Risk prediction model for precancerous gastric lesions based on magnifying endoscopy combined with narrow-band imaging features

BackgroundThis study aimed to construct and validate diagnostic models for the Operative Link on Gastritis Assessment (OLGA) and Operative Link on Gastric Intestinal Metaplasia Assessment (OLGIM) staging systems using three different methodologies based on magnifying endoscopy with narrow-band imagi...

Full description

Saved in:
Bibliographic Details
Main Authors: Jingna Tao, Zhongmian Zhang, Linghan Meng, Liju Zhang, Jiaqi Wang, Zhihong Li
Format: Article
Language:English
Published: Frontiers Media S.A. 2025-04-01
Series:Frontiers in Oncology
Subjects:
Online Access:https://www.frontiersin.org/articles/10.3389/fonc.2025.1554523/full
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1850272161689763840
author Jingna Tao
Zhongmian Zhang
Linghan Meng
Liju Zhang
Jiaqi Wang
Zhihong Li
author_facet Jingna Tao
Zhongmian Zhang
Linghan Meng
Liju Zhang
Jiaqi Wang
Zhihong Li
author_sort Jingna Tao
collection DOAJ
description BackgroundThis study aimed to construct and validate diagnostic models for the Operative Link on Gastritis Assessment (OLGA) and Operative Link on Gastric Intestinal Metaplasia Assessment (OLGIM) staging systems using three different methodologies based on magnifying endoscopy with narrow-band imaging (ME-NBI) features, to evaluate model performance, and to analyse risk factors for high-risk OLGA/OLGIM stages.MethodsWe enrolled 356 patients who underwent white-light endoscopy and ME-NBI at the Department of Gastroenterology, Dongzhimen Hospital, Beijing University of Chinese Medicine, between January 2022 and September 2023. Clinical data were recorded. Chi-square or Fisher’s exact tests were used to analyse differences in endoscopic features between OLGA/OLGIM stages. Variables showing statistical significance underwent collinearity diagnosis before model inclusion. We constructed predictive models using Bayesian stepwise discrimination, random forest, and XGBoost algorithms. Receiver operating characteristic (ROC) curves were plotted using Python 3.12.4. Model accuracy, area under the ROC curve (AUC), sensitivity, and specificity were calculated for comprehensive validation.ResultsAll three models demonstrated excellent diagnostic performance, with random forest and XGBoost models showing marginally superior accuracy, AUC values, and sensitivity compared with the Bayesian stepwise discrimination model. For OLGA staging, the AUC values were 0.928, 0.958, and 0.966, with accuracies of 0.854, 0.902, and 0.918 for Bayesian, random forest, and XGBoost models, respectively. For OLGIM staging, the corresponding AUC values were 0.924, 0.975, and 0.979, with accuracies of 0.910, 0.938, and 0.927. Risk factors for high-risk OLGA included lesion location (subcardial and lower body greater curvature), intestinal metaplasia patches, lesion size, demarcation line (DL), and margin regularity of micro-capillary demarcation line (MCDL). Risk factors for high-risk OLGIM included Helicobacter pylori infection status, mucosal condition, lesion location (lesser curvature and lower body greater curvature), erosion, lesion size, DL, vessel and epithelial classification (VEC), white globe appearance (WGA), and MCDL margin regularity.ConclusionsAll three models demonstrated robust accuracy and predictive capability, confirming that conventional white-light endoscopy combined with ME-NBI features provides valuable diagnostic reference for clinical risk assessment of precancerous gastric lesions.
format Article
id doaj-art-cb1d1c4042ec4c65a383f0ec10cdae2e
institution OA Journals
issn 2234-943X
language English
publishDate 2025-04-01
publisher Frontiers Media S.A.
record_format Article
series Frontiers in Oncology
spelling doaj-art-cb1d1c4042ec4c65a383f0ec10cdae2e2025-08-20T01:51:55ZengFrontiers Media S.A.Frontiers in Oncology2234-943X2025-04-011510.3389/fonc.2025.15545231554523Risk prediction model for precancerous gastric lesions based on magnifying endoscopy combined with narrow-band imaging featuresJingna Tao0Zhongmian Zhang1Linghan Meng2Liju Zhang3Jiaqi Wang4Zhihong Li5Dongzhimen Hospital, Beijing University of Chinese Medicine, Beijing, ChinaBeijing Hospital of Traditional Chinese Medicine, Capital Medical University, Beijing, ChinaGuang’anmen Hospital, China Academy of Chinese Medical Sciences, Beijing, ChinaDongzhimen Hospital, Beijing University of Chinese Medicine, Beijing, ChinaDongzhimen Hospital, Beijing University of Chinese Medicine, Beijing, ChinaDongzhimen Hospital, Beijing University of Chinese Medicine, Beijing, ChinaBackgroundThis study aimed to construct and validate diagnostic models for the Operative Link on Gastritis Assessment (OLGA) and Operative Link on Gastric Intestinal Metaplasia Assessment (OLGIM) staging systems using three different methodologies based on magnifying endoscopy with narrow-band imaging (ME-NBI) features, to evaluate model performance, and to analyse risk factors for high-risk OLGA/OLGIM stages.MethodsWe enrolled 356 patients who underwent white-light endoscopy and ME-NBI at the Department of Gastroenterology, Dongzhimen Hospital, Beijing University of Chinese Medicine, between January 2022 and September 2023. Clinical data were recorded. Chi-square or Fisher’s exact tests were used to analyse differences in endoscopic features between OLGA/OLGIM stages. Variables showing statistical significance underwent collinearity diagnosis before model inclusion. We constructed predictive models using Bayesian stepwise discrimination, random forest, and XGBoost algorithms. Receiver operating characteristic (ROC) curves were plotted using Python 3.12.4. Model accuracy, area under the ROC curve (AUC), sensitivity, and specificity were calculated for comprehensive validation.ResultsAll three models demonstrated excellent diagnostic performance, with random forest and XGBoost models showing marginally superior accuracy, AUC values, and sensitivity compared with the Bayesian stepwise discrimination model. For OLGA staging, the AUC values were 0.928, 0.958, and 0.966, with accuracies of 0.854, 0.902, and 0.918 for Bayesian, random forest, and XGBoost models, respectively. For OLGIM staging, the corresponding AUC values were 0.924, 0.975, and 0.979, with accuracies of 0.910, 0.938, and 0.927. Risk factors for high-risk OLGA included lesion location (subcardial and lower body greater curvature), intestinal metaplasia patches, lesion size, demarcation line (DL), and margin regularity of micro-capillary demarcation line (MCDL). Risk factors for high-risk OLGIM included Helicobacter pylori infection status, mucosal condition, lesion location (lesser curvature and lower body greater curvature), erosion, lesion size, DL, vessel and epithelial classification (VEC), white globe appearance (WGA), and MCDL margin regularity.ConclusionsAll three models demonstrated robust accuracy and predictive capability, confirming that conventional white-light endoscopy combined with ME-NBI features provides valuable diagnostic reference for clinical risk assessment of precancerous gastric lesions.https://www.frontiersin.org/articles/10.3389/fonc.2025.1554523/fullprecancerous gastric lesionsprediction modelBayesianrandom forestXGBoost
spellingShingle Jingna Tao
Zhongmian Zhang
Linghan Meng
Liju Zhang
Jiaqi Wang
Zhihong Li
Risk prediction model for precancerous gastric lesions based on magnifying endoscopy combined with narrow-band imaging features
Frontiers in Oncology
precancerous gastric lesions
prediction model
Bayesian
random forest
XGBoost
title Risk prediction model for precancerous gastric lesions based on magnifying endoscopy combined with narrow-band imaging features
title_full Risk prediction model for precancerous gastric lesions based on magnifying endoscopy combined with narrow-band imaging features
title_fullStr Risk prediction model for precancerous gastric lesions based on magnifying endoscopy combined with narrow-band imaging features
title_full_unstemmed Risk prediction model for precancerous gastric lesions based on magnifying endoscopy combined with narrow-band imaging features
title_short Risk prediction model for precancerous gastric lesions based on magnifying endoscopy combined with narrow-band imaging features
title_sort risk prediction model for precancerous gastric lesions based on magnifying endoscopy combined with narrow band imaging features
topic precancerous gastric lesions
prediction model
Bayesian
random forest
XGBoost
url https://www.frontiersin.org/articles/10.3389/fonc.2025.1554523/full
work_keys_str_mv AT jingnatao riskpredictionmodelforprecancerousgastriclesionsbasedonmagnifyingendoscopycombinedwithnarrowbandimagingfeatures
AT zhongmianzhang riskpredictionmodelforprecancerousgastriclesionsbasedonmagnifyingendoscopycombinedwithnarrowbandimagingfeatures
AT linghanmeng riskpredictionmodelforprecancerousgastriclesionsbasedonmagnifyingendoscopycombinedwithnarrowbandimagingfeatures
AT lijuzhang riskpredictionmodelforprecancerousgastriclesionsbasedonmagnifyingendoscopycombinedwithnarrowbandimagingfeatures
AT jiaqiwang riskpredictionmodelforprecancerousgastriclesionsbasedonmagnifyingendoscopycombinedwithnarrowbandimagingfeatures
AT zhihongli riskpredictionmodelforprecancerousgastriclesionsbasedonmagnifyingendoscopycombinedwithnarrowbandimagingfeatures