Diagnosing Developmental Dysplasia of the Hip Using Deep Transfer Learning

Developmental dysplasia of the hip (DDH) requires a timely and accurate diagnosis to prevent long‐term complications. Conventional assessment relies on manual measurement of the acetabular index from X‐rays, a labor‐intensive process subject to inter‐observer variability. To address these limitations, we propose AcetabulaVision, an automated method for DDH diagnosis. AcetabulaVision integrates a fine‐tuned YOLOv8 model for hip joint landmark detection with a new angle estimation mechanism to compute the acetabular index directly from X‐rays. A specialized dataset collected at Jordan University Hospital was used for training and evaluation. Experimental results show that AcetabulaVision achieves a mean absolute error of 4.76° in acetabular index prediction and a classification accuracy of 73.8% for DDH detection. The results demonstrate AcetabulaVision's potential as a consistent and efficient decision‐support tool in clinical practice.