Artificial intelligence assistance using deep metric learning vs. object detection in classifying lumbar disc degeneration on magnetic resonance images.

Abstract

OBJECTIVE: This study aimed to assess the performance of an image retrieval system based on the deep metric learning (DML) approach in discriminating between early and late stages of degenerative intervertebral disc degeneration (IDD). MATERIALS AND METHODS: A total of 2,341 sagittal-plane lumbar spinal MR images were obtained and labeled as early and late degeneration using the Pfirrmann classification. Both the DML model and the Object Detection Model based on a state-of-the-art YOLOv7tiny were trained and tested using the labeled data. Then, performance parameters, such as sensitivity and precision, were computed and compared. RESULTS: The trained DML model achieved both sensitivity and precision levels of approximately 93% and 95%, respectively, and an area under the receiver operating characteristic curve of at least 0.96. The trained Object Detection Model based on modified YOLOv7tiny achieved a precision of 92.6%, a sensitivity of 85.9%, and a mean average precision (mAP) of 0.851. CONCLUSIONS: These results showed that DML yielded a state-of-the-art performance and could be used as a diagnostic tool for discriminating the severity of IDD via MRI. Graphical Abstract: https://www.europeanreview.org/wp/wp-content/uploads/Graphical-abstract-15.jpg.