AUTOMATIC DETECTION OF COVID-19 CORONAVIRUS INFECTION BASED ON THE CHEST X-RAYS ANALYSIS WITH DEEP LEARNING MODELS

Abstract

Early detection of the patients with COVID-19 coronavirus infection is essential to ensure adequate treatment and reduce the burden on the healthcare system. An effective method of detecting COVID-19 is computer analysis of chest X-rays. Changes caused by COVID-19 can be detected by them even in asymptomatic patients, so deep learning models have great potential as the screening tools. The paper proposes a methodology consisting of a stage of preliminary processing of X-ray images, images augmentation and multiclass classification using deep convolutional neural networks Xception, MobileNetV2, DenseNet121, ResNet50 and VGG16, previously trained on the ImageNet dataset. The results of computer experiments showed that VGG16 model demonstrated the best performance of the patients with COVID-19 classification with accuracy of 94,12 %, sensitivity (recall) 95,76 % and AUC = 98 %.