Application of multi-class deep learning technique in detection of Covid-19 and other four lung diseases using X-ray images

Abstract

Deep learning technique have been effectively used in resolving computer vision issues including medical image analysis. Since chest X-rays are the most frequently ordered and less expensive diagnostic imaging test, they are used as the first imaging technique to diagnose COVID-19 disease. In medical image analysis and classification, Convolutional Neural Networks (CNNs) and transfer learning are a highly effective mechanism for efficiently sharing knowledge from generic to domain-specific object recognition tasks. This work deals with the deep learning modelling as a precise tool for diagnosing and classification of five Lung Diseases (covid-19, healthy, viral pneumonia, bacterial pneumonia, lung opacity, and Tuberculosis) quickly and accurately. In this study, x-ray image dataset of covid-19 and healthy cases was collected from various locations in Iraq. The other x-ray images of other diseases were obtained from multiple publicly available x-ray datasets, totalling 150 images for each class. Utilizing deep and transfer learning techniques such as ResNet18, ResNet50, MobileNetv2, GoogleNet, and DenseNet201. The application and evaluating of these models are done using five-fold cross-validation the AUC (Area Under the Receiver Operating Characteristic Curve) and confusion matrices. Comparison results of these five proposed models showed that the pre-trained DenseNet201 model outperforms the other models and achieve an accuracy rate of 92%.