Time varying EEG Bandpower Estimation Improves 3D Hand Motion Trajectory Prediction Accuracy

Abstract

Motion trajectory prediction (MTP) employs a time-series of band-pass filtered EEG potentials for reconstructing the three-dimensional (3D) trajectory of limb movements with a multiple linear regression (mLR) block. While traditional multiclass classification methods use power values of mu (8-12Hz) and beta (12-30Hz) bands for limb movement based classification, recent MTP brain-computer interface (BCI) studies report the best accuracy using a 0.5-2Hz band-pass filter. We recently introduced a novel approach for MTP BCIs where the time-series of band-pass filtered EEG potentials were replaced with the time-series of power values of subject-specific frequency band(s) prior to the application of mLR. Here we present an analysis of three subjects performing 3D arm movements and comparing the accuracy rates of the standard EEG potential model and the proposed spectrum power-based approach.