Researchers develop AI models to detect hidden medical conditions in CT scans

Medical researchers have developed artificial intelligence models capable of identifying subtle health risks and diseases from routine CT scans that may be missed during conventional interpretation.

At Penn Medicine, researchers developed Percival, a vision-language foundation model trained on over 400,000 CT images from nearly 52,000 patients. Unlike traditional AI designed for specific conditions, Percival analyzes broad imaging patterns and relationships between imaging findings and clinical information to identify potential future health risks. The model is currently in preclinical testing.

In a separate study published in Gut, researchers introduced REDMOD, an AI system designed to detect pancreatic cancer signals. The system uses radiomics to examine tissue texture patterns and automatically outlines the pancreas. In testing, REDMOD identified 73% of future pancreatic cancer cases, compared to 39% by experienced radiologists, with a median lead time of 475 days before clinical diagnosis. While promising for early detection, researchers noted it has not yet been proven that routine use leads to improved survival rates.