A Lesion Classification Method Using Deep Learning Based on NICE Classification for Computer-Aided Diagnosis System in Colorectal NBI Endoscopy

Currently, video image diagnosis by NBI (Narrow Band Imaging) is commonly used for colonoscopy. The purpose of this paper is to develop a CAD (computer-aided diagnosis) system that can reduce the variability of diagnosis due to differences in clinical doctor's experience by presenting quantitative inference results to the clinical doctor. As a part of this system development, we create a classifier that classifies lesion images into NICE (NBI International Colorectal Endoscopic) classification using deep learning. We can achieve to develop a CNN (Convolutional Neural Network) based classifier in which five performance indicators (Accuracy, Recall, Specificity, PPV, and NPV) are satisfied more than 90 % quality.