Lung Cancer Detection and Severity Level Classification Using Sine Cosine Sail Fish Optimization Based Generative Adversarial Network with CT Images

Abstract

Abstract This paper develops a lung nodule detection mechanism using the proposed sine cosine Sail Fish (SCSF) based generative adversarial network (GAN). However, the proposed SCSF-based GAN is designed by integrating the sine cosine algorithm with the SailFish optimizer, respectively. By using pre-processing, lung nodule segmentation, feature extraction, lung cancer detection, and severity level classification methods detection and classification are performed. The pre-processed computed tomography (CT) image is fed to the lung nodule segmentation phase, where the CT image is segmented into different sub-images to exactly detect the abnormal region. The segmented result after segmentation is fed to the feature extraction phase, where the features like mean, variance, entropy and hole entropy, are extracted from the nodule region. The affected regions are accurately detected using the loss function of the discriminator component. Finally, the lung nodules are detected and classified using the proposed SCSF-based GAN. The proposed approach obtained better performance with the accuracy of 96.925%, sensitivity of 96.900% and specificity of 97.920% for the first-level classification, and the accuracy of 94.987%, the sensitivity of 94.962% and specificity of 95.962% for second-level classification, respectively.