抑制心血管图像序列中运动伪影的无监督深度学习方法

PDF下载HTML

Title:

Unsupervised deep learning method for suppressing motion artifacts in cardiac vascular image sequences

作者:

王茹¹, 孙正^1,2, 姚越¹

1. 华北电力大学 电子与通信工程系, 河北 保定 071003;
2. 华北电力大学 河北省电力物联网技术重点实验室, 河北 保定 071003

Author(s):

WANG Ru¹, SUN Zheng^1,2, YAO Yue¹

1. Department of Electronic and Communication Engineering, North China Electric Power University, Baoding 071003, China;
2. Hebei Key Laboratory of Power Internet of Things Technology, North China Electric Power University, Baoding 071003, China

关键词:

心血管成像; 冠状动脉内成像; 图像序列; 运动伪影; 运动补偿; 神经网络; 无监督学习; 弹性配准

Keywords:

cardiac vascular imaging; intracoronary imaging; image sequences; motion artifact; motion compensation; neural network; unsupervised learning; elastic registration

分类号:

TP391.41; R445

DOI:

10.11992/tis.202408014

文献标志码:

2025-2-25

摘要:

血管内超声(intravascular ultrasound, IVUS)和光学相干断层成像(optical coherence tomography, OCT)是诊断冠状动脉粥样硬化性病变的重要手段，但心脏运动和血流搏动会产生运动伪影，影响图像质量。为解决这一问题，本文提出一种无监督深度学习方法，用于抑制IVUS/OCT图像序列中的运动伪影。设计一个深度神经网络，包含特征提取、上采样、运动估计和运动校正模块，实现从连续回撤导管采集的图像序列到去伪影图像序列的映射。利用临床IVUS/OCT图像进行无监督训练，实验结果表明，该方法能显著提高纵向视图中管壁边缘的平滑度，标准差和帧间差异度指标得到显著改善。与其他方法相比，本文方法可使平均帧间差异度降低约23%，标准差降低约24%。该方法有效解决了IVUS/OCT图像序列因运动伪影造成的质量下降问题，优化了图像视觉效果。

Abstract:

Intravascular ultrasound (IVUS) and optical coherence tomography (OCT) are important tools for diagnosing coronary atherosclerotic lesions. However, cardiac motion and pulsatile blood flow can cause motion artifacts that affect image quality. To address this issue, an unsupervised deep learning method for suppressing motion artifacts in IVUS/OCT image sequences is proposed in this paper. A deep neural network consisting of feature extraction, upsampling, motion estimation, and motion correction modules is designed, and it can map continuous pullback image sequences to artifact-free image sequences. The network is trained in an unsupervised manner using clinical IVUS/OCT images. Experimental results demonstrate that this method remarkably improves the smoothness of the vessel wall boundaries in the longitudinal view, with substantial improvements in standard deviation and interframe dissimilarity metrics. Compared with other methods, the proposed approach reduces average interframe dissimilarity and standard deviation by approximately 23% and 24%, respectively. This method effectively solves the image quality degradation caused by motion artifacts in IVUS/OCT image sequences and optimizes the visual quality of the images.