Deep learning pipeline detects lumbar disc herniation on MRI with F1-score up to 0.73

A deep learning pipeline combining segmentation (Dice 0.898) and classification (F1-score up to 0.73) detected lumbar disc herniation on MRI, though limited by sample size and class imbalance.

• Retrospective development of a multi-stage pipeline using 484 lumbar spine MRI exams, with 165 studies (5,200 slices) annotated for segmentation training.
• Best segmentation: MA-Net with EfficientNet-B2, Dice 0.898 for three-class segmentation (background, vertebrae, discs).
• Classification yielded F1-scores 0.66–0.73, best with ResNet18; authors note limitations of small sample size and class imbalance.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.