Prospective Multicenter Study of Artificial Intelligence-assisted Quality Evaluation System for Colonoscopy

Colonoscopy is a key technique for detection and diagnosis of Lower Gastrointestinal Diseases. High-quality endoscopy leads to better outcomes, However, the skill level of different endoscopists is quite different. The rate of missed diagnosis of colorectal ADENOMAS was 22%, the rate of missed diagnosis ranged from 7.4% to 52.5% . A series of studies showed that 12.1% of patients did not have cancer during colonoscopy but were diagnosed with colorectal cancer (post-colonoscopy Colorectal Cancer, PCCRC) within 5 years, of which 71%-86% could be attributed to the negligence of the endoscopist 6, seven. Therefore, it is very important to improve the quality of routine endoscopic examination while expanding the endoscopic technique.

The quality control index of colonoscope can reflect the operation level of endoscopy doctors, and it is a very important tool to ensure the quality of colonoscope. The examination and feedback of the quality control data of the endoscopist can effectively improve the detection rate of lesions and guarantee the high level of endoscopic quality. It is the key to guarantee the quality of enteroscopy to put forward the quality control index which can reflect the quality of endoscopist operation. Barclay proposed in 2006 that there was a significant correlation between colonoscopy withdrawal time and the rate of Adenomas in patients, and their findings were published in the The New England Journal of Medicine, with an impact factor of 70.67. Then, there were 367 papers on the study of the speed of enteroscopy withdrawal. Aslinia made a retrospective analysis of the correlation between the incidence of blindness reached by enteroscopy and the prognosis of patients, and proposed the incidence of blindness reached as a quality control index for enteroscopy. Other quality control indicators, such as bowel cleanliness and polyp detection rate, have been published in international well-known journals and included in the guidelines to regulate the quality of daily work of endoscopic physicians, which has great clinical significance.

However, despite the above-mentioned quality control indicators, the existing endoscopic quality control indicators are seriously inadequate. According to a multicenter clinical study, current quality control indicators only reflect about 40% of the difference in lesion detection levels between endoscopic physicians. The existence of this phenomenon indicated that more quality control indexes are needed to fully reflect the quality of endoscopic procedures. As a quality control index reflecting intestinal preparation, intestinal cleanliness is a fatal defect with strong subjectivity and poor consistency among observers, therefore, the existing colonoscopy need more comprehensive, objective quality control indicators to reflect the level of endoscopic doctors, so as to ensure the quality of colonoscopy.

In the past decade, artificial intelligence (AI) has made remarkable progress in the medical field. Andre Esteva used a deep neural network (DNN) to classify skin cancer with expert-level accuracy. In the field of digestive endoscopy, the project team in the early is made a huge breakthrough. In 2018, the artificial intelligence gastroscopy blind spot monitoring model developed by the project team significantly reduced the endoscopy blind spot in randomized controlled clinical trials, the findings were published in the journal Gut with an impact factor of 17.94. In 2019, the team's artificial intelligence model for evaluating gut cleanliness, published in the journal gastrogastrotestinal endoscopy with an impact factor of 7.23, showed better picture accuracy than that of the endoscopist in the human computer competition. In the same year, the project developed a model for colonoscopy speed monitoring based on Hasche perception algorithm and a model for colonoscopy slide mirror monitoring based on image classification, and conducted randomized controlled clinical trials, the model significantly improved The ability of colonoscopists to detect lesions, The study was published in The Lancet Gastroenterology 18 with an impact factor of 12.87.

Based on the above-mentioned rich foundation of early work, as well as the current great demand in the field of quality control of enteroscopy, we intend to validate the artificial intelligence-based enteroscopy exit velocity, bowel cleanliness, and enteroscopy slide model developed by the project team through a multi-center clinical study, to evaluate the feasibility and accuracy of doctor's operation quality as a quality control index.