Smartphone-based AI platform for home-based neurorehabilitation monitoring

Thesis Abstract

This study addresses the persistent gap in continuous, accessible neurorehabilitation following stroke by evaluating a smartphone-based AI platform designed for home-based monitoring and personalized intervention delivery. The problem centers on limited access to regular therapy, variable adherence, and the need for objective, scalable metrics to guide remote rehabilitation. The aim is to assess the feasibility, effectiveness, and user experience of an AI-enabled mobile platform that integrates multimodal sensor data (accelerometer, gyroscope, touch, and camera-based motion analysis) with machine learning-driven feedback for home neurorehabilitation. Specific objectives include (1) evaluating changes in motor function using standardized measures (Fugl-Meyer Assessment for Upper Extremity, Box and Block Test) over a 12-week period, (2) examining improvement in functional independence as captured by the Functional Independence Measure, (3) assessing platform adherence, engagement, and user satisfaction, (4) identifying predictors of adherence and outcomes via regression analyses, and (5) exploring clinicians’ and patients’ perceptions of usability and safety through qualitative interviews. The method adopts a mixed-methods design with longitudinal quasi-experimental elements. The population comprises adults aged 40–75 years in the subacute to chronic phase post-stroke, recruited from three urban rehabilitation centers. A target sample of 180 participants is planned, with 120 completing the 12-week program, determined through a power analysis (? = 0.80, ? = 0.05) to detect a medium effect size (Cohen’s d = 0.5) on motor outcomes. Participants are allocated to two groups an intervention arm using the smartphone AI platform plus standard care, and a control arm receiving standard care alone. Data collection instruments include wearable sensor data (actigraphy-derived activity counts; kinematic metrics from the smartphone’s camera-based pose estimation), clinical assessments (FMA-UE, BBT, FIM), the User Engagement Qualitative Survey, and the System Usability Scale. The platform’s AI components employ supervised learning models (random forest and gradient boosting) for real-time motion analysis, anomaly detection, and adaptive exercise prescription, informed by established motor learning theories such as Fitts’ Law and the Challenge Point Framework. Qualitative data are gathered via semi-structured interviews with 25–30 participants and 10 clinicians, analyzed using thematic analysis framed by the Technology Acceptance Model. Quantitative analyses include descriptive statistics, repeated-measures ANOVA to examine within- and between-group changes across time, multivariate linear regression to identify predictors of motor improvement and adherence, and time-series analyses of sensor-derived metrics to characterize recovery trajectories. Model validation includes cross-validation for predictive accuracy and receiver operating characteristic analysis to determine adherence thresholds. Qualitative data are analyzed inductively, with coding conducted by two independent researchers and triangulated with quantitative findings to yield a comprehensive interpretation of usability, acceptability, and perceived safety. Missing data are addressed using multiple imputation under the assumption of missing at random. Expected findings include statistically significant greater gains in FMA-UE and BBT scores in the intervention group versus control, improved FIM performance, higher adherence rates (?75% weekly active days), and positive user satisfaction with perceived usefulness and ease of use. Sensor-derived metrics are anticipated to correlate with clinical improvements, supporting the platform’s validity as an objective monitoring tool. Regression analyses are expected to identify factors such as prior technology familiarity, social support, and initial impairment level as predictors of adherence and outcome magnitude. Qualitative insights are projected to reveal themes around autonomy, confidence in AI guidance, perceived safety, and the impact of remote monitoring on motivation and therapy consistency. The study contributes to knowledge by providing empirical evidence on the efficacy and feasibility of AI-enhanced mobile rehabilitation platforms, integrating objective digital biomarkers with conventional clinical outcomes, and offering a framework for scalable, remote neurorehabilitation. It informs best practices for design, validation, and deployment of home-based rehabilitation technologies, including considerations for data privacy, clinician workflow integration, and equitable access. The main conclusion anticipates that the smartphone-based AI platform can safely augment standard care, improve motor outcomes, and enhance adherence, with recommendations emphasizing iterative AI personalization, user-centered design, and broader multicenter trials to confirm generalizability across diverse populations and healthcare settings.