[Sportschosun reporter Jang Jong-ho] An Artificial Intelligence (AI) diagnostic support model has been developed to detect free air in the abdominal cavity, which indicates the need for emergency surgery, from abdominal computed tomography (CT) images.
A surgical team led by Professor Kim Dong-jin of the Department of Surgery at The Catholic University of Korea Eunpyeong St. Mary's Hospital, together with Professor Park Shin-hye of Eunpyeong St. Mary's Hospital, doctoral researcher Kim Sang-wook of the University of Toronto, researcher Lee Joong-hyup of MasAuto, and Professor Lee Ha-ye-min of Bucheon St. Mary's Hospital, recently published related findings in the International Journal of Surgery, a top international journal in the field of surgery with an impact factor of 9.0.
Perforation, in which a hole forms in the gastrointestinal tract, can occur when gastric or duodenal ulcers caused by inflammation or Helicobacter pylori infection are not treated in time. Although its incidence has declined thanks to advances in medicine, the condition still carries a high risk of death and complications, making it a representative surgical emergency that requires immediate operation.
When a perforation develops in the gastrointestinal tract, gas from inside the tract escapes into the abdominal cavity. This is called free air. Free air is one of the most important imaging findings suggesting gastrointestinal perforation and a key clue in deciding whether surgery should be performed quickly. It can be identified on contrast-enhanced abdominal CT, but it is not easy to detect when it is hidden near the liver or in the upper abdomen, or when the amount is very small. In emergency rooms especially, concerns have persisted that tiny amounts of free air may be missed because of limited time, heavy workloads, and differences in clinicians' experience.
To address this, Professor Kim's team trained AI on abdominal CT images from 127 patients who underwent surgery for gastric ulcer perforation between April 2019 and April 2022, and developed a first-stage model called FA-NET (Free Air Net) that automatically identifies free-air regions. The team then added CT images from 76 patients with appendicitis, a condition that can be easily confused with free air because of similar imaging findings. By applying a negative training technique to distinguish the two diseases and reduce false positives, the researchers completed an advanced model called FA-NET-NT (Free Air Net Negative).
The FA-NET-NT model recorded a Dice score of 0.87, indicating diagnostic accuracy, and achieved 85% sensitivity and 96% specificity in image-level analysis using representative CT slices.
In validation tests involving 215 new patients examined over time, the FA-NET-NT model detected actual gastric ulcer perforations with 95% to 96% sensitivity. It also distinguished gastric ulcer perforation from other conditions without free air, including appendicitis, pancreatitis, and cholecystitis, with 82% to 92% specificity.
The team also conducted additional multicenter external validation on 237 patients from other hospitals, where the imaging equipment and scanning protocols were entirely different, to assess the model's real-world clinical applicability. The results showed that sensitivity for gastric ulcer perforation remained at 95%, while specificity was also high for appendicitis (88%), pancreatitis (88%), cholecystitis (82%), and intestinal obstruction (80%), confirming stable performance across different CT devices and clinical environments.
Professor Kim said, "This model is not meant to replace clinicians' judgment. Its value lies in serving as a second set of eyes that can recheck free-air findings that are easy to miss in busy emergency situations. It could be especially useful as a support tool for making quick surgical decisions in nighttime or holiday emergency rooms, where specialized staff are in short supply," adding, "We will continue follow-up research through multicenter prospective clinical studies and direct comparisons with radiology specialists so that it can be applied in real emergency care settings."
Professor Kim Dong-jin and Professor Park Shin-hye received an award for an outstanding oral presentation at the International Congress of the Korean Society of Gastrointestinal Surgery in 2024 for the paper. In 2025, they received the same award again at the same congress for research demonstrating the superiority of robot-assisted minimally invasive esophagectomy.
Jang Jong-ho bellho@sportschosun.com
This article has been translated by GripLabs Mingo AI.
