VGGCovidNet: A deep convolutional neural network to predict COVID-19 Cases from X-Ray images

Faruk, F.

The novel Coronavirus disease (COVID-19) has become an epidemic by affecting many countries globally. It has severely affected both the human lifestyle and public health. The number of tests should be increased among suspicious people to reduce the prevalence. However, the conventional test method is both time-consuming and costly. So, a computer-based automatic system may resolve these issues. In this study, a deep learning-based model was proposed to automatically detect COVID-19 cases using radiography chest x-ray images. The VGG16 network was redesigned to localize the abnormality features in a chest x-ray. This proposed modified network was renamed as VGGCovidNet. However, the class imbalance problem may arise due to the smaller number of COVID-19 positive cases. A combination of dataset re-sampling and image augmentation technique was addressed to generate a balanced dataset. Then, the gradient weighted class activation maps (Grad-CAMs) were produced to prove that how our model localized and learned the distinguishable abnormality features for COVID-19, normal, and pneumonia cases from x-ray images. The proposed model produced a positive predictive value-PPV (Precision of COVID-19 class) of 97.41% and true positive rate for COVID-19 class (Recall) of 97.41%. © 2021 IEEE.