Implementation of CNN Voting based Technique for Classification of Lung Images

Abstract

Lung-related disorders such as pneumonia, cancer, and tuberculosis remain significant health concerns in recent decades. This research proposes a voting-based ensemble of pretrained CNN models to accurately classify lung disorders, such as pneumonia, tuberculosis, and COVID-19, from medical images. The suggested method improves diagnostic performance by aggregating predictions using majority voting, resulting in enhanced accuracy, sensitivity, and robustness compared to traditional techniques. This study presents a CNN-based multi-tier classification framework for the diagnosis of lung disorders utilizing transfer learning with ResNet50, AlexNet, and VGG19. A voting-based fusion method integrates results from separate models to improve diagnostic precision. The suggested CNN voting-based classifier was evaluated on a dataset of more than 900 lung pictures, encompassing pneumonia, tuberculosis, COVID-19, and normal cases. Three pretrained models—ResNet50, AlexNet, and VGG19—were employed utilizing a voting-based ensemble approach to augment classification robustness. The experimental findings indicated that the fusion model surpassed individual CNNs, with an accuracy of 92.30% and a sensitivity of 100%. Performance was assessed utilizing measures including accuracy, precision, specificity, sensitivity, and AUC, and compared to state-of-the-art approaches to illustrate its efficacy.