| 摘 要: 脂质斑块的准确分割对于冠状动脉粥样硬化斑块的风险评估具有重要意义。血管内光学相干断层成像(Optical Coherence Tomography,OCT)能够提供高分辨率的血管壁微结构信息,但脂质成分在OCT图像中常呈现斑块形状不规则、边界模糊等特点,导致传统分割方法分割精度受限。针对上述问题,提出一种基于衰减先验与语义融合的U-Net模型,称为DASF-UNet(Dual-branch Attenuation-guided Semantic Fusion U-Net)。由于脂质成分对OCT信号的衰减效应显著强于斑块中的其他成分,衰减梯度图能够有效反映脂质成分的空间分布特征,为分割任务提供重要的先验信息。因此,模型将原始OCT图像与衰减梯度图作为双通道输入,利用卷积编码器与Transformer编码器分别提取局部细节与全局特征,并引入SENet(Squeeze-and-Excitation Network)增强通道特征的判别能力。实验结果表明,所提方法在提升脂质斑块分割精度,减少区域误分以及还原边界形态方面均取得了良好效果,整体性能优于当前主流的方法。 |
| 关键词: 脂质斑块 冠状动脉 OCT 衰减先验 语义分割 |
|
中图分类号:
文献标识码:
|
| 基金项目: 江苏省研究生科研与实践创新计划项目 (SJCX240280) |
|
| A Method for Lipid Plaque Segmentation Based on Attenuation Prior and Dual-branch Semantic Fusion UNet |
|
yangyi
|
Nanjing University of Posts and Telecommunications
|
| Abstract: Accurate segmentation of lipid plaques is of great significance for the risk assessment of coronary atherosclerotic plaques. Intravascular optical coherence tomography (OCT) can provide high-resolution microstructural information of the vessel wall. However, lipid components in OCT images often exhibit characteristics such as irregular plaque shapes and blurred boundaries, which limit the segmentation accuracy of traditional methods. To address the above issues, this paper proposes a U-Net model based on attenuation prior and semantic fusion, termed DASF-UNet (Dual-branch Attenuation-guided Semantic Fusion U-Net). Since the attenuation effect of lipid components on OCT signals is significantly stronger than that of other components in the plaque, the attenuation gradient map can effectively reflect the spatial distribution characteristics of lipid components, providing important prior information for the segmentation task. Therefore, the model uses the original OCT image and the attenuation gradient grayscale map as input, and combines convolutional and Transformer encoders to extract local details and global features. In addition, SENet(Squeeze-and-Excitation Network) is applied to enhance the discriminative ability of channel features. Experimental results show that the proposed method achieves favorable performance in improving segmentation accuracy, reducing regional misclassification, and restoring boundary morphology, with overall performance surpassing current mainstream methods. |
| Keywords: lipid plaque coronary artery OCT attenuation prior semantic segmentation |
