Classification of COVID-19 Disease Severity using CT Scans via Deep Convolutional Neural Networks

ABSTRACT

The respiratory virus Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2), commonly known as COVID-19, has caused wide concern and a need to be able to accurately determine its effects on individual lung capacities. Computerized Tomography (CT) scans were chosen as the main datatype for determining whether a patient had COVID-19 or had normal lung capacities. The rationale is that there is an inherent lack of CT scan experts, especially in counties with low socio-economic status (SES). Thus, an automated and objective artificially intelligent (AI) algorithm can assist in the hope to provide more efficacious use CT scans for classification of COVID-19 and monitoring of disease progression. In this study, a wide variety of CT scans with different formats were tested to determine the best approach for binary classification models using deep learning (DL) techniques. A total of three publicly available COVID-19 datasets were tested using a 2-dimensional and a 3-dimensional algorithm, where each dataset had its own subsets and unique parameters. A finalized version of the 3-dimensional model was shown to achieve a high accuracy, precision, sensitivity, and F1 Score of 86.6% each. The developed method provides an objective measure to automatically classify COVID-19 patients from CT scans © 2022 IEEE.