Development and Evaluation of an Explainable AI Model for Diagnosing Early Gastric Cancer Using Multimodal Endoscopic Imaging

What this Trial Is About

Researchers are investigating an explainable multimodal artificial intelligence AI model to improve the diagnosis of early gastric cancer EGC using endoscopic imaging. The study addresses challenges in accurately detecting EGC due to subtle features and variability among endoscopists. The goal is to develop a reliable AI tool that combines multiple imaging types and clinical data, enhancing diagnostic accuracy and interpretability during endoscopic examinations. The study collects data from patients who underwent endoscopic submucosal dissection ESD with confirmed early gastric cancer or non-cancerous lesions. It uses white-light imaging WLI and image-enhanced endoscopy to gather representative images. The process includes developing a lesion detection model, extracting quantitative image features, and using deep learning to analyze images. These features are integrated with clinical information to build a multimodal AI prediction model. Participants provide endoscopic images and clinical data, which are divided into training, validation, and testing groups for model evaluation. The AI models diagnostic performance is measured using metrics like the area under the receiver operating characteristic curve AUROC, sensitivity, specificity, and accuracy. The study also applies methods to explain and visualize the AIs decision-making process, aiming to support clinical use of AI-assisted diagnosis. Evaluation occurs up to 14 days after endoscopy when histopathological results are available.

AI Model for Early Gastric Cancer Diagnosis Using Endoscopic Imaging