Related Experiment Video
Updated: Jun 1, 2025
Author Spotlight: Bridging Gaps in Anatomy and Establishing a Foundation for Algorithmic Studies
Published on: December 15, 2023
Benefit from public unlabeled data: A Frangi filter-based pretraining network for 3D cerebrovascular segmentation
1School of Engineering Medicine and School of Biological Science and Medical Engineering, Beihang University, Beijing, 100191, China; Key Laboratory of Big DataBased Precision Medicine (Beihang University), Ministry of Industry and Information Technology of China, Beijing, 100191, China; CAS Key Laboratory of Molecular Imaging, Institute of Automation, Chinese Academy of Sciences, Beijing, 100190, China.
This study introduces a new method for segmenting brain blood vessels in Time-of-Flight Magnetic Resonance Angiography (TOF-MRA) data. Leveraging a large unlabeled dataset and the Frangi filter, it significantly improves segmentation accuracy for computer-aided diagnosis.
Area of Science:
- Medical Imaging
- Neuroscience
- Computer Vision
Background:
- Precise cerebrovascular segmentation in Time-of-Flight Magnetic Resonance Angiography (TOF-MRA) is vital for computer-aided diagnosis.
- Manual labeling of TOF-MRA data is costly due to sparse cerebrovascular structures.
Purpose of the Study:
- To develop an effective pretraining strategy for 3D cerebrovascular segmentation using large-scale unlabeled TOF-MRA data.
- To enhance model performance by leveraging the Frangi filter for vessel-like structure enhancement.
Main Methods:
- Construction of the largest preprocessed unlabeled TOF-MRA dataset (1510 subjects).
- Development of a Frangi filter-based preprocessing workflow for unlabeled data.
- Implementation of a multi-task pretraining strategy for efficient knowledge extraction.
Main Results:
- The pretrained model achieved superior performance on four cerebrovascular segmentation datasets.
- Demonstrated an improvement of approximately 2%-3% in the clDice metric compared to state-of-the-art methods.
- Ablation studies confirmed the generalizability and effectiveness across different backbone structures.
Conclusions:
- The proposed pretraining strategy effectively utilizes unlabeled TOF-MRA data for improved 3D cerebrovascular segmentation.
- This approach offers a cost-effective solution for enhancing computer-aided diagnosis systems.
- Open-sourced code and data facilitate further research and development.
