Multimodal Behavioral Data Predict Stroke Patient's Response to BCI Treatment Through Explainable AI

Area of Science:

• Neuroscience
• Artificial Intelligence
• Rehabilitation Medicine

Background:

• Brain-computer interface (BCI) neurorehabilitation shows potential for stroke motor recovery but yields varied patient responses.
• Identifying predictors of BCI therapy success is crucial for optimizing treatment strategies.

Purpose of the Study:

• To investigate the predictive power of multimodal behavioral data for BCI therapy response in subacute stroke patients.
• To utilize explainable artificial intelligence (XAI) methods to understand the contribution of different data types.

Main Methods:

• Forty-two subacute stroke patients with lower-limb impairment received BCI training and underwent behavioral assessments.
• Machine learning models (linear regression, elastic net, artificial neural network) were developed using multimodal data.
• XAI techniques (stepwise method, Shapley additive explanation) were employed for model interpretation.

Main Results:

• A multivariate model combining optimal behavioral data (R²=0.852) significantly outperformed a univariate model (R²=0.758).
• Elastic net and artificial neural network models showed high prediction performance (accuracies 0.810 and 0.762, AUCs 0.782 and 0.771).
• Baseline motor impairment, muscle spasticity, and balance function emerged as primary predictors of BCI response.

Conclusions:

• Multimodal behavioral data offer unique insights into predicting stroke patient response to BCI therapy.
• XAI methods are vital for precision medicine, enabling clinicians to tailor neurorehabilitation strategies.
• Identifying individual recovery potential through data-driven approaches can optimize treatment planning.