FDA grants breakthrough status to AI-based gastric cancer diagnostic

Dive Brief:

FDA granted breakthrough device designation to AI Medical Service's technology for diagnosing gastric cancer, the company said Monday.
The technology uses artificial intelligence to analyze endoscopy images for signs of gastric cancer, a hard-to-diagnose disease associated with a high rate of false negatives.
AI Medical Service, which raised $42.9 million last month, is working to bring the technology to the U.S. and its home market of Japan.

Dive Insight:

Gastric cancer is difficult to diagnose in the early stages of the disease. Studies have linked the diagnostic endoscopic procedure esophagogastroduodenoscopy (EGD) with a false negative rate as high as 25%.

EGD is a procedure during which a small flexible endoscope is introduced through the mouth or nose and advanced through the pharynx, esophagus, stomach and duodenum. The rate of false negatives stems from the resemblance between gastric cancer and atrophic gastritis.

Last year, AI Medical Service collaborated on a research paper that detailed a new, potentially better way to diagnose gastric cancer. The authors of the paper believe the study marked the first time a convolutional neural network had been used to diagnose gastric cancer from endoscopy images.

Engineers at AI Medical Service and their collaborators trained the CNN-based diagnostic system on more than 13,000 endoscopic images of confirmed cases of gastric cancer. The trained AI was then tested on more than 2,000 images taken from 69 patients.

AI Medical Service's technology correctly diagnosed 71 out of the 77 gastric cancer lesions, resulting in a sensitivity of 92%. According to the authors, "all missed lesions were superficially depressed and differentiated-type intramucosal cancers that were difficult to distinguish from gastritis even for experienced endoscopists."

The CNN also wrongly flagged 161 non-cancerous lesions as gastric cancer. That level of false positive results pushed the probability of a positive result indicating a patient had gastric cancer down toward 30%. The authors said that probability would be "clinically acceptable" given the accuracy of biopsies and belief that false negatives, not false positives, are the big problem in cancer.

AI Medical Service has since collaborated on other studies that have used CNN to detect small‐bowel angioectasia and analyze small‐bowel capsule endoscopy videos.

The research positions AI Medical Service to push its technology toward the market, joining a wave of companies including Medtronic and Caption Health that want to use AI to improve imaging. Last month, AI Medical Service outlined plans to use the proceeds from a financing round to run clinical trials with a view to winning regulatory authorizations around the world.

Securing the breakthrough status will support the U.S. part of that strategy by enabling AI Medical Service to receive additional input from FDA and benefit from an expedited regulatory review.