Abstract
Meniscus tears can be difficult to diagnose on magnetic resonance imaging (MRI) due to the various types of tears. As artificial intelligence (AI) continues to advance, it could serve as a valuable tool that assists physicians with diagnostic accuracy and efficiency. The purpose of this study was to analyze the performance of AI in diagnosing meniscus tears on MRI using an array of studies and to compare the performance of AI to that of physicians. A literature search was conducted on PubMed and Embase for articles regarding the use of AI for the detection of meniscus tears. AI model, number of MRI studies, sensitivity, specificity, accuracy, AUC, and comparison to physicians encompassed the data that was extracted. A total of 18 studies, comprising 47,621 MRIs, were included in this review. The AUC of these AI models ranged from 0.781 to 0.984, averaging 0.912±0.053. The pooled sensitivity and specificity were 0.831±0.097 and 0.868±0.088, respectively. Three of these studies compared their results to those of 10 radiologists and two orthopedic surgeons. When comparing the sensitivity and specificity of AI models to those of physicians, Cochrane's Q was statistically significant (p < 0.01) with large heterogeneity amongst the two groups (I(2) = 84.72%). The results suggest that AI's ability to detect meniscus lesions on MRI was relatively strong. When this performance was compared to that of physicians, the results were comparable, highlighting the potential benefits in a clinical setting.