Diagnostic Accuracy of Deep Learning-Assisted Multimodal Endocytoscopy for Superficial Esophageal Neoplasia and Precancerous Lesions: A Multicenter ，Prospective Study

In recent years, endocytoscopy (EC) has undergone rapid development as an advanced endoscopic technique. Characterized by real-time histological imaging and detection capabilities, EC enables visualization of living cells in the digestive mucosal layer through methylene blue staining via the endoscopic instrument channel. This technology provides ultra-magnified endocytoscopic images, allowing for identification of key cellular features including epithelial cell morphology and the shape of methylene blue-stained epithelial nuclei, thereby facilitating clear in vivo observation of microstructural characteristics in mucosal tissues - a technique commonly referred to as "optical biopsy". However, current clinical application of EC in China remains limited due to insufficient practical experience and the absence of standardized diagnostic criteria for image interpretation. This study aims to comprehensively evaluate the diagnostic accuracy of domestically developed high-magnification electronic upper gastrointestinal endoscopy for esophageal carcinoma and precancerous lesions, using histopathology as the gold standard. The research objectives include establishing standardized diagnostic evaluation protocols and clarifying the clinical value of this technology through systematic validation.