A DEEP LEARNING FRAMEWORK FOR AUTOMATED SEGMENTATION AND CLASSIFICATION OF INTERVERTEBRAL DISC DEGENERATION

Abstract

 Intervertebral disc (IVD) degeneration is a major contributor to chronic low back pain, yet its diagnosis through magnetic resonance imaging (MRI) remains time-consuming and subjective. This study proposes a robust deep learning framework integrating three state-of-the-art architectures U-Net, V-Net, and a novel IVD-ResNet for automated segmentation and classification of IVD degeneration into three categories: normal, mild, and severe. A multi-institutional dataset of 218 patient MRIs encompassing 3,535 discs was used for training and validation. The IVD-ResNet achieved superior classification performance with an overall accuracy of 98.30%, outperforming U-Net (83.31%) and V-Net (76.66%). Evaluation included precision, recall, F1-score, and confusion matrix analyses, with the IVD-ResNet demonstrating strong generalization and reliability across different degeneration levels. Comparative experiments highlight the benefits of hybrid feature integration and the potential of deep learning in clinical workflows. The proposed framework offers an accurate, efficient, and scalable solution for IVD degeneration analysis, with significant implications for early detection, personalized treatment planning, and reducing radiologist workload.