Adherence to prescribed physiotherapy is essential for successful treatment of rotator cuff pathology. Patients are typically responsible for performing the majority of physiotherapy independently at home, but methods for measuring adherence to home physiotherapy such as patient self-reporting lack objectivity and are prone to biases. Using state-of-the art machine learning (ML) techniques, our team has developed a Smart Physiotherapy Activity Recognition System (SPARS) for monitoring home shoulder physiotherapy adherence with inertial sensors in a commercial smart watch.

The objectives of this thesis research are to advance ML research in physiotherapy adherence monitoring using smart watch inertial data, validate SPARS and its algorithms with clinical patient data, and to develop and deploy an ethically-guided SPARS-powered rehabilitation program.

The newly developed SPARS technology was tested with a cohort of healthy volunteers, which allowed for the refinement of the technology prior to clinical use. An out-of-distribution (OOD) ML analysis was performed with the data collected, which advanced understanding of the use of ML algorithms for detecting activities of interest with human activity recognition (HAR) inertial data.

A non-interventional validation study was performed with rotator cuff pathology patients, wherein smart watch inertial sensor data was collected using the SPARS system during supervised physiotherapy and independent home exercise sessions. An ML analysis of data collected by patients in the home setting was performed, yielding insights into ML physiotherapy exercise detection as well as home exercise performance. An analysis is presented comparing objective SPARS-generated measures of at-home physiotherapy participation and adherence with clinical outcomes and patient factors.

Policy and ethics research guided the development of a SPARS-powered rehabilitation pilot program and SPARS-powered mobile apps providing real-time feedback on adherence, progress, and technique to both patients and treating physiotherapists. Output from this research and development work is presented, as well an analysis of data collected from a small patient cohort from this ongoing study.

This research has advanced the field of ML HAR with inertial data, generated clinical and policy insights, and yielded an objective tool for monitoring and feedback of home shoulder physiotherapy adherence to inform individualized patient care.