Design of Magnetoencephalography-based Brain–Machine Interface Control Methodology Through Time-Varying Cortical Neural Connectivity and Extreme Learning Machine

Abstract

Introduction: Human-machine interfaces (HMIs) can improve the quality of life for physically disabled users. This study proposes a noninvasive BMI design methodology to control a robot arm using MEG signals acquired during the user's imagined wrist movements in four directions. Methods: The BMI uses the partial directed coherence measure and a time-varying multivariate adaptive autoregressive model to extract task-dependent features for mental task discrimination. An extreme learning machine is used to generate a model with the extracted features, which is used to control the robot arm for rehabilitation or assistance tasks for motor-impaired individuals. Results: The classification results show that the proposed BMI methodology is a feasible solution with good performance and fast learning speed. Discussion: The proposed BMI methodology is a promising solution for rehabilitation or assistance systems for motor-impaired individuals. The BMI provides satisfactory classification performance at a fast learning speed.