JOA - 2026-08-06 - Journal Article
Diagnostic Accuracy of Artificial Intelligence Systems in Early Prediction of Dislocation and Aseptic Loosening After Total Hip Arthroplasty.
Abul MS, Sevim ÖF, Kılıç NC, Agir M, Tuncay İ
Topics
Key Takeaway
The best-performing AI system achieved AUC 0.91 for dislocation and 0.90 for aseptic loosening prediction from first-day postoperative AP pelvic radiographs in primary THA at mean 11.6-year follow-up.
Summary Depth
Choose how much analysis to show on this article page.
Summary
This study asked whether AI systems could predict dislocation and aseptic loosening from immediate postoperative AP pelvic radiographs in 1,045 primary THA patients followed for a mean of 11.6 years. Three AI systems were compared; the top performer achieved AUC 0.91/0.90 for dislocation/aseptic loosening, with the lowest-performing system reaching only AUC 0.79/0.77. Over follow-up, 34 patients dislocated and 22 developed aseptic loosening, with 989 remaining complication-free.
Key Limitation
The AI systems are not named or described, and it is unclear whether the validation was performed on a held-out dataset or the same cohort used for model development, raising serious concern for overfitting and precluding clinical adoption.
Original Abstract
BACKGROUND
Artificial intelligence (AI) is increasingly used in total hip arthroplasty (THA), yet its role in postoperative imaging remains limited. This study evaluated the diagnostic performance of AI systems in interpreting postoperative radiographs following THA.
METHODS
A total of 1,045 patients who underwent primary THA between 2011 and 2015 were retrospectively analyzed. The first-day postoperative antero-posterior pelvic radiographs were assessed using AI systems to estimate the risk of dislocation and aseptic loosening. During a mean follow-up of 11.6 years (range, 10 to 14), 34 patients developed dislocation, 22 developed aseptic loosening, and 989 patients remained free of complications. Actual complications were confirmed through clinical and radiological records. Sensitivity, specificity, and area under the curve (AUC) values were calculated and compared.
RESULTS
There was one AI system that demonstrated the highest diagnostic accuracy (AUC 0.91 for dislocation and 0.90 for aseptic loosening), followed by the other systems (0.84 and 0.82) and (0.79 and 0.77), respectively. Repeated analyses produced identical outputs across sessions, confirming reproducibility.
CONCLUSION
The AI-based analyses can accurately and consistently evaluate postoperative THA radiographs. There was one system that showed the best balance between sensitivity and specificity, indicating that AI-assisted interpretation may serve as a reliable adjunct to clinical follow-up for early complication prediction.