Latent Polyp: Variational Representation Learning for Accurate Colonic Polyp Detection and Intelligent Classification to Assist Colonoscopy and Early Cancer Screening

Abstract

Colorectal cancer is a leading cause of cancer-related deaths, making early detection through colonoscopy critical for patient survival. LatentPolyp presents a novel deep learning framework that employs β-VAE architecture for variational representational learning to achieve accurate polyp classification. The system learns robust 256-dimensional latent representations of colonoscopy images to classify multiple polyp types including adenomas, hyperplastic polyps, and gastrointestinal conditions. Integrated with a T5 transformer-based clinical suggestion system, LatentPolyp analyzes patient medical history to generate personalized treatment recommendations, risk assessments, and follow up schedules, providing comprehensive AI-assisted diagnostic support for gastroenterologists in early cancer screening.