Globally, knee lesions, including anterior cruciate ligament (ACL), menisci, cartilage, and osteoarthritis, are a major clinical burden. Clinical examination and medical imaging are key diagnostic methods, but they are inaccurate, inefficient, and rarely reproducible. Musculoskeletal imaging diagnostic precision and workflows may improve with artificial intelligence (AI) and machine learning (ML) technologies. To comprehensively review AI and ML applications for knee lesion diagnosis, compare their diagnostic performance to conventional methods, and identify key challenges and future clinical directions, a systematic literature review was conducted on PubMed, Scopus, and IEEE Xplore for studies from 2015 to the present. Keywords included "artificial intelligence", "machine learning", "deep learning", "knee", "MRI", "diagnosis", and "lesion detection". Performance metrics, methodological rigor, and relevance to AI/ML knee lesion diagnosis were considered when selecting studies. Analyses included peer-reviewed articles and regulatory guidance. There were 14 AI/ML knee lesion diagnosis studies in the review. Convolutional neural networks (CNNs) were the most popular architecture, with ResNet, VGG, DenseNet, and custom architectures performing well. Deep learning models were accurate diagnostically. AI-assisted diagnosis excelled at ACL tear, osteoarthritis grading, and meniscal tear detection. AI systems outperformed radiologists in diagnostic accuracy, interpretation time, and inter-reader agreement. In many applications, AI and ML technologies can diagnose knee lesions with human-level accuracy. However, dataset diversity, model generalizability, regulatory approval, and clinical integration remain issues. Clinical deployment requires ongoing research on standardized protocols, diverse training datasets, and real-world validation.
Velitsikakis et al. (Sun,) studied this question.