Deep Learning-based Mobile Platform for Remote Functional Recovery Monitoring After Total Hip Arthroplasty

Ensuring functional recovery and early complication detection after Total Hip Arthroplasty (THA) requires reliable follow-up systems. Digital health technologies offer new opportunities to support patients remotely and improve clinical decision-making. This thesis presents My Mobility, a mobile client–server application designed to assist patients and orthopaedic clinicians during post-operative rehabilitation. Through continuous and multimodal telemonitoring, the system aims to reduce the incidence of early dislocation, one of the leading causes of THA revision surgery. The telemonitoring process integrates three complementary data sources: videos recording the patient’s weekly performance in hip mobility exercises, quantitative gait metrics extracted via smartphone integration with Google Fit, and responses to a validated recovery questionnaire. These data allow the orthopaedist to track progress remotely and tailor the rehabilitation pathway accordingly. Meanwhile, patients benefit from clear guidance, reduced need for in-person follow-ups, and a stronger sense of medical supervision. The exercise assessment is performed involving YOLOv8 Pose, a lightweight deep learning model capable of extracting anatomical keypoints from video frames. The system supports two targeted rehabilitation exercises: hip flexion–extension and abduction–adduction. To determine which exercise is being performed, the algorithm evaluates the patient’s orientation by computing the ratio between the horizontal and vertical distances of the shoulder keypoints: a predominance of horizontal distance indicates a frontal view, associated with abduction–adduction, while a greater vertical component suggests a lateral view, corresponding to flexion–extension. Once the exercise type is identified, the system computes ROM using anatomically relevant keypoints on the shoulder, hip, and knee. A dedicated Python watcher monitors uploads on Google Drive and processes new videos without user intervention. The proposed solution is based on a dual-client architecture, with distinct interfaces for patients and clinicians. This work proposes an AI-powered telemonitoring framework to enhance postoperative care, enabling continuous assessment, personalized rehabilitation, and reduced dependence on in-person follow-up.