Muir Duncan, Elgebaly Ahmed, Kim Woo Jae, Althaher Ahmad, Narvani Ali, Imam Mohamed A
Royal Berkshire Hospitals NHS Trust, Reading, UK.
Smart Health Centre, University of East London, London, UK.
Eur J Orthop Surg Traumatol. 2025 Feb 22;35(1):73. doi: 10.1007/s00590-025-04197-5.
Tendinopathy, a degenerative condition of tendon collagen protein, is a common sports injury among elite athletes. Despite its prevalence, the manifestation and progression of tendinopathy remain unclear, and the efficiency of diagnosis and treatment modalities is uncertain. Artificial intelligence and machine learning (ML) have shown positive results in disease diagnosis and treatment evaluation. This systematic review examined many ML methods and their diagnostic yield in predicting tendinopathy.
A comprehensive search of electronic databases, including Ovid Medline, EMBASE, PubMed and the Web of Science, was conducted. The quality of the studies was assessed using the Newcastle-Ottawa scale. The statistical analysis was performed using mada package on R software.
Four studies were considered eligible for this meta-analysis, constituting outcomes from 12,611 patients. The ML methods used in the selected studies included random forest, convolutional neural networks and linear support vector machines. The results showed that all selected studies demonstrated the relevance of ML in moderately (Reviewer 2, Comment 2) predicting tendinopathy. The pooled diagnostic yield of the ML algorithms estimated an overall sensitivity of 0.74 (95% CI: 0.64 to 0.82, p = < 0.001) and an overall specificity of 0.69 (95% CI: 0.49 to 0.85, p = 0.06). The diagnostic odds ratio was 6.01 (95% CI: 1.8 to 20.13), with substantial heterogeneity (I = 97.6%).
ML methods can predict tendinopathy accurately in elite and non-elite athletes. However, further research is needed to establish the specific clinical features associated with tendinopathy prevalence.
肌腱病是一种肌腱胶原蛋白的退行性疾病，是精英运动员中常见的运动损伤。尽管其发病率很高，但肌腱病的表现和进展仍不清楚，诊断和治疗方式的有效性也不确定。人工智能和机器学习在疾病诊断和治疗评估方面已显示出积极成果。本系统评价研究了多种机器学习方法及其在预测肌腱病方面的诊断效能。
对包括Ovid Medline、EMBASE、PubMed和Web of Science在内的电子数据库进行全面检索。使用纽卡斯尔-渥太华量表评估研究质量。使用R软件中的mada包进行统计分析。
四项研究被认为符合本荟萃分析的条件，涵盖了12,611名患者的结果。所选研究中使用的机器学习方法包括随机森林、卷积神经网络和线性支持向量机。结果表明，所有所选研究均证明了机器学习在中度（审稿人2，评论2）预测肌腱病方面的相关性。机器学习算法的合并诊断效能估计总体敏感性为0.74（95%CI：0.64至0.82，p = <0.001），总体特异性为0.69（95%CI：0.49至0.85，p = 0.06）。诊断比值比为6.01（95%CI：1.8至20.13），存在显著异质性（I = 97.6%）。
机器学习方法可以准确预测精英和非精英运动员的肌腱病。然而，需要进一步研究以确定与肌腱病患病率相关的具体临床特征。
