Predicting clinically significant motor function improvement after contemporary task-oriented interventions using...

Area of Science:

• Neuroscience
• Rehabilitation Medicine
• Artificial Intelligence

Background:

• Accurate prediction of motor recovery post-stroke is crucial for effective treatment planning.
• Machine learning (ML) shows promise for outcome prediction due to its accuracy and data processing capabilities.
• ML's effectiveness in predicting rehabilitation outcomes for chronic stroke patients undergoing contemporary interventions is under-explored.

Purpose of the Study:

• To evaluate the accuracy and performance of ML models in predicting motor function improvements in chronic stroke patients after task-oriented interventions.
• To identify key predictors for developing robust ML-based prediction models for stroke recovery.

Main Methods:

• Secondary analysis of data from 239 chronic stroke patients who received task-oriented training (e.g., constraint-induced movement therapy, bilateral arm training).
• Utilized k-nearest neighbor (KNN) and artificial neural network (ANN) ML approaches.
• Models were trained and validated using training and test datasets with cross-validation, assessing outcomes via the Fugl-Meyer Assessment scale (FMA).

Main Results:

• Identified time since stroke, baseline Functional Independence Measure (FIM), and baseline FMA scores as significant predictors.
• The KNN model achieved 85.42% prediction accuracy and an AUC-ROC of 0.89.
• The ANN model demonstrated 81.25% prediction accuracy and an AUC-ROC of 0.77.

Conclusions:

• Integrating ML with key predictors like time since stroke and baseline function can help identify patients likely to benefit from task-oriented interventions.
• KNN and ANN models show potential utility in predicting clinically significant motor recovery in chronic stroke survivors.