AI development for medical imaging detection

Researchers have developed artificial intelligence models designed to identify subtle medical conditions in routine CT scans that may be missed by human experts.

One system, REDMOD, utilizes radiomics to analyze minute texture changes in pancreatic tissue. In studies published in the journal Gut, REDMOD demonstrated the ability to detect signs of pancreatic ductal adenocarcinoma an average of 475 days before a formal clinical diagnosis. The system achieved a 73% sensitivity rate, which was significantly higher than the 39% sensitivity rate observed in experienced radiologists. While the results are promising, researchers noted that the technology requires further validation through prospective testing in broader, more diverse populations.

Additionally, researchers at Penn Medicine developed Percival, a vision-language foundation model trained on over 400,000 CT images. Unlike models designed for specific conditions, Percival analyzes broad imaging patterns and the relationships between imaging findings and clinical information to identify potential future health risks. This model is currently in the preclinical testing phase.