DEVELOPMENT AND VALIDATION OF A DEEP-LEARNING MODEL TO PREDICT TOTAL HIP REPLACEMENT ON RADIOGRAPHS: THE TOTAL HIP REPLACEMENT PREDICTION (THREP) MODEL

The methods for accurately assessing risk of total hip replacement (THR) in patients with osteoarthritis are scarce. A novel and reliable method should be investigated to play a significant role in research and clinical routine. Purpose: To develop a deep learning model that can reliably predict the risk of THR based on radiographic images and clinical symptoms. This retrospective, multi-center, case-control study assessed hip joints on weighted-bearing anterior-posterior pelvic radiographs from the Osteoarthritis Initiative (OAI) participants. Participants received THR were matched to controls according to age, sex, ethnicity, body mass index and ethnicity. Cases and controls were uniformly split at 72% ( n = 528), 14% ( n = 104), 14% ( n = 104) into training, validation, and testing data sets. Images and clinical symptoms were passed through a detection model and a deep convolutional neural network (DCNN) model to predict the probability of THR within 9 years and the most likely time period of THR (0-3 years, 3–6 years, 6–9 years). Model performance was assessed using area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, and specificity in the testing set. A total of 736 participants were evaluated, including 184 cases and 552 controls. The prediction model achieved an overall accuracy, sensitivity, and specificity of 91.35%, 92.59% and 86.96%, with an AUC of 0.944 within 9 years. Meanwhile, the AUC of the DCNN model for assessing the most likely time period was 0.907 for 0-3 years, 0.916 for 3–6 years, and 0.841 for 6–9 years. Gradient-weighted class activation mapping closely corresponded to regions affecting the prediction of the DCNN model. Conclusion: The proposed DCNN model is a reliable and valid method to predict the probability of THR within limitations. It could assist clinicians in patient counselling and decision-making about the timing of the intervention. In the future, by increasing the size of the dataset, enhancing the ethnic and socioeconomic diversity of the participants, and improving the follow-up rate, the quality of the conclusions can be further improved.