Integrating AI into Stroke Physical Therapy: A Narrative Review of Evidence and Emerging Trends

Abstract

Background: Artificial Intelligence (AI) has rapidly emerged as a transformative force across healthcare, offering automation, predictive analytics, and personalized interventions. Stroke remains a leading cause of adult disability worldwide, with post-stroke survivors often experiencing persistent motor deficits despite conventional physical therapy. AI applications including robotics, machine learning, computer vision, and telerehabilitation have the potential to augment traditional rehabilitation, improve assessment accuracy, optimize therapy delivery, and expand access to care. However, evidence on AI’s clinical utility in stroke physical therapy is scattered across diverse domains, making it difficult for clinicians and policymakers to gain a coherent understanding of its scope, benefits, and limitations.

Objective: This scoping review aims to map and synthesize existing evidence on the applications, effectiveness, and challenges of AI in stroke physical therapy and rehabilitation, while identifying emerging trends and gaps to inform clinical practice, research, and policy makers.

Methods: This review followed Arksey and O’Malley’s five-stage framework (2005), refined by Levac et al. (2010), and was reported in accordance with PRISMA-ScR guidelines. A comprehensive search was conducted in PubMed/MEDLINE, PEDro, and Cochrane Library up to August 2025 using terms related to AI (e.g., machine learning, robotics, computer vision) and stroke rehabilitation. Eligible studies included quantitative, qualitative, mixed-methods research, and reviews involving post-stroke patients in physical therapy or rehabilitation contexts. Non-AI digital tools and purely technical studies without clinical relevance were excluded. Data were extracted for study design, AI applications (assessment, intervention, decision support), clinical outcomes, and barriers.

Results:
Out of 245 records, 14 studies published between 2020–2025 met inclusion criteria. Evidence demonstrated that AI enhances assessment and monitoring through computer vision, wearable sensors, and predictive modeling, enabling automated gait analysis, motion tracking, and recovery forecasting. In therapy delivery, AI-powered robotics (e.g., Lokomat, MIT-Manus), virtual reality integration, and telerehabilitation platforms improved motor function, adherence, and patient engagement, while reducing therapist workload. Decision support tools applied AI algorithms for outcome prediction, patient stratification, and treatment optimization, supporting more personalized rehabilitation. Reported advantages included personalization, objectivity, scalability, efficiency, and increased patient motivation. However, barriers such as limited high-quality datasets, privacy concerns, high financial costs, insufficient infrastructure in low- and middle-income countries, algorithmic bias, and limited clinician training hinder widespread adoption. Emerging trends highlight hybrid models combining therapist-guided and AI-driven therapy, explainable AI to improve trust, integration with brain-computer interfaces, and home-based wearable systems tailored for resource-limited settings.

Conclusion:
AI holds transformative potential in stroke rehabilitation by improving diagnostic accuracy, personalizing therapy, and extending access through robotics and digital platforms. While short-term studies report encouraging outcomes, large-scale clinical trials, cost-effectiveness analyses, and ethical evaluations are urgently needed. A hybrid rehabilitation model—integrating AI technologies with therapist expertise—appears most promising. Future research should prioritize patient-centered, context-specific approaches, particularly in low-resource settings, to ensure equitable, safe, and sustainable integration of AI into stroke physical therapy.