Development of an AI-Driven Tele-Assessment Platform for Acute Musculoskeletal Pain

Thesis Abstract

Acute musculoskeletal pain presents a substantial barrier to timely assessment and appropriate management, particularly in remote or underserved settings where access to in-person physiotherapy is limited and conventional triage may be inconsistent. This study develops and evaluates an AI-driven tele-assessment platform designed to standardize remote evaluation, triage risk stratification, and guided self-management for individuals with acute musculoskeletal complaints. The aim is to determine whether the platform improves diagnostic accuracy, reduces time to appropriate intervention, and enhances patient-reported outcomes within a 6-week follow-up window. Specific objectives are (i) to design an AI-powered decision-support module that integrates patient-reported symptoms, video-based functional tests, and wearable-derived movement data; (ii) to validate the platform against gold-standard clinical assessments in a cross-sectional cohort; (iii) to evaluate its impact on clinical decision-making, triage accuracy, and referral appropriateness in a randomized controlled feasibility study; (iv) to assess user experience, engagement, and adherence; and (v) to examine the platform’s cost-effectiveness relative to standard tele-physiotherapy pathways. A mixed-methods approach is employed. The methodological core comprises a two-phase design Phase I focuses on iterative development and validation of the AI models, while Phase II conducts a prospective feasibility trial. Phase I involves 150 participants presenting acute musculoskeletal pain at primary-care or emergency settings, with data collected on patient-reported outcome measures, standardized physical tests captured via smartphone video, and accelerometer data from wearable sensors. A multimodal neural network and Bayesian updating framework are trained to predict clinical diagnosis, red-flag risk, and recommended management. In Phase II, 180 participants are randomized to either the AI-driven tele-assessment platform or standard remote assessment by clinicians. Data collection instruments include the Numeric Pain Rating Scale, QuickDASH/Oswestry Disability Index, the CONSORT-compliant triage form, and a user experience questionnaire (System Usability Scale). Validity and reliability of instruments follow established psychometric properties, with test-retest reliability assessed for remote measures. Ethical approval is obtained, with informed consent, data anonymization, and robust data-security protocols. Data analysis comprises predictive validity and calibration metrics for the AI module, including sensitivity, specificity, area under the ROC curve, and Brier score. In Phase II, group comparisons utilize regression analyses to adjust for confounders, with ANOVA or ANCOVA applied to continuous outcomes such as pain intensity and functional scores. Time-to-intervention analyses employ Kaplan-Meier estimates and Cox proportional hazards models. Qualitative data from semi-structured interviews with participants and clinicians are analyzed thematically using Braun and Clarke’s approach to elucidate acceptability, perceived barriers, and facilitators. A cost-effectiveness analysis adopts a health-system perspective, calculating incremental cost-effectiveness ratios based on quality-adjusted life years (QALYs) derived from the EQ-5D-5L. Expected findings include improved diagnostic concordance between remote AI-assisted assessments and face-to-face evaluations, faster determination of appropriate care pathways, and higher adherence to prescribed home exercises due to tailored feedback. The AI module is anticipated to maintain robust predictive performance across diverse patient subgroups, with calibration indicating reliability in real-world settings. Qualitative insights are expected to reveal high acceptability among patients and clinicians, contingent on clear explanations of AI reasoning and transparent data governance. The study is poised to demonstrate that remote AI-driven assessments can reduce unnecessary imaging, streamline triage, and lower indirect costs through reduced travel and wait times. The study contributes to knowledge by integrating multimodal data fusion, explainable AI, and tele-rehabilitation within a unified platform for acute musculoskeletal pain, addressing gaps in scalable remote assessment, standardization of care, and economic evaluation of AI-assisted pathways. It informs clinical guidelines on when tele-assessment is sufficient versus when in-person evaluation remains essential and provides a framework for implementing AI-supported protocols in primary care and emergency settings. Recommendations include establishing standardized data-sharing agreements, continuous-model monitoring for drift, user-centered interface refinements, and broader multi-site trials to confirm generalizability. The overarching conclusion is that an AI-driven tele-assessment platform can enhance the accuracy, efficiency, and patient experience of acute musculoskeletal pain management when integrated with clinician oversight, with scalable implications for global health systems.