| 摘 要: 冠状动脉疾病（Coronary Artery Disease, CAD）是全球主要致死病因，CT血管造影（CT Angiography, CTA）虽能清晰显示血管解剖，但其图像常存在分辨率低、对比度差及伪影干扰，导致血管自动分割面临挑战。为此，本文提出一种基于nnUNet改进的分割架构CBAMnnUNet，通过集成卷积块注意力模块，动态增强对关键区域及多尺度特征的聚焦能力，以提升对细微血管结构的分割精度。在包含200例3D冠脉造影数据集的实验表明，该模型在各项指标上均显著优于常见模型，尤其在细小血管分割与边界优化方面表现突出，验证了注意力机制在nnUNet模型中对冠脉自动化分割的有效性与临床潜力。 |
| 关键词: 冠状动脉分割 医学图像分割 深度学习 nnUNet 注意力机制 三维图像处理 CBAM模块 |
|
中图分类号:
文献标识码:
|
| 基金项目: 浙江省自然科学基金资助项目（No.ZCLQN26F0205）；浙江省教育厅一般项目(No.Y202354089) |
|
| CBAMnnUNet:3D Coronary artery image segmentation algorithm based on improved nnUNet with CBAM Module |
|
ZHANG Haifeng, JIANG Lurong, LI Shuo, ZOU Yutao, JIANG Nan
|
Zhejiang Sci-Tech University, School of Information Science and Engineering
|
| Abstract: Coronary Artery Disease is a leading global cause of mortality. While Computed Tomography Angiography provides clear visualization of vascular anatomy, its images often suffer from low resolution, poor contrast, and artifacts, posing significant challenges for automated vessel segmentation. To address this, this paper proposes an enhanced segmentation architecture named CBAMnnUNet, based on nnUNet. By integrating the Convolutional Block Attention Module (CBAM), the model dynamically enhances focus on key regions and multi-scale features, thereby improving segmentation accuracy for fine vascular structures. Experiments conducted on a dataset of 200 3D coronary angiography images demonstrate that the proposed model significantly outperforms commonly used models across all evaluation metrics, particularly in segmenting small vessels and refining boundary delineation. This validates the effectiveness of the attention mechanism in nnUNet for automated coronary artery segmentation and highlights its clinical potential.. |
| Keywords: Coronary artery segmentation Medical image segmentation Deep learning nnUNet Attention mechanism 3D image processing CBAM module |
