Muscle activity reconstruction from EEG cortical current source

 The spatial resolution of electroencephalography (EEG) is known to be not enough to decode detailed information in the human brain. Therefore, most brain-machine interfaces for motor control use a phenomenon called event-related desynchronization (ERD) that is observed when people start and stop executing or thinking of movements. The ERD-based brain-machine interfaces (BMI) can decode which “body-parts” movements people are thinking about by identifying the approximate position where ERD is observed. To overcome the limit of spatial resolution, I estimated EEG current sources that were computationally defined on the cortex using a variational Bayesian method from EEG and fMRI data (Fig. 1, https://vbmeg.atr.jp). Using time-series signals of the estimated EEG current sources, I could first succeed in reconstructing the activity of flexor and extensor muscles (Neuroimage, 2012). This technique was then applied to develop electromyography (EMG) based arm or leg robot control. Fig. 2 shows how the estimated muscle activity signals from EEG current source (see the middle photo in Fig.2) can successfully control a wrist exoskeleton. In contrast muscle activity estimated from EEG sensor signals (the right photo) failed to realize the same flexion and extension motion as EMG signals (the left photo) could not discriminate the difference between flexor and extensor activities (Advanced Robotics 2016, Best paper award).