CVD19-Net: An Automated Deep Learning Model for COVID-19 Screening using Chest CT Images

ABSTRACT

COVID-19 is the most recent coronavirus-related disease that has been declared a pandemic by the World Health Organization (WHO), causing a global emergency that has resulted in a large number of deaths and is rapidly spreading around the world. It causes respiratory illness and is highly contagious, putting a strain on health and medical systems worldwide. With the help of various deep learning (DL) techniques, chest CT scans are considered an effective tool for diagnosing COVID-19 because it directly affects the lungs. In addition, the visual similarities between COVID-19 and pneumonia make identification even more challenging, as COVID-19 is also a virus. In this paper, we designed a unique lightweight DL model named CVD19-Net with fewer layers as an accurate diagnostic method for COVID-19. Different regularization techniques such as dropout layers, batch normalization layers and data augmentation are injected into the CVD19-Net model to improve classification accuracy and reduce overfitting. We considered three different publicly available datasets for our experiments. (1) Dataset 1 2482 CT images were collected;(2) Dataset 2 7544 CT images were collected;(3) Dataset 3 3190 CT images were collected. The experimental results show that the proposed model achieves 98.59% accuracy on dataset 1, 98.21% on dataset 2, and 95.61% on dataset 3, which is better than the existing methods. The proposed model requires less training time and storage space, which makes it computationally efficient while maintaining a high level of accuracy, which can help clinicians quickly identify COVID-19 patients. © 2022 IEEE.