融合冠脉造影三维重建信息的血管内超声图像处理关键技术研究

Intravascular ultrasound (IVUS) is playing an important role in clinical routine for the early diagnosis and the subsequent therapeutics of coronary heart disease, which relies heavily on medical image processing technology constantly updated. During the acquisition IVUS images, speckle noise and motion artifact frequently occur, which might reduce the contrast of IVUS images, as well as causing misalignment to the adjacent image frames in the pullback series. These phenomena might seriously impact both the diagnosis quality and the analysis efficiency. The project aims to combine IVUS image sequential information with the 3D coronary angiography reconstructed from the multiple angiographic views, and to solve some key problems including speckle reduction, image enhancement, motion artifact suppression, and border detection in IVUS image processing. The main research contents are as follow: 1).A new 2D/3D bilateral filter is designed and implemented with its fast algorithm. Weighted parameters in the filter are adaptively regulated or optimally selected in order to reduce speckle noise meanwhile to preserve or enhance edge details.2).Image-based cardiac phase retrieval in intravascular ultrasound sequences. Sequential stabilization is achieved by sampling sequences always at the same part of the cardiac phase. Image registration is carried out in IVUS sub-sequences adjacent stable frame for the sake of suppressing motion artifact. 3).Based on plaque characteristics and speckle texture information, a novel cost matrix for border detection is constructed with the purpose of detecting the media-adventitia border and intima border. Developing the research project is able to provide theoretical support and key algorithms for the diagnosis of coronary artery disease and for the quantitative analysis of IVUS images.

血管内超声(IVUS)在冠心病早期诊断与治疗中发挥重要的临床应用价值，这很大程度上依赖于医学图像处理技术不断推陈出新。在图像采集过程中，常因斑点噪声和运动伪影，造成血管组织病变区域不规则且对比度低、相邻帧错位等现象，严重影响IVUS分析质量与诊断效率。本项目以IVUS图像序列为研究对象，融合多视角冠脉造影血管三维重建信息，拟解决IVUS图像处理中的去噪增强、伪影消除、边缘检测等难点问题。主要研究内容包括:1).设计新的2D/3D双滤波器及其快速算法，自适应调节并优化滤波器的加权参数，实现去斑点噪声、保持或增强边缘细节；2).基于图像心动相位检索，提取序列稳定帧，相邻稳定帧子序列图像配准，抑制IVUS序列运动伪影；3).利用斑块特征和斑点纹理信息，构造全新的边缘检测代价矩阵，实现IVUS序列中-外膜和内膜的边缘检测。本项目开展将为冠状动脉疾病诊断与量化分析影像系统提供理论支撑与关键算法。

血管内超声(IVUS)是评价冠状动脉狭窄性病变的主要技术手段之一，在成像过程中，常因超声斑点、血管运动等因素干扰，影响IVUS分析质量与诊断效率。本项目以IVUS图像序列为对象，致力于斑点去噪、伪影消除和边缘检测等问题研究，其算法推广于掌静脉识别和MRI组织分割，并提出快速计算冠状动脉血流储备分数(Fractional Flow Reserved，FFR)方法。内容为6个方面：1) 研究超声图像斑点去噪方法。首先斑点模拟与多分类概率混合模型分析；其次，提出基于非局部全变分(NLTV)的改进算法，一是用皮尔森(Pearson)距离加权函数代替高斯加权函数，二是用对称KL距离代替欧式距离，引入Split-Bregman迭代算法解NLTV去噪模型；最后，实验表明，SNR和SSIM等指标优于现有的超声去斑点算法。2).针对IVUS图像序列运动伪影，研究IVUS序列图像关键帧提取方法：基于图像灰度信息、基于流形学习。通过计算门控前后序列的血管容积、管腔容积、平均斑块负荷等参数，评价伪影抑制方法效果。3). IVUS序列图像的中-外膜、血管内腔等边缘检测。完成三种IVUS图像分割方法研究：结合先验形状信息和序贯学习的心血管内超声外弹力膜检测、基于深度全卷积网络的IVUS血管内腔与中-外膜自动边缘检测、基于局部形状结构分类的心血管内超声图像中-外膜边缘自动检测。结果表明，我们的算法与医生手工标注方法具有良好的一致性与重复性。4). 融合CAG和IVUS图像信息，初步完成对动脉血管三维重建研究。5).提出了一种基于冠脉造影三维重建在线快速计算FFR的新方法(即定量血流分数，Quantitative Flow Ratio，QFR)，在判断临界病变的精度达到88%，受到心血管介入领域同行的关注。6).针对掌静脉识别，提出基于主方向特征的掌静脉识别分类方法，减少数据搜索范围，缩短数据库查询响应时间，设计实现一种基于Zedboard的掌静脉快速识别认证系统。针对MRI图像组织分割，提出基于多模态3D卷积神经网络(CNNs)特征提取、改进的混合核函数支持向量机(SVM)方法。解决了MRI图像中肿瘤分割精度问题。本项目的研究成果提供了有效的医学图像处理技术，为医学成像设备高端化、智能化提供了关键技术。