Ultrasound based diagnosis of musculoskeletal tendon pathology using artificial intelligence: Systematic review

Introduction Musculoskeletal (MSK) tendon pathologies contribute to global non-fatal disability, causing pain, functional limitation, and reduced quality of life. Early and accurate diagnosis is essential to guiding management and preventing progression to rupture. Ultrasound (US) is the preferred first-line imaging modality due to its non-invasive, real-time dynamic capabilities, excellent soft-tissue resolution, accessibility, and low cost, enabling correlation with clinical findings. However, US suffers from substantial inter-operator variability and requires intensive training. Artificial intelligence (AI) offers the potential to mitigate these limitations, but its diagnostic performance in MSK tendon pathology requires evaluation. This review aimed to assess the performance of AI models applied to US for diagnosing MSK tendon pathologies. Methods A systematic review was conducted in accordance with PRISMA guidelines. Inclusion criteria for articles searched from March 2014 to July 2025 were AI-assisted ultrasound diagnosis of MSK tendon pathologies in adults, using the keywords ‘tendon', ‘pathologies', ‘ultrasound', and either ‘machine learning' or ‘deep learning'. Exclusion criteria were articles other than in English, or non-peer-reviewed literature or scholarly commentary. Databases searched on 5th March 2024 were Embase, Google Scholar, Scopus, IEEE Xplore, PubMed, PLoS ONE, and Web of Science. The ‘Must Artificial Intelligence Criteria-10 (MAIC 10)' framework was used to assess risk of bias. Due to the methodological heterogeneity of included articles, the results were synthesised narratively. Results Of 335 screened papers, 10 studies met the inclusion criteria; five used deep learning and five used machine learning. Reported metrics included accuracy (87.9%–97.21%), sensitivity (82%–98.1%), specificity (79%–100%), recall (94%–95.4%), F1 score (0.95), and AUC (0.82–0.99). The main limitation was the absence of external validation across all studies. Conclusion Despite study heterogeneity, AI-based tools show promise for enhancing US diagnostic accuracy in MSK tendon pathology. Future research should prioritise methodological standardisation and robust external validation.