A histopathologically verified dataset of magnifying narrow-band imaging endoscopy for classifying the gastric precancerous cascade.

Gastric cancer is a leading cause of mortality worldwide, yet the development of computer-aided diagnosis (CAD) systems for its early detection is hindered by the scarcity of high-quality datasets utilizing Magnifying Endoscopy with Narrow-Band Imaging (ME-NBI). Addressing this gap, we present EndoWLI-NBI, a large-scale dataset comprising 11,971 ME-NBI images from 868 patients, specifically curated to cover the full gastric precancerous cascade-including Intestinal Metaplasia & Chronic Gastritis, Low- and High-Grade Intraepithelial Neoplasia, and Early Gastric Cancer. Unlike existing public datasets, every image in EndoWLI-NBI is rigorously mapped to a biopsy-confirmed histopathological diagnosis, providing a definitive gold standard for label reliability. Technical validation using state-of-the-art deep learning benchmarks demonstrates the dataset's quality and separability, while highlighting the intrinsic visual challenges in distinguishing high-grade neoplasia from early cancer. This dataset serves as a critical, medically verified resource for advancing fine-grained lesion classification, domain adaptation research, and endoscopic training in gastroenterology.