Related Experiment Video
Updated: Aug 29, 2025
08:05
Lung CT Segmentation to Identify Consolidations and Ground Glass Areas for Quantitative Assesment of SARS-CoV Pneumonia
Published on: December 19, 2020
14.3K
Automated deep learning-based segmentation of COVID-19 lesions from chest computed tomography images
Mohammad Salehi1, Mahdieh Afkhami Ardekani2,3, Alireza Bashari Taramsari4
1Department of Medical Physics, School of Medicine, Iran University of Medical Sciences, Tehran, Iran.
Polish Journal of Radiology
|September 12, 2022
Summary
Deep learning models like U-Net accurately segment COVID-19 lesions on chest CT scans. These 2D algorithms aid radiologists in faster screening and quantification of infected regions for improved patient treatment.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Radiology
Background:
- The COVID-19 pandemic caused a global health crisis.
- Chest computed tomography (CT) is crucial for COVID-19 detection.
- Automated segmentation of COVID-19 lesions in CT scans is challenging.
Purpose of the Study:
- To develop and evaluate 2D deep learning algorithms for automated segmentation of COVID-19 lesions in chest CT slices.
- To compare the performance of U-Net, U-Net++, and Res-Unet for this task.
Main Methods:
- Trained three deep learning networks (U-Net, U-Net++, Res-Unet) from scratch.
- Utilized a dataset of 20 labeled COVID-19 chest CT volumes (2112 images).
- Assessed segmentation performance using Dice similarity coefficient, ASSD, MAE, sensitivity, specificity, and precision.
Main Results:
- All models demonstrated good performance, with mean Dice values over 84.0%.
- U-Net and U-Net++ outperformed Res-Unet.
- U-Net achieved the highest performance with 86.0% sensitivity and 2.22 mm ASSD.
Conclusions:
- 2D deep learning models can accurately segment COVID-19 lesions from chest CT images.
- These models can assist radiologists in faster screening and quantification.
- Further studies are needed to evaluate clinical performance and robustness.
