Ensemble Convolutional Layers and Modified Wild Horse Optimized Learning Framework for Kidney Tumor Classification from CT Images

Abstract

Our society is affected by the most common disease Kidney Tumor (KT) in humans. The early diagnosis of KT may reduce the risk of death rates. Preventive measures can be taken to reduce the severe effects and overcome the tumor progression. Traditional methods consume time and tedious task. Deep Learning (DL) methods are emerging now to save time for diagnosis, to improve the accuracy of detection and reduce the physician’s manpower. In this research work, detection system is developed to detect KT in Computed Tomography (CT) images. Convolutional Neural Networks (CNNs) integrated with Modified Wild Horse Optimization (MWHO) is developed to test and train the network. The images from the Kaggle dataset are taken for the experiment. The dataset is divided into two as 80% is for training and 20% is for testing purpose. The accuracy values provided by CNN, CNN combined with particle swarm optimization (PSO) and CNN combined with genetic algorithm (GA) models are 86.6%, 86.8%, and 96%, respectively. The accuracy value provided by the proposed classification method is 100%. The proposed model achieved promising results when the number of classes able to be predicted (K) is equal to 5.