Accurate optical assessment of submucosal invasive cancer (SMIC) in large non-pedunculated colorectal polyps (LNPCPs) is crucial for selecting appropriate treatment. Current classification systems are complex and vary in accuracy. We developed a simple diagnostic algorithm combining gross morphological assessment (Blink impression: spontaneous bleeding, extra redness, depression, fold deformation, ulceration, chicken-skin mucosa), the presence of a ≥10 mm Paris 0-Is nodule and a validated simplified vascular pattern score (Colorectal Regular-Irregular Score, CRIS)¹. This multicentre prospective study evaluated the real-time accuracy of this algorithm for SMIC detection by trained endoscopists.

We conducted a prospective study across four centres (three Belgian, one Italian; August 2023–November 2025). Consecutive LNPCPs were assessed in vivo before resection. For each lesion, a structured approach was followed: (1) Blink impression, positive when ≥2 gross morphologic features present using high definition white light; (2) presence of a ≥10mm Paris 0-Is nodule; (3) CRIS as regular (R), regularly irregular (RI), or irregular (I), positive when RI or I using virtual chromoendoscopy plus up to 80x magnification. Blinded histopathologic identification of SMIC served as the reference standard. Diagnostic performance and odds ratios (OR) with 95% confidence intervals (CI) were calculated using generalized linear mixed models.

Overall, 240 LNPCPs from 240 patients were assessed by 10 endoscopists (3 experts, 7 fellows). SMIC prevalence was 20.8% (50/240), of which 70% (35/50) had deep SMIC. Median polyp size was 30mm (IQR 40–20); 33.8% were rectal.

For SMIC detection, Blink (≥2 features) achieved sensitivity of 76.0% (95%CI 61.8–86.9), specificity of 69.2% (62.0–75.7), and accuracy of 70.6% (64.4–76.4), with OR 2.3 (1.7–3.0; P<0.001). CRIS (RI/I) achieved sensitivity of 82.0% (68.6–91.4), specificity of 71.6% (64.6–77.9), and accuracy of 73.7% (67.7–79.2), with OR 11.5 (5.4–26.7; P<0.001).

The ≥10mm Paris 0-Is nodule showed poor diagnostic performance: sensitivity 40.0% (26.4–54.8), specificity 63.1% (55.9–70.0), accuracy 58.3% (51.8–64.6), with non-significant OR 1.1 (0.6–2.1; P=0.682) and was excluded from the final algorithm.

When either Blink or CRIS was positive, sensitivity increased to 84.0% (70.8–92.8), specificity was 61.6% (54.2–68.7), and accuracy 66.4% (59.9–72.4), with OR 8.4 (3.9–20.3; P<0.001).

This novel, simple diagnostic algorithm, where either Blink impression or vascular pattern assessment (CRIS) was positive achieved high sensitivity for SMIC detection. For general endoscopists evaluating LNPCPs, prioritizing high sensitivity is crucial to minimize the risk of missing malignancy. A sensitivity-first approach can naturally trigger thorough photo-documentation and clear signposting for multidisciplinary review, while the accompanying modest drop in specificity may be a reasonable and clinically acceptable trade-off. In this way, the algorithm functions as a practical, easy-to-apply aid for real-time decision-making, with the potential to shorten diagnostic pathways, reduce delays, and lessen the need for repeat procedures, ultimately easing the burden on patients.