A domestic research team has developed an artificial intelligence (AI) model that analyzes X-ray images to determine whether a newborn has an intestinal perforation. Intestinal perforation is a life-threatening condition in which a hole forms in the intestine due to necrotizing enterocolitis and other causes.
Seoul Asan Medical Center said on the 1st that a team led by Yoon Hee-mang, professor of radiology, Kim Nam-kook, professor of convergence medicine, and Lee Byung-sup, professor of neonatology, developed an AI reading model to determine whether a newborn has an intestinal perforation.
To diagnose intestinal perforation in newborns, X-ray tests are used to check for the presence of free air in the abdominal cavity, but the imaging findings of perforation are not always distinct, making accurate interpretation difficult. There have been AI reading models, but they were developed based on adult data, making them hard to apply to newborns.
In response, the research team developed a training model that classifies whether a newborn has an intestinal perforation using X-ray images and simultaneously learns and highlights areas with free air in the abdominal cavity. To do this, they collected about 2.6 million pediatric X-ray images from Seoul Asan Medical Center from Jan. 1995 to Aug. 2018, and ultimately selected 294 perforation images and 252 control images to train the model.
The AI model also showed high accuracy, suggesting strong potential for clinical use. In internal validation, the intestinal perforation AI reading model achieved a diagnostic accuracy of 94.9% and accurately identified areas with free air in the abdominal cavity. Diagnostic accuracy validated with external data was 84.1%, showing a level similar to that of specialists.
They also evaluated the assistive effect when clinicians used the AI reading model, and diagnostic accuracy improved from 82.5% to 86.6%. In particular, inter-reader agreement increased significantly from 71% to 86%.
Yoon Hee-mang, professor of radiology at Seoul Asan Medical Center, said, "Intestinal perforation in newborns is highly emergent, so a prompt diagnosis is paramount, but imaging findings are ambiguous and differ from those of adults, so diagnosis varies greatly depending on reading experience," adding, "The AI reading model for neonatal intestinal perforation not only demonstrated specialist-level accuracy but also improved inter-clinician reading agreement."
Kim Nam-kook, professor of convergence medicine at Seoul Asan Medical Center, said, "We are focusing on developing technologies that are essential in clinical settings, such as neonatal intestinal perforation, but are still under-researched," and added, "We will develop and apply various models that can aid early diagnosis in neonatal intensive care units where swift decisions are required, contributing to improving newborn survival rates."
The results of this study were published in the latest issue of the renowned international journal in biomedicine, "Computers in Biology and Medicine" (impact factor 6.3).
References
Computers in Biology and Medicine (2025), DOI: https://doi.org/10.1016/j.compbiomed.2025.110945