EEG Based Foot Movement Onset Detection with the Probabilistic Classification Vector Machine

A critical issue in designing a self-paced brain computer interface (BCI) system is onset detection of the mental task from the continuous electroencephalogram (EEG) signal to produce a brain switch. This work shows significant improvement in a movement based self-paced BCI by applying a new sparse learning classification algorithm, probabilistic classification vector machines (PCVMs) to classify EEG signal. Constant-Q filters instead of constant bandwidth filters for frequency decomposition are also shown to enhance the discrimination of movement related patterns from EEG patterns associated with idle state. Analysis of the data recorded from seven subjects executing foot movement using the constant-Q filters and PCVMs shows a statistically significant 17% (p<0.03) average improvement in true positive rate (TPR) and a 2% (p<0.03) reduction in false positive rate (FPR) compared with applying constant bandwidth filters and SVM classifier.