Detection for Pulmonary Nodules using RGB Channel Superposition Method in Deep Learning Framework

Abstract

The detection of pulmonary nodules is a very important research field in computer-aided diagnosis. In order to help doctors to identify pulmonary nodules more conveniently, especially for some small pulmonary nodules, a method based on RGB channel superposition to detect pulmonary nodules is proposed in this paper. We put the same ROI (region of interest) from three sequential lung CT slices into RGB channels to gain a pseudo-color image for deep learning. AlexNet and GoogLeNet is used as the deep learning network. We use 10000 patches of healthy tissues and 12000 patches of pulmonary nodules in LIDC-IDRI dataset for training and get a prediction model. The model is tested on 176 patients’ CT images and gain the sensitive of 95.0% at 5.62 false positives per scan. The experimental results show that the proposed method can improve the detection rate of pulmonary nodules compared with some traditional feature extraction methods.