A deep learning approach for lung cancer classification and nodule identification using CT-images

This work devises an efficient technique deep learning enabled hybrid Shepard convolutional Kronecker network (ShCKN) for lung cancer classification and nodule identification using computed tomography (CT)-images. Initially, the image input is taken from the specific database and the acquired images are fed into an image pre-processing unit, where the Laplacian filter removes unnecessary noise. Thereafter, segmentation of the lung lobe is performed using K-Net. Then, nodules are detected using grid-based schemes. After that, feature extraction is performed and essential features are extracted using entropy measures. Finally, lung cancer classification is accomplished by the devised ShCKN, which is obtained by combining Shepard convolutional neural networks (ShCNN) and deep kernel networks (DKN). The performance estimation of ShCKN is validated based on early devised approaches and performance measures; the ShCKN achieves accuracy, F-measure, precision, recall, and specificity at 92%, 92%, 91%, 94%, and 93%.