Methods
We conducted a PRISMA-compliant systematic review of PubMed, IEEE Xplore, and Scopus (2015-2024). Included studies evaluated AI-driven interventions (machine learning/computer vision) for home-based upper limb rehabilitation, reported adherence metrics (completion rates, engagement) or clinical outcomes (FMA, WMFT), and involved remote monitoring. Two reviewers independently screened studies, extracted data, and assessed bias using ROB-2. Meta-analysis was performed where feasible.
Results
From 1,420 records, 29 studies (n=1,832) met criteria. AI systems improved therapy completion rates by 18-35% versus conventional programs (p<0.01), with 82% of studies reporting ?80% adherence. Gamified AI increased exercise duration by 42% (p=0.003). Clinically, AI-guided therapy showed significant FMA-UE improvements (?=4.1 points) and reduced WMFT times (?=-9.7 sec, p=0.02). Wearable sensors (68% IMUs) and video-based AI (24%) were most common, though technical literacy (31%) and data security (19%) posed challenges.
Conclusion
AI significantly enhances adherence and functional outcomes in home-based upper limb rehabilitation. Future implementation should prioritize user-friendly design, robust security, and cost-effectiveness studies to support widespread clinical adoption. These findings advocate for integrating AI into standard post-stroke care to overcome current adherence barriers.
