A dictionary learning approach for spatio-temporal characterization of absence seizures.

OBJECTIVE: This research explores absence seizures using data recorded from different layers of somatosensory cortex of four genetic absence epilepsy rats from Strasbourg (GAERS). Localizing the active layers of somatosensory cortex (spatial analysis) and investigating the dynamics of recorded seizures (temporal analysis) are the main goals of this research. APPROACH: We model the spike discharges of seizures using a generative spatio-temporal model. We assume that there are some states under first-order Markovian model during seizures, and each spike is generated when the corresponding state is activated. We also assume that a few specific epileptic activities (or atoms) exist in each state which are linearly combined and form the spikes. Each epileptic activity is described by two characteristics: (1) its spatial topography which shows the organization of current sources and sinks generating the epileptic activity, and (2) its temporal representation which illustrates the activation function of the epileptic activity. We show that the estimation of the model parameters, i.e. states and their epileptic activities (atoms), is similar to solving a dictionary learning problem for sparse representation. Instead of using classical dictionary learning algorithms, a new approach, taking into account the Markovian nature of the model, is proposed for estimating the models parameters, and its efficiency is experimentally verified. MAIN RESULTS: Experimental results show that there are one dominant and one unstable state with two epileptic activities in each during the seizures (temporal analysis). It is also found that the top and bottom layers of the somatosensory cortex are the most active layers during seizures (spatial analysis). The structural model is similar for all rats with a spatial topography which is the same for all rats but a temporal activation which changes according to the rat. SIGNIFICANCE: The proposed framework can be applied on any database acquired from a small area of the brain, and can provide valuable spatio-temporal analysis for neuroscientists.

A closed-form unsupervised geometry-aware dimensionality reduction method in the Riemannian Manifold of SPD matrices.

Riemannian geometry has been found accurate and robust for classifying multidimensional data, for instance, in brain-computer interfaces based on electroencephalography. Given a number of data points on the manifold of symmetric positive-definite matrices, it is often of interest to embed these points in a manifold of smaller dimension. This is necessary for large dimensions in order to preserve accuracy and useful in general to speed up computations. Geometry-aware methods try to accomplish this task while respecting as much as possible the geometry of the original data points. We provide a closed-form solution for this problem in a fully unsupervised setting. Through the analysis of three brain-computer interface data bases we show that our method allows substantial dimensionality reduction without affecting the classification accuracy.

Integrated Analysis of EEG and fMRI Using Sparsity of Spatial Maps.

Integration of electroencephalography (EEG) and functional magnetic resonance imaging (fMRI) is an open problem, which has motivated many researches. The most important challenge in EEG-fMRI integration is the unknown relationship between these two modalities. In this paper, we extract the same features (spatial map of neural activity) from both modality. Therefore, the proposed integration method does not need any assumption about the relationship of EEG and fMRI. We present a source localization method from scalp EEG signal using jointly fMRI analysis results as prior spatial information and source separation for providing temporal courses of sources of interest. The performance of the proposed method is evaluated quantitatively along with multiple sparse priors method and sparse Bayesian learning with the fMRI results as prior information. Localization bias and source distribution index are used to measure the performance of different localization approaches with or without a variety of fMRI-EEG mismatches on simulated realistic data. The method is also applied to experimental data of face perception of 16 subjects. Simulation results show that the proposed method is significantly stable against the noise with low localization bias. Although the existence of an extra region in the fMRI data enlarges localization bias, the proposed method outperforms the other methods. Conversely, a missed region in the fMRI data does not affect the localization bias of the common sources in the EEG-fMRI data. Results on experimental data are congruent with previous studies and produce clusters in the fusiform and occipital face areas (FFA and OFA, respectively). Moreover, it shows high stability in source localization against variations in different subjects.

Single-Trial Classification of Multi-User P300-Based Brain-Computer Interface Using Riemannian Geometry.

The classification of electroencephalographic (EEG) data recorded from multiple users simultaneously is an important challenge in the field of Brain-Computer Interface (BCI). In this paper we compare different approaches for classification of single-trials Event-Related Potential (ERP) on two subjects playing a collaborative BCI game. The minimum distance to mean (MDM) classifier in a Riemannian framework is extended to use the diversity of the inter-subjects spatio-temporal statistics (MDM-hyper) or to merge multiple classifiers (MDM-multi). We show that both these classifiers outperform significantly the mean performance of the two users and analogous classifiers based on the step-wise linear discriminant analysis. More importantly, the MDM-multi outperforms the performance of the best player within the pair.

From molecular model to sparse representation of chromatographic signals with an unknown number of peaks.

Analysis of a fluid mixture using a chromatographic system is a standard technique for many biomedical applications such as in-vitro diagnostic of body fluids or air and water quality assessment. The analysis is often dedicated towards a set of molecules or biomarkers. However, due to the fluid complexity, the number of mixture components is often larger than the list of targeted molecules. In order to get an analysis as exhaustive as possible and also to take into account possible interferences, it is important to identify and to quantify all the components that are included in the chromatographic signal. Thus the signal processing aims to reconstruct a list of an unknown number of components and their relative concentrations. We address this question as a problem of sparse representation of a chromatographic signal. The innovative representation is based on a stochastic forward model describing the transport of elementary molecules in the chromatography column as a molecular random walk. We investigate three methods: two probabilistic Bayesian approaches, one parametric and one non-parametric, and a determinist approach based on a parsimonious decomposition on a dictionary basis. We examine the performances of these 3 approaches on an experimental case dedicated to the analysis of mixtures of the micro-pollutants Polycyclic Aromatic Hydrocarbons (PAH) in a methanol solution in two cases of high and low signal to noise ratio (SNR).

A robust sensor-selection method for P300 brain-computer interfaces.

A brain-computer interface (BCI) is a specific type of human-computer interface that enables direct communication between human and computer through decoding of brain activity. As such, event-related potentials like the P300 can be obtained with an oddball paradigm whose targets are selected by the user. This paper deals with methods to reduce the needed set of EEG sensors in the P300 speller application. A reduced number of sensors yields more comfort for the user, decreases installation time duration, may substantially reduce the financial cost of the BCI setup and may reduce the power consumption for wireless EEG caps. Our new approach to select relevant sensors is based on backward elimination using a cost function based on the signal to signal-plus-noise ratio, after some spatial filtering. We show that this cost function selects sensors' subsets that provide a better accuracy in the speller recognition rate during the test sessions than selected subsets based on classification accuracy. We validate our selection strategy on data from 20 healthy subjects.

Directed differential connectivity graph of interictal epileptiform discharges.

In this paper, we study temporal couplings between interictal events of spatially remote regions in order to localize the leading epileptic regions from intracerebral EEG (iEEG). We aim to assess whether quantitative epileptic graph analysis during interictal period may be helpful to predict the seizure onset zone of ictal iEEG. Using wavelet transform, cross-correlation coefficient, and multiple hypothesis test, we propose a differential connectivity graph (DCG) to represent the connections that change significantly between epileptic and nonepileptic states as defined by the interictal events. Postprocessings based on mutual information and multiobjective optimization are proposed to localize the leading epileptic regions through DCG. The suggested approach is applied on iEEG recordings of five patients suffering from focal epilepsy. Quantitative comparisons of the proposed epileptic regions within ictal onset zones detected by visual inspection and using electrically stimulated seizures, reveal good performance of the present method.

Comparison of five directed graph measures for identification of leading interictal epileptic regions.

Directed graphs (digraphs) derived from interictal periods of intracerebral EEG (iEEG) recordings can be used to estimate the leading interictal epileptic regions for presurgery evaluations. For this purpose, quantification of the emittance contribution of each node to the rest of digraph is important. However, the usual digraph measures are not very well suited for this quantification. Here, we compare the efficiency of recently introduced local information (LI) measure and a new measure called total global efficiency with classical measures like global efficiency, local efficiency and node degree. For evaluation, the estimated leading interictal epileptic regions based on five measures are compared with seizure onset zones obtained by visual inspection of epileptologists for five patients. The comparison revealed the superior performance of the LI measure. We showed efficiency of different digraph measures for the purpose of source and sink node identification.

Model-based Bayesian filtering of cardiac contaminants from biomedical recordings.

Electrocardiogram (ECG) and magnetocardiogram (MCG) signals are among the most considerable sources of noise for other biomedical signals. In some recent works, a Bayesian filtering framework has been proposed for denoising the ECG signals. In this paper, it is shown that this framework may be effectively used for removing cardiac contaminants such as the ECG, MCG and ballistocardiographic artifacts from different biomedical recordings such as the electroencephalogram, electromyogram and also for canceling maternal cardiac signals from fetal ECG/MCG. The proposed method is evaluated on simulated and real signals.

Filtering electrocardiogram signals using the extended Kalman filter.

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.

Maximum likelihood neural approximation in presence of additive colored noise.

In many practical situations, the noise samples may be correlated. In this case, the estimation of noise parameters can be used to improve the approximation. Estimation of the noise structure can also be used to find a stopping criterion in constructive neural networks. To avoid overfitting, a network construction procedure must be stopped when residual can be considered as noise. The knowledge on the noise may be used for "whitening" the residual so that a correlation hypothesis test determines if the network growing must be continued or not. In this paper, supposing a Gaussian noise model, we study the problem of multi-output nonlinear regression using MLP when the noise in each output is a correlated autoregressive time series and is spatially correlated with other output noises. We show that the noise parameters can be determined simultaneously with the network weights and used to construct an estimator with a smaller variance, and so to improve the network generalization performance. Moreover, if a constructive procedure is used to build the network, the estimated parameters may be used to stop the procedure.