Research Article
Semantic segmentation of chest X-ray images based on the severity of COVID-19 infected patients
@ARTICLE{10.4108/eai.7-7-2021.170287, author={D. Štifanić and J. Musulin and Z. Jurilj and S. Baressi Šegota and I. Lorencin and N. Anđelić and S. Vlahinić and T. Šušteršič and A. Blagojević and N. Filipović and Z. Car}, title={Semantic segmentation of chest X-ray images based on the severity of COVID-19 infected patients}, journal={EAI Endorsed Transactions on Bioengineering and Bioinformatics}, volume={1}, number={3}, publisher={EAI}, journal_a={BEBI}, year={2021}, month={7}, keywords={Artificial Intelligence, COVID-19, DeepLabv3+, Semantic segmentation, X-ray images}, doi={10.4108/eai.7-7-2021.170287} }
- D. Štifanić
J. Musulin
Z. Jurilj
S. Baressi Šegota
I. Lorencin
N. Anđelić
S. Vlahinić
T. Šušteršič
A. Blagojević
N. Filipović
Z. Car
Year: 2021
Semantic segmentation of chest X-ray images based on the severity of COVID-19 infected patients
BEBI
EAI
DOI: 10.4108/eai.7-7-2021.170287
Abstract
INTRODUCTION: As a result of this global health crisis caused by the COVID-19 pandemic, the medical industry is searching for innovations that have the potential to automate the diagnostic process of COVID-19 and serve as an assistive tool for clinicians.
OBJECTIVES: X-ray images have shown to be useful in the diagnosis of COVID-19. The goal of this research is to demonstrate an approach for automatic segmentation of lungs in chest X-ray images.
METHODS: In this research DeepLabv3+ with Xception_65, MobileNetV2, and ResNet101 as backbones are used in order to perform lung segmentation.
RESULTS: The proposed approach was experimented on X-ray images and has achieved an average mIOU of 0.910, F1 of 0.925, accuracy of 0.968, precision of 0.916, sensitivity of 0.935, and specificity of 0.977.
CONCLUSION: Based on the obtained results, the proposed approach proved to be successful in terms of lung segmentation in chest X-ray images and has a great potential for clinical use.
Copyright © 2021 D. Štifanić et al., licensed to EAI. This is an open access article distributed under the terms of the Creative Commons Attribution license, which permits unlimited use, distribution and reproduction in any medium so long as the original work is properly cited.