Multi-Class Independent Common Spatial Patterns: Exploiting Energy Variations of Brain Sources

This paper presents a method to recover task-related sources from a multi-class Brain-Computer Interface (BCI) based on motor imagery. Our method gathers two common approaches to tackle the multi-class problem: 1) the supervised approach of Common Spatial Pattern (CSP) to discriminate between different tasks; 2) the criterion of statistical independence of non-stationary sources used in Independent Component Analysis (ICA). We show that the resulting spatial filters have to be adapted to each subject and that the combined use of intra-trial and inter-class energy variations of brain sources yield an increase of classification rates for four among eight sub jects.