Asan Medical Center Develops AI to Detect Deadly Neonatal Intestinal Perforation

A research team from Asan Medical Center (AMC), South Korea’s largest hospital, has created a deployable artificial intelligence system designed to detect intestinal perforation—a potentially fatal condition in premature newborns—through the analysis of X-ray images. The model applies deep multi-task learning to identify the presence and location of air-filled lesions in the abdominal cavity, a key diagnostic indicator of the condition.

Trained on 294 intestinal perforation images and 252 control samples from AMC’s 23-year database of 2.6 million pediatric X-rays, the AI demonstrated a 94.9% accuracy rate on internal data. When validated externally using 378 images from 11 hospitals, it achieved an 84.1% accuracy rate, comparable to that of experienced clinicians. The system also improved diagnostic accuracy among medical staff from 82.5% to 86.6% and raised inter-interpreter agreement from 71% to 86%.

The findings, published in Computers in Biology and Medicine, highlight how the AI enhances both detection accuracy and diagnostic consistency within neonatal intensive care units (NICUs). According to Dr. Yoon Hee-mang, professor of radiology at AMC, “Neonatal intestinal perforation presents a high level of urgency, making rapid diagnosis paramount. However, imaging findings are ambiguous and differ from those in adults, making the diagnostic rate significantly dependent on the interpreter’s experience.” Dr. Yoon added that the model not only achieves expert-level precision but also improves consensus among clinicians.

Intestinal perforation poses a severe risk for premature infants, as delayed or missed diagnoses can rapidly lead to life-threatening complications. Conventional AI tools, typically trained on adult datasets, are poorly suited for neonatal imaging due to anatomical and physiological differences. AMC’s newborn-focused model addresses this gap, offering a tailored diagnostic support system for NICU settings.

This innovation aligns with broader efforts across the Asia-Pacific region to integrate AI into neonatal care. In Australia, AI-driven imaging systems are being used to remotely monitor vital signs and automate jaundice diagnosis in infants, while in India, a SAP-supported screening initiative employs machine learning to identify retinopathy of prematurity, a preventable cause of blindness among preterm babies.

By demonstrating robust performance and practical utility, AMC’s AI represents a meaningful advancement in clinical decision support for neonatal medicine.

Click here for the original news story.