Applying Combined Action Observation and Motor Imagery to Enhance Classification Performance in a Brain-Computer Interface System for Stroke Patients
Issued Date
2022-01-01
Resource Type
eISSN
21693536
Scopus ID
2-s2.0-85134208795
Journal Title
IEEE Access
Volume
10
Start Page
73145
End Page
73155
Rights Holder(s)
SCOPUS
Bibliographic Citation
IEEE Access Vol.10 (2022) , 73145-73155
Suggested Citation
Rungsirisilp N., Wongsawat Y. Applying Combined Action Observation and Motor Imagery to Enhance Classification Performance in a Brain-Computer Interface System for Stroke Patients. IEEE Access Vol.10 (2022) , 73145-73155. 73155. doi:10.1109/ACCESS.2022.3190798 Retrieved from: https://repository.li.mahidol.ac.th/handle/20.500.14594/84630
Title
Applying Combined Action Observation and Motor Imagery to Enhance Classification Performance in a Brain-Computer Interface System for Stroke Patients
Author(s)
Author's Affiliation
Other Contributor(s)
Abstract
Motor imagery (MI) and action observation (AO) are mental practices commonly applied in brain-computer interface (BCI) systems for stroke rehabilitation. However, previous studies have reported that combined AO and MI (AOMI) is more effective than MI or AO alone in terms of enhanced event-related desynchronization (ERD), which expresses cortical excitability and improves the classification performance of the BCI system in healthy subjects. Nonetheless, evidence the use of this strategy in stroke patients is still lacking. Hence, this study aimed to investigate the effect of AOMI on ERD and classification performance in chronic stroke patients. Ten chronic stroke participants were recruited for this study. Each participant was asked to perform both MI (control condition) and AOMI (experimental condition) tasks. For the MI task, the participants requested to perform MI while gazing at a static arrow picture. For the AOMI task, the participants were given a video-guided movement while executing MI. An array of 16 Ag/AgCl electrodes were used to record electroencephalographic (EEG) data during the mental tasks to analyze ERD amplitudes. Common spatial patterns (CSPs) combined with support vector machines (SVMs) were employed to evaluate the classification performance (offline analysis) of the baseline and imagery classes under each condition. Our results indicated that the ERD values and classification accuracy in AOMI were significantly greater than those under MI conditions. Moreover, a significant negative correlation between ERD values and classification performance was also found. In other words, enhanced ERD values (more negative values) also increased classification performance.