Development of Computer-Aided Diagnosis System Using Single FCN Capable for Indicating Detailed Inference Results in Colon NBI Endoscopy

Abstract

In this paper, we propose a single fully convolutional network (FCN) capable of indicating the detailed inference results for Computer-Aided Diagnosis (CAD) in colon Narrow Band Imaging (NBI) endoscopy. The proposed CAD system is capable of real-time processing with a latency of 0.05 seconds and 20 frames per second and can detect more than 80% of lesions even for non-magnified images. Classification results at the pixel with the highest confidence level at resulted in a diagnosis with 73% agreement with histopathologic findings.