Creating an anthropomorphic digital MR phantom-an extensible tool for comparing and evaluating quantitative imaging algorithms
Creating an anthropomorphic digital MR phantom-an extensible tool for comparing and evaluating quantitative imaging algorithms
复制标题
DOI：
10.1088/0031-9155/61/2/974
复制
发表时间：
2016-01-21
影响因子：
3.5
通讯作者：
Jackson, Edward F.
中科院分区：
文献类型：
--
作者：
Bosca, Ryan J.;Jackson, Edward F.
评估和减少与图像量化算法相关的各种偏差和方差的来源对于在临床研究和实践中使用此类算法是至关重要的。评估通常是通过基于网格的数字参考对象(DRO)或最近基于正常人体解剖的数字拟人幻影来完成的。公开可用的数字拟人幻影可以为生成包含病理学中常见的异质性的真实的基于模型的Dros提供基础。利用公共可用的血管输入功能(VIF)和正常人脑的数字拟人模型，开发了一种基于代表真实和异质增强病理的通用动力学模型(GKM)的DRO生成方法。根据明确的临床动态对比增强(DCE)MRI检查估计GKM参数。这一临床成像体积与离散组织模型共同配准，并使用从临床图像估计的模型参数来合成由正常脑组织和不均匀增强的脑肿瘤组成的DCE-MRI检查。空间平滑的一个应用实例被用来说明在评估定量成像算法方面的潜在应用。体素Bland-Altman分析表明，在使用空间平滑和不使用空间平滑(使用小半径高斯核)的情况下，估计的参数之间的差异可以忽略不计。在这项工作中，我们报告了一种可扩展的方法来生成基于模型的拟人Dros，该Dros包含正常和病理组织，可以用于评估定量成像算法。
Assessing and mitigating the various sources of bias and variance associated with image quantification algorithms is essential to the use of such algorithms in clinical research and practice. Assessment is usually accomplished with grid-based digital reference objects (DRO) or, more recently, digital anthropomorphic phantoms based on normal human anatomy. Publicly available digital anthropomorphic phantoms can provide a basis for generating realistic model-based DROs that incorporate the heterogeneity commonly found in pathology. Using a publicly available vascular input function (VIF) and digital anthropomorphic phantom of a normal human brain, a methodology was developed to generate a DRO based on the general kinetic model (GKM) that represented realistic and heterogeneously enhancing pathology. GKM parameters were estimated from a deidentified clinical dynamic contrast-enhanced (DCE) MRI exam. This clinical imaging volume was co-registered with a discrete tissue model, and model parameters estimated from clinical images were used to synthesize a DCE-MRI exam that consisted of normal brain tissues and a heterogeneously enhancing brain tumor. An example application of spatial smoothing was used to illustrate potential applications in assessing quantitative imaging algorithms. A voxel-wise Bland-Altman analysis demonstrated negligible differences between the parameters estimated with and without spatial smoothing (using a small radius Gaussian kernel). In this work, we reported an extensible methodology for generating model-based anthropomorphic DROs containing normal and pathological tissue that can be used to assess quantitative imaging algorithms.