基于CTA影像数据的3D冠脉狭窄自动检测及其量化评估研究

Coronary computed tomography angiography (CCTA) is a common means of diagnosis in coronary artery diseases, and quantitative analysis of coronary artery is important for computer-aided diagnosis. Currently, the multiplanar reformation or the curve planner reformation is widely used in the clinical diagnosis. However, this method requires a large amount of manual intervention, and it is difficult to automatically analyse coronary artery stenoses. To address the above-mentioned problems,we will study the key techniques of automatic detection and quantification of 3D coronary artery stenoses in the project, and establish the general theoretical framework for stenosis quantitative analysis based on partial differential equations. We combine the object description, object similarity measure, variational model and parameter relationship model to resolve the problems of detection and quantification of coronary artery stenoses, such as blur vessel boundaries, thin vessels, and intensity inhomogeneity.Specific studies are as follows: Study the segmentation of objects with complex shapes, and construct the variational model with a priori constraint and a shape probabilistic model; Study the description of coronary shape, define a vesselness measure based on tensor analysis, build the search and connection criterions for the branches of coronary artery tree, and propose the segmentation algorithm for the 3D coronary artery with accuracy and efficiency, and the extraction algorithm for the centerlines with accuracy; Study the parameter estimation model for diameter parameter of the lumen, clear the corresponding relationship between fuzzy distance transform and observed diameter, determine the coherence analysis strategy of the observed data, build the functional relationship model of the expected diameter and observed diameter, detect the stenosis locations, quantify the stenoses degree, and realize the stenoses grade; Employ the probabilistic model and the lastest standard evaluation framework and platform, and verify and evaluate the proposed algorithms, to satisfy the requirments of clinical application.

冠状动脉CT血管造影术(CCTA)是当前诊断冠心病的常用手段，而冠脉量化分析是计算机辅助诊断该类疾病的重要依据；目前常用的多平面重组或曲面重组技术需大量人工干预，难以实现狭窄的自动分析。针对上述问题，本项目拟研究3D冠脉狭窄检测及其量化评估的关键技术，建立冠脉量化分析的理论框架，将目标描述、目标相似性测度、变分模型和参数关系模型结合起来，解决边界模糊、结构细小和灰度分布不均匀冠脉的狭窄检测及量化评估问题。具体包括：研究复杂形状的目标分割，构造先验约束变分模型和形状概率模型；研究冠脉形状描述方法，定义血管测度，建立冠脉分支搜索、连接准则，提出3D冠脉精确高效分割和中心线精确抽取算法；研究管腔直径参数估计方法，确定观测数据一致性策略，建立直径参数关系模型，检测血管狭窄位置，量化狭窄程度，实现狭窄分级；采用概率模型及最新标准评估框架和平台，验证所提算法，以满足对冠心病辅助诊疗的临床需要。

冠状动脉量化分析是当前计算机辅助诊断心脏疾病的重要依据，然而，当前常用的基于影像数据的多平面重组或曲面重组技术需大量人工干预，诊断结果易受医生主观因素影响，难以实现狭窄的自动精确分析。本项目面向冠脉螺旋CT 血管造影术（coronary computed tomography angiography, CCTA）在冠心病诊断中的应用，研究了冠脉狭窄检测和量化的新技术。具体包括：.1) 建立静态图像鲁棒性分割结果的初始形状模型；将非参数密度估计与水平集的形状表达相结合，构建满足线性函数空间约束的曲线族以描述水平集形状；通过非参数估计技术，从形状训练样本集中得到多种先验形状的混合统计概率模型；提出了基于形状先验约束的变分模型，实现底层数据驱动能量项和高层形状约束能量项的自适应调节。.2) 构建了心脏图像的多图谱集合和冠脉的局部区域统计信息模型，将血管形状信息同多尺度理论结合起来，基于定义的血管测度，实现了三维冠脉的精确自动分割；提出了一种基于改进的符号压力函数和局部图像拟合模型相结合的新型主动轮廓模型，通过图像局部信息和全局信息的有效结合，解决了血管图像灰度分布不均匀的分割问题；提出了一种基于先验形状和局部统计信息的血管分割方法，该方法有效地解决了多数血管分割技术难以处理的孤立和冗余点问题，实现了在保持血管分支的连通性的同时去除异常值。.3) 结合模糊距离变换理论，通过对中心线上模糊距离值的追踪和拟合，借助回归方法对冠脉进行狭窄进行分析，实现了基于模糊距离变换的狭窄检测和量化。该方法不仅能够定位狭窄的位置，而且还能量化狭窄的严重程度。.4) 结合项目研究成果，借助 MeVisLab医学图像处理与可视化框架，构建了冠状动脉分割与狭窄检测原型系统。.本项目的研究成果对于计算机辅助诊疗冠心病及相关疾病教学培训方面具有重要的理论意义和应用价值。