Optimizing ICA-Based Spatial Filters for Long-Term BCI Users

Independent Component Analysis (ICA) has been proved as an effective method for separating non-neural artifacts from task-related neural activities. In most cases, the "bad data", contaminated with the obvious artifacts, was recognized by visual inspection. This paper proposed a fresh artifact-detected method, which can identify the artifact trials automatically. Instead of using the standard original codes, the self-written computer codes of Infomax algorithm have been applied in the three class motor imagery Electroencephalographic (EEG) datasets. Meanwhile, a new strategy was thereby proposed for automatic selection of the three Movement-related Independent Components (MRICs), which is very crucial for ICA-based BCI system. Then the different brain states were discriminated by the variances criterion based on the Event-related Desynchronization (ERD) phenomena. The results suggested that our approach of optimizing the ICA spatial filters can achieve higher average classification accuracy, compared to that based on random selected EEG data.