Machine Learning to Recognise ACL Tears: A Systematic Review

Machine learning-based tools are becoming increasingly popular in clinical practice. They offer new possibilities but are also limited in their reliability and accuracy. The present systematic review updates and discusses the existing literature regarding machine learning algorithm-based identification of cruciate ligament injury on radiographic images. PubMed was searched for articles containing machine learning algorithms related to cruciate ligament injury recognition. No additional filters or time constraints were used. All eligible studies were accessed by hand. From the 115 articles initially retrieved, 29 articles were finally included. Only one study included the posterior cruciate ligament (PCL). Deep learning algorithms in the form of convolutional neural networks (CNNs) were most frequently used. Many studies presented CNNs that identified binary decision classes of regular and torn anterior cruciate ligaments (ACLs) with a best sensitivity of 0.98, a specificity of 0.99, and an AUC ROC of 1.0. Other studies expanded the decision classes to partially torn ACLs or reconstructed ACLs, usually at the cost of sensitivity and specificity. Deep learning algorithms are excellent for identifying ACL injuries, tears, or postoperative status after reconstruction on MRI images. They are much faster but only sometimes better than the human reviewer. While the technology seems ready, barriers to ethical and legal issues and clinicians’ refusals must be overcome to some extent. It can be firmly assumed that artificial intelligence will have a future contribution in the diagnosis of cruciate ligament injuries.