Artificial intelligence assisted standard white light endoscopy accurately characters early colorectal cancer: a multicenter diagnostic study

Colorectal cancer (CRC) is the third in incidence and mortality of cancer. Screening with colonoscopy has been shown to reduce mortality by 40-60%. Challenge for screening indistinguishable precancerous and noninvasive lesion using conventional colonoscopy was still existing. We propose to establish a propagable artificial intelligence assisted high malignant potential early CRC characterization system (ECRC-CAD). 4,390 endoscopic images of early CRC were used to establish the model. The diagnostic accuracy of high malignant potential early CRC was 0.963 (95% CI, 0.941-0.978) in the internal validation set and 0.835 (95% CI, 0.805-0.862) in external datasets. It achieved better performance than the expert endoscopists. Spreading of ECRC-CAD to regions with different medical levels can assist in CRC screening and prevention.