Filter Bank Common Spatial Pattern Algorithm on BCI Competition IV Datasets 2a and 2b

Kai Keng Ang,Zheng Yang Chin,Haihong Zhang,Cuntai Guan,Chuanchu Wang

doi:10.3389/fnins.2012.00039

Abstract

The Common Spatial Pattern (CSP) algorithm is an effective and popular method for classifying 2-class motor imagery electroencephalogram (EEG) data, but its effectiveness depends on the subject-specific frequency band. This paper presents the Filter Bank Common Spatial Pattern (FBCSP) algorithm to optimize the subject-specific frequency band for CSP on Datasets 2a and 2b of the Brain-Computer Interface (BCI) Competition IV. Dataset 2a comprised 4 classes of 22 channels EEG data from 9 subjects, and Dataset 2b comprised 2 classes of 3 bipolar channels EEG data from 9 subjects. Multi-class extensions to FBCSP are also presented to handle the 4-class EEG data in Dataset 2a, namely, Divide-and-Conquer (DC), Pair-Wise (PW), and One-Versus-Rest (OVR) approaches. Two feature selection algorithms are also presented to select discriminative CSP features on Dataset 2b, namely, the Mutual Information-based Best Individual Feature (MIBIF) algorithm, and the Mutual Information-based Rough Set Reduction (MIRSR) algorithm. The single-trial classification accuracies were presented using 10 × 10-fold cross-validations on the training data and session-to-session transfer on the evaluation data from both datasets. Disclosure of the test data labels after the BCI Competition IV showed that the FBCSP algorithm performed relatively the best among the other submitted algorithms and yielded a mean kappa value of 0.569 and 0.600 across all subjects in Datasets 2a and 2b respectively.

Highlights

The challenge in Motor Imagery-based Brain-Computer Interface (BCI) (MI-BCI), which translates the mental imagination of movement to commands, is the huge inter-subject variability with respect to the characteristics of the brain signals (Blankertz et al, 2007)
In this paper, the Filter Bank Common Spatial Pattern (FBCSP) algorithm is presented to classify singletrial EEG data for 2-class as well as 4-class motor imagery, where results using different feature selection algorithms and multiclass extensions to the FBCSP algorithm were compared with the Common Spatial Pattern (CSP) algorithm and other entries submitted to the BCI Competition IV Dataset 2a and Dataset 2b
Other algorithms were not included in this study, prior studies on the 2-class motor imagery data of the BCI Competition III Dataset IV showed that a modified SPEC-CSP algorithm using Support Vector Machines (SVM) yielded a 10 × 10-fold cross-validation classification accuracy of 89.5% (Wu et al, 2008) averaged over the 5 subjects, while the FBCSP algorithm yielded a 10 × 10-fold cross-validation classification accuracy of 90.3% (Ang et al, 2008)

Summary

INTRODUCTION

The challenge in Motor Imagery-based BCI (MI-BCI), which translates the mental imagination of movement to commands, is the huge inter-subject variability with respect to the characteristics of the brain signals (Blankertz et al, 2007). For effective use of the CSP algorithm, several parameters have to be specified, namely, the frequency for band-pass filtering of the EEG measurements, the time interval of the EEG measurements taken relative to the stimuli, and the subset of CSP filters to be used (Blankertz et al, 2008b). The Filter Bank Common Spatial Pattern (FBCSP) algorithm is presented to enhance the performance of the CSP algorithm by performing autonomous selection of discriminative subject-specific frequency range for band-pass filtering of the EEG measurements (Ang et al, 2008). This paper investigates the performance of the FBCSP algorithm on Dataset 2b (Leeb et al, 2007) using 2 mutual information-based feature selection algorithms.

FILTER BANK COMMON SPATIAL PATTERN

CLASSIFICATION

DIVIDE-AND-CONQUER

EXPERIMENTAL RESULTS

CONCLUSION