The Application Study Of Deep Learning Assisted Magnetically Controlled Capsule Endoscopy Quality Control

Background:At present,the incidence of upper gastrointestinal diseases such as gastric cancer or ulcer is increasing,and gastroendoscopy is an increasingly important examination in surveillance.Nowadays gastroscopy and gastroscopic biopsy are the gold standard for the diagnosis of gastric lesion,but limitations of gastroscopy,such as pain and uncomfort,impose restrictions on its application in large-scale screening of the population.Besides,conventional endoscopy can’t inspect small intestine.Capsule endoscopy was firstly invented by Israeli scientists in 2000.After nearly 20 years of development,it has become a first-line diagnostic tool of gastrointestinal diseases.With the emergence of new capsule endoscopy,indications of capsule endoscopy has been extended from small intestine to esophagus,stomach and colon.At present,magnetically controlled capsule gastroscopy has been used to screen digestive tract diseases in many kinds of population.However,there is a temporary lack of effective supervision and quality control system for the operation of magnetically controlled capsule gastroscopy in China,and the operator can only improve the operation quality by communicating with endoscopists who are responsible for after-check reading.Therefore,it is of great significance to suggest an objective quality control system for the monitoring of the quality of magnetically controlled capsule gastroscopy.Objective:To construct a real-time quality-control system during the stage of magnetically controlled capsule endoscopy for supervision based on deep learning.Further randomized controlled trial was conducted to verify the clinical value of artificial intelligence technology for the quality control of magnetically controlled capsule gastroscopy.Methods:380 Patients from Qilu Hospital of Shandong University and Meizhao healthy technology and Mei nian One health Healthcare Holdings of Jvxian were retrospectively recruited.After labeling by experienced endoscopists according to the guidelines published in China,a realtime quality-control system including completeness with cleanliness and operation time report was developed based on the Resnet-50 model.With the help of real-time quality-control system,the operator can modulate the inspection process in terms of feedback information provided by quality-control system.The model was integrated to the operation system and a multi-center prospective randomized controlled trial was carried out to verify whether the quality control system intervention could improve the completeness of the inspection and calculate the inspection time as well as cleanliness.Results:The auxiliary quality control system of magnetically controlled capsule gastroscope was successfully integrated with the operating system of magnetically controlled capsule gastroscope.The average accuracy of the real-time quality-control system in identifying anatomical parts of the gastric is 85.35%,the average accuracy of system in recognizing mucosal cleanliness is 86.34%;the accuracy of timing magnet control time was 100%.Meanwhile this system can be integrated into the magnetic control operating system safely and efficiently.In a prospective randomized controlled trial,it was confirmed that the inspection time of magnetically controlled capsule gastroscopy was increased with the assistance of auxiliary quality control system,and the consistency rate of cleanliness judgment with gastroenterologists was 96.95%.Conclusion:The real-time quality-control system of magnetically controlled capsule endoscopy can achieve the quality control function in the real-time inspection process,and provide real-time feedback to the endoscopic operator besides provide guidance and help for endoscopic operation.