Automatic Segmentation of COVID-19 Infection on Lung CT Scans using Mask R-CNN

Abstract

The novel of coronavirus disease (COVID-19) emerged as an infection threatening all humanity and later became a pandemic. The most common known symptoms of COVID-19 infection are dry cough, sore throat/inflammation and fever. The disease continues as a form of extreme pneumonia in the lung in the later stages and may cause permanent damage to the lung. Therefore, automated computer-assisted methods can assist in diagnosing COVID-19 infection at an early stage. In this study, we propose a robust method based on Mask R-CNN for automatic segmentation of COVID-19 infections and lung abnormalities on a publicly-available dataset. Experimental studies for segmentation of COVID-19 infections using CT scans achieved Dice similarity score (DSC) of 81.93% on the dataset. As a result, in this study, it is revealed that Mask R-CNN method for segmentation of COVID-19 infections is successful and can help physicians in decision-making. © 2022 IEEE.