Application of multivariate autoregressive modeling for analyzing the interaction between EEG and EMG in humans

Understanding the network of the human motor control system in noninvasive ways is beneficial not only for designing human interfaces, but also to clinical applications. This article presents applications of multivariate autoregression (MVAR) modeling for analyzing the interaction between EEG and EMG, which is challenging because of their different modalities. In contrast to previous research employing the WAR modeling by means of frequency-domain analysis, our approach emphasizes time-domain analysis. We examined one normal subject and one mirror-movement (MM) patient. The task was a weak isotonic contraction of the right abductor pollicis brevis muscle in the normal subject, and the left extensor carpi radialis brevis in the MM patient. For each subject, three channels consisting of two EEG signals and one EMG signal were analyzed. The EMG signals were from the bilateral primary sensorimotor cortices. By using the Bayesian Information Criterion (BIC), and by choosing the appropriate data length, the model order was determined in a stable fashion. Our results provided plausible information on EEG-EMG networks: (1) Information-transmission-delay time that seems physiologically appropriate, and (2) relative contribution from the ipsi- and contralateral corticospinal pathway, which is opposite in the normal subject in comparison to MM patients. (C) 2004 Elsevier B.V All rights reserved.