Use of AI to study CT scans shows promise in finding pancreatic cancer

Researchers are looking to use artificial intelligence to take a more exacting look at computed tomography scans in the hope of detecting pancreatic cancer earlier.

A study underway at Johns Hopkins University has shown promising results in training computers to detect pancreatic tumors in CT scans, as reported in findings in the American Journal of Roentgenology.

The use of AI to detect difficult-to-find variants of pancreatic cancer holds the promise of identifying characteristics of the disease with greater accuracy than can be done with the naked eye.

Computed tomography is the most commonly used imaging modality for the initial evaluation of suspected pancreatic ductal adenocarcinoma (PDAC). Indications of PDAC can be subtle and missed by even experienced radiologists.

In efforts to improve the accuracy of detection, the Lustgarten Foundation has funded research in AI with hopes of using AI and computers to detect pancreatic cancer earlier. The Johns Hopkins project uses sophisticated programs that “teach” themselves to read CT scans. The approach, called radiomics, attempts to extract many quantitative features from medical images using data characterization algorithms.

johns-hopkins-istock-357.jpg
View of the Johns Hopkins University School of Medicne building in Baltimore city, Maryland.

In this initiative, one of the biggest challenges is to get the computer to recognize a healthy pancreas vs. one with a tumor. The retrospective case study reported in the Journal all 60 PDAC cases were correctly classified and differentiated from normal, healthy pancreases.

"This work demonstrates that radiomics features extracted from the whole pancreas can be used to differentiate between CT cases from patients with PDAC and healthy control subjects," says Elliot Fishman, MD, professor of radiology and radiological science, who conducted the study with Alan Yuille, professor of cognitive science and computer science at Johns Hopkins University.

"The results are very encouraging," says David Tuveson, MD, Lustgarten's chief scientist and director of the Cancer Center at Cold Spring Harbor Laboratory. "It reinforces that we are on the right track. The goal of this research is for computers to be trained to look for any slight abnormality within the pancreas before radiologists would consider this to be a worrisome finding. This could result in earlier detection of pancreatic tumors, as well as improve the number of accurate diagnoses so that the proper care can be initiated sooner." Lustgarten Foundation, based in Woodbury, N.Y., is the largest private funder of pancreatic cancer research in the world.

For reprint and licensing requests for this article, click here.