Deep convolutional neural network-based detection of meniscus tears: comparison with radiologists and surgery as standard of reference

Abstract

Objective To clinically validate a fully automated deep convolutional neural network (DCNN) for detection of surgically proven meniscus tears. Materials and methods One hundred consecutive patients were retrospectively included, who underwent knee MRI and knee arthroscopy in our institution. All MRI were evaluated for medial and lateral meniscus tears by two musculoskeletal radiologists independently and by DCNN. Included patients were not part of the training set of the DCNN. Surgical reports served as the standard of reference. Statistics included sensitivity, specificity, accuracy, ROC curve analysis, and kappa statistics. Results Fifty-seven percent (57/100) of patients had a tear of the medial and 24% (24/100) of the lateral meniscus, including 12% (12/100) with a tear of both menisci. For medial meniscus tear detection, sensitivity, specificity, and accuracy were for reader 1: 93%, 91%, and 92%, for reader 2: 96%, 86%, and 92%, and for the DCNN: 84%, 88%, and 86%. For lateral meniscus tear detection, sensitivity, specificity, and accuracy were for reader 1: 71%, 95%, and 89%, for reader 2: 67%, 99%, and 91%, and for the DCNN: 58%, 92%, and 84%. Sensitivity for medial meniscus tears was significantly different between reader 2 and the DCNN (p = 0.039), and no significant differences existed for all other comparisons (all p ≥ 0.092). The AUC-ROC of the DCNN was 0.882, 0.781, and 0.961 for detection of medial, lateral, and overall meniscus tear. Inter-reader agreement was very good for the medial (kappa = 0.876) and good for the lateral meniscus (kappa = 0.741). Conclusion DCNN-based meniscus tear detection can be performed in a fully automated manner with a similar specificity but a lower sensitivity in comparison with musculoskeletal radiologists.

Clinical and Research Medical Applications of Artificial Intelligence

“…applied deep learning to 100 MRIs and found the algorithm detected meniscus tears with similar specificity but lower sensitivity than musculoskeletal radiologists. 15 In a study of 260 patients with knee MRIs, Chang et al. reported anterior cruciate ligament tears were detected with 96% accuracy using a deep learning architecture.…”

Artificial intelligence for MRI diagnosis of joints: a scoping review of the current state-of-the-art of deep learning-based approaches

“…In 2020, a fully automated CNN was clinically evaluated to detect and differentiate medial and lateral meniscus tears on coronal and sagittal fat-suppressed fluid-sensitive MR images [32]. In contrast to other published studies evaluating AI for meniscus tear detection, arthroscopic surgery was used as the reference standard.…”

“…To estimate the diagnostic performance of deep learning algorithms in clinical practice, the use of an independent standard of reference may be critically important. There is a paucity of studies using a surgical reference standard [32], whereas most studies reference radiological interpretation. While this is a practical approach that compares diagnostic performance levels to radiologists, the clinical standard of reference for MRI interpretations is surgical inspection.…”

“…Right lower cut-out boxes represent a magnification of the left upper area (dashed box). In A and B: gray = DL algorithm of Fritz et al [32]; red = DL algorithm of Rizk et al [33]; yellow = metaanalysis of Phelan et al [21]; blue = meta-analysis of Oei et al [18]. Note: Only studies reporting 95% CI and differentiating between the medial and lateral meniscus were included.…”

Applications of Artificial Intelligence in Musculoskeletal Imaging: From the Request to the Report

“…Some examples include the When the radiologist dictates the fracture description, the natural language processor identifies the disease (fracture) and anatomy (tibia) concepts from the unstructured text and proceeds to retrieve pertinent fracture knowledge (eg, classification, diagnostic criteria, disease probability). Reprinted Figure 1 from Do et al 78 identification of anterior cruciate ligament 65 and meniscal 66 tears on MRI of the knee, detection of epidural masses on CT, 67 evaluation of the joint space, 68 and presence of erosions 69 in rheumatoid arthritis, sex estimation from sacrum and coccyx CT 70 or from hand and wrist radiographs, 71 assessment of Achilles tendon healing, 72 and automated segmentation of peripheral nerves of the thigh, 73 supraspinatus muscle, 74 proximal femur, 37 and bones of the wrist 75 on MRI. Additional repetitive and time-consuming tasks suitable to automation will likely be tackled in coming years.…”