King Martin, Giger Maryellen L, Suzuki Kenji, Pan Xiaochuan
Department of Radiology, Committee on Medical Physics, The University of Chicago, Chicago, Illinois 60637, USA.
Med Phys. 2007 Dec;34(12):4860-75. doi: 10.1118/1.2794172.
In coronary calcium scoring, motion artifacts affecting calcified plaques are commonly characterized using descriptive terms, which incorporate an element of subjectivity in their interpretations. Quantitative indices may improve the objective characterization of these motion artifacts. In this paper, an automated method for generating 12 quantitative indices, i.e., features that characterize the motion artifacts affecting calcified plaques, is presented. This method consists of using the rapid phase-correlated region-of-interest (ROI) tracking algorithm for reconstructing ROI images of calcified plaques automatically from the projection data obtained during a cardiac scan, and applying methods for extracting features from these images. The 12 features include two dynamic, six morphological, and four intensity-based features. The two dynamic features are three-dimensional (3D) velocity and 3D acceleration. The six morphological features include edge-based volume, threshold-based volume, sphericity, irregularity, average margin gradient, and variance of margin gradient. The four intensity-based features are maximum intensity, mean intensity, minimum intensity, and standard deviation of intensity. The 12 features were extracted from 54 reconstructed sets of simulated four-dimensional images from the dynamic NCAT phantom involving six calcified plaques under nine heart rate/multi-sector gating combinations. In order to determine how well the 12 features correlated with a plaque motion index, which was derived from the trajectory of the plaque, partial correlation coefficients adjusted for heart rate, number of gated sectors, and mean feature values of the six plaques were calculated for all 12 features. Features exhibiting stronger correlations ([r] epsilon [0.60,1.00]) with the motion index were 3D velocity, maximum intensity, and standard deviation of intensity. Features demonstrating stronger correlations ([r] epsilon [0.60, 1.00]) with other features mostly involved intensity-based features. Edge-based volume/irregularity and average margin gradient/variance of margin gradient were the only two feature pairs out of 12 with stronger correlations that did not involve intensity-based features. Automatically extracted features of the motion artifacts affecting calcified plaques in cardiac computed tomography images potentially can be used to develop models for predicting image assessability with respect to motion artifacts.
在冠状动脉钙化评分中，影响钙化斑块的运动伪影通常使用描述性术语来表征，这些术语在解释中包含主观性因素。定量指标可能会改善对这些运动伪影的客观表征。本文提出了一种自动生成12个定量指标的方法，即表征影响钙化斑块的运动伪影的特征。该方法包括使用快速相位相关感兴趣区域（ROI）跟踪算法，从心脏扫描期间获得的投影数据中自动重建钙化斑块的ROI图像，并应用从这些图像中提取特征的方法。这12个特征包括两个动态特征、六个形态特征和四个基于强度的特征。两个动态特征是三维（3D）速度和3D加速度。六个形态特征包括基于边缘的体积、基于阈值的体积、球形度、不规则度、平均边缘梯度和边缘梯度方差。四个基于强度的特征是最大强度、平均强度、最小强度和强度标准差。这12个特征是从涉及六个钙化斑块的动态NCAT体模的54组模拟四维图像重建集中提取的，这些图像处于九种心率/多扇区门控组合下。为了确定这12个特征与从斑块轨迹得出的斑块运动指数的相关性有多好，针对所有12个特征计算了针对心率、门控扇区数量和六个斑块的平均特征值进行调整的偏相关系数。与运动指数表现出更强相关性（[r]∈[0.60,1.00]）的特征是3D速度、最大强度和强度标准差。与其他特征表现出更强相关性（[r]∈[0.60,1.00]）的特征大多涉及基于强度的特征。基于边缘的体积/不规则度和平均边缘梯度/边缘梯度方差是12个特征对中仅有的两个具有较强相关性且不涉及基于强度特征的特征对。心脏计算机断层扫描图像中影响钙化斑块的运动伪影的自动提取特征可能可用于开发预测运动伪影图像可评估性的模型。
