A deep learning technology trained to read MRIs identified anterior cruciate ligament (ACL) tears in the knee with accuracy similar to clinical radiologists, reported authors of a study published in Radiology: Artificial Intelligence.
“Detecting an ACL tear relies on evaluating an obliquely oriented structure on multiple image sections with different tissue contrasts using a combination of MRI findings including fiber discontinuity, changes in contour, and signal abnormality within the injured ligament,” wrote Fang Liu, PhD, with the University of Wisconsin School of Medicine and Public Health in Madison, and colleagues. “Investigation of the ability of a deep learning approach to detect an ACL tear would be useful to determine whether deep learning could aid in the diagnosis of complex musculoskeletal abnormalities at MRI.”
Liu and colleagues created their system using three deep convolutional neural networks—one to isolate the ACL on MRIs and another to detect structural problems in the ligament. The method analyzed T2-weighted fast spin-echo MRIs of the knee in 175 patients with a full-thickness ACL tear and another 175 with a healthy ligament.
The deep learning approach achieved a sensitivity and specificity of 0.96, and an area under the ROC curve (AUC) score of 0.98.
In comparison, five clinical radiologists independently reviewed 100 patient MRIs from a hold-out dataset and achieved a sensitivity that ranged from 0.96 to 0.98 and a specificity spanning 0.90 to 0.98.
There were several limitations, according to the researchers. One was that their CNNs were not a single end-to-end network, but rather three connected in a “cascaded fashion.” Training the networks individually also increased the training burden of their study.
Larger training datasets in future studies would likely improve the performance of their CNNs, the authors noted.
“In summary, our study has demonstrated the feasibility of using a deep learning–based approach to detect a full-thickness ACL tear within the knee joint at MRI, with arthroscopy as the reference standard,” Liu et al. concluded.