Semantic segmentation of chest computed tomography scans for quantifying pathologically altered lung parenchyma in patients with post-COVID syndrome

АУТОР(И) / AUTHOR(S): Daniel Štifanić, Tin Nadarević, Jelena Štifanić, Igor Barković, Zlatan Car

Download Full Pdf   

DOI:  10.46793/ICCBIKG25.398S

САЖЕТАК / ABSTRACT:

The term „post-COVID syndrome“ refers to a variety of long-lasting symptoms that have been reported in a significant number of patients after SARS-CoV-2 infection. Computed tomography (CT) scans of the chest in patients with post-COVID syndrome often reveal various pathological changes. A visual evaluation by radiologists serves as the foundation for CT examination of the chest in modern clinical practice; yet this diagnostic process may lead to subjectivity, interpretation variability, and is often time-consuming due to the complexity and volume of data that must be thoroughly assessed. For this reason, the medical industry is searching for innovations that have the potential to automate the diagnostic process of CT analysis and serve as an assistive tool for radiologists. CT scans have shown to be useful for quantifying pathologically altered lung parenchyma in patients with post-COVID syndrome. The aim of this research is to demonstrate an approach for automatic segmentation of pathologically altered lung parenchyma based on the chest CT scans. In this research DeepLabv3+ with Xception_65, MobileNetV2, and ResNet101 as backbones are used in order to perform semantic segmentation. Performances of the proposed approach based on CT scans have achieved an average mIOU of 0.86299 ± 0.00614, F1 of 0.92259 ± 0.00397, accuracy of 0.98421 ± 0.00106, precision of 0.91931 ± 0.00453, sensitivity of 0.92614 ± 0.0047, and specificity of 0.99144 ± 0.00062. Based on the obtained results, the proposed approach proved to be successful in terms of quantifying pathologically altered lung parenchyma in patients with post-COVID syndrome and has promising potential for clinical use. 

КЉУЧНЕ РЕЧИ / KEYWORDS:

ct scan, lungs, post-covid, deep learning, semantic segmentation

ПРОЈЕКАТ / ACKNOWLEDGEMENT:

This research was (partly) supported by Erasmus+ AISE, under grant 2023-1-EL01- KA220-SCH-000157157; and by the CZI ‘BrainClock’ project under grant NPOO.C3.2.R3- I104.0089.

ЛИТЕРАТУРА / REFERENCES:

• Akande, O. W., & Akande, T. M. (2020). COVID-19 pandemic: A global health burden. Nigerian Postgraduate Medical Journal, 27(3), 147-155.
• Maltezou, H. C., Pavli, A., & Tsakris, A. (2021). Post-COVID syndrome: an insight on its pathogenesis. Vaccines, 9(5), 497.
• Solomon, J. J., Heyman, B., Ko, J. P., Condos, R., & Lynch, D. A. (2021). CT of post-acute lung complications of COVID-19. Radiology, 301(2), E383-E395.
• Alarcón-Rodríguez, J., Fernández-Velilla, M., Ureña-Vacas, A., Martín-Pinacho, J. J., Rigual- Bobillo, J. A., Jaureguízar-Oriol, A., & Gorospe-Sarasúa, L. (2021). Radiological management and follow-up of post-COVID-19 patients. Radiología (english edition), 63(3), 258-269.
• Panahi, O. (2025). Deep Learning in Diagnostics. Journal of Medical Discoveries, 2(1), 1-6.