Skip to main content
Back

Copenhagen Researchers Blueprint AI to Predict Deadly Bowel Surgery Leaks Before They Happen

A new AI protocol maps bowel blood supply from preoperative CT scans to predict anastomotic leak risk before colorectal surgery

Paper, Full Access
  • The problem: Anastomotic leak after colorectal surgery affects up to 30% of patients and carries mortality rates up to 35%, yet no validated preoperative prediction tool exists.
  • The proposal: A new protocol paper from Copenhagen and Odense University Hospital outlines an AI framework to predict leak risk from routine preoperative CT scans.
  • The approach: Paired pre- and post-contrast CT scans are analyzed to map bowel blood supply; poor perfusion at the anastomotic site is the key risk signal.
  • Two deliverables: A deep learning risk-scoring module and a CBMIR (Content-Based Medical Image Retrieval) module that surfaces similar historical cases to explain the score.
  • Explainability first: The CBMIR component shows surgeons real analogous patient cases alongside outcomes, rather than a standalone black-box score.
  • Status: This is a protocol paper -- no trained model or benchmark results yet; no code or model weights have been released.

When a surgeon joins two ends of the bowel after removing a cancerous segment, the resulting connection is called an anastomosis. If it fails and leaks, the consequences are severe. Anastomotic leakage remains one of the most potentially life-threatening sequelae in colorectal cancer surgery, with an incidence ranging from 1% to 30%. Mortality among patients who develop a leak can reach 22%, with figures in the literature ranging from 10% to 35% depending on severity. Despite decades of surgical refinement, the rate has barely budged.

The frustrating part is not just that leaks happen -- it is that surgeons currently have no reliable way to see them coming. Current preoperative assessment relies only on clinical judgment, a process that is subjective, error-prone, and highly dependent on individual expertise. Existing diagnostic methods, including CT scans, contrast enemas, and endoscopic examinations, can confirm a leak after the fact but lack strong predictive value, and a definitive predictive test is still lacking.

A gap nobody has closed yet

CT imaging is already standard of care before colorectal cancer surgery, yet its predictive potential has gone largely untapped. Despite progress in the field, there is still no standardized or validated pipeline for CT-based preoperative risk prediction of colorectal anastomotic leak. That is the gap a new protocol paper from researchers at the University of Copenhagen and Odense University Hospital aims to close.

The paper does not report a trained model with benchmark numbers -- it is a protocol paper, meaning it lays out the full methodological blueprint for building and validating such a system. Think of it as the architectural plan before construction begins. The value is in the rigor and reproducibility of the roadmap itself, designed so that other hospitals and research groups can follow the same steps.