Bispectrum-based feature extraction technique for devising a practical brain-computer interface

Shahjahan Shahid, G Prasad

Research output: Contribution to journalArticle

27 Citations (Scopus)

Abstract

The extraction of distinctly separable features from electroencephalogram (EEG) is one of the main challenges in designing a brain–computer interface (BCI). Existing feature extraction techniques for a BCI are mostly developed based on traditional signal processing techniques assuming that the signal is Gaussian and has linear characteristics. But the motor imagery (MI)-related EEG signals are highly non-Gaussian, non-stationary and have nonlinear dynamic characteristics. This paper proposes an advanced, robust but simple feature extraction technique for a MI-related BCI. The technique uses one of the higher order statistics methods, the bispectrum, and extracts the features of nonlinear interactions over several frequency components in MI-related EEG signals. Along with a linear discriminant analysis classifier, the proposed technique has been used to design an MI-based BCI. Three performance measures, classification accuracy, mutual information and Cohen's kappa have been evaluated and compared with a BCI using a contemporary power spectral density-based feature extraction technique. It is observed that the proposed technique extracts nearly recording-session-independent distinct features resulting in significantly much higher and consistent MI task detection accuracy and Cohen's kappa. It is therefore concluded that the bispectrum-based feature extraction is a promising technique for detecting different brain states.
LanguageEnglish
Pages1-12
JournalJournal of Neural Engineering
Volume8
Issue number2
DOIs
Publication statusPublished - Apr 2011

Fingerprint

Brain computer interface
Feature extraction
Electroencephalography
Higher order statistics
Power spectral density
Discriminant analysis
Brain
Signal processing
Classifiers

Cite this

@article{36623557ec6f46978395f49caae5fd82,
title = "Bispectrum-based feature extraction technique for devising a practical brain-computer interface",
abstract = "The extraction of distinctly separable features from electroencephalogram (EEG) is one of the main challenges in designing a brain–computer interface (BCI). Existing feature extraction techniques for a BCI are mostly developed based on traditional signal processing techniques assuming that the signal is Gaussian and has linear characteristics. But the motor imagery (MI)-related EEG signals are highly non-Gaussian, non-stationary and have nonlinear dynamic characteristics. This paper proposes an advanced, robust but simple feature extraction technique for a MI-related BCI. The technique uses one of the higher order statistics methods, the bispectrum, and extracts the features of nonlinear interactions over several frequency components in MI-related EEG signals. Along with a linear discriminant analysis classifier, the proposed technique has been used to design an MI-based BCI. Three performance measures, classification accuracy, mutual information and Cohen's kappa have been evaluated and compared with a BCI using a contemporary power spectral density-based feature extraction technique. It is observed that the proposed technique extracts nearly recording-session-independent distinct features resulting in significantly much higher and consistent MI task detection accuracy and Cohen's kappa. It is therefore concluded that the bispectrum-based feature extraction is a promising technique for detecting different brain states.",
author = "Shahjahan Shahid and G Prasad",
year = "2011",
month = "4",
doi = "10.1088/1741-2560/8/2/025014",
language = "English",
volume = "8",
pages = "1--12",
journal = "Journal of Neural Engineering",
issn = "1741-2560",
number = "2",

}

Bispectrum-based feature extraction technique for devising a practical brain-computer interface. / Shahid, Shahjahan; Prasad, G.

In: Journal of Neural Engineering, Vol. 8, No. 2, 04.2011, p. 1-12.

Research output: Contribution to journalArticle

TY - JOUR

T1 - Bispectrum-based feature extraction technique for devising a practical brain-computer interface

AU - Shahid, Shahjahan

AU - Prasad, G

PY - 2011/4

Y1 - 2011/4

N2 - The extraction of distinctly separable features from electroencephalogram (EEG) is one of the main challenges in designing a brain–computer interface (BCI). Existing feature extraction techniques for a BCI are mostly developed based on traditional signal processing techniques assuming that the signal is Gaussian and has linear characteristics. But the motor imagery (MI)-related EEG signals are highly non-Gaussian, non-stationary and have nonlinear dynamic characteristics. This paper proposes an advanced, robust but simple feature extraction technique for a MI-related BCI. The technique uses one of the higher order statistics methods, the bispectrum, and extracts the features of nonlinear interactions over several frequency components in MI-related EEG signals. Along with a linear discriminant analysis classifier, the proposed technique has been used to design an MI-based BCI. Three performance measures, classification accuracy, mutual information and Cohen's kappa have been evaluated and compared with a BCI using a contemporary power spectral density-based feature extraction technique. It is observed that the proposed technique extracts nearly recording-session-independent distinct features resulting in significantly much higher and consistent MI task detection accuracy and Cohen's kappa. It is therefore concluded that the bispectrum-based feature extraction is a promising technique for detecting different brain states.

AB - The extraction of distinctly separable features from electroencephalogram (EEG) is one of the main challenges in designing a brain–computer interface (BCI). Existing feature extraction techniques for a BCI are mostly developed based on traditional signal processing techniques assuming that the signal is Gaussian and has linear characteristics. But the motor imagery (MI)-related EEG signals are highly non-Gaussian, non-stationary and have nonlinear dynamic characteristics. This paper proposes an advanced, robust but simple feature extraction technique for a MI-related BCI. The technique uses one of the higher order statistics methods, the bispectrum, and extracts the features of nonlinear interactions over several frequency components in MI-related EEG signals. Along with a linear discriminant analysis classifier, the proposed technique has been used to design an MI-based BCI. Three performance measures, classification accuracy, mutual information and Cohen's kappa have been evaluated and compared with a BCI using a contemporary power spectral density-based feature extraction technique. It is observed that the proposed technique extracts nearly recording-session-independent distinct features resulting in significantly much higher and consistent MI task detection accuracy and Cohen's kappa. It is therefore concluded that the bispectrum-based feature extraction is a promising technique for detecting different brain states.

U2 - 10.1088/1741-2560/8/2/025014

DO - 10.1088/1741-2560/8/2/025014

M3 - Article

VL - 8

SP - 1

EP - 12

JO - Journal of Neural Engineering

T2 - Journal of Neural Engineering

JF - Journal of Neural Engineering

SN - 1741-2560

IS - 2

ER -