Artificial Intelligence Assisted Standard White Light Endoscopy Accurately Characters Early Colorectal Cancer: A Multicenter Diagnostic Study
Sijun Meng,Yueping Zheng,Ruizhang Su,Wangyue Wang,Yu Zhang,Hang Xiao,Zhaofang Han,Wen Zhang,Wenjuan Qin,Chen Yang,Yan Luo,Haineng Xu,Yemei Bu,Yuhuan Zhong,Yi Zhang,Yulong He,Hesong Qiu,Wen Xu,Hong Chen,Siqi Wu,Jiang Zhenghua,Yongxiu Zhang,Chao Dong,Yongchao Hu,Lizhong Xie,Xugong Li,Jianping Jiang,Huafen Zhu,Wenxia Li,Zheng Wen,Xiaofang Zheng,Yuanlong Sun,Limin Ding,Xiaolu Zhou,Changhua Zhang,Wensheng Pan,Shuisheng Wu,Yiqun Hu
DOI: https://doi.org/10.2139/ssrn.3514689
2020-01-01
Abstract:Background: Early detection and treatment under endoscopy have effectively decreased the mortality of colorectal cancer (CRC). But it is difficult to detect early CRC under white light endoscopy (WLE). Artificial intelligence (A.I.) assisted diagnose system are accumulated, but early CRC assisting diagnosis system is lack due to clinical design and sample size limitation. We propose to establish a propagable AI-assisted early CRC diagnosis and characterization system (ECRC-CAD).Methods: Establishing algorithms were implemented in four (i.e., municipal, provincial, rural) hospitals of China. All patients who completed WLE from January 2016 to July 2019 were enrolled. Early CRC is diagnosed by pathological examination. Images of intraepithelial neoplasm (IN-ECRC) were included in the case group, and precancerous conditions and serrate lesion without intraepithelial neoplasm (NIN-ECRC) were randomly sampled in corresponding quantity included to control group. These images were randomly assigned (8:1:1) to the training, optimizing datasets and internal datasets. The evaluating indexes of diagnostic performance include sensitivity, specificity, positive predictive value, and negative predictive value.Findings: The largest new CRC endoscopic image database to date were established, which included 4,390 images of IN-ECRC. The diagnostic accuracy was 0·963 (95% CI, 0·941-0·978) in the internal validation set, and 0·835 (95% CI, 0·805-0·862) of identifying IN-ECRC in external datasets. ECRC-CAD achieved better diagnostic accuracy than the expert endoscopist (0·885 [95% CI, 0·859-0·907] vs 0.505 [95% CI, 0·411–0·598]; p<0.0001). The sensitivity, specificity, positive, and negative predictive value were significantly higher for ECRC-CAD than the expert endoscopist (p<0.0001).Interpretation: ECRC-CAD can identify early CRC and characterize neoplasms with high malignant potential; the performance is better comparing with expert endoscopists. Spreading of ECRC-CAD to regions with different medical levels can assist in CRC screening and prevention. Funding Statement: This research was supported by Major Science and Technology Project of Zhejiang Provincial (2020C03030), Natural Science Foundation of Fujian Province (2019-CXB-31), Science and technology program of Guangdong Province (2016A020213002), and National Natural Science Foundation of China (81773956).Declaration of Interests: The authors declare no competing interests.Ethics Approval Statement: Each participating institute has ethical approval by independent ethics Committee according to the Helsinki declaration. The informed consent of retrospective cohort was exempted by the institutional review boards as the secondary use of clinical data without any intervention. In the prospective study, each patient was informed and signed informed consent.