A Prospective Multicenter Study of AI-assisted Gastroscopic Varices Diagnosis

This study is a multi-center prospective study. Patients who meet the inclusion and exclusion criteria of the study will collect their basic information during outpatient and hospitalization. The designated doctor will introduce the significance of this study to the patients and sign the informed consent. After the endoscopy of the patient, the video of the operation under the endoscope was collected and saved, and the AI was imported to obtain its conclusions on the above characteristics, as well as stratification and treatment recommendations. Follow-up requirements for third-party physicians who meet the inclusion criteria, record their various diagnosis and treatment recommendations after watching the video (mainly in the form of a selection questionnaire to avoid subjective bias), and the content of the questionnaire includes the veins observed in the endoscopic operation video. Varicose related features (with or without red sign, bleeding sign, varicose vein shape, diameter, color, mucosal sign), and corresponding risk stratification and treatment recommendations are required. For videos with inconsistent conclusions between AI and physicians, physicians will be asked to watch these videos with AI prompts again to give their opinions, and there will be a two-week buffer period between the first time and the second viewing. The reference standard for diagnosis will be obtained by three endoscopists watching videos without AI prompts, and the conclusions drawn by the three experts will be used as the reference standard. When the three experts have inconsistent conclusions, the three experts will discuss A consensus shall prevail. If no consensus can be reached in the end, the video will be discarded.

Primary and secondary endpoints were obtained by comparing the accuracy of diagnosis made by physicians and AI. All data were analyzed by SPSS statistical software.