The Japan NBI Expert Team (JNET) classification provides a structured approach to predicting histology based on narrow-band imaging features that can be utilised in combination with the Paris  and laterally spreading tumour (LST) morphological classification for the assessment of advanced colorectal polyps. The accuracy and interobserver variability amongst United Kingdom (UK) clinicians for JNET and descriptors of morphology is uncertain. We aimed to 1) assess the diagnostic accuracy of the JNET classification among endoscopists, 2) evaluate interobserver variability for the JNET and morphology classification systems, 3) determine whether grouping the morphology categories improves agreement among endoscopists and 4) assess whether the level of endoscopic experience influences diagnostic accuracy and interobserver agreement.

An online image-based survey was distributed for 12 weeks in 2025 to practising endoscopists. The survey contained a total of 15 polyps derived from a previously validated database of images with a JNET classification consensus by JNET core members. Each polyp was displayed sequentially in white-light, standard narrow-band, and magnified narrow-band imaging modes. Respondents were asked to assign a JNET classification and morphology (six predefined options) for each case. The JNET consensus diagnosis served as the gold standard for JNET classification. The survey was advertised through The Joint Advisory Group on Gastrointestinal Endoscopy (JAG) newsletter, personal communications, as well as through social media pages.

Responses were analysed for diagnostic accuracy and interobserver agreement using Fleiss’ Kappa (κ). Subgroup analyses were performed by endoscopic experience (<300, 300–3000, >3000 lifetime procedures). Lastly, we assessed whether grouping the morphology into polypoid (Is, Is+IIa and LSTGM), flat (IIa, LSTNGFE and LSTGH) and those with a depressed component (IIc,IIc+IIa and LSTNG PD) would impact agreement.

A total of 129 endoscopists from the UK completed the survey. Distribution by experience was: <300 procedures (n=18), 300–3000 (n=68) and >3000 (n=42).

JNET classification:

Overall diagnostic accuracy was 58% when compared with the JNET consensus. Accuracy by experience group was: <300 (56%), 300–3000 (56%) and >3000 (59%).

Overall interobserver agreement for JNET classification was fair (κ = 0.34). Agreement generally improved with increasing experience: <300 (k=0.25), 300–3000 (k=0.32) and >3000 (k=0.42). 

Morphology classification:

Initial interobserver agreement was poor (κ=0.15 overall). Agreement improved with experience, but this improvement was primarily observed in the most experienced group: <300 κ=0.16;  300-3000 κ=0.14; and  >3000 κ=0.21.

After grouping the morphology, overall agreement increased to κ = 0.27, with improvements seen across all experience levels. The most significant gains were observed at the highest experience level: <300 cases, κ = 0.23; 300-3000 κ = 0.24; and >3000 κ = 0.37.

In this UK survey, the diagnostic accuracy of JNET classification among endoscopists was moderate (58%), with fair interobserver agreement (κ = 0.34). Accuracy and agreement both tended to improve with greater endoscopic experience. Morphology assessment showed only slight agreement (κ = 0.16), though grouping of the classification improved reproducibility. This suggests that reducing the category complexity enhances consistency amongst endoscopists.

These findings highlight persistent variability in the optical diagnosis of colorectal polyps with the use of JNET and morphological classification. This illustrates the need for standardized national training with focused image interpretation workshops and feedback. In addition, a grouping of morphological description may improve diagnostic consistency.