Anterior Cruciate Ligament Injury Classification from MRI Scans Using Deep Learning

One of the most common injuries incurred through physical activity is the anterior cruciate ligament tear or ACL tear. An ACL injury is at risk for developing knee instability, meniscal damage, and long-term osteoarthritis of the knee. Prompt surgical intervention and post-operative rehabilitation are usually required to lessen the risk of further meniscal and cartilage damage. Currently, the imaging modality of choice to diagnose an ACL injury is based on the knee magnetic resonance imaging (MRI). The study aims to investigate the diagnostic capacity of CNN via transfer learning in the recognition of the ACL injury from knee MRI scans. Different network parameters are tested to find the optimum network configuration and best-performing model. AlexNet model utilizing oversampling with data augmentation with an input size of 75x75, 30 epochs, and a batch size of 8 was the best performing model configuration with the highest weighted average area under the receiver operating characteristic curve at 85.70. A web-based decision support tool for assessing and diagnosing ACL injury was also developed integrating the best performing CNN model. The early and immediate recognition of ACL injury with transfer learning methods can cascade prompt referral to an orthopedic expert as well as the institution of appropriate therapeutic measures to minimize the risk of worsening knee instability.