BCDU-Net and chronological-AVO based ensemble learning for lung nodule segmentation and classification

Abstract

ABSTRACT Lung cancer is the deadly disorder occurring worldwide among all age groups and it causes high mortality rate. Traditionally, Computed Tomography (CT) is employed as a non-invasive tool to detect the position of tumour in the lung and its severity level. It is quintessential to detect the nodules concealed in the lung that is mainly responsible for deadly cancer disease. Lung nodule segmentation has been recognised as a significant topic for research discussion among researchers and major obstacle faced in this lung nodule segmentation is accuracy, which is mainly deteriorated because of the visual variations and diversity in lung nodules. Therefore, to enhance the segmentation performance level of lung nodule detection, this research proposes an effective strategy for lung nodule segmentation and classification using newly devised method named Chronological-African Vultures Optimisation (CAVO) algorithm. Here, lung nodule segmentation is carried out using Bi-directional ConvLSTM-U-Net (BCDU-Net) and lung nodule classification is done using ensemble classifier, such as Multi-Layer Perceptron (MLP), Convolutional Neural Network (CNN), and Deep Residual Network (DRN). Meanwhile, the proposed CAVO-based ensemble learning has provided better insights with high testing accuracy, maximum sensitivity, and maximum specificity with measures of 0.946, 0.930, and 0.890 using dataset-1.