller paradigm varied between 90% and 100%. In particular, all of the inferred symbols were correct for the BCI competition dataset

Stochastic analysis of gradient adaptive identification of nonlinear systems with memory for Gaussian data and noisy input and output measurements

This paper investigates the statistical behavior of two gradient search adaptive algorithms for identifying an unknown nonlinear system comprised of a discrete-time linear system H followed by a zero-memory nonlinearity g(/spl middot/). The input and output of the unknown system are corrupted by additive independent noises. Gaussian models are used for all inputs. Two competing adaptation schemes are analyzed. The first is a sequential adaptation scheme where the LMS algorithm is first used to estimate the linear portion of the unknown system. The LMS algorithm is able to identify the linear portion of the unknown system to within a scale factor. The weights are then frozen at the end of the first adaptation phase. Recursions are derived for the mean and fluctuation behavior of the LMS algorithm, which are in excellent agreement with Monte Carlo simulations. When the nonlinearity is modeled by a scaled error function, the second part of the sequential gradient identification scheme is shown to correctly learn the scale factor and the error function scale factor. Mean recursions for the scale factors show good agreement with Monte Carlo simulations. For slow learning, the stationary points of the gradient algorithm closely agree with the stationary points of the theoretical recursions. The second adaptive scheme simultaneously learns both the linear and nonlinear portions of the unknown channel. The mean recursions for the linear and nonlinear portions show good agreement with Monte Carlo simulations for slow learning. The stationary points of the gradient algorithm also agree with the stationary points of the theoretical recursions.

Support vector EEG classification in the Fourier and time-frequency correlation domains

We use support vector machines (SVM) for classifying EEG signals corresponding to imagined motor movements. The parameters of an SVM Kernel are optimized for minimizing a theoretical error bound. Fourier features and correlative time-frequency based features are extracted from EEG signals and compared with respect to their discriminatory power.