Related Experiment Video
Updated: Aug 6, 2025
06:48
Automated Segmentation of Cortical Grey Matter from T1-Weighted MRI Images
Published on: January 7, 2019
9.0K
Integration- and separation-aware adversarial model for cerebrovascular segmentation from TOF-MRA.
Cheng Chen1, Kangneng Zhou1, Tong Lu2
1School of Computer and Communication Engineering, University of Science and Technology Beijing, Beijing 100083, China.
Computer Methods and Programs in Biomedicine
|March 17, 2023
Summary
This study introduces a novel adversarial model for enhanced cerebrovascular segmentation from time-of-flight magnetic resonance angiography (TOF-MRA). The method improves texture and edge detection, achieving superior accuracy in segmenting complex cerebrovascular structures.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Neuroscience
Background:
- Cerebrovascular segmentation from time-of-flight magnetic resonance angiography (TOF-MRA) is crucial for studying cerebrovascular diseases but remains challenging.
- Current deep learning methods often lack global awareness due to reliance on local voxel or regional optimization.
- Existing models struggle with texture and edge interpretation, limiting the refinement of segmented cerebrovascular structures.
Purpose of the Study:
- To develop a novel cerebrovascular segmentation method that overcomes the limitations of existing deep learning approaches.
- To enhance global awareness and texture/edge interpretation in cerebrovascular segmentation.
- To achieve more refined and accurate segmentation of cerebrovascular structures from TOF-MRA.
Main Methods:
- A new adversarial model integrating a segmentation model with a discriminator for result filtering was proposed.
- Time-of-flight magnetic resonance angiography (TOF-MRA) images were separated into high- and low-frequency components to improve texture and edge representation, addressing sample imbalance.
- Encoder weight sharing and diversified discrimination were employed to enhance model integration, separation correlation, robustness, and regularization.
Main Results:
- The proposed adversarial model achieved top rankings on two cerebrovascular datasets, scoring 82.26% and 73.38%.
- The method demonstrated superior performance compared to recent specialized cerebrovascular segmentation models.
- The model also outperformed commonly used adversarial models in segmentation tasks.
Conclusions:
- The adversarial model enhances texture and edge extraction, leading to global cerebrovascular topology awareness and accurate, robust segmentation.
- This framework offers potential applications in various imaging fields, especially for datasets with sample imbalance.
- The developed code is publicly available, facilitating further research and application.
