<- Back to digest

Arthroscopy - 2026-10-09 - Journal Article; Review

Artificial Intelligence Models Based on Medical Imaging Achieve Accurate Detection of Rotator Cuff Tear and Initial Predictive Capability for Postoperative Outcomes: A Systematic Review.

Zhan H, Liang Q, Zhao Z, Liao C, Yang Z, Kang X, Zheng J, Zhang L

systematic reviewLOE IIIn = 37 studies, 46,593 patientsN/A

Topics

sportsshoulder elbow
PMID: 42852970DOI: 10.1002/arj.70584View on PubMed ->

Key Takeaway

AI models detect rotator cuff tears on MRI with mean AUC 0.93 and accuracy 89.5%, matching senior clinician performance, while predicting postoperative retear at AUC 0.83 across 37 studies and 46,593 patients.

Summary Depth

Choose how much analysis to show on this article page.

Summary

This PRISMA-registered systematic review evaluated AI diagnostic and predictive performance for rotator cuff tears across PubMed, Embase, Web of Science, and Cochrane Library through July 2025. MRI-based AI achieved mean AUC 0.93 and accuracy 89.5% for tear detection; ultrasound-based models reached AUC 0.97 and accuracy 91%; radiograph-based models achieved AUC 0.86 and accuracy 84%. AI retear prediction after arthroscopic repair yielded mean AUC 0.83 and accuracy 83.4%, with older age and female sex as the most consistently identified risk factors.

Key Limitation

The review pools performance metrics across studies with heterogeneous AI architectures, training datasets, imaging protocols, and tear classification systems without formal meta-analytic synthesis, making the reported average AUC and accuracy figures unreliable for direct clinical benchmarking.

Original Abstract

PURPOSE

To assess the performance of artificial intelligence (AI) compared with physicians in detecting rotator cuff tear (RCT) on medical imaging and evaluating the ability of AI to predict outcomes after arthroscopic rotator cuff repair.

METHODS

A comprehensive search was conducted in the PubMed, Embase, Web of Science, and Cochrane Library databases from their inception to July 15, 2025. The criteria were as follows: (1) original studies employing AI to detect RCT or predict outcomes after arthroscopic rotator cuff repair and (2) case reports, reviews, and editorials were excluded. This review followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines and was registered on the International Prospective Register of Systematic Reviews.

RESULTS

A total of 37 studies met the inclusion and exclusion criteria, encompassing 46,593 patients. The mean age of the patients was 55.8 years, with an average female representation of 52.2%. AI models developed for diagnosing RCT using magnetic resonance imaging, ultrasound, and radiographic images showed excellent performance, achieving average area under the curves and accuracies of 89.5% and 0.93, 91% and 0.97, and 84% and 0.86, respectively. AI models using radiographs and magnetic resonance imaging have shown performance in diagnosing RCT that surpasses that of junior clinicians, with magnetic resonance imaging-based models performing at a level comparable to senior clinicians. AI models developed to predict retear after arthroscopic rotator cuff repair showed an average accuracy of 83.4% and an area under the curve of 0.83. Older and women were the most common risk factors for postoperative retear identified in the included studies.

CONCLUSIONS

AI models significantly improved diagnostic accuracy based on medical imaging and showed promising predictive performance, particularly regarding rotator cuff retear. However, predictions related to functional scores exhibited relatively lower accuracy.

LEVEL OF EVIDENCE

Level III, systematic review of Level II and III studies.