Linear Dynamical Systems Modeling for EEG-Based Motor Imagery Brain-Computer Interface

Abstract

Motor imagery-based Brain Computer Interfaces (MI-BCI) has attracted more and more attention due to its effectivity for stroke and spinal cord injury patients’ rehabilitation. Common Spatial Pattern (CSP) and other spatio-spectral feature extraction methods become the most effective and principle successful solutions for MI-BCI pattern recognition in the recent few years. This paper applies Linear dynamical systems (LDS) referring to control field for EEG signals feature extraction and classification. Compared to other state-of-the-art methods, this model has lots of obvious advantages, such as simultaneous generation spatial and temporal feature matrix, without complex preprocessing or post-processing, ease of use, and low cost. A study is shown to program by computer and assess the performance of feature selection and classification algorithms for use with the LDS. Extensive experimental results are presented on public dataset from ‘BCI Competition III Data Sets IVa’. The results show that LDS, using Martin Distance and k-Nearest Neighbors classification algorithm, yields higher accuracies compared to prevailing approaches.