NAVT-net neuron attention visual taylor network for lung cancer detection using CT images

Abstract

Lung Cancer is regarded as a common fatal disease affecting humans throughout the entire world. Early diagnosis is vital to save the patient’s life and Computed Tomography (CT) scans are referred to as the major imaging modes but, the manual examination of a CT scan is time-consuming and results in errors. Hence, a novel system of Neuron Attention Visual Taylor Network (NAVT-Net) is developed to detect lung cancer. At first, the CT image is acquired, and then, the input image is filtered based on homomorphic filtering. Then, the lung nodule is segmented using the Dual-Branch-UNet (DB-UNet). Later, the image augmentation is achieved by resizing, flipping, as well as rotation. Next, the shape-based features are extracted and subjected to the last stage of lung cancer detection, which is done by the NAVT-Net system that is established on the basis of Neuron Attention Stage-by-Stage Network (NASNet), Visual Geometry Group-16 (VGG16), and Taylor series. Hence, the experimental results of the developed NAVT-Net system achieved high values of 92.176 % accuracy, 93.997 % of True Positive Rate (TPR), 92.189 % of True Negative Rate (TNR), F1-score of 90.999 %, and precision of 91.998 %, computational time, and memory usage of 37.879 s, and 41.100MB at K-values of 9.