Hand gesture decoding using ultra-high-density EEG

The exploration of high EEG electrode densities is of great interest in current BCI research. This paper investigates the neural representation of high-level hand gestures using ultra-high-density EEG electrodes. The EEG of two subjects was recorded using a total of 352 and 256 electrodes placed over the sensorimotor cortex, respectively. Subject S1 performed motor execution, and subject S2 performed motor imagery of the hand gestures. Band power features for the mu (8-12 Hz) and beta (13–30 Hz) frequency bands were exploited for pairwise classifications for the respective gestures using a linear SVM. Topography plots were generated using the reconstructed head models from T1-weighted MRI scans of both subjects. The spatiotemporal dynamics from the topographies allow a detailed understanding of the associated brain patterns. Subjects S1 and S2 achieved a mean classification accuracy of 72.7% and 71.3% across all classification pairs, respectively. The classification of the gestures “rock” vs. “scissors” resulted in the highest classification accuracy for both subjects, with 78.9% for subject S1 and 73.9% for subject S2 on average. Classification models based on electrode subsets reflecting the 10–10 and the extended 10–10 systems performed worse than the ultra-high-density EEG system resulting in a decrease of about 10 and 8% (on median). The implementation of feature reduction algorithms, as well as real-time BCI feedback, are expected to enhance performance in future applications.