Pulmonary Lung Nodule Detection from Computed Tomography Images Using Two-Stage Convolutional Neural Network

Abstract

Abstract Lung cancer is one of the leading causes of cancer-related death in people all over the world. Lung cancer screening is a crucial part of the diagnosis of cancer. The initial sign of lung cancer is the pulmonary nodules that can be detected based on the computed tomography (CT) scan images. In some cases, the nodules are not obvious and may take a trained eye and a considerable amount of time to detect. The automatic detection of the nodules can save considerable time and money, thus opening prescreening accessibility, ultimately saving lives. Hence, a two-stage convolutional neural network (CNN) model is proposed to segment and detect pulmonary lung nodules from CT scan images. In the first stage, we have used U-net for nodule segmentation, and in the second stage, a CNN model is designed to reduce the false positives (FPs) generated in the previous stage, which enhances the overall efficiency and correctness of the system. The proposed method’s performance is evaluated on the publicly available Lung Image Database Consortium and Image Database Resource Initiative dataset. The proposed two-stage CNN model achieved an accuracy of 84.4 %, where the FP per true positive is reduced from 11.1 to 0.97. The proposed model shows its superiority over the existing methods for nodule detection.