Three-dimensional vessel segmentation in magnetic resonance angiography using mask modeling(PDF)
《中国医学物理学杂志》[ISSN:1005-202X/CN:44-1351/R]
- Issue:
- 2025年第10期
- Page:
- 1361-1368
- Research Field:
- 医学影像物理
- Publishing date:
Info
- Title:
- Three-dimensional vessel segmentation in magnetic resonance angiography using mask modeling
- Author(s):
- LI Dexuan1; WANG Chenglong1; ZHANG Qi1; ZHANG Xuefeng2; YANG Guang1
- 1. Shanghai Key Laboratory of Magnetic Resonance/Institute of Magnetic Resonance and Molecular Imaging in Medicine, East China Normal University, Shanghai 200062, China 2. The First Hospital Affiliated to Naval Medical University, Shanghai 200433, China
- Keywords:
- Keywords: deep learning vessel segmentation magnetic resonance angiography topological connectivity
- PACS:
- R318；TP391.41
- DOI:
- DOI:10.3969/j.issn.1005-202X.2025.10.014
- Abstract:
- Abstract: Magnetic resonance angiography (MRA) is a non-invasive imaging technique used to observe blood vessels. Quantitative analysis of MRA images enables visualization of vascular pathways, condition, and blood flow dynamics, which is essential for diagnosing vascular diseases such as vascular lesions, stenosis, and occlusions. Vessel segmentation serves as the fundamental basis for quantitative vascular analysis. However, the complex morphology of vessels, difficulties in labeling, and scarcity of accurate 3D vascular annotations pose significant challenges for MRA-based vessel segmentation. A strategy of selectively occluding vessels during model training is proposed to enhance the algorithms capacity to capture the topological structure of blood vessels, thereby improving the continuity of vessel segmentation results. Additionally, a Refine network is incorporated to refine the binary segmentation results of the segmentation network, thereby further improving segmentation accuracy. Model training and testing are carried out using 42 cases of 3D MRA data from the public MIDAS dataset. For the test set, the 3D U-Net baseline model with vessel occlusion strategy shows a β0 Error of 1.274 2±0.210 3 and a β1 Error of 0.339 3±0.081 8, respectively, which are 0.113 6 and 0.028 0 lower than the baseline. The model integrating vessel occlusion strategy and Refine network achieves an average Dice score of 0.710 5 ± 0.012 5, which is 0.002 8 higher than the baseline. These results demonstrate that the proposed method effectively improves both vascular connectivity and segmentation accuracy.
Last Update: 2025-10-29