Automatic diagnosis of COVID-19 using deep learning

The COVID-19 pandemic is an unprecedented global health crisis. Given the delays in obtaining RT-PCR results, the reference diagnostic technique, CT (computed tomography) plays a central role in triaging patients arriving in the emergency department, allowing them to be admitted to "COVID" or "non-COVID" services. To make reliable automatic diagnoses and to quantify the lesion extent, we used deep learning methods for the diagnosis of COVID-19. In this work, we implemented six CNN architectures, namely VGG16, ResNet50, MobileNet, GoogLeNet, Xception, and DenseNet121, and evaluated their performance using appropriate metrics on CT images from freely available public databases. The ResNet50 model was found to give the best results for acceptable computational complexity.