Clinical validation of a deep learning tool for characterizing spinopelvic mobility in total hip arthroplasty

Radiographic measurement of patient-specific spinopelvic mobility is standard-of-care in total hip arthroplasty (THA). Current manual methods are time-consuming, subjective, and error-prone. This study utilized deep learning (DL) to classify functional positions and measure pelvic tilt (PT), sacral slope (SS) and lumbar lordotic angle (LLA). A DL pipeline integrating image classification, vertebra detection, and landmark detection was developed using data from an international joint registry, comprising 52,772 images for classification, 9,875 for object detection, and 25,249 for landmark detection. Performance was evaluated using area under the curve (AUC), F1 score, and mean absolute error (MAE). Accuracy was compared to annotations by three expert engineers and validated by two senior engineers and a surgeon. Radiographs were processed in 1.96 ± 0.04 s, achieving precision, recall, Receiver-operator-characteristic-AUC, and Precision-Recall-AUC metrics above 0.994. Anatomical landmark predictions resulted in errors of: PT: 1.6°±2.1°, SS: 3.3°±2.6°, LLA: 4.2°±3.2°. There was no significant difference in PT and LLA between expert engineers and the DL pipeline, and 0.5° difference in SS (p = 0.043). Clinical validation showed no difference in landmark rejection rates (p > 0.05). We developed and clinically validated a DL pipeline that accurately measures patient-specific spinopelvic mobility from lateral functional radiographs, providing a scalable method for routine characterization in THA patients.