Radu Ranta

Blind source separation, wavelet denoising and discriminant analysis for EEG artefacts and noise cancelling

Abstract This paper proposes an automatic method for artefact removal and noise elimination from scalp electroencephalogram recordings (EEG). The method is based on blind source separation (BSS) and supervised classification and proposes a combination of classical and news features and classes to improve artefact elimination (ocular, high frequency muscle and ECG artefacts). The role of a supplementary step of wavelet denoising (WD) is explored and the interactions between BSS, denoising and classification are analyzed. The results are validated on simulated signals by quantitative evaluation criteria and on real EEG by medical expertise. The proposed methodology successfully rejected a good percentage of artefacts and noise, while preserving almost all the cerebral activity. The “denoised artefact-free” EEG presents a very good improvement compared with recorded raw EEG: 96% of the EEGs are easier to interpret.

Seizure lateralization in scalp EEG using Hjorth parameters

OBJECTIVE This paper describes and assesses a new semi-automatic method for temporal lobe seizures lateralization using raw scalp EEG signals. METHODS We used the first two Hjorth parameters to estimate quadratic mean and dominant frequency of signals. Their mean values were computed on each side of the brain and segmented taking into account the seizure onset time identified by the electroencephalographist, to keep only the initial part of the seizure, before a possible spreading to the contralateral side. The means of segmented variables were used to characterize the seizure by a point in a (frequency, amplitude) plane. Six criteria were proposed for the partitioning of this plane for lateralization. RESULTS The procedure was applied to 45 patients (85 seizures). The two best criteria yielded, for the first one, a correct lateralization for 96% of seizures and, for the other, a lateralization rate of 87% without incorrect lateralization. CONCLUSIONS The method produced satisfactory results, easy to interpret. The setting of procedure parameters was simple and the approach was robust to artifacts. It could constitute a help for neurophysiologists during visual inspection. SIGNIFICANCE The difference of quadratic mean and dominant frequency on each side of the brain allows lateralizing the seizure onset.

EEG Ocular Artefacts and Noise Removal

The general framework of this research is the pre-processing of the electroencephalographic (EEG) signals. The goal of this paper is to compare several combinations of wavelet denoising (WD) and independent component analysis (ICA) algorithms for noise and artefacts removal. These methods are tested on simulated EEG, using different evaluation criteria. According to our results, the most effective method consists in source separation by SOBI-RO [1], followed by wavelet denoising by SURE thresholding [2].

Digestive Activity Evaluation by Multichannel Abdominal Sounds Analysis

This paper introduces a complete methodology for abdominal sounds analysis, from signal acquisition to statistical data analysis. The goal is to evaluate if and how phonoenterograms can be used to detect different functioning modes of the normal gastrointestinal tract, both in terms of localization and of time evolution during the digestion. After the description of the acquisition protocol and the employed instrumentation, several signal processing steps are presented: wavelet denoising and segmentation, artifact suppression, and source localization. Next, several physiological features are extracted from the processed signals issued from a database of 14 healthy volunteers, recorded during 3 h after a standardized meal. Data analysis is performed using a multifactorial statistical method. Based on the introduced approach, we show that the abdominal regions of healthy volunteers present statistically significant phonoenterographic characteristics, which evolve differently during the normal digestion. The most significant feature allowing us to distinguish regions and time differences is the number of recorded sounds, but important information is also carried by sound amplitudes, frequencies, and durations. Depending on the considered feature, the sounds produced by different abdominal regions (especially stomach, ileocaecal, and lower abdomen regions) present a specific distribution over space and time. This information, statistically validated, is usable in further studies as a comparison term with other normal or pathological conditions.

Iterative wavelet-based denoising methods and robust outlier detection

The goal of this letter is to study convergence conditions for a previously presented iterative wavelet denoising method and to shed light on its relationship with outlier rejection. This method involves a user-defined parameter, which must fulfill certain conditions in order to ensure denoising. Using generalized Gaussian modeling for the wavelet coefficients distribution, we obtain a lower bound for this parameter, and the resulting threshold, both adapted to the shape of the distribution. The properties of this threshold are examined, and the proposed method is compared with other classical rejection methods.

EEG montage analysis in the Blind Source Separation framework

Abstract Blind Source Separation (BSS) is a relatively recent technique, more and more applied in electroencephalographic (EEG) signal processing. Still, the classical mixing model of the BSS does not take into account the real recording set-up. In fact, a major problem in electrophysiological recording systems (e.g. ECG, EEG, EMG) is to find a region in the human body whose bio-potential activity can be considered as neutral as possible i.e., a quasi-inactive reference place. Nowadays, it is well known that it is impossible to find a “zero-potential” site on the human body. In particular, the most common way of performing EEG recordings is by using as a common reference an electrode placed somewhere on the head. Starting from this Common Reference Montage (CRM), several other montages can be constructed to obtain alternative interpretation or processing solutions. Regardless of the chosen montage, the reference electrode intervenes in the mixing model of the BSS. The objective of this work is to analyse the influence of the montage on the mixing matrix and the quality of the BSS solution. This paper proposes to formalize the source separation problem in a non zero-potential reference context and shows that the Average Reference Montage (ARM), augmented by a virtual “average measure”, leads to better source separation results (separability index IS ). This conclusion is supported by simulated EEGs using the most common montages i.e., Common Reference Montage, Average Reference Montage and Bipolar-Longitudinal Montage, as well as by real EEG examples.

Interpretation and improvement of an iterative wavelet-based denoising method

The goal of this paper is to shed new light on a wavelet-based denoising method developed by Hadjileontiadis et al. (1997, 2000) which is derived from an iterative denoising algorithm by Coifman and Wickerhauser (1995, 1998). The underlying algorithm is revisited and interpreted as a fixed-point algorithm. This allows us to derive a new version of the algorithm largely increasing computational efficiency.

Wavelet-based bowel sounds denoising, segmentation and characterization

The general framework of this communication is phonoenterography. The ultimate goal is the development of a clinical diagnostic tool based on abdominal sound monitoring. Bowel sounds are recorded using several microphones. Unsupervised data processing should lead to diagnosis assessment. We address here the early stages of data processing, i.e., denoising, segmentation and characterization of detected events. The denoising algorithm is based on former work by Coifman and Wickerhauser [1998] and Hadjileontiadis et al. [1997], [2000]. Their wavelet-based algorithm is revisited, allowing to significantly reduce the computational burden. Sound segmentation and event characterization are based on the wavelet representation of the phonoenterogram. Real data processing examples are given.

Denoising Depth EEG Signals During DBS Using Filtering and Subspace Decomposition

In difficult epileptic patients, the brain structures are explored by means of depth multicontact electrodes [stereoelectroencephalography (SEEG)]. Recently, a novel diagnostic technique allows an accurate definition of the epileptogenic zone using deep brain stimulation (DBS). The stimulation signal propagates in the brain and thus it appears on most of the other SEEG electrodes, masking the local brain electrophysiological activity. The objective of this paper is the DBS-SEEG signals detrending and denoising in order to recover the masked physiological sources. We review the main filtering methods and put forward an approach based on the combination of filtering with generalized eigenvalue decomposition (GEVD). An experimental study on simulated and real SEEG shows that our approach is able to separate DBS sources from brain activity. The best results are obtained by an original singular spectrum analysis-GEVD approach.

A unified treatment of the reference estimation problem in depth EEG recordings

The starting point of this paper is the analysis of the reference problem in intra-cerebral electroencephalographic (iEEG) recordings. It is well accepted that both surface and depth EEG signals are always recorded with respect to some unknown time-varying signal called reference. This article discusses different methods for determining and reducing the influence of the reference signal for the iEEG signals. In particular, we derive optimal approaches for the estimation of the reference signal in iEEG recording setups and demonstrate their relation to the well-known minimum power/variance distortionless response approaches derived for general array and antenna signal processing applications. We show that the proposed approaches achieve optimal performance in terms of estimation error and that they outperform other reference identification methods proposed in the literature. The developed algorithms are illustrated on simulated examples and on real iEEG signals.