Lung Nodule Classification of CT Images Based on the Deep Learning Algorithms

Abstract

Early diagnosis and treatment of lung cancer patients can help to reduce a lung cancer mortality rate that contributes to more than 11.6% of the total cancer-associated deaths. Usually, for patients' diagnosis, Computed Tomography (CT) images are manually diagnosed by radiologists, which is a huge burden to them and sometimes leads to inaccuracy. Therefore, an automated computer-aided diagnosis (CAD) system that can assist radiologists in decision-making is preferable to speed up patients' diagnosis and overcome inaccuracy. This study aims to develop an automated CAD system algorithm that uses a deep Convolutional Neural Network (CNN) for in-depth feature extraction to classify benign and malignant lung nodules from a chest CT scan. The CSPNet and PANet modules are used to combine with the ResNet model for CT scan feature extraction. Due to the multidimensional (3D) nature of CT scans, we propose to use 3D convolutional kernels. We validated the proposed network on CT scan from The Lung Nodule Analysis 2016 (LUNA16) public dataset, which comprises of 888 CT scans. Experimental results show that the proposed algorithm gives a specificity, sensitivity, accuracy, and AUC of 98.98%, 94.42%, 97.09%, and 0.995, respectively. In addition, the results of our design show competitiveness compared to other studies.