MRI-Based Anterior Cruciate Ligament Injury Detection Using Transfer Learning

It is critical for the timely discovery of abnormalities or pathologies from radiographic scans, such as magnetic resonance images. Unfortunately, such a task is often rather cumbersome, time-consuming, and error-prone. With the development of machine learning, computer-aided diagnoses systems have been increasingly applied to the field of medical imaging. These auxiliary diagnostic systems are designed to help physicians for better diagnoses. In this work, a neural model for human Anterior Cruciate Ligament (ACL) injury detection is proposed. The network incorporates transfer learning technique with automatic deep layer feature extraction from 3D sagittal knee magnetic resonance images to detect the possibility of mild ACL injury that does not necessitate surgery, and complete ACL rupture that requires surgical intervention. Compared to traditional machine learning methods, the area under the curve has increased by 3%, i.e., the performance of the proposed model has reached a higher classification accuracy rate, which can reduce the misdiagnosis of ACL injury detection. Therefore, the proposed method can be applied for medical diagnosis, such as diabetic retinopathy and pneumonia detection, and can also be used as an early warning system for patients and hospitals to reduce medical costs.