Deep Learning-Based Quantitation for Lumbar Intervertebral Disc Degeneration from MRI

Lumbar disc degeneration (LDD), as one of the leading causes of chronic low back pain and sciatica, has become a critical focus in clinical research. In this study, a comprehensive 3D dataset of lumbar intervertebral disc degeneration was constructed using 603 lumbar MRI cases and corresponding expert annotations collected from the Suzhou Hospital of Traditional Chinese Medicine. To identify the most suitable segmentation method, two 3D segmentation models were systematically evaluated: 3D U-Net and nnFormer. The experimental results demonstrated that nnFormer outperformed the other models, achieving a Dice coefficient of 61.3% and an IoU of 44.2%, with a good capability to handle complex anatomical boundaries and subtle degenerative regions. Consequently, nnFormer was selected as the final segmentation model. Based on this, an end-to-end visualization and analysis system, iSpineQuant, was developed. The system integrates segmentation results, volumetric changes, and clinical metadata to support automated quantification and comparative analysis of treatment response. This study not only introduces a novel benchmark dataset tailored for lumbar disc degeneration but also provides a clinically applicable tool that facilitates standardized and data-driven evaluation of the efficacy of traditional Chinese medicine (TCM) treatment, thus promoting the integration of AI into traditional clinical workflows.