A Series Of Studies On Improving Colonoscopy Quality And Validation Of Artificial Intelligence Assisted Colorectal Polyp Detection System

Section 1 Impact of Wechat and Short Message Service on Bowel Preparation:an Endoscopist-blinded,Randomized Controlled TrialBackground: Adequate bowel preparation is essential for a successful colonoscopy;clinical studies suggest reinforced education can improve the preparation process.However,there have been no trials to compare Wechat directions(the most widely used social media app in China)with those of the Short Message Service(SMS).This study is aimed to assess the impact of Wechat and SMS on the bowel preparation quality.Methods: This was a single-center,prospective,endoscopically-blinded,randomized,controlled study.Patients in reinforced education groups received additional reminder messages by Wechat and SMS two days before colonoscopy.The primary outcome was bowel preparation quality evaluated by the Boston Bowel Preparation Scale(BBPS)score and the rate of adequacy(BBPS score ? 6).Secondary outcomes included polyp detection rate(PDR),adenoma detection rate(ADR),and mean total adenomas detected.Patient tolerance level and subjective feelings were also collected.Results: The total BBPS score and the percentage of adequacy were significantly higher in the reinforced education groups compared to control(Wechat vs.control,P<0.001;SMS vs.control,P<0.001).Moreover,statistically significant differences between the two interventions were found in the total BBPS score,but not in the rate of adequacy(P=0.007,P=0.561).The detection of adenomas,using multiplicity detection rate,advanced ADR,was much higher in the intervention groups(P=0.039,0.037,and 0.019).Conclusion: Wechat was superior to SMS for bowel preparation,although both of them may help improving the detection of adenomas.Section 2 Effect of left lateral tilt-down position on cecal intubation time: a 2-center,pragmatic,randomized controlled trialBackground: Colonoscopy insertion is technically challenging,time-consuming and painful,especially for the sigmoid.Several pilot studies indicated(left)tilt-down position could facilitate insertion procedure,but no formal trials have been published to demonstrate its efficacy.We performed the study to verify the benefits of left lateral tilt-down position(LTDP)on insertion process.Methods: The two-center prospective trial(NCT02842489)randomized unsedated patients to LTDP or left lateral horizontal position(LHP)to aid insertion.The primary outcome measure was cecal intubation time(CIT).The secondary outcome measures included decending colon intubation time(DIT),pain score of insertion,acceptance to unsedated colonoscopy in the future exam,difficulty score of insertion,and the complication rates of colonoscopy.Results: 258 patients were randomized to the LTDP(128)or LHP(130)in two centers.The median CIT and DIT were shorter with patients positioned in LTDP than in LHP(CIT: 280.0 vs.339.5 s,P<0.001;DIT: 53.0 vs.69.0 s,P<0.001,respectively)and patients with high and low body mass index(BMI)benefited more from LTDP than from LHP,as opposed to patients with normal BMI.In addition,colonoscopy insertion in LTDP was also less painful(3.4±1.6 vs.4.0±1.7,P=0.02)and less difficult(3.1±1.9 vs.3.7±1.4,P ? 0.001),showing a higher tendency to accept unsedated colonoscopy(82.9% vs.73.8%,P=0.08).The rates of complications were extremely low and did not differ significantly in two groups.Conclusions: LTDP for colonoscopy insertion can reduce insertion time and pain,and potentially improves patients' acceptance of unsedated colonoscopy.Clinical Trials.gov number,NCT02842489.Section 3 Establishment and Real-world Validation of Computer-assisted Polyp Identification and Localization System Based on Deep LearningBackground: Several deep learning based artificial intelligence(AI)assisted-diagnosis systems(ADSs)were reported to real-time identify polyps in colonoscopic images and selected videos,while their diagnostic performance and real value has not been validated in clinical practice.Therefore,we trained and tested an ADS with the largest high-quality dataset of polyps,and preliminarily validated it in real-world colonoscopy.Methods: We trained and tested the ADS in separate datasets randomly divided from the of 116040 colonoscopic images.In real-world colonoscopy,we preliminarily validated its diagnostic performance different scenarios,verified the real value in detecting additional polyps and adenomas,and analyzed potential causes of false positives and negatives.Results: In the test dataset of 12419 images,ADS could identify and localize the colorectal polyps with a 95.0% sensitivity,99.1% specificity,96.9% accuracy and 0.972 area under curve within 30 milliseconds.In preliminarily validation of real-world colonoscopy,the sensitivity of the ADS in identifying polyps was significantly higher that of colonoscopists(98.4% vs.91.0%,P=0.03).With the aid of ADS,colonoscopists could detect more polyps and adenomas(0.90 vs.0.82,P? 0.001;0.32 vs.0.30,P?0.05),particularly for polyps ?5 mm and flat(0.65 vs.0.57,P? 0.001;0.74 vs.0.67,P=0.001,respectively),but its efficacy could not be reproduced in patients with inadequate bowel preparation quality and withdrawal time(P=0.32;P=0.16,respectively).Conclusions: In preliminarily validation of real-world colonoscopy,ADS can identify polyps with a high sensitivity,help detect more polyps and adenomas,and potentially reduce polyp miss rate.