Machine Learning Methods Predict Individual Upper-Limb Motor Impairment Following Therapy in Chronic Stroke.

Area of Science:

• Neuroscience
• Rehabilitation Medicine
• Artificial Intelligence

Background:

• Predicting upper-extremity motor function recovery in chronic stroke patients is crucial for effective therapy.
• Machine learning (ML) offers a promising approach to enhance prediction accuracy in clinical practice.

Purpose of the Study:

• To evaluate five ML methods for predicting post-intervention upper-extremity motor impairment in chronic stroke survivors.
• To identify key input variables for accurate prediction models.

Main Methods:

• 102 chronic stroke patients assessed using the upper-extremity Fugl-Meyer Assessment (UE-FMA) pre- and post-intervention.
• Five ML methods (Elastic Net, SVM, ANN, CART, Random Forest) applied to predict post-intervention UE-FMA.
• Model performance compared using cross-validated R².

Main Results:

• Elastic Net (EN) demonstrated superior performance using demographic and clinical data.
• Pre-intervention UE-FMA and interhemispheric motor threshold (MT) difference were the strongest predictors.
• The MT difference was more significant than motor-evoked potential (MEP) presence/absence.

Conclusions:

• ML methods show potential for accurately predicting post-intervention UE-FMA in chronic stroke.
• Interhemispheric MT difference is a vital predictor of patient response to therapy and warrants inclusion in future research.