Related Experiment Video
Updated: Sep 2, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.9K
COVID-19 CT image segmentation based on improved Res2Net.
Shangwang Liu1,2, Xiufang Tang1, Tongbo Cai1
1School of Computer and Information Engineering, Henan Normal University, Xinxiang, Henan, China.
Medical Physics
|August 2, 2022
Summary
This study introduces a new segmentation network for COVID-19 detection in CT scans. The method accurately identifies infected regions, improving upon existing models and aiding in rapid screening of coronavirus disease 2019 (COVID-19).
Area of Science:
- Medical Imaging
- Computer-Aided Diagnosis
- Artificial Intelligence in Healthcare
Background:
- Coronavirus disease 2019 (COVID-19) poses a significant global health and economic threat.
- Computed tomography (CT) image segmentation is crucial for identifying COVID-19 infected areas.
- Accurate segmentation aids in the screening and management of confirmed COVID-19 cases.
Purpose of the Study:
- To design and evaluate a novel segmentation network for precise identification of COVID-19 infected regions in CT images.
- To improve the accuracy of COVID-19 segmentation compared to existing methods.
- To develop a tool that assists clinicians in screening and managing COVID-19 patients.
Main Methods:
- A segmentation network utilizing Res2Net for multilayered feature extraction.
- Incorporation of an edge attention module to extract low-level edge features.
- Development of an attention position module (APM) for high-level feature extraction and region detection.
- Implementation of a context exploration module to minimize false positives and negatives.
Main Results:
- The proposed method achieved a Dice similarity coefficient of 0.755, sensitivity of 0.751, and specificity of 0.959.
- Compared to Inf-Net, the Dice similarity coefficient increased by 7.3% and sensitivity by 5.9%.
- The mean absolute error (MAE) was reduced by 2.2%.
Conclusions:
- The developed method demonstrates strong performance in segmenting COVID-19 infected regions on CT images.
- The network's portability allows integration with various popular deep learning architectures.
- This approach offers an effective tool for screening COVID-19, potentially reducing clinician workload.
