Comparison between model-based RSA and an AI-based CT-RSA: an accuracy study of 30 patients

Abstract

Background and purpose: Radiostereometry (RSA) is the current gold standard for evaluating early implant migration. CT-based migration analysis is a promising method, with fewer handling requirements compared with RSA and no need for implanted bone-markers. We aimed to evaluate agreement between a new artificial intelligence (AI)-based CT-RSA and model-based RSA (MBRSA) in measuring migration of cup and stem in total hip arthroplasty (THA). Patients and methods: 30 patients with THA for primary osteoarthritis (OA) were included. RSA examinations were performed on the first postoperative day, and at 2 weeks, 3 months, 1, 2, and 5 years after surgery. A low-dose CT scan was done at 2 weeks and 5 years. The agreement between the migration results obtained from MBRSA and AI-based CT-RSA was assessed using Bland–Altman plots. Results: Stem migration (y-translation) between 2 weeks and 5 years, for the primary outcome measure, was –0.18 (95% confidence interval [CI] –0.31 to –0.05) mm with MBRSA and –0.36 (CI –0.53 to –0.19) mm with AI-based CT-RSA. Corresponding proximal migration of the cup (y-translation) was 0.06 (CI 0.02–0.09) mm and 0.02 (CI –0.01 to 0.05) mm, respectively. The mean difference for all stem and cup comparisons was within the range of MBRSA precision. The AI-based CT-RSA showed no intra- or interobserver variability. Conclusion: We found good agreement between the AI-based CT-RSA and MBRSA in measuring postoperative implant migration. AI-based CT-RSA ensures user independence and delivers consistent results.

Keywords

• AI
• Arthroplasty
• Artificial intelligence
• CT-based migration analysis
• CT-RSA
• Hip