A deep learning artificial intelligence (AI) system has been developed to detect gastric cancers, and achieved over 90 percent accuracy in analysis of more than 2,000 endoscopic images in under a minute, a team made up of mostly Saitama and Tokyo-based researchers has announced in the journal "Gastric Cancer."
Not only is the AI system expected to lessen the examination burden on endoscopy physicians, but because the AI makes decisions in real time based on the images, it can also possibly be utilized so doctors have more time to investigate suspected cancerous areas in more detail.
The AI was a joint project developed by Tomohiro Tada of the Saitama-based Tada Tomohiro Institute of Gastroenterology and Proctology, the Tokyo-based Cancer Institute Hospital in Ariake, Tokyo, and other facilities. The team used 13,584 endoscopic images from the two facilities to have the AI learn to identify cancer by image recognition through a process known as deep learning. After teaching the AI, it was tasked to process 2,296 endoscopic images from 69 patients, and it accurately identified cancerous areas in the patients at a rate of 92.2 percent. When extremely early stage cancer with a diameter of 5 millimeters or less was excluded, the AI's accuracy rate jumped to 98.6 percent. It took the AI only 47 seconds for the diagnoses.
In an endoscopic examination for gastric cancers, physicians must carefully distinguish between reddish, swollen or slightly bumpy inflammation of the stomach wall and cancerous areas. Gastric cancer 5 millimeters or less in size is reportedly difficult to spot even for doctors who specialize in endoscopy.
"There is a high possibility that the AI's diagnostic accuracy has exceeded that of expert physicians," said Tada. "We will begin clinical trials this spring, and aim to put the AI system into practical use at an early date." The team has already developed an AI system that can recognize stomach cancer-precursor gastritis caused by the bacteria Helicobacter pylori on the same level as field experts.