Using Deep Learning Approaches for the Automatic Detection of COVID-19 and Assessing Disease Severity through Chest CXR and CT Scan Image Processing

The novel coronavirus has been rapidly spreading globally since its first appearance in China at the end of 2019 and has become a worldwide pandemic. Early detection and isolation of infected cases are major factor in decreasing the spread of the virus. The most popular and effective method for detecting the disease is RT-PCR, but it is expensive and time-consuming. Radiography is a diversty option for detecting the infected cases and determining its severity. However, this process requires experienced radiologists to manually interpret the images. This paper aims to evaluate the effectiveness of using DL models in the automatic classification of the COVID-19 and quantitative measurement of the severity of the spread of the virus in the lungs through CXR and CT scan image processing. Two models have been introduced, one specifically designed for processing CXR images using CNN and DL (Xcov_model), and another based on the structure of the first model but intended for processing CT scan images (CTcov_model). Both models were supported by the Grad-Cam algorithm to create a heat map indicating where the disease is expected to spread. The dataset consisted of 9000 images, which were evenly divided into three classes for both CXR and CT scans. 80% of these images were used for training and 20% for testing, in addition to the implementation of DA technology. The models were implemented and evaluated using Python on the Google Collaboratory platform, with the Xcov_model achieving an F1-Score of 98% for both the COVID-19 class and the normal class, and the CTcov_model having a test accuracy of about 98%.