Lung Nodule Classification in CT Images Using 3D DenseNet
Ge Zhang, Lan Lin, Jingxuan Wang
Abstract
Abstract Lung cancer is the main malignant tumour affecting the health of residents in China. Automatically discriminating benign and malignant pulmonary nodules can facilitate the early detection of lung cancer, which reduces lung cancer mortality. The rising quantity of public available lung CT datasets made it possible to use deep learning approaches for lung nodules malignancy classification. Unlike most of the previous models that focused on 2D convolutional neural nets (CNN), here we explore the use of the DenseNet architecture with 3D filters and pooling kernels. The performance of the proposed nodule classification was evaluated on publicly available LUNA16 dataset, a subset of lung image database consortium and image database resource initiative dataset (LIDC/IDRI). It achieved a 92.4% classification accuracy. The proposed method provides an independent module with encouraging prediction accuracy that can be easily incorporated with a lung cancer computer-aided diagnosis system.