Efficient Chest X-Ray Investigation Using Firefly Algorithm Optimized Deep and Handcrafted Features

Abstract

Radiological imaging of the chest (X-ray) is a cost-effective, widely accepted method of examining the lungs and abnormalities. During this research, a clinically significant Convolutional-Neural-Network (CNN) framework will be proposed for the examination of lung abnormalities. The proposed scheme aims to achieve better detection accuracy from the selected chest X-ray data. The following stages are included in these techniques: (i) CNN segmentation of the lung section, (ii) Deep-feature extraction utilizing selected CNN schemes, (iii) Handcrafted-feature extraction, (iv) Optimizing features using Firefly Algorithms, and (v) Binary classification and cross-validation of fivefold cross-validation. By implementing the pre-trained VGG-UNet, this framework is capable of extracting lungs sections from X-ray images. Using this lung segment, handcrafted features such as Local Binary Patterns (LBP) and Pyramid Histograms of Oriented Gradients (PHOG) are obtained. DFs and HFs are obtained using the FA, and then a serial concatenation is performed in order to obtain a hybrid feature vector. This feature vector is used to classify X-ray images into healthy and diseased groups. For examination in this study, X-ray images of disease classes, such as tuberculosis, COVID19, pneumonia, lung masses, and effusions, are considered. Based on the experimental results of this study, >98% of disease detection accuracy was confirmed using the proposed scheme combined with the SoftMax classifier.