Your next colonoscopy could be analysed by AI

A new artificial intelligence system can spot the early signs of cancer from a colonoscopy. The system was developed in Japan, and spotted 90% of colorectal adenomas, tumours that can be precursors to cancer, in recent tests.

The computer-aided system looked at 250 images, analysing them for bowel polyps. It is hoped eventually the system will be used in real time during a colonoscopy, to help diagnose diseases like bowel cancer.

“This AI is designed to instantly analyse roughly 300 image-features from a colorectal polyp, outputting its pathological prediction with an accuracy of approximately 90%,” Yuichi Mori, assistant professor at Showa University Northern Yokohama Hospital, in Yokohama, Japan, told Alphr.

Overall, 306 polyps were assessed real-time by using the AI-assisted system, providing a sensitivity of 94% and accuracy of 86%. Mori and his colleagues developed the AI system to be used alongside a super-magnifying endoscope, which magnifies the image of the inside of the colon to 500 times its size.

“We expect this system can be used on patients by the end of 2018 in Japan,” Mori said. “To achieve this goal, we are currently conducting a multicentre clinical trial for obtaining the drug regulatory approval.”

This could be the start of a wider trend of AI being used in medicine for diagnosis of various test results, the authors say.

“In my view, AI is especially beneficial for non-expert endoscopists,” Mori said. “However, we should also consider the limitation of AI technology for medical imaging.”

AI learns by being fed data, and the data that are available to feed medical AI are lacking compared to other industries that have been using AI for longer.

“The amount of the learning material for AI is significantly smaller in the medical field than that in the other fields such as automatic driving,” Mori said. “From this point of view, whether AI for endoscopy exceeds experts’ performance is still a questionable issue.”