Artificial intelligence for postoperative plain radiographic assessment after reverse total shoulder arthroplasty: current evidence, clinical readiness, and limitations.
Authors
Affiliations (6)
Affiliations (6)
- Department of Orthopaedic Surgery, Konkuk University School of Medicine, Konkuk University Medical Center, Seoul, Korea.
- Department of Orthopaedic Surgery, School of Medicine, Kyungpook National University, Kyungpook National University Hospital, Daegu, Korea.
- Department of Radiology, School of Medicine, Kyungpook National University, Kyungpook National University Hospital, Daegu, Korea.
- Department of Orthopedic Surgery, Keimyung University Dongsan Hospital, Keimyung University School of Medicine, Daegu, Korea.
- Department of Orthopaedic Surgery, Daegu Catholic University College of Medicine, Daegu, Korea.
- Department of Orthopaedic Surgery, School of Medicine, Kyungpook National University, Kyungpook National University Hospital, Daegu, Korea. [email protected].
Abstract
Postoperative radiographic follow-up after reverse total shoulder arthroplasty (RTSA) plays an essential role in the early detection of implant-related complications and longitudinal risk stratification. However, interpretation of plain radiographs remains limited by interobserver variability and low sensitivity to subtle or progressive mechanical changes. This review was intended to survey the current evidence regarding artificial intelligence (AI) applications for postoperative plain radiograph-based assessment after RTSA, with a focus on clinical readiness, validated performance, and existing limitations. A narrative review was conducted with a specific focus on AI studies relating to shoulder arthroplasty and postoperative plain radiographic analysis. Applications outside shoulder arthroplasty or those based primarily on computed tomography were excluded. Among AI applications in RTSA imaging, implant identification and automated measurement of glenosphere orientation demonstrated the highest level of clinical readiness, with reproducible accuracy reported in multiple studies. In contrast, AI-based detection of scapular notching progression, component loosening, baseplate migration, and acromial or scapular spine stress reactions remains exploratory, with limited shoulder-specific validation. Across these domains, the principal barrier to clinical translation is not algorithmic capability but the lack of high-quality, longitudinally annotated RTSA radiographic datasets. Current AI applications in postoperative RTSA radiographs primarily serve to augment existing radiographic assessment, rather than replace established clinical interpretation. While select tasks are approaching clinical usability, broader adoption will require shoulder-specific longitudinal data, validated outcome-linked thresholds, and integration into routine clinical workflows.