Validation study for development of the Japan NBI Expert Team classification of colorectal lesions

TLDR

The Japan narrow‑band imaging (NBI) Expert Team (JNET) was formed to unify four earlier magnifying NBI classifications (Sano, Hiroshima, Showa, and Jikei). The authors conducted a multicenter validation study of the NBI scale to develop the JNET classification of colorectal lesions. The JNET working group created the NBI scale for vessel and surface pattern evaluation and validated it in a multicenter study where 25 expert colonoscopists assessed 100 still NBI images with and without magnification to determine the necessity of each criterion for diagnosis. The study found that surface pattern was necessary in over 60 % of polyp diagnoses and vessel pattern in about 90 %, identified three key findings for type 2B and three for type 3, yielding sensitivities of 44.9 % and 54.7 % and a 97.4 % specificity for type 3, with macroscopic type having only a minor effect, leading to consensus on the JNET classification.

Abstract

Background and Aim The Japan narrow‐band imaging (NBI) Expert Team (JNET) was organized to unify four previous magnifying NBI classifications (the Sano, Hiroshima, Showa, and Jikei classifications). The JNET working group created criteria (referred to as the NBI scale) for evaluation of vessel pattern (VP) and surface pattern (SP). We conducted a multicenter validation study of the NBI scale to develop the JNET classification of colorectal lesions. Methods Twenty‐five expert JNET colonoscopists read 100 still NBI images with and without magnification on the web to evaluate the NBI findings and necessity of the each criterion for the final diagnosis. Results Surface pattern in magnifying NBI images was necessary for diagnosis of polyps in more than 60% of cases, whereas VP was required in around 90%. Univariate/multivariate analysis of candidate findings in the NBI scale identified three for type 2B (variable caliber of vessels, irregular distribution of vessels, and irregular or obscure surface pattern), and three for type 3 (loose vessel area, interruption of thick vessel, and amorphous areas of surface pattern). Evaluation of the diagnostic performance for these three findings in combination showed that the sensitivity for types 2B and 3 was highest (44.9% and 54.7%, respectively), and that the specificity for type 3 was acceptable (97.4%) when any one of the three findings was evident. We found that the macroscopic type (polypoid or non‐polypoid) had a minor influence on the key diagnostic performance for types 2B and 3. Conclusion Based on the present data, we reached a consensus for developing the JNET classification.