Comparing Methods for Decoding Movement Trajectory from ECoG in Chronic Stroke Patients

Decoding the neural activity based on ECoG signals is widely used in the field of Brain-Computer Interfaces (BCIs) to predict movement trajectories or control a prosthetic device. However, there are only few reports of using ECoG in stroke patients. In this paper, we compare different methods for predicting contralateral movement trajectories from epidural ECoG signals recorded over the lesioned hemisphere in three chronic stroke patients. The results show that movement trajectories can be predicted with correlation coefficients ranging from 0.24 to 0.64. Depending on the intended application, either the use of Support Vector Regression (SVR) or Canonical Correlation Analysis (CCA) obtained the best results. By investigating how ECoG based decoding performs in comparison with EMG based decoding it becomes visible that abnormal muscle activation patterns affect the prediction and that using activity of only the forearm muscles, there is no significant difference between ECoG and EMG for predicting wrist movement trajectory.