Methods
We conducted a PRISMA-compliant systematic review of PubMed, IEEE Xplore, and Cochrane Library (2015-2024), focusing on RCTs comparing AI-driven exoskeleton therapy with conventional rehabilitation. Included studies reported validated motor outcomes (Fugl-Meyer Assessment, Action Research Arm Test) and detailed AI implementation (adaptive algorithms, biofeedback systems). Two reviewers independently assessed study quality using PEDro scale and performed meta-analyses for primary outcomes.
Results
Analysis of 18 RCTs (n=892 participants) revealed AI-exoskeleton systems significantly improved FMA-UE scores (mean difference 6.2 points, 95% CI 4.1-8.3, p<0.001) compared to conventional therapy, with effect sizes largest in subacute stroke (Hedges' g=0.78). Active patient participation through AI-adjusted assistance increased training intensity by 38% (p=0.004) while reducing compensatory movements by 29% (p=0.01). Shoulder-elbow exoskeletons with EMG-based control showed particular benefits for proximal recovery (?FMA-proximal=4.9 vs 2.3 points distal, p=0.03). However, cost and setup complexity limited accessibility, with only 22% of studies reporting long-term (>6 month) follow-up data.
Conclusion
AI-powered exoskeletons demonstrate superior efficacy for upper limb motor recovery compared to conventional therapy, particularly when combining adaptive assistance with biofeedback. Future development should prioritize cost reduction, user-friendly interfaces, and standardized AI protocols to facilitate clinical implementation. These findings support integrating intelligent exoskeletons into rehabilitation programs for patients with moderate-to-severe impairment.
