3D computed tomography-based framework for automated grading of COVID-19 infections

Abstract

The COVID-19 outbreak caused by SARS-Cov-2, has affected several million people across the world. Artificial intelligence (AI) techniques have aided in the management of the COVID-19 epidemic in a variety of ways. These include diagnosis, estimation of epidemic trends, prognosis, and exploration of effective and safe drugs and vaccines. The primary goal of this chapter is to present a 3D computed tomography (CT)-based system for the automated diagnosis of COVID-19 severity levels. First, the lung regions are delineated, followed by the application of a Markov–Gibbs random field (MGRF) model to extract features that differentiate between COVID-19 severity levels. These features are derived by proposing three distinct MGRF models, each of which calculates Gibbs energy by tuning the model using a specific severity category. Subsequently, a neural network model is employed to classify the three Gibbs energy values, thereby identifying COVID-19 severity levels. The system is evaluated on a dataset of 76 COVID-19 patients using a hold-out validation approach. It achieves an overall accuracy of 83.33% and a Cohen’s kappa score of 73.98%, demonstrating the potential of such a system for grading COVID-19 severity.