Automatic features selection in BCI-supported motor imagery practice for stroke rehabilitation

Brain Computer Interfaces (BCIs) can support motor imagery practice during the neuromotor rehabilitation of post-stroke patients. The practical application of this approach in a clinical setting implies to simplify configuration procedures: the EEG activity to be employed in the BCI setting, and thus reinforced via the training, should be selected taking into account neurophysiological evidence and rehabilitation principles. In this study, we tested an automatic procedure to select the BCI control parameters (i.e. specific EEG signal’s characteristics) based on neurophysiological and rehabilitation principles. To this aim, we compared the classification’s performance of an algorithm for the automatic features selection (stepwise linear discriminant analysis) with a configuration procedures based on human choices. Preliminary results did not show significant differences of the proposed method with respect to the selection of features performed by highly skilled medical doctors and therapists.