IDDF2023-ABS-0296 Three-tiered risk stratification for gastric varices using ai technology

Background Gastric variceal bleeding is a serious complication for patients with cirrhosis. Identifying high-risk patients with gastric varices and providing timely prophylactic treatment is crucial. However, previous stratification systems, such as the Sarin classification, did not provide a direct grading of the risk and severity of gastric varices. The AI-aided gastric varices risk stratification system aims to classify patients into three risk categories, which may provide a better way to predict the risk of bleeding and severity of gastric varices. Methods Gastric variceal images were collected from the endoscopy room of the Second Hospital of Hebei Medical University during the period from 2017 to 2022 (registration number: ChiCTR2200065160). A total of 1678 gastroscopy images from 411 patients were jointly annotated by three senior clinical endoscopists into three risk categories. The gastric varices risk stratification system is as follows: a. Mild: low risk of bleeding, usually with a diameter less than 5mm and low degree of protrusion. b. Moderate: moderate risk of bleeding, and endoscopic treatment is necessary, with relatively low endoscopic treatment difficulty, usually with a diameter between 5mm and 10mm and a moderate degree of protrusion. c. Severe: high risk of bleeding and endoscopic treatment is necessary, with high endoscopic treatment difficulty. The varices are usually with a high degree of protrusion and a diameter > 10mm or less than 10mm but with positive red signs. We trained a deep learning model to automatically predict the risk. Specifically, we first pretrained the CNN encoder using a large amount of normal images and varices images without annotations, then a classification model was trained for risk stratification based on the pretrained weights (IDDF2023-ABS-0296 Figure 1). Results Our model achieved an accuracy rate of more than 70% in classifying gastric varices into their corresponding risk categories, which can be a valuable tool for young doctors in diagnosing and stratifying the risk factors of gastric varices. Conclusions Our research team has developed the first three-tiered classification system for gastric varices, which holds the potential to improve clinical management by allowing for more precise risk stratification and treatment decisions for gastric varices.