Deep learning in ventral hernia imaging: automated multi-structure CT segmentation for surgical planning.
Authors
Affiliations (15)
Affiliations (15)
- Department of Pediatric Surgery, Dr. Mehta Hospital, Chennai, India.
- Department of General Surgery, Sheffield Teaching Hospitals NHS Foundation Trust, Sheffield, United Kingdom.
- Department of General Surgery, Grant Government Medical College and Sir J J Group of Hospitals, Mumbai, India.
- Department of General Surgery, Madras Medical College and Government General Hospital, Chennai, India.
- Department of General Surgery, Dayanand Medical College and Hospital, Ludhiana, India.
- Department of General Surgery, Sri Ramachandra Medical College and Research Institute, Chennai, India.
- Department of General Surgery, Chennai Hernia Centre, Chennai, India.
- Department of GI Surgery, KIMS-Sunshine Hospital, Hyderabad, India.
- Department of General Surgery, Zen Multi Speciality Hospital, Chembur, India.
- Department of General Surgery, Army Medical Corps Centre and College, Lucknow, India.
- Department of GI Surgery, Yashoda Hospital, Hyderabad, India.
- Department of Radiology, Advantage Imaging and Research Institute, Chennai, India.
- King Edward Memorial Hospital and Seth Gordhandas Sunderdas Medical College, Mumbai, India.
- Department of Pediatrics, Saint Peter's University Hospital, New Brunswick, NJ, United States.
- Curium Life Tech, Chennai, India.
Abstract
Accurate preoperative assessment of ventral hernia defects remains time-intensive and subject to inter-observer variability. Current manual CT analysis for surgical planning is time-consuming, with inconsistent measurements affecting operative decision-making. 215 CT scans of adults with ventral hernias were analyzed using TransUNet-inspired deep learning models. Expert annotations of anatomical landmarks and hernia features served as ground truth. Models were trained to automate segmentation of hernia defects and other critical anatomical structures. Automated segmentation achieved IoU values of 0.85 for hernia defects, 0.89 for rectus abdominis muscles, 0.87 for lateral abdominal wall muscles, and 0.91 for psoas muscles. Deep learning automation provides rapid, standardized hernia assessment for surgical planning. The system delivers objective measurements with significant time savings, demonstrating technical feasibility as a proof-of-concept that warrants further prospective clinical validation before deployment in operative decision-making.