The Japan NBI Expert Team (JNET) classification is widely used to support optical diagnosis and guide treatment strategies for colorectal lesions [1]. However, most available evidence derives from Asian cohorts [2-3], and real-world performance across the full JNET spectrum in Western populations remains unclear [4]. This study aimed to evaluate the diagnostic accuracy of JNET subtypes in predicting histologic features and depth of invasion in a large multicenter European cohort of colorectal lesions treated with ESD.

We conducted a retrospective multicenter study including consecutive patients who underwent colorectal ESD in 19 European referral centers between 2010 and 2025. Lesions were eligible if they had a pre-resection chromoendoscopic assessment with documented JNET classification and complete post-ESD histopathology. Lesions without JNET classification or complete histologic data were excluded. Lesions were categorized as: JNET 1 (non-neoplastic/serrated), JNET 2A (low-grade intramucosal neoplasia), JNET 2B (high-grade intramucosal neoplasia or superficial submucosal invasion), and JNET 3 (deep submucosal invasion). Diagnostic performance metrics (sensitivity, specificity, PPV, NPV, accuracy) were calculated using histology as the reference standard.

A total of 2193 lesions were analyzed (49 JNET 1, 1281 JNET 2A, 817 JNET 2B, 46 JNET 3). For JNET type 1, the sensitivity, specificity, PPV, and NPV for predicting non-neoplastic lesions were 31%, 98%, 24%, and 99%, respectively, with an overall accuracy of 97%. Notably, among JNET 1 lesions, 26.5% showed high-grade dysplasia and 10.2% superficial submucosal adenocarcinoma (sm1).For JNET type 2A predicting low-grade dysplasia, the sensitivity and specificity were 77% and 68%, with a PPV of 58%, NPV of 84%, and accuracy of 71%. Importantly, 34.2% of JNET 2A lesions harbored high-grade dysplasia or sm1 carcinoma, and 3.7% showed deep submucosal invasion (sm2-sm3).For JNET type 2B predicting high-grade dysplasia or sm1 carcinoma, sensitivity was 52%, specificity 79%, PPV 61%, NPV 72%, and accuracy 69%. Among these lesions, 12.4% showed deep submucosal adenocarcinoma. Histology was overestimated in 26.4% of cases (final diagnoses included sessile serrated lesions with LGD, adenomas with LGD, or traditional serrated lesions).For JNET type 3 predicting deep submucosal invasion, sensitivity was 18%, specificity 99%, PPV 72%, NPV 93%, and accuracy 93%.En-bloc resection was achieved in 90.1%, ensuring high reliability of histopathologic assessment.

The JNET classification is widely used to classify colorectal lesions and guide therapeutic decisions. However, in this large real-world European cohort, diagnostic performance in the intermediate categories (JNET 2A and 2B) was suboptimal, with a non-negligible risk of underestimating advanced neoplasia, including deep submucosal invasion. These findings suggest that treatment decisions should not rely solely on JNET assessment, but rather integrate complementary diagnostic tools and clinical judgment to avoid inappropriate management.