Detection of Lung Tumor Using ASPP-Unet with Whale Optimization Algorithm

Abstract

The unstructured growth of abnormal cells in the lung tissue creates tumor. The early detection of lung tumor helps the patients avoiding the death rate and gives better treatment. Various medical image modalities can help the physicians in the diagnosis of disease. Many research works have been proposed for the early detection of lung tumor. High computation time and misidentification of tumor are the prevailing issues. In order to overcome these issues, this paper has proposed a hybrid classifier of Atrous Spatial Pyramid Pooling (ASPP)-Unet architecture with Whale Optimization Algorithm (ASPP-Unet -WOA). To get a fine tuning detection of tumor in the Computed Tomography (CT) of lung image, this model needs pre-processing using Gabor filter. Secondly, feature segmentation is done using Guaranteed Convergence Particle Swarm Optimization. Thirdly, feature selection is done using Binary Grasshopper Optimization Algorithm. This proposed (ASPP-Unet -WOA) is implemented in the dataset of National Cancer Institute (NCI) Lung Cancer Database Consortium. Various performance metric measures are evaluated and compared to the existing classifiers. The accuracy of Deep Convolutional Neural Network (DCNN) is 93.45%, Convolutional Neural Network (CNN) is 91.67%, UNet obtains 95.75% and ASPP-UNet-WOA obtains 98.68%. compared to the other techniques.

Visualization for Explanation of Deep Learning-Based Fault Diagnosis Model Using Class Activation Map

“…It ignores the part or leaves the input unchanged, adding the weight of the convolution kernel parameter to 0, thereby expanding the acceptance domain, convolving with a value more excellent than one can achieve the same effect. However, downsampling will be performed simultaneously with the convolution, which will reduce the feature map's size and is unsuitable for use [41].…”