Related Experiment Video
Updated: Jan 18, 2026
04:25
Author Spotlight: Bridging Gaps in Anatomy and Establishing a Foundation for Algorithmic Studies
Published on: December 15, 2023
3.8K
Cerebrovascular segmentation network based on fast fourier convolution and Mamba.
Chaozhi Yang1,2, Mingzhe Cao2, Jinbao Zhu3
1School of Optics and Photonics, Beijing Institute of Technology, Beijing 100081, People's Republic of China.
Biomedical Physics & Engineering Express
|September 8, 2025
Summary
This study introduces F-Mamba-YNet, a new deep learning model for segmenting cerebral vessels in Time-of-Flight Magnetic Resonance Angiography (TOF-MRA) scans. It improves vessel segmentation accuracy and continuity for better cerebrovascular disease diagnosis.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Neuroscience
Background:
- Cerebrovascular segmentation is vital for diagnosing and treating cerebrovascular diseases.
- Accurate segmentation of cerebral vessels from Time-of-Flight Magnetic Resonance Angiography (TOF-MRA) is challenging due to complex anatomy and topology.
Purpose of the Study:
- To develop a novel deep learning network, F-Mamba-YNet, for improved cerebrovascular segmentation from TOF-MRA.
- To enhance the accuracy, completeness, and connectivity of cerebral vessel segmentation.
Main Methods:
- Proposed F-Mamba-YNet, a Y-shaped network utilizing a dual-encoder architecture with Fast Fourier Convolution and Mamba modules.
- Spectral encoder captures high-frequency vessel edge details; spatial encoder captures long-range dependencies.
- Multi-scale Feature Selection Module in the decoder for adaptive feature enhancement and reuse.
Main Results:
- F-Mamba-YNet achieved a Dice Similarity Coefficient (DSC) of 86.28% on the IXI-A-SegAN dataset.
- Achieved a DSC of 72.24% on the MIDAS dataset.
- Demonstrated more connected and continuous segmentation results compared to existing methods.
Conclusions:
- F-Mamba-YNet offers competitive performance and superior segmentation continuity for cerebral vessels.
- The proposed network shows strong generalization capabilities for cerebrovascular segmentation tasks.
