ABSTRACT
The automatic diagnosis of the Coronavirus has become essential to minimize the workload of the health system in the face of the epidemic. Today, the coronavirus is spreading at an accelerated rate. It is Therefore essential to identify infected persons in order to avoid any contamination. This virus can be detected using real-time polymerase chain reaction (RT-PCR) kits. These kits are expensive and it takes several hours to confirm infection;another weakness is that in the early stages of the disease, the positivity rate of this test was considered to be very low, however, high international demand has revealed a gap in this kit. Therefore, an alternative strategy is advocated to assist in effective diagnosis. This study examines the issue of automatic classification systems in lung disease, including the new COVID-19. To accurately track this virus, Artificial Intelligence (AI) combined with radiography that can detect radiological symptoms, represents an ideal alternative solution. In this sense, we first propose a deep-transfer learning approach to extract high dimensional features from radiological images and then we apply a machine learning technique to classify the obtained features. The proposed classification model, i.e. DenseNet121 plus SVM, achieved an accuracy and f1 score of 96.09% and 97% respectively, for COVID-19 detection based on three classes. This model is superior to other models in the literature. We implemented 10 CNNs to compare our methodology. The data available in GitHub and Kaggle repository are used as a basis to produce the result. © 2022, The Author(s), under exclusive license to Springer Nature Switzerland AG.
ABSTRACT
The world has experienced the spread of a dangerous virus, Coronavirus (COVID-19), that has caused the death of millions of people worldwide at an extremely rapid rate, many studies have confirmed that the virus can be detected effectively using medical images. However, it takes a long time to analyze each image by radiologists who suffer from high pressures, especially due to the high similarity of symptoms between this virus and other respiratory diseases, which can lead to the confusion of cases and, consequently, the inability to identify them quickly, which could be a problem in a pandemic situation. In this paper, a methodology is proposed for the rapid and automatic diagnosis of this virus from chest radiographic images through the use of Artificial Intelligence (AI) techniques. There are two stages of the proposed model. The first step is data augmentation and preprocessing;the second step is the detection of COVID-19 with a transfer learning technique using a pre-trained deep convolutional network (CNN) architecture to extract features, Then, the obtained feature vectors are classified into three classes: COVID-19, Normal, and pneumonia, from two open medical repositories. In the experimentation phase of our model, we evaluate a set of common metrics to measure the performance of the architecture. Experimental conclusions show an accuracy of 96.52% for all classes, then a comparison with existing models in literature demonstrates that our proposed model achieves better classification accuracy © 2022. IAENG International Journal of Computer Science.All Rights Reserved.