Automatic Detection of Meniscus Tears Using Backbone Convolutional Neural Networks on Knee MRI

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Timely diagnosis of meniscus injuries is crucial for preventing knee joint dysfunction and improving patient outcomes.
• Early detection facilitates treatment planning and reduces morbidity associated with knee injuries.

Purpose of the Study:

• To develop and validate a deep learning model for automated meniscus tear detection using knee MRI.
• To evaluate the model's performance against human radiologists.

Main Methods:

• A deep learning model (improved YOLOv4 with Darknet-53) was trained and tested on 584 knee MRI studies.
• The model's performance was assessed using sensitivity, specificity, and accuracy, with surgical reports as ground truth.
• Model performance was compared to three radiologists of varying experience levels.

Main Results:

• The model achieved high overall accuracies of 95.4% (internal testing) and 95.8% (internal validation).
• External validation accuracy was 78.8%, and the model demonstrated significantly better performance than one radiologist (95.8% vs. 90.2%).

Conclusions:

• The proposed deep learning model demonstrates high sensitivity, specificity, and accuracy in detecting meniscus tears on knee MRIs.
• This AI-driven approach shows potential for improving the efficiency and accuracy of meniscus tear diagnosis.