Journal of minimal access surgery · 2026
Evaluation of an artificial intelligence model in assessing the severity of gastritis during gastroscopy: A pilot study.
СтатьяНазвание и аннотация: Перевод пока не подготовлен
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Rajaram Krishnamoorthy, Sakthivel Palaniappan, Easwaramoorthy Sundaram, Sakthivel Chandrasekar
Аннотация
BACKGROUND: Gastritis diagnosis via endoscopy remains highly subjective, often leading to inter-observer variability and the misdiagnosis of early inflammatory changes as congestive portal gastropathy. While Computer-Aided Diagnosis (CAD) shows promise, the "paucity of labelled data" remains a significant barrier to clinical implementation.
AIM AND OBJECTIVES: This study aimed to develop and validate a scalable deep-learning solution using a hybrid CNN-Vision Transformer (ViT) architecture to assist endoscopists in the real-time severity stratification of gastritis (Normal, Mild, Moderate, and Severe).
MATERIALS AND METHODS: We utilized 4,462 high-definition, anonymised endoscopic images annotated by expert gastroenterologists using the Kyoto Classification as a benchmark. A multi-branch parallel processing architecture was developed, combining a CNN-based branch for local feature extraction (texture and erythema) and a Vision Transformer (ViT) for global context learning. Data augmentation expanded the training set to over 68,000 images to prevent overfitting. Performance was evaluated using Accuracy, Precision, Recall, and the Matthews Correlation Coefficient (MCC).
RESULTS: The hybrid model achieved an Overall Classification Accuracy of 91.4% and a Matthews Correlation Coefficient (MCC) of 0.90, indicating a near-perfect correlation with clinical ground truth. The system demonstrated exceptional performance in the "Normal" category with 99% specificity, providing a highly dependable negative result. Precision (0.8892) and Recall (0.8481) remained robust across all categories, significantly reducing the risk of over-diagnosing gastropathy as gastritis.
CONCLUSION: The proposed hybrid AI architecture provides a robust, real-time "second opinion" tool for endoscopists. By integrating local and global mucosal patterns, the system achieves high-precision severity stratification, potentially reducing unnecessary biopsies and minimizing diagnostic errors caused by inter-observer variability in the endoscopy suite.
AIM AND OBJECTIVES: This study aimed to develop and validate a scalable deep-learning solution using a hybrid CNN-Vision Transformer (ViT) architecture to assist endoscopists in the real-time severity stratification of gastritis (Normal, Mild, Moderate, and Severe).
MATERIALS AND METHODS: We utilized 4,462 high-definition, anonymised endoscopic images annotated by expert gastroenterologists using the Kyoto Classification as a benchmark. A multi-branch parallel processing architecture was developed, combining a CNN-based branch for local feature extraction (texture and erythema) and a Vision Transformer (ViT) for global context learning. Data augmentation expanded the training set to over 68,000 images to prevent overfitting. Performance was evaluated using Accuracy, Precision, Recall, and the Matthews Correlation Coefficient (MCC).
RESULTS: The hybrid model achieved an Overall Classification Accuracy of 91.4% and a Matthews Correlation Coefficient (MCC) of 0.90, indicating a near-perfect correlation with clinical ground truth. The system demonstrated exceptional performance in the "Normal" category with 99% specificity, providing a highly dependable negative result. Precision (0.8892) and Recall (0.8481) remained robust across all categories, significantly reducing the risk of over-diagnosing gastropathy as gastritis.
CONCLUSION: The proposed hybrid AI architecture provides a robust, real-time "second opinion" tool for endoscopists. By integrating local and global mucosal patterns, the system achieves high-precision severity stratification, potentially reducing unnecessary biopsies and minimizing diagnostic errors caused by inter-observer variability in the endoscopy suite.