Monday, November 30 | 9:30 a.m.-10:30 a.m. | M3-SSGI03-3 | Room E451B
An open-source AI system could detect pancreatic cancer on routine CT exams nearly a year before clinical diagnosis, according to this study.
Pancreatic ductal adenocarcinoma (PDAC) is the deadliest gastrointestinal cancer and is often diagnosed when curative surgery is no longer possible. Although opportunistic detection on routine CT could make a difference, radiologists routinely miss small tumors and early lesions, noted presenter Wenxuan Li of Johns Hopkins University and colleagues. To address this, the group developed and externally validated ePAI, an AI system that detects and localizes PDAC on contrast-enhanced CT.
After training ePAI on scans from nearly 1,600 patients, the researchers performed internal testing on more than 1,000 patients and over 7,000 at six centers in North America, Europe, and Asia. They also evaluated it on scans from 159 patients acquired three to 36 months before diagnosis, all of which had been read as negative at the time, and compared it with 30 radiologists in a reader study.
ePAI showed high accuracy in both internal and external testing, including for tumors 2 cm or smaller. On the scans performed prior to diagnosis, it detected cancer in 75 of the 159 patients, with a median lead time of 347 days. It also outperformed the radiologists in sensitivity by a wide margin.
“ePAI enables opportunistic detection of early pancreatic cancer on routine abdominal CT, identifying tumors at a stage when curative surgery is still feasible,” the researchers wrote. “By flagging subtle lesions and ductal changes, it supports proactive surveillance rather than passive detection in high-volume radiology settings.”
Make plans to attend this Monday morning session.
