Prediction of Joint angle from Muscle Activities decoded from Electrocorticograms in Primary Motor Cortex

摘要

Electrocorticography (ECoG) has drawn attention as an effective recording approach for less invasive brain-macnine interfaces (BMI). Previous studies succeeded m classitying the movement direction and predicting hand trajectories from ECoGs. Despite such successful studies, there still remain considerable works for the purpose of realizing an ECoG-based BMI robot. We developed a method to predict multiple muscle activities from ECoG measurements. We also verified that ECoG signals could be effective for predicting muscle activities in time varying series for preforming sequential movements. Each ECoG signal was filtered by different bandpass filters for sensorimotor rhythms, normalized by the standard z-score, and smoothed by a Gaussian filter. We used sparse linear regression to find the best fit between frequency bands of ECoG and electromyogram (EMG). We also predicted angle of 4 DOF robot arm from the decoded EMG using 3-layer neural network. Consequently, this study shows that it could derive online prediction of angle of robot arm from ECoG signals.