Fully Automatic Fine-Grained Grading of Lumbar Intervertebral Disc Degeneration Using Regional Feature Recalibration...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Spinal Diagnostics

Background:

• Accurate grading of lumbar intervertebral disc (LIVD) degeneration is crucial for managing low back pain.
• Current grading methods often overlook subtle degenerative changes in disc structures surrounding the nucleus pulposus.

Purpose of the Study:

• To develop a novel Regional Feature Recalibration Network (RFRecNet) for accurate and reliable fine-grained LIVD degeneration grading.
• To enhance diagnostic capabilities for low back pain by improving the precision of disc degeneration assessment.

Main Methods:

• Utilized Detection Transformer (DETR) for LIVD detection.
• Developed a Regional Cube-based Feature Boosting and Suppression (RC-FBS) module for adaptive feature recalibration.
• Implemented a Feature Diversification (FD) module to integrate multi-scale semantic information.

Main Results:

• Achieved an average grading accuracy of 90.5% on a dataset of 10,225 LIVDs from 500 MR scans.
• Demonstrated high specificity (97.5%) and sensitivity (90.8%).
• Obtained a Cohen's kappa correlation coefficient of 0.876, indicating strong agreement.

Conclusions:

• The proposed RFRecNet framework offers a promising solution for reliable, quantitative, fine-grained evaluation of LIVD degeneration.
• This technology can aid clinicians in designing optimal surgical plans for low back pain patients.