Introduction: In an analysis of videos recorded in a RCT comparing water exchange (WE) and air insufflation using computer-aided detection (CADe) algorithm, we showed that the strengths of CADe and WE complemented the weaknesses of each other (GIE. 2022;95:1198). CADe increased WE polyp and adenoma detection rate (ADR) and WE reduced false positive of CADe compared with the air insufflation. In the current study, we adopted adenomas per colonoscopy (APC) as an improved quality indicator (Dig Liver Dis. 2014;46:176; Transl Gastroenterol Hepatol. 2023;8:24). The APC is defined as the total number of adenomas resected divided by the total number of colonoscopies; higher rate decreases the incidence of post-colonoscopy colon cancer (GIE. 2023;20:S0016). CADe has been shown to increase APC compared with conventional air insufflation colonoscopy without CADe. In a US randomized trial, CADe improved adenoma detection (APC) for screening and surveillance colonoscopy without significantly increasing ADR (Gastroenterology. 2022;163:732). We applied the CADe model to conduct a randomized controlled real-time study to compare the ADR and APC between the WE+CADe and air insufflation+CADe in Taiwan subjects.
Method: By mimicking the methods established to identify diseases in plants, we developed a CADe model using convolutional neural network (CNN) with transfer learning approach (YOLOv4) to detect colon polyps. The CADe algorithm achieved a mean average precision of 94.0% and the area under receiver operating characteristic curve of 0.98. The sensitivity and specificity were set to 0.96 and 0.97, respectively. The positive predictive value was 0.98 and negative predictive value was 0.93. The algorithm was validated by a video analysis study. To assess the real-time application, 228 eligible patients were prospectively recruited to undergo real-time CADe colonoscopies inserted with either WE or air insufflation method. The resected polyps were evaluated by a blinded pathologist. (ClinicalTrial.gov number NCT05448300)
Results: The real-time CADe colonoscopies included patients inserted with either WE (n=114) or air insufflation (n=114) methods. Table 1. shows that the WE group had significantly higher adenoma detection rate (ADR), 60.5% versus 47.4%, P=0.046. The number of adenomas is higher in WE (n=174) compared with the air insufflation (n=125). The APC was significantly higher in WE group compared with the air insufflation group (1.53 versus 1.10, P=0.043), respectively.
Conclusion: This first real-time CADe comparison revealed ADR was significantly higher in WE than the air insufflation. The APC was significantly lower in the air insufflation group. The CADe algorithms confirmed our hypothesis of higher ADR and APC in the WE group in the randomized controlled real-time colonoscopy.

Chia-Pei Tang

Appearances