Reconstruction of hand movements from EEG signals based on non-linear regression

Brain-Computer Interface (BCI) systems allow users to control external devices using their thoughts. In particular, brain signals can be used to decode the trajectory of hand movements for neurorehabilitation or control of arm prostheses. Previous studies have decoded hand movement velocity during simple tasks. However, under real world conditions, patients need to control artificial limbs with more degrees of freedom in order to accomplish everyday tasks such as drinking water or eating food. In this work we decode hand movement velocity from electroencephalography (EEG) signals based on linear and nonlinear regression during complex trajectories. We considered two types of movement trajectories: one with low variation in movement velocity and one with high variation in hand movement velocity. Two decoding strategies are compared, linear and non-linear regression. Our results show that linear models can yield state-of-the-art decoding performance on the simple task with low variations in movement velocity, in the more difficult task with large variations in movement velocity, nonlinear regression techniques can improve decoding of movement trajectories.