A Real-time Full-screen Diagnostic Support Method with Simultaneous Detecting and Classifying for Colonoscopy Lesion Areas

In this paper, we propose a Computer-Aided Diagnosis (CAD) system for full-screen lesion recognition using detecting and classifying method simultaneously based on deep learning architecture. The proposed CAD system used full-screen images token by colorectal Narrow Band Imaging (NBI) endoscopy as the inputs, which can detect the lesion areas and classify the areas with different levels of progress at the same time. Our system aims to reduce the variation and burden of diagnoses due to the experience of diagnosing doctors. As the result, the model has detected all the lesions in test dataset, and the current classification accuracy has reached a reasonable value, 85% for magnified images in our test dataset.