Identification of Motor Functions Based on an EEG Analysis

Abstract

A combination of several techniques is necessary for a reliable identification of activities based on EEG signals. A separation of the overlapping patterns in the EEG signals is often performed first. These separated patterns are then analysed by some artificial intelligence methods in order to identify the activity. As pattern separation and activity identification are often linked, the two processes must be tuned to a specific problem, thus losing some generality of the procedure. The complexity of the patterns in EEG signals is often too great for completely automated pattern recognition. In this case, phase demodulation was introduced as a procedure for the extraction of the phase properties of the EEG signals. These phase shifts are known to correlate with the brain activity; therefore, phase-demodulated EEG signals were used to predict the motor activity. Three studies with off-line identification of the motor activities have been performed so far. In the first study, a continuous gripping force was predicted. In the second study, index- and middle-finger activation was predicted, and in the final study, wrist movements were analysed. The presented procedure can be used for designing a continuous brain-computer interface.