Artificial intelligence revolution in shoulder magnetic resonance imaging: current evidence and future directions for rotator cuff diagnosis

Abstract

Background: Rotator cuff tears (RCTs) are a leading cause of shoulder pain. Magnetic resonance imaging (MRI) is the gold standard for diagnosis, but interpretation is limited by variable sensitivity, interobserver variability, and increasing workload. Recent advances in artificial intelligence (AI), including deep learning and large language models (LLMs), have been explored to enhance diagnostic accuracy and efficiency. Methods: A structured search of PubMed, Embase, Scopus, and Cochrane Central databases identified studies published from 2019 or later applying AI to shoulder MRI for RCT detection, classification, segmentation, or reporting. Eligible studies reported quantitative outcomes. Of 732 records, 584 were screened, 121 underwent full-text review, and 19 were included in this narrative review. Results: Deep learning models demonstrated diagnostic accuracies ranging from 71% to 100%, with sensitivities of 73%-100% and specificities of 70%-100%. About 20% of included studies used external validation, with 5%-15% performance reductions, underscoring limited generalizability. Visual Geometry Group (VGG)-based architectures and convolutional neural networks (CNNs) using multiplanar inputs achieved higher performance for full-thickness tear detection, while radiomics contributed to assessment of tear severity and muscle quality. Regulatory-approved software remains limited, and no fully automated diagnostic system has received Food and Drug Administration (FDA) clearance. LLMs demonstrated potential in patient education and report drafting, though informational quality was variable (median DISCERN score, 40/80), requiring expert oversight. Conclusions: AI applications in shoulder MRI demonstrate promising diagnostic accuracy and potential workflow benefits, but current evidence is constrained by limited external validation, dataset heterogeneity, and lack of regulatory clearance. Future research should prioritize multicenter validation, clinical integration, and explainability to enable safe clinical use of AI.

Ähnliche Arbeiten