AI System Beats Endoscopists for Detecting Early Neoplasia in Barrett's

Bruce Jancin

April 02, 2021

One of the top publications in gastroenterology in 2020 was a Dutch study demonstrating that a computer-aided system suitable for real-time use in clinical practice detected early neoplasia in patients with Barrett's esophagus with impressively greater accuracy than did a group of general endoscopists, according to Douglas A. Corley, MD, PhD.

Dr Douglas Corley

It's not just his personal opinion that this was one of the major studies of the past year, either. Analytic tools showed the Dutch report was one of the most frequently downloaded studies in 2020 by both clinical gastroenterologists and researchers, said Corley, director of delivery science and applied research at Kaiser Permanente of Northern California, Oakland, and a faculty gastroenterologist at the University of California, San Francisco.

The deep-learning system developed, evaluated, and externally validated by the Dutch investigators is designed to reduce the rate of failed detection of high-grade dysplasia and early adenocarcinoma in patients undergoing surveillance by general practice gastrointestinal endoscopists.

The false-negative rate in looking for the sometimes subtle mucosal surface abnormalities indicative of early neoplasia is known to be higher among these general endoscopists than that among expert endoscopists, and yet it's the general endoscopists who perform the majority of cancer surveillance in patients with Barrett's esophagus.

The Dutch group developed the computer-aided detection system by applying artificial intelligence methods to analyze nearly one half-million endoscopic images of confirmed early neoplasia. Once the system was ready to go, they compared its diagnostic accuracy in 80 patients to that of 53 general, nonexpert endoscopists.

The deep-learning system had 93% sensitivity and 83% specificity for identification of early neoplasia, significantly better than the 72% sensitivity and 74% specificity for the general endoscopists. The overall accuracy of the computer-assisted detection system was 88%, compared to 73% for the general endoscopists. Moreover, the deep-learning system achieved greater accuracy than did any single one of the endoscopists.

"I think this will be a really helpful addition, the equivalent of a second endoscopist raising a yellow flag to take a closer look at a particular area. It'll be complementary," Corley said at the Gastroenterology Updates, IBD, Liver Disease Conference.

An audience member said he's aware that other computer-assisted detection systems have also shown outstanding performance for the detection of early neoplasia in Barrett's esophagus. He asked, why aren't these being deployed yet in routine clinical practice?

Two reasons, Corley replied. One is that some of those systems aren't capable of working during real-time endoscopy. Also, industry seems to be taking a wait-and-see approach. The field of applied artificial intelligence is moving incredibly rapidly, and none of the endoscopic equipment manufacturers wants to incorporate a computer-assisted detection system into their gear when rumor has it that an even better system is going to be announced 6 months later. The manufacturers want to make sure they're operationalizing the right one.

He suspects the major players in the endoscopic imaging industry are waiting to find a computer-assisted detection system that's been published and widely accepted as clearly a winner. Then they'll introduce it into their equipment.

"I do think we're probably going to be seeing these increasingly. Some computer-assisted detection systems for colon cancer are starting to be put into equipment," he observed.

Corley reported having no financial conflicts regarding his presentation.

This article originally appeared on, part of the Medscape Professional Network.


Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.