Baishideng Publishing Group Inc, 7041 Koll Center Parkway, Suite 160, Pleasanton, CA 94566, USA
Share the Article
Zhang Y, Yan S, Li YJ, Li QX, Wei LJ. Deep learning for predicting high-grade dysplasia in colorectal polyps using white-light endoscopy. World J Gastroenterol 2027; 33(3): 125206 [DOI: 10.3748/wjg.125206]
Yue Zhang, Shuang Yan, Li-Juan Wei, Department of Gastroenterology and Digestive Endoscopy Center, The Second Hospital of Jilin University, Changchun 130000, Jilin Province, China
Ying-Jie Li, Department of Gastroenterology, The First Affiliated Hospital of Jinzhou Medical University, Jinzhou 121001, Liaoning Province, China
Qian-Xun Li, Department of Gastroenterology, Jilin Provincial People’s Hospital, Changchun 130000, Jilin Province, China
Author contributions: Zhang Y and Wei LJ designed and analyzed the study; Li QX and Li YJ critically revised the article for important intellectual content; Yan S approved the final manuscript. All authors contributed to the article and approved the submitted version.
AI contribution statement: Not applicable.
Institutional review board statement: This study was approved by the Medical Ethics Committee of the Second Hospital of Jilin University, No. 2025-471-1.
Informed consent statement: Patients were not required to give informed consent to the study because the analysis used anonymous clinical data that were obtained after each patient agreed to treatment by written consent.
Conflict-of-interest statement: All the authors report no relevant conflicts of interest for this article.
STROBE statement: The authors have read the STROBE Statement-checklist of items, and the manuscript was prepared and revised according to the STROBE Statement-checklist of items.
Data sharing statement: The datasets and source code are available from the corresponding author upon reasonable request.
Corresponding author: Li-Juan Wei, MD, PhD, Director, Full Professor, Department of Gastroenterology and Digestive Endoscopy Center, The Second Hospital of Jilin University, No. 218 Ziqiang Street, Nanguan District, Changchun 130000, Jilin Province, China. 1835529876@qq.com
Received: July 3, 2026 Revised: July 31, 2026 Accepted: September 20, 2026 Published online: January 21, 2027 Processing time: 138 Days and 5.2 Hours
Abstract
BACKGROUND
Accurate identification of high-grade dysplasia (HGD) in colorectal polyps is important for optimizing endoscopic management and surveillance strategies. Although deep learning has shown potential for colorectal polyp characterization, the generalizability of white-light imaging (WLI)-based models across independent clinical settings remains insufficiently established.
AIM
To develop and externally validate a deep learning model based on routine WLI images for predicting HGD in colorectal polyps, to explore the predictive contribution of routinely available clinical variables, and to compare model performance with endoscopists with different levels of experience.
METHODS
In this multicenter retrospective observational study, a MobileNetV2-based deep learning model was developed using routine WLI images of colorectal polyps and externally validated in two independent tertiary hospitals. Each polyp was represented by one representative high-quality WLI image, and histopathological diagnosis was used as the reference standard. Model discrimination, calibration, decision curve analysis, and diagnostic performance were evaluated. An exploratory clinical model based on routinely available clinical and lesion characteristics was developed as a baseline comparator. The model was further compared with endoscopists using binary HGD/non-HGD classifications in a blinded reader study.
RESULTS
The WLI-based model achieved an area under the receiver operating characteristic curve of 0.879 [95% confidence interval (CI): 0.795-0.952] in the internal test cohort and maintained consistent discrimination in two external validation cohorts [area under the receiver operating characteristic curve, 0.814 (95%CI: 0.728-0.890) and 0.803 (95%CI: 0.719-0.896)]. The exploratory clinical model provided no clear evidence of incremental predictive value beyond image-based prediction. In the reader study, the artificial intelligence model demonstrated comparable classification performance with senior endoscopists, and superior performance compared with junior endoscopists based on binary diagnostic metrics. However, the external validation cohorts contained a limited number of HGD-positive cases, and performance estimates should therefore be interpreted with caution.
CONCLUSION
A deep learning model based on routine WLI images showed promising and consistent performance for HGD risk stratification in colorectal polyps across three tertiary hospitals. The model may provide complementary information to support endoscopic assessment, particularly for less experienced clinicians. Prospective real-time validation in broader populations and healthcare settings is required before clinical implementation.
Core Tip: We developed and externally validated a deep learning model based on routine white-light endoscopy for predicting high-grade dysplasia in colorectal polyps. The model maintained stable performance across two independent external cohorts and achieved diagnostic accuracy comparable to experienced endoscopists. Because it relies only on routinely acquired white-light images, the proposed approach may facilitate broad clinical implementation as a decision-support tool for colorectal polyp risk stratification.
