CNN-based automatic classification of anterior talofibular ligament tears on MRI shows high accuracy

A 2D ResNet model automatically classified anterior talofibular ligament (ATFL) tears on MRI with 92.6% accuracy, detecting abnormal ligaments with 95.0% sensitivity and 93.6% specificity. Performance was comparable to a junior radiologist, but the workflow was faster (3 min vs…

• A YOLO11 segmentation model achieved a median Dice similarity coefficient of 0.95 on external validation data.
• For abnormal ATFL detection, the model showed 95.0% sensitivity and 93.6% specificity, while the junior radiologist showed 97.0% sensitivity and 95.5% specificity (P = 0.22).
• The automatic workflow (segmentation plus classification) took 3 minutes for the test set versus 14 minutes for manual review by a junior radiologist.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.