Abstract
AbstractTreatment of non‐small cell lung cancer depends on detecting the cancer stage. The oncologist decides the cancer stage based on the tumour‐node‐metastasis (TNM) staging suggested by the American Joint Committee on Cancer (AJCC). This study simplifies the complicated problem of classifying computed tomography (CT) images into TNM‐based classes using deep learning algorithms at various levels. In the first level, an optimised conditional generative adversarial network (c‐GAN) network is developed for automatic lung segmentation, including nodules within the lung and juxtapleural nodules. Earlier studies used time‐consuming manual identification of the region of interest patches from the lung CT image before applying the deep learning classification algorithm. At the next level, three different deep learning algorithms, along with three support vector machine classifiers, are used for the classification of Tumour, Node and Metastasis as per the AJCC staging nomenclature. The specially designed c‐GAN network's performance is maximised using the Taguchi approach, which helps automatically preprocess CT images by removing unwanted background noises. Further, three different pre‐trained Resnet50 networks are trained using transfer learning for extracting the deep features for finally applying to three different classifiers, resulting in three different classes. The comparative segmentation performance assessment in the form of the average dice similarity coefficient and Jaccard index indicates that the proposed c‐GAN gives the best segmentation performance of the lung without losing the nodule compared to other segmentation algorithms. The proposed approach gives the classification performance for the Tumour as 91.94%–97.32%, the Nodule as 91.99%–100%, and the Metastasis as 99.25%–100.00%.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have