Machine learning-based adaptive personalization in virtual reality stroke rehabilitation: a systematic review

Area of Science:

• Neurorehabilitation
• Virtual Reality Technology
• Machine Learning Applications

Background:

• Stroke is a leading cause of global disability, necessitating innovative therapeutic strategies.
• Virtual Reality (VR) is emerging as a promising tool in stroke rehabilitation.
• Existing VR systems often use rule-based difficulty adjustments, which may not fully address the complex, non-linear recovery process after stroke.

Purpose of the Study:

• To systematically review evidence on Machine Learning (ML)-based adaptive personalization in VR stroke rehabilitation.
• To analyze ML algorithms, adaptation strategies, clinical outcomes, and implementation considerations.
• To synthesize current research on the efficacy and safety of ML-driven VR for stroke recovery.

Main Methods:

• A systematic literature search was conducted across six databases (January 2015-December 2025) following PRISMA 2020 guidelines.
• Studies utilizing ML algorithms for VR in stroke rehabilitation were included.
• Data extraction covered ML algorithms, adaptation mechanisms, therapeutic parameters, clinical outcomes, and implementation factors; risk of bias was assessed, and meta-analysis was performed.

Main Results:

• Twenty-five studies were analyzed, with Reinforcement Learning (n=8) and Deep Learning (n=7) being prominent ML algorithms.
• Approximately 80% of studies employed real-time closed-loop adaptation.
• Meta-analysis revealed a significant improvement in upper limb motor function (FMA-UE) by 7.47 points (p<0.001), exceeding the minimal clinically important difference. Usability scores were high (SUS=72.5), with minimal adverse events.

Conclusions:

• ML-based adaptive VR rehabilitation is clinically effective and safe for stroke recovery.
• Significant improvements in motor function and positive usability highlight the potential of this approach.
• Implementation challenges in resource-limited settings warrant further attention for broader accessibility.