Abstract
BACKGROUND: Predicting the progression of hip osteoarthritis (OA) remains challenging, and no reliable predictive method has been established. This study aimed to develop an artificial intelligence (AI) model to predict hip OA progression via plain radiographs and patient data and to determine its accuracy. METHODS: This retrospective study utilized anteroposterior pelvic radiographs of consecutive patients with hip OA who underwent primary unilateral total hip arthroplasty. Radiographs diagnosed with Kellgren-Lawrence (KL) grade 0-2 were extracted from 361 patients and 1697 images. This AI model was developed to predict whether OA would progress from KL grade 0-2 to KL grade ≥ 3 within n years (n = 3, 4, 5). A gradient-boosting decision tree approach was utilized according to feature extractions obtained by a convolutional neural network from radiographs and patient data (height, body weight, sex, age, and KL grade given by an orthopedic surgeon) with five-fold cross-validation. The model performance was assessed using accuracy, specificity, sensitivity, and the area under the receiver operating characteristic curve (AUC). RESULTS: The mean accuracy, specificity, sensitivity, and AUC of our prediction model were, respectively, 81.8%, 88.0%, 66.7%, and 0.836 for 3 years; 79.8%, 85.0%, 71.6%, and 0.836 for 4 years; and 78.5%, 80.4%, 76.9%, and 0.846 for 5 years. CONCLUSIONS: The proposed AI model performed adequately in predicting hip OA progression and may be clinically applicable with additional datasets and validation.