An source-based common spatial filter selection for improving mis-triggering problem in brain-computer interface based on motor imagery

Motor imagery based brain-computer interface (MI-BCI) has been proven to be effective in post-stroke rehabilitation. However, the low specificity of MI-BCI is usually associated with a mis-triggering problem, which means the irrelevant task would induce a false positive output. Traditional classification method collects more training data for interferences which overburdens patients. To reduce mis-triggering rate in MI-BCI, we proposed a source selection common spatial pattern (SS-CSP) method to build the MI classifier without extra training sessions on irrelevant tasks. Using this approach, we were able to acquire a mis-triggering rate for interference task as low as 36%, showing a decrease of nearly 23% compared to traditional method. Meanwhile, the accuracy of two methods only had difference of 6%. The results suggest that our method can significantly depress the mis-triggering rate and has potential effect on post-stroke rehabilitation.