Predicting rehabilitation outcomes of unilateral stroke after brain-computer interface training based on magnetic resonance imaging data

Stroke remains a significant cause of disability globally, with a noticeable prevalence in China. Post-stroke rehabilitation, particularly through brain-computer interface (BCI) methods, plays a vital role in enhancing motor function recovery. However, the efficacy of BCI rehabilitation might be hindered by challenges in individualized program of prognosis prediction. This study aimed to develop prognostic prediction models for unilateral hemiplegia after BCI rehabilitation, utilizing both clinical and functional magnetic resonance imaging (fMRI) data, in order to enhance treatment efficiency and optimize patient outcomes. The study included 40 stroke patients (22 left hemisphere affected and 18 right hemisphere affected) who underwent BCI rehabilitation training at the Beijing Tsinghua Changgung Hospital (Beijing, China). Data related to patients' demographics, disease duration, and assessment scores were collected. Based on the improvement in the Fugl-Meyer assessment of the upper extremity (FMA-UE) rating scale, patients were categorized into responder and non-responder groups. Linear regression and its variants, including multivariate logistic regression and optimal subset regression, were utilized to predict the post-treatment scores based on both fMRI and clinical data. The accuracy and R-squared value of the models were assessed using leave-one-out cross-validation (LOOCV). The linear regression model using imaging data exhibited a remarkable performance with a classification accuracy of 100% and R2 (LOOCV) exceeding 0.94. In contrast, the model relying solely on clinical data achieved a classification accuracy of