Can AI opportunistically detect gastric cancer on noncontrast CT?

Sunday, November 26 | 3:20 p.m.-3:30 p.m. | S5-SSGI04-6 | Room S405

This gastrointestinal imaging-related session will discuss the effectiveness of using an AI model in combination with noncontrast CT exams for the detection of gastric cancer tumors.

Ling Zhang, MD, and a team from Moffitt Cancer Center in Tampa, FL, trained an AI model to perform gastric cancer screening using 1,891 3D noncontrast CT volumes from a single hospital, including 687 patients with pathology-confirmed gastric cancer and 1,204 normal controls. They then tested the model on a hold-out test set comprised of 100 gastric cancer cases and 148 normal cases, as well as an external test set of 903 normal cases.

The model produced an area under the curve of 0.939, which was better than the two radiologists, according to the researchers. In addition, it outperformed the best-performing radiologist in detecting early-stage gastric cancer (T1 60% vs. 50%, T2 77.8% vs. 55.6%), according to the researchers. After adjusting the model's threshold to achieve the same level of specificity of established gastric cancer screening tests such as a blood test and endoscopy, the researchers found that their model achieved comparable sensitivity to a blood test and was more sensitive than endoscopy.

"Compared to traditional screening tools, such as blood tests and endoscopy, our findings suggest that noncontrast CT scans demonstrate favorable accuracy as a new clinical tool for the detection of gastric cancer," the authors wrote. "This approach may be particularly promising due to its simplicity, accessibility, and no additional costs.

Those weighing the viability of opportunistic screening for gastric cancer using noncontrast CT scans will want to join this Sunday afternoon session.

Page 1 of 2
Next Page