Detection of wear in hip prostheses using deep learning

This study presents a fully automatic, user‐independent measurement and reporting framework that determines the direction and magnitude of wear in total hip arthroplasty by comparing postoperative and follow-up radiographs. The system models the femoral head and acetabular cup with circular parameters; following YOLOv5-based region proposals, it runs edge-based analysis, Hough-circle detection, and RANSAC fitting in sequence to select the best circles without any manual annotation. This yields repeatable results despite variations in acquisition conditions and quantifies subtle changes in a standardized output format. The fully automated pipeline markedly reduces time cost and inter-observer variability. Pixel-based measurements can be converted to millimeters using magnification information, enabling reliable comparison across different examinations. Produced overlay visuals and numeric tables consistently compare two time points, clearly exposing directional wear patterns most notably superior and medial. In sum, the proposed system provides an early-warning, fully automatic, fast, and standardized follow-up tool using routine radiographs.