Département de physique, Université de Montréal, Complexe des sciences, 1375 Avenue Thérèse-Lavoie-Roux, Montréal, Québec H2V 0B3, Canada. Centre de recherche du Centre hospitalier de l'Université de Montréal, 900 Rue Saint-Denis, Montréal, Québec, H2X 3H8, Canada.
Phys Med Biol. 2020 Jul 28;65(15):155001. doi: 10.1088/1361-6560/ab8107.
The purpose of this work is, firstly, to propose an optimized parametrization of the attenuation coefficient to describe human tissues in the context of projection-based material characterization with multi-energy CT. The approach is based on eigentissue decomposition (ETD). Secondly, to evaluate its benefits in terms of accuracy and precision of radiotherapy-related parameters against established parametrizations. The attenuation coefficient is parametrized as a linear combination of virtual materials, eigentissues, obtained by performing principal component analysis on a set of reference tissues in order to optimally represent human tissue composition. Two implementations of ETD are compared with other pre-reconstruction formalisms established for dual-energy and photon-counting CT in a simulation framework. The first implementation uses a single set of eigentissues to describe all human tissues, while the second uses different sets of eigentissues to characterize soft tissues and bones, and includes a post-reconstruction classification step. The simulation framework evaluates the reconstruction accuracy of various radiotherapy-related quantities over a range of 71 human tissues for various noise levels. Compared to conventional parametrizations, the first implementation of ETD reduces the mean error and root-mean-square error (RMSE) in two radiotherapy-related quantities (the proton stopping power and the mass energy absorption coefficient of 21 keV photons from Pd seeds used in brachytherapy) for all noise levels and modalities investigated. This illustrates that a decomposition basis selected with principal component analysis is superior to an arbitrary pair of materials to describe human tissues. The mean error on radiotherapy-related parameters can be further reduced with the classification-based approach. In the context of pre-reconstruction material characterization with multi-energy CT, parametrizing the attenuation coefficient with eigentissues provides a more accurate and precise evaluation of human tissues properties for radiotherapy. Accurate quantification can thus be achieved without the need to parametrize tissues using unphysical parameters, such as the energy-dependent effective atomic number.
这项工作的目的首先是提出一种优化的衰减系数参数化方法，用于在基于投影的多能 CT 物质特征化中描述人体组织。该方法基于特征组织分解（Eigentissue Decomposition，ETD）。其次，评估其在放疗相关参数的准确性和精密度方面相对于已有参数化方法的优势。衰减系数被参数化为虚拟物质的线性组合，通过对一组参考组织进行主成分分析来获得特征组织，以最佳地表示人体组织组成。在模拟框架中，将两种 ETD 实现与其他为双能和光子计数 CT 建立的预重建形式主义进行了比较。第一种实现使用单个特征组织集来描述所有人体组织，而第二种实现使用不同的特征组织集来描述软组织和骨骼，并包括后重建分类步骤。模拟框架评估了各种放疗相关数量在各种噪声水平下 71 个人体组织范围内的重建准确性。与传统参数化方法相比，第一种 ETD 实现降低了所有噪声水平和模态下两种放疗相关数量（质子阻止本领和用于近距离治疗的 Pd 种子的 21keV 光子的质量能量吸收系数）的平均误差和均方根误差（RMSE）。这表明，与任意一对材料相比，使用主成分分析选择的分解基更好地描述了人体组织。基于分类的方法可以进一步降低放疗相关参数的平均误差。在多能 CT 预重建物质特征化的背景下，用特征组织参数化衰减系数可以更准确、更精确地评估放疗中的人体组织特性。因此，无需使用不真实的参数（如能量相关有效原子数）对组织进行参数化，就可以实现准确的量化。
