Diagnosis of Anterior Cruciate Ligament injuries by means of Deep Learning approach using Magnetic Resonance Imaging

The anterior cruciate ligament is an essential organ in stabilizing the knee joint and unfortunately, it has a higher potential for injury, compared with other ligaments of the human knee. Generally, ligament injuries can be divided into three categories in terms of severity: partial tear or sprain, complete tear and complete tear with contraction. Each of these categories require specific treatment. Therefore, accurate detection of these injuries is a major medical challenge. With magnetic resonance imaging of the knee, radiologists are able to investigate and use their experience to diagnose the injury. But still, their diagnosis may be subject to error. In such situations, artificial intelligence can be used to reduce these medical errors and help the medical community. In this study, the modified ResNet-14 architecture of convolutional neural networks is used to classify the ACL injuries into three mentioned classes. The performance of the model has been evaluated using accuracy metric, after applying 5-fold cross-validation and class-balancing, with the result of 88.8%. The diagnostic results show that this model can be used for automatic detection and evaluation of the ACL injuries using deep learning technique.