Lung Nodule Classification using A Novel Two-stage Convolutional Neural Networks Structure

Lung cancer is one of the most fatal cancers in the world. If the lung cancer can be diagnosed at an early stage, the survival rate of patients post treatment increases dramatically. Computed Tomography (CT) diagram is an effective tool to detect lung cancer. In this paper, we proposed a novel two-stage convolution neural network (2S-CNN) to classify the lung CT images. The structure is composed of two CNNs. The first CNN is a basic CNN, whose function is to refine the input CT images to extract the ambiguous CT images. The output of first CNN is fed into another inception CNN, a simplified version of GoogLeNet, to enhance the better recognition on complex CT images. The experimental results show that our 2S-CNN structure has achieved an accuracy of 89.6%.