A Random Tree Forest decision support system to personalize upper extremity robot-assisted rehabilitation in stroke:...

Area of Science:

• Neuroscience
• Rehabilitation Engineering
• Machine Learning in Healthcare

Background:

• Robotic rehabilitation with serious games offers customized therapy for individual patient needs.
• Current systems often lack long-term outcome data, hindering clinical decision-making during therapy setup.

Purpose of the Study:

• To develop and validate a predictive system for post-stroke rehabilitation outcomes.
• To utilize clinical scale scores and robotic measurements for predicting patient recovery at discharge.

Main Methods:

• A Random-Forest model was trained and tested using data from 25 post-stroke patients.
• Input features included clinical scale scores (FMA, ARAT, MI) and robotic system measurements at enrollment.
• The system predicted outcomes for the Fugl-Meyer Assessment (FMA), Action Research Arm Test (ARAT), and Motor Index (MI).

Main Results:

• The system achieved prediction accuracy ranging from 60% to 73% for the selected clinical scales.
• Key predictors identified included clinical scores (FMA, ARAT, MI, NRS, PCS, MCS) and robotic measurements of patient effort and time.
• The findings highlight the importance of both clinical assessments and objective robotic data for outcome prediction.

Conclusions:

• A Random-Forest based system can effectively predict rehabilitation outcomes in post-stroke patients undergoing robotic therapy.
• This predictive capability can serve as a valuable decision support tool for clinicians, enhancing therapy customization and management.
• Integrating robotic data with clinical assessments improves the prediction of therapeutic success.