BCI-FES Rehabilitation Training Platform And Stroke Subjects’ Motor Imagery EEG Data Analysis

BrainComputerInterfacesprovideacommunicationsystembetweenhumanbrainand external devices. The goal, transmitting control commands and feedbacks frombraincortexwithoututilizinghumanneuralpathways,hasbeenaccomplishedbysignalrecording and processing techniques based on BCI.In this study, we design an efective rehabilitation training system based on BrainComputer Interface. Cooperating with the Huashan Hospital of Fudan University, weconduct three months’ experiment in clinical environment. During the experiment, wecollected stroke subject’s Motor Imagery data, and revealed the potential mechanismsby analyzing the Motor Imagery patterns. A complete rehabilitation training paradigmis proposed and evaluated, and the quantitative comparison results are provided for fur-ther research. Additionally, in this paper, the classical CSP-SVM scheme is improved,anda WeaklySupervisedLearning (WSL) algorithmbased on Gaussian Mixture Mod-el (GMM) is utilized for depicting stroke subjects’ EEG patterns. Compared to tradi-tional CSP, our proposed algorithm can better handle the EEG data produced by lesioncortexofstrokesubjectswithextremelylowSignal-to-NoiseRatio(SNR),thusachievea higher classifcation accuracy. Sufcient observations and test cases on patients’ MIdatasetshavebeenimplementedforvalidatingtheGMM-WSLmodel. Theresultsalsoreveal some working mechanisms and recovery appearances of impaired cortex duringthe rehabilitation training period.