Development of a AI-guided Telerehabilitation Platform for Post-stroke PT

Thesis Abstract

Post-stroke rehabilitation faces substantial barriers including limited access to timely therapy, transportation challenges, and inconsistent adherence to conventional in-clinic programs, leading to suboptimal functional recovery. This study addresses the problem by developing and evaluating an AI-guided telerehabilitation platform designed to deliver individualized, remote post-stroke physical therapy, monitor progress, and enhance engagement through real-time feedback. The aim is to determine whether the platform improves motor outcomes, adherence, and patient engagement compared with standard in-person care. The specific objectives are to (i) design an AI-driven decision-support module that recommends personalized exercise regimens based on baseline impairment, progress data, and biometrics; (ii) implement remote sensing and computer vision components to assess upper and lower limb function accurately in home environments; (iii) evaluate usability, accessibility, and adherence among stroke survivors and caregivers; (iv) examine the platform’s effectiveness on motor impairment, activities of daily living, and quality of life; and (v) identify ethical, legal, and data security considerations pertinent to telerehabilitation. A mixed-methods research design will be employed. The quantitative strand will use a randomized controlled trial with 120 post-stroke participants (6 weeks to 6 months post-stroke) assigned to either AI-guided telerehabilitation or standard in-clinic therapy for 12 weeks, with follow-up at 24 weeks. Primary outcomes will include changes in Fugl-Meyer Assessment (FM-UE) for upper-extremity function and the 6-Minute Walk Test (6MWT) for mobility, analyzed via repeated-measures ANOVA and regression analyses to identify predictors of improvement. Secondary outcomes include the Barthel Index, Stroke Impact Scale, adherence rates, and device satisfaction, analyzed with multivariate ANOVA and hierarchical linear modeling to account for clustering by site and caregiver involvement. The qualitative strand will involve semi-structured interviews with 30 participants (and 15 caregivers) post-intervention, analyzed using thematic analysis to elucidate user experience, barriers to use, and perceived value. The theoretical framework will integrate the Technology Acceptance Model (TAM) to explain adoption and the Self-Determination Theory (SDT) to interpret motivation and adherence, with the AI components grounded in reinforcement learning to optimize therapy dosing over time. Data collection will utilize (i) wearable sensors and radar-based motion capture integrated within a consumer-grade smartphone app to quantify range of motion, speed, and symmetry; (ii) clinician-rated scales administered remotely by trained therapists via secure teleconsultations; (iii) in-app analytics on usage patterns, completion rates, and response to AI-generated prompts; and (iv) validated questionnaires for quality of life, fatigue, and mental health. Instrument validity and reliability will be established through pilot testing (n=15) and Cronbach’s alpha analysis for multi-item scales. Missing data will be addressed using multiple imputation, and sensitivity analyses will assess the robustness of findings. Expected findings anticipate that the AI-guided telerehabilitation group will show clinically meaningful greater gains on FM-UE and 6MWT at 12 weeks, with sustained effects at 24 weeks, compared with standard care. Higher adherence, stronger intrinsic motivation, and greater user satisfaction are anticipated in the AI-supported group, moderated by caregiver engagement and digital literacy. The study is expected to reveal that AI-driven personalization reduces therapy non-compliance and improves functional independence, while maintaining data security and ethical integrity via encryption, role-based access controls, and informed consent aligned with GDPR/local equivalents. The study contributes to knowledge by integrating AI-based decision support with telerehabilitation for post-stroke care, providing empirical evidence on motor and functional outcomes, adherence determinants, and user experience in home-based therapy. It informs best practices for designing scalable, equitable, and secure digital rehabilitation solutions that can be deployed across diverse healthcare settings. The main conclusion is that AI-guided telerehabilitation can augment post-stroke recovery by delivering personalized, accessible, and engaging therapy, with implications for policy, clinical guidelines, and future research on adaptive rehabilitation ecosystems. Recommendations include refining AI dosing algorithms with larger, more diverse populations; exploring cost-effectiveness analyses; and developing standardized protocols for remote assessment to complement in-clinic care.