Personalized rehabilitation approach for reaching movement using reinforcement learning.

Area of Science:

• Rehabilitation Medicine
• Computer Science
• Biomedical Engineering

Background:

• Musculoskeletal disorders significantly impair daily activities and quality of life.
• Traditional physical therapy faces limitations in duration, accessibility, and patient motivation.
• Current virtual reality (VR) rehabilitation systems often lack personalization and clinical validation.

Purpose of the Study:

• To develop and validate an adaptive VR rehabilitation game for upper-limb reaching movements.
• To integrate a machine learning algorithm for patient-customized therapeutic interventions.
• To address the limitations of traditional therapy and current VR systems through personalization and flexibility.

Main Methods:

• Development of an immersive VR system featuring a bubble-popping game for reaching rehabilitation.
• Implementation of a Q-learning algorithm to adapt game difficulty based on patient kinematic data.
• Simulation of a 10-day training program to evaluate the algorithm's effectiveness.

Main Results:

• The Q-learning algorithm successfully adapted to simulated patient capabilities and kinematic variations.
• The system demonstrated responsiveness to performance changes and adherence to therapist-defined reward policies.
• Simulations validated the algorithm's ability to personalize the rehabilitation process effectively.

Conclusions:

• The proposed VR system offers adaptive capabilities and high flexibility for personalized upper-limb rehabilitation.
• The integration of reinforcement learning in VR shows promise for enhancing therapeutic interventions.
• Further clinical trials are needed to demonstrate the concept's efficacy in real-world patient populations.