Artificial intelligence empowering the diagnosis and treatment of rotator cuff tears: A systematic review from image optimization to prognosis prediction.
Abstract
Background
Artificial intelligence (AI) is being developed for image acquisition, quantitative tissue assessment, diagnosis, and clinical prediction in rotator cuff tears. The breadth of this literature and its methodological limitations have not been integrated across the complete care pathway.
Methods
We searched PubMed, Embase, Web of Science, Scopus, and the Cochrane Library from inception to July 5, 2026 for English-language original studies applying AI to rotator cuff-related reconstruction, segmentation or quantitative imaging, diagnosis or classification, anatomical measurement, or prognosis. Two reviewers screened reports and extracted study, data, model, validation, reference standard, and performance characteristics. Diagnostic accuracy studies were assessed with QUADAS-2 and prediction-model studies with PROBAST. Because clinical tasks, modalities, models, reference standards, and reported metrics were not sufficiently comparable, evidence was synthesized narratively rather than pooled statistically.
Results
Sixty-six reports were included: 35 diagnostic accuracy or classification studies, 20 segmentation or quantitative imaging studies, 9 prognosis or outcome prediction studies, 1 Magnetic Resonance Imaging reconstruction or image-optimization study, and 1 exploratory unsupervised-phenotyping study. Applications spanned Magnetic Resonance Imaging, radiography, ultrasound, computed tomography, clinical or multimodal data, motion video, and emerging sensing data. QUADAS-2 classified overall risk-of-bias as high in 27/36 diagnostic studies, unclear in 6/36, and low in 3/36. PROBAST classified all 8/8 completed prediction-model studies as high risk-of-bias; 1 protocol without model results was not rated. Promising internal performance was recurrent, but selection methods, imperfect or incompletely described reference standards, limited external validation, small or selected datasets, and analysis-stage weaknesses constrained clinical inference.
Conclusion
AI has proof-of-concept value across rotator cuff imaging and decision support, but the current evidence does not establish routine clinical readiness or replacement of clinician judgment. Prospective multicenter studies, transparent reference standards, independent external validation, calibration and clinical-utility assessment, and standardized task-specific reporting are required.
