Deep learning automates multi-structure CT segmentation for ventral hernia surgical planning

Deep learning CT segmentation of ventral hernias and abdominal muscles achieved IoU up to 0.91 (hernia defects 0.85), offering rapid, standardized measures. Proof-of-concept; prospective validation needed.

• Study used 215 adult CT scans with expert annotations to train a TransUNet-inspired model.
• The model simultaneously segmented hernia defects, rectus abdominis, lateral abdominal wall, and psoas muscles with high overlap scores.
• Limitation: Proof-of-concept lacking prospective or external validation before potential clinical deployment.

Automated summary

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