An Upper-Limb Motor Imagery EEG Dataset of Chronic Stroke Patients

Motor imagery (MI)-based brain-computer interface (BCI) systems offer a promising approach for post-stroke motor rehabilitation. However, their clinical translation is limited by the scarcity of large, clinically relevant electroencephalography (EEG) datasets. This study presents the HS Stroke dataset, consisting of 57,902 left- and right-hand MI EEG trials collected across 278 sessions from 14 chronic stroke patients, along with comprehensive clinical assessments. Under the cross-trial evaluation setting, validation results show that state-of-the-art deep learning models achieve up to 82.65% accuracy in MI classification, confirming the quality and discriminability of the collected EEG signals. Regression analyses further suggest that MI-related EEG features are predictive of individual Fugl-Meyer Assessment scores, highlighting their potential as non-invasive markers of motor recovery. The HS Stroke dataset is expected to support the development of MI-BCI decoding methods and facilitate research in post-stroke rehabilitation.