Automatic disease detection in endoscopy with light weight transformer

Gastric and intestine cancers have a high incidence and mortality rate, posing a significant threat to human health and life expectancy. The most effective way to prevent these diseases is to identify and treat them in their early stages through gastroscopy and colonoscopy. However, traditional diagnosis methods for these procedures are not always accurate. Recent advances in deep learning-based object detection techniques have shown promise in improving the accuracy of disease detection in the stomach and intestine through gastroscopy and colonoscopy images. In this paper, we present a novel deep learning architecture that can accurately localize and identify diseases in these images. Our design includes a lightweight transformer detection head that improves inductive ability and a plain FPN that significantly speeds up the inference process. Our experiments show that our architecture performs at the state-of-the-art level in disease detection on gastroscopy and colonoscopy images, as measured by F1, precision, and recall.