Classification of Lung Nodule Malignancy on CT Images Using Convolutional Neural Network

Abstract

This paper developed an integrated lung cancer diagnosis program on Android using a three-dimensional convolutional neural network (CNN). The CNN is trained with CT images from the LUNA16 dataset, which are prepossessed to improve the efficiency of the training process. To maximize the accuracy of the diagnosis, we propose novel 3D LeNet-5 and FishNet to apply to 3D medical image processing. The experiments validate the effectiveness of our methods. Additional ways to increase the accuracy of the model are discussed.