Meniscus tear detection on knee magnetic resonance images with U-net model

Motivation: Making an accurate diagnosis of meniscus tears from knee magnetic resonance imaging (MRI) is difficult. Goal(s): To evaluate the accuracy of the 3D/2D deeply supervised U-net model for meniscus tears detection on knee MRI. Approach: A total of 391 adult knee MRI scans were annotated in the tear regions of the menisci in coronal and sagittal images as ground truth. Tear detection was performed as a segmentation within the exam in the 679 test dataset. Results: The accuracy of the coronal model and sagittal model were 0.76 and 0.74, respectively. Impact: The diagnostic model for meniscus tears on knee magnetic resonance imaging might be useful as a screening tool and diagnostic aid for radiologists but further improvement is required for more accurate detection.