Multicenter Observational Study of a Multimodal AI Model Using EUS, White-Light Endoscopy, and Clinical Data for Diagnosis of Upper GI Mesenchymal Tumors and Risk Stratification of Gastric GISTs

What this Trial Is About

Researchers are evaluating a multimodal artificial intelligence (AI) model designed to support the diagnosis of upper gastrointestinal (GI) mesenchymal tumors and the risk classification of gastric gastrointestinal stromal tumors (GISTs). This observational study combines retrospective image data with prospectively recruited patients to test the AI model's diagnostic accuracy compared to expert endoscopists. No multimodal AI model has yet reported performance for these tasks, and the study aims to meet specific statistical requirements with a planned enrollment of 130 patients. The study collects endoscopic ultrasound (EUS) images, white-light endoscopy (WLE) images, and clinical data meeting strict quality standards. The AI model integrates these data using a multi-branch fusion strategy. Prospectively recruited patients undergo standard diagnostic evaluation, including histopathological confirmation. Each participant's diagnostic images are independently analyzed by the AI model and expert endoscopists, who are blinded to each other's results. No additional interventions or costs are involved, ensuring participants receive optimal diagnostic and treatment options. Participants will provide informed consent and complete required examinations to confirm eligibility. Researchers will compare diagnostic performance using measures such as sensitivity, specificity, predictive values, accuracy, and area under the curve. Statistical analyses include paired tests and group comparisons to evaluate the AI model against human experts. The primary outcomes focus on the AI model's accuracy in classifying GI subepithelial lesions and predicting GIST risk. Data privacy measures are in place to protect participant information throughout the study period ending in June 2026.

Multicenter Observational Study of Multimodal AI for Upper GI Mesenchymal Tumor Diagnosis