Related Experiment Video
Updated: Oct 22, 2026
Comparison of Predictive Performance of Three Lymph Node Staging Systems in Colorectal Signet Ring Cell Carcinoma Based on Machine Learning Model
Published on: April 18, 2025
A white-light endoscopy-based mixed-effects nomogram for predicting small colorectal polyp neoplasia in
Introduction:
Accurate optical characterization of small colorectal polyps relies on narrow-band imaging (NBI) or artificial intelligence, both of which are unavailable in resource-limited settings.
Objectives:
We aimed to develop and validate a web-based nomogram using white-light endoscopy (WLE) and routine clinical variables to predict neoplastic polyp histology.
Patients And Methods:
This prospective single-center study (Vietnam, September 2020-April 2021) included 301 patients with 451 small colorectal polyps (≤10 mm). Polyps were assessed in real time using WLE and NBI before histological confirmation. Data were split at the patient level into training (254 patients, 383 polyps) and validation (47 patients, 68 polyps) cohorts. L1-penalized generalized linear mixed-effects models were used for feature selection and model development.
Results:
Neoplastic histology was present in 55.9% of polyps. Four predictors were identified: sessile/pedunculated morphology (0-Is/0-Ip), age ≥50 years, polyp size >5 mm, and neoplastic WLE diagnosis. In internal validation, the WLE+3F model achieved an area under the curve (AUC) of 0.849, outperforming WLE alone (AUC, 0.760; P = 0.009) and comparable with NBI (AUC, 0.858; P = 0.86), with sensitivity of 65.8%, specificity of 90.0%, and good calibration (Brier score, 0.160). Decision curve analysis showed clinical net benefit.
Conclusions:
A WLE-based nomogram combined with 3 clinical factors achieved performance comparable with NBI, offering an accessible tool for real-time risk stratification in resource-limited settings; external validation is warranted.
