Using of Convolutional Neural Networks on X-Ray Images in the Diagnosis of COVID-19
COVID-19 is a highly contagious disease that can spread rapidly and strain healthcare systems if not controlled in a timely manner. RT-PCR is widely used to diagnose COVID-19, but the sensitivity of RT-PCR is low and existing PCR-based tests can be time-consuming. On the other hand, radiographic imaging methods such as chest X-ray for the diagnosis of COVID-19 and convolutional neural network (CNN), a subtype of deep learning, are frequently used for their analysis. The aim of this study is to provide automatic diagnosis of COVID-19 on X-ray images using CNN architectures and to help decision makers and clinicians. In this study, Xception, DenseNet121, DenseNet169, DenseNet201, ResNet50, VGG16, VGG19, InceptionResNetV2 and InceptionV3 architectures were used to diagnose COVID-19 on X-ray images. The performance of the architectures was evaluated using accuracy, precision, recall and F1 score. Data analyses were performed with Python programming language. The highest performing architecture was Xception with accuracy of 96%. The performance measures obtained with this architecture for patients with COVID-19 are precision of 100%, recall of 97%, F1 score of 98%; precision of 93%, recall of 92%, F1 score of 93% for normal patients; and precision of 97%, recall of 98%, F1 score of 97% for patients with pneumonia. The lowest performing architecture was VGG16 with accuracy of 60%. In this study, it was demonstrated that the CNN is an effective method that will help clinicians in diagnosing COVID-19 using X-ray images, if it is worked with data-appropriate architectures, and in this way, the disease can be detected as soon as possible, the effectiveness of treatment can be increased, and the negativities caused by the disease can be prevented.
Index Terms– Deep learning, Convolutional neural networks, Chest X-ray, Medical image classification, COVID-19
