Construction of artificial intelligence cloud platform for multi-center digestive endoscopy in Shandong Province (with video) / 中华消化杂志

ABSTRACT

Objective:Based on the artificial intelligence (AI) technology in endoscopy and the internet platform, to explore and construct a safe, standardized, scientific and rigorous database for digestive endoscopy, and to provide reference and evidence for the data quality control of AI in digestive endoscopy in China.Methods:After referring to relevant guidelines and standards, data collection and labelling standards of digestive endoscopy of 12 common gastrointestinal diseases were determined. The software of online collection and labelling of multi-center digestive endoscopy data in Shandong Province was developed. Endoscopic equipment with a domestic market share of >5% was used and dozens of experienced endoscopists from 9 medical centers in Shandong Province were uniformly trained for data labelling. From July 2019 to July 2020, the endoscopic examination data from 9 medical centers including Qilu Hospital of Shandong University, Shandong Provincial Hospital , Liaocheng People′s Hospital, Linyi People′s Hospital, Weihai Municipal Hospital, Taian City Central Hospital, Binzhou Medical University Hospital, Yantai Yuhuangding Hospital and Qilu Hospital of Shandong University (Qingdao) were prospectively and continuously collected and labeled. The optimized, desensitized, and generalized data were uploaded to the server. After the file synchronization, data processing, and expert review, a multi-center digestive endoscopy AI database with standard data collection and labelling in Shandong Province was constructed, namely cloud platform. Descriptive methods were used for statistical analysis.Results:The collection and labelling standards for multi-center digestive endoscopy AI data in Shandong province was established. The software of online collection and labelling of multi-center digestive endoscopy AI data in Shandong province was developed. The database in Shandong province was successfully constructed. In the database, 43 010 lesions, 40 353 images, and 11 289 examinations were labeled. Among them, there were 2 906 cases of early esophageal cancer, 2 912 cases of early gastric cancer, 2 397 cases of early colorectal cancer, and 9 773 cases of colorectal polyps (5 539 cases of adenomatous polyps, 1 161 cases of non-adenomatous polyps and 3 073 case of undetermined polyps).Conclusions:The multi-center AI cloud platform for digestive endoscopy in Shandong Province adopts unified standards and collection and labeling software, which ensures the safety and standardization of endoscopy data. It provides a reference and basis for the construction of a quality control system for standardized data collection and labelling of digestive endoscopy AI data in our country and for the third-party data supervision.