Diagnostic performance of the Japan Narrow-Band Imaging Expert Team and pit pattern classifications for small polyps using next-generation endoscopy

Background and study aims Optical diagnosis of small colorectal lesions is important for the potential implementation of “resect-and-discard” and “leave-in-situ” strategies. We evaluated the performance of the Japan Narrow-Band Imaging Expert Team (JNET) and pit pattern classifications in differentiating small adenomatous and serrated colorectal lesions using a modern high-definition endoscopy system (EVIS X1), with outcomes stratified by endoscopist expertise. Patients and methods This secondary analysis of the prospective multicenter deTXIon trial included 1,830 colorectal lesions for JNET evaluation, of which 92.9% measured <10 mm, and 1,601 lesions for pit pattern evaluation. Adenocarcinomas were excluded. Sensitivity, specificity, positive and negative predictive values, and likelihood ratios were calculated. To account for multiple lesions within the same patient, 95% confidence intervals (CIs) were estimated using a cluster bootstrap approach with patients as the resampling unit. Exploratory expert–non-expert differences in sensitivity, specificity, and negative predictive value were also estimated. Results JNET showed high diagnostic performance, with no clear between-group differences in sensitivity, specificity, or negative predictive value. For pit pattern classification, expert–non-expert differences were 5.01 percentage points for sensitivity (95% CI, 1.23–9.08) and 10.00 percentage points for negative predictive value (95% CI, 1.37–18.56). Conclusions JNET showed similar observed diagnostic performance across expertise levels, whereas pit pattern classification showed greater between-group differences in sensitivity and negative predictive value. These findings suggest that JNET may serve as a practical framework for routine optical diagnosis of small lesions.