Related Experiment Video
Updated: May 6, 2026
3D Whole-heart Myocardial Tissue Analysis
Published on: April 12, 2017
MHASegNet: A multi-scale hybrid aggregation network of segmenting coronary artery from CCTA images
Shang Li1,2, Yanan Wu3, Bojun Jiang1
1College of Medicine and Biological Information Engineering, Northeastern University, Shenyang, China.
Insights
A new deep learning model, MHASegNet, combined with refinement techniques, significantly improves coronary artery segmentation in CCTA images for better coronary artery disease diagnosis.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Cardiovascular Disease Research
Background:
- Accurate segmentation of coronary arteries in Coronary Computed Tomography Angiography (CCTA) is vital for diagnosing coronary artery disease (CAD).
- Challenges include small vessel size, uneven contrast, and segmentation errors like over-segmentation or omissions.
- Existing methods struggle to achieve consistent accuracy.
Purpose of the Study:
- To enhance coronary artery segmentation in CCTA images.
- To develop a robust method combining deep learning and conventional techniques.
- To improve the accuracy and reliability of CAD diagnosis through better image analysis.
Main Methods:
- Proposed MHASegNet, a lightweight deep learning network utilizing multi-scale hybrid attention for feature extraction.
- Integrated a 3D context anchor attention module to focus on coronary artery structures and reduce background noise.
- Employed an iterative, region-growth-based refinement strategy to address segmentation discontinuities and false positives.
Main Results:
- MHASegNet with refinement achieved a Dice Similarity Coefficient (DSC) of 0.867 on an in-house dataset.
- Performance on public datasets included DSC of 0.875 (ASOCA) and 0.827 (ImageCAS).
- The method demonstrated superior performance compared to state-of-the-art algorithms.
Conclusions:
- The tailored refinement effectively reduces false positives and resolves discontinuities, benefiting even other segmentation networks.
- MHASegNet and its refinement show promise for improving CAD diagnosis and quantification.
- Further validation is recommended for clinical application.
Background:
Segmentation of coronary arteries in Coronary Computed Tomography Angiography (CCTA) images is crucial for diagnosing coronary artery disease (CAD), but remains challenging due to small artery size, uneven contrast distribution, and issues like over-segmentation or omission.
Objective:
The aim of this study is to improve coronary artery segmentation in CCTA images using both conventional and deep learning techniques.
Methods:
We propose MHASegNet, a lightweight network for coronary artery segmentation, combined with a tailored refinement method. MHASegNet employs multi-scale hybrid attention to capture global and local features, and integrates a 3D context anchor attention module to focus on key coronary artery structures while suppressing background noise. An iterative, region-growth-based refinement addresses crown breaks and reduces false alarms. We evaluated the method on an in-house dataset of 90 subjects and two public datasets with 1060 subjects.
Results:
MHASegNet, coupled with tailored refinement, outperforms state-of-the-art algorithms, achieving a Dice Similarity Coefficient (DSC) of 0.867 on the in-house dataset, 0.875 on the ASOCA dataset, and 0.827 on the ImageCAS dataset.
Conclusion:
The tailored refinement significantly reduces false positives and resolves most discontinuities, even for other networks. MHASegNet and the tailored refinement may aid in diagnosing and quantifying CAD following further validation.
