АВТОМАТИЗИРОВАННОЕ ОБНАРУЖЕНИЕ И ОЦЕНКА СТЕПЕНИ ГРЫЖИ МЕЖПОЗВОНОЧНОГО ДИСКА ПОЯСНИЧНОГО ОТДЕЛА ПОЗВОНОЧНИКА С ИСПОЛЬЗОВАНИЕМ ГЛУБОКОГО ОБУЧЕНИЯ НА ОСНОВЕ ДАННЫХ МРТ

Abstract

Introduction. Assessment of the severity of lumbar intervertebral disc herniation (LIDH) based on magnetic resonance imaging (MRI) is crucial for selecting appropriate candidates for surgical treatment. However, manual interpretation of MRI scans is time-consuming and labor-intensive. The aim of this study was to develop and evaluate a deep learning–based diagnostic model for automated detection and classification of LIDH on axial T2-weighted lumbar spine MRI images.

Materials and methods. In this retrospective study, data from 1,115 patients were analyzed, including a development dataset (1,015 patients, 15,249 images) and an external test dataset (100 patients, 1,273 images). All images were jointly annotated by experts according to the Michigan State University (MSU) classification criteria, and the consensus results were considered the reference standard. The automated diagnostic model incorporated Faster R-CNN and ResNeXt101 as the detection and classification networks, respectively. Diagnostic performance was evaluated using mean intersection over union (IoU), accuracy, precision, sensitivity, specificity, F1 score, area under the receiver operating characteristic curve (AUC), and intraclass correlation coefficient (ICC) with 95% confidence intervals (CI).

Results. High detection consistency was achieved in both the internal test dataset (mean IoU = 0.82, accuracy = 98.4%, sensitivity = 99.4%) and the external test dataset (mean IoU = 0.70, accuracy = 96.3%, sensitivity = 97.8%). The overall classification accuracy of LIDH was 87.70% (95% CI: 86.59%–88.86%) and 74.23% (95% CI: 71.83%–76.75%) in the internal and external test datasets, respectively. During internal testing, the proposed model demonstrated high classification agreement (ICC = 0.87, 95% CI: 0.86–0.88, P < 0.001), which was slightly lower in the external test dataset (ICC = 0.79, 95% CI: 0.76–0.81, P < 0.001). The AUC values for classification were 0.965 (95% CI: 0.962–0.968) and 0.916 (95% CI: 0.908–0.925) for the internal and external test datasets, respectively.

Conclusions. The proposed automated diagnostic model demonstrated high performance in detecting and classifying lumbar intervertebral disc herniation and showed substantial agreement with expert-based classification.