Performance of EEG Motor-Imagery based spatial filtering methods: A BCI study on Stroke patients

Abstract

Abstract The study reports the performance of stroke patients to operate Motor-Imagery based Brain-Computer Interface (MI-BCI) in early post-stroke neurorehabilitation and compares three different BCI spatial filtering techniques. The experiment was conducted on five stroke patients who performed a total of 15 MI-BCI sessions targeting paretic limbs. The EEG data were collected during the initial calibration phase of each session, and the individual BCI models were made by using Source Power Co-Modulation (SPoC), Spectrally weighted Common Spatial Patterns (SpecCSP), and Filter-Bank Common Spatial Patterns (FBCSP) BCI approaches. The accuracy of FBCSP was significantly higher than the accuracy of SPoC (85.1±1.9 % vs. 83.0±1.9 %; p=0.002), while the accuracy of FBCSP was slightly higher than the accuracy of SpecCSP (85.1±1.9 % vs. 83.8±2.0 %; p=0.068). No significant difference was found between SPoC and SpecCSP (p=0.616). The average false positive ratio was 16.9%, 17.1%, 14.3%, while the average false negative was 15.5 %, 16.9 %, 15.5 % for SpecCSP, SPoC, FBCSP, respectively. In conclusion, we demonstrated that the stroke patients were capable of controlling MI-BCI, with high accuracy and that FBCSP may be used as the MI-BCI approach for complementary neurorehabilitation during early stroke phases.