Continuous Detection of Motor Imagery in a Four-Class Asynchronous BCI

Asynchronous brain computer interface (BCI) is an important class of BCI systems that has not received enough attention from the BCI community. In this work we introduce for the first time a system for classification of four different motor imageries in the context of an asynchronous BCI system which distinguishes between periods of movement imagination occurrence and idling or resting periods of ongoing EEG signal as well as classifying the 4 class motor imageries. We used two multi class extensions of the method of Common Spatial Patterns (CSP) for feature extraction and LDA, SVM, and MDA well known classifiers for combination purposes. We have applied our procedure to data set Ilia from BCI Competition III [2]. Offline evaluation of a prototype system demonstrated true positive rates in the range of 56%- 88% with corresponding false positive rates in the range of 18%-9%.