Convolutional neural network in the classification of COVID-19

Abstract

Covid-19 spread out rapidly around the world, forcing many countries to full shutdown, and economical and social consequences. Resulting in rapid need for new and effective methods to deal with this crisis and control it. X-ray lung images is considered one of the most effective and safe method for diagnosing Covid-19, since it could provide solid proof of the existing of the disease, and it has limited effect on the health of the human comparing with other radiography methods. In this proposed work, CNN model is designed and trained to classify Covid-19 X-ray images, by using the COVID-19 Radiography Database, which is published and available online. This database is collected by researchers and experts from various universities around the world. The database contains total of 15153 lung x-ray images, divided into three classes. The classification classes are: Normal, Covid-19, and Viral Pneumonia. The model is trained and tested on publicly available dataset. The dataset is divided into three parts: training, validation, and testing datasets. The model is evaluated based on the three of these datasets. Totally, the evaluation metrics include Accuracy, F1-score, Area Under Curve (AUC), Precision, and Recall, with values of greater than 98% for all of the evaluation metrics. Comparing the results with state of arts publications, which used the same dataset, the proposed method outperformed the state of arts publications depending on the evaluation metrics. The number of the trainable parameters in the proposed CNN model is about 25.4 millions.