Multimodal Neural and Behavioral Data Predict Response to Rehabilitation in Chronic Poststroke Aphasia.

Area of Science:

• Neuroscience
• Medical Imaging
• Machine Learning

Background:

• Poststroke recovery and response to language rehabilitation vary significantly among individuals.
• Identifying predictors of rehabilitation outcomes is crucial for personalized treatment strategies.

Purpose of the Study:

• To investigate the prognostic role of patient-related factors in predicting response to language rehabilitation after stroke.
• To evaluate the independent and complementary predictive power of behavioral, demographic, and neuroimaging data using machine learning.

Main Methods:

• Fifty-five individuals with chronic poststroke aphasia were assessed using standardized measures and MRI.
• Support vector machine and random forest models were developed to predict treatment responsiveness.
• Models utilized pretreatment behavioral, demographic, and neuroimaging data (structural and functional MRI).

Main Results:

• A support vector machine model using aphasia severity, demographics, and resting-state functional connectivity achieved the highest prediction performance (F1=0.94).
• This multimodal model significantly outperformed models using all features (F1=0.82) or single feature sets (F1 range=0.68-0.84).
• Random forest models showed that resting-state functional connectivity data yielded the best prediction score (F1=0.87).

Conclusions:

• Behavioral, demographic, and multimodal neuroimaging data offer complementary information for predicting rehabilitation response in poststroke aphasia.
• Resting-state functional connectivity is a particularly important predictor of treatment responsiveness, both independently and in combination with other factors.