Improvement of Discrimination Performance Using Temporal Smoothing for Brain–Machine Interface Based Rehabilitation System

The aim of our study is to develop a brain– machine interface rehabilitation system for patients with leg paralyzed. Using this system, the patient’s paralyzed legs are forcibly moved according to his intention of motion. This may activate a damaged neural circuit and improve the rehabilitation effect. In this study, we proposed a motion discrimination method for actual pedal exercise using electroencephalography (EEG) measured at several positions of the parietal region, and the discrimination performance was verified with healthy subjects. Although this method was uses the spatial EEG information, this often causes false detection owing to the sudden noise included in the measured EEG signals. In order to improve the discrimination performance, smoothing of the motion discriminator output was considered using temporal information. Thus, we developed a spatiotemporal filterbased discrimination method and its parameter determination method. Experimental results indicated that the discrimination performance of this method is over 10 percentage points higher than that of the general linear discriminant analysis method.