Related Experiment Video
Updated: Jul 27, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.8K
Automated semantic lung segmentation in chest CT images using deep neural network
M Murugappan1,2,3, Ali K Bourisly4, N B Prakash5
1Intelligent Signal Processing (ISP) Research Lab, Department of Electronics and Communication Engineering, Kuwait College of Science and Technology, Block 4, Doha, Kuwait.
Neural Computing & Applications
|June 5, 2023
Summary
This study developed a robust deep learning model for lung segmentation in COVID-19 patients using chest CT scans. The DeepLabV3+ network with ResNet-18 and ResNet-50 achieved superior performance for two-class and four-class segmentation, respectively.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Computer Vision
Background:
- Lung segmentation is crucial for analyzing lung infections in CT images.
- Deep learning models offer potential for automated and accurate lung segmentation.
- COVID-19 diagnosis can be aided by precise segmentation of lung abnormalities.
Purpose of the Study:
- To develop a computationally efficient and robust deep learning model for lung segmentation in COVID-19 patients.
- To evaluate the performance of the DeepLabV3+ network with various pretrained models for two-class and four-class lung segmentation.
- To propose a unified model for automatic delineation of lung regions in CT images.
Main Methods:
- Utilized the DeepLabV3+ network architecture with five pretrained networks (Xception, ResNet-18, Inception-ResNet-v2, MobileNet-v2, ResNet-50).
- Trained the model on a publicly available dataset of 750 COVID-19 chest CT images with pixel-level annotations.
- Assessed segmentation performance using Intersection of Union (IoU), Weighted IoU, Balance F1 score, pixel accuracy, and global accuracy.
Main Results:
- DeepLabV3+ with ResNet-18 achieved higher performance for two-class segmentation (background and lung field) with a batch size of 8.
- DeepLabV3+ with ResNet-50 demonstrated superior results for four-class segmentation (including ground-glass opacities and consolidation) with a batch size of 16.
- ResNet-based models offered a better balance of performance and computational complexity compared to Xception.
Conclusions:
- The developed DeepLabV3+ model provides an effective automated solution for segmenting lung regions in COVID-19 CT images.
- The model's efficiency and accuracy support its potential use in clinical diagnosis systems for COVID-19.
- This automated segmentation can assist clinicians in providing accurate second opinions for COVID-19 diagnosis.
