Decoding Motor Imagery Movements using Area of 2-D Phase Space Reconstruction

Amyotrophic lateral sclerosis (ALS) is a common neurological disorder where voluntary muscle movements of the patient stop functioning and resulting paralyzed person. One of the solution of ALS is motor imagery (MI) based brain computer interface (BCI) which helps motor disabled patient to interact with the external world through their brain signal. But it has limited real-time applications due to its lower performance. The performance depends on feature extraction technique which extract relevant feature related to MI movements. The extraction of significant feature is challenging task in MI based BCI system. To improve the performance, this paper introduces an efficient feature extraction technique known as phase space reconstruction (PSR) for decoding various MI movements. First, filter bank technique was applied to MI signal and sets of sub-bands were generated. To study MI activities effectively, PSR was applied to each sub-band. The features (area of 2-D PSR pattern) of all sub-bands were combined and the significant features (p <0.05) were extracted using one-way analysis of variance (ANOVA). The significant features were fed into multi-class support vector machine (SVM) for decoding MI movements. The proposed method and classifier were tested on BCI competition 2008 dataset-II-a. The performance of proposed method was based on Cohen’s kappa coefficient (K). The results show that the SVM improved mean kappa coefficient (K=0.60) and outperformed existing methods presented in the literature.