Hybrid Trajectory Decoding from ECoG Signals for Asynchronous BCIs

Brain-Computer Interfaces (BCIs) are systems which convert brain neural activity into commands for external devices. BCI users generally alternate between No Control (NC) and Intentional Control (IC) periods. Numerous motor-related BCI decoders focus on the prediction of continuously-valued limb trajectories from neural signals. Although NC/IC discrimination is crucial for clinical BCIs, continuous decoders rarely support NC periods. Integration of NC support in continuous decoders is investigated in the present article. Two discrete/continuous hybrid decoders are compared for the task of asynchronous wrist position decoding from ElectroCorticoGraphic (ECoG) signals in monkeys. One static and one dynamic decoder, namely a Switching Linear (SL) decoder and a Switching Kalman Filter (SKF), are evaluated on high dimensional time-frequency-space ECoG signal representations. The SL decoder was found to outperform the SKF for both NC/IC class detection and trajectory modeling.