Skip to main content

Distinguished Professor of Computer Science Pierre Baldi recently published a new paper, “Deep Learning Localizes and Identifies Polyps in Real Time With 96% Accuracy in Screening Colonoscopy,” in the journal Gastroenterology. Working in a collaboration with Dr. William Karnes and his team in UCI’s Department of Medicine, Baldi and Ph.D. student Gregor Urban designed and trained deep convolutional neural networks (CNNs) to detect the polyps in colonoscopies in order to help doctors improve the adenoma detection rate (ADR). Using a set of 8,641 colonoscopy images containing 4,088 polyps, the trained CNN was able to identify polyps with a cross-validation accuracy of 96.4 percent. The system has the potential to increase ADR and reduce interval colorectal cancers.

“AI and machine learning can address many problems in biomedical imaging today, and across a spectrum of modalities — microscopy, X-ray, MRI and, as in this case, video,” says Baldi. “While there remain significant obstacles to gathering training data and deploying these technologies in clinical settings, over the longer term, AI will help improve the precision and reduce the cost and turn-around time of all these procedures.”

— Shani Murray