Material Decomposition in Spectral CT Using Deep Learning: A Sim2Real Transfer Approach
Material Decomposition in Spectral CT Using Deep Learning: A Sim2Real Transfer Approach
复制标题
DOI：
10.1109/access.2021.3056150
复制
发表时间：
2021-01-01
期刊：
影响因子：
3.9
通讯作者：
Peyrin, Francoise
中科院分区：
文献类型：
--
作者：
Abascal, Juan F. P. J.;Ducros, Nicolas;Peyrin, Francoise
用于解决光谱计算机断层扫描中的非线性材料分解问题的现有技术是基于变分方法，但是这些方法在计算上是缓慢的并且严重依赖于正则化泛函的特定选择。卷积神经网络已经被提出来解决这些问题。然而，学习算法需要大量的实验数据集。我们提出了一种基于U-Net架构和Sim 2 Real迁移学习方法的深度学习策略来解决材料分解问题，其中我们从合成数据中学到的知识被转移到现实世界的场景中。为了使这种方法发挥作用，合成数据必须是真实的，并代表实验数据。为此，从KiTS 19挑战数据集的人体CT体积生成数值体模，分割成特定材料（软组织和骨）。将这些体积投影到正弦图空间中，以模拟光子计数数据，同时考虑扫描仪的能量响应。我们比较了基于投影和基于图像的分解方法，其中训练网络在投影或图像域中分解材料。提出的Sim 2 Real传输策略进行了比较，正则化高斯-牛顿（RGN）方法的合成数据，实验体模数据和人类胸部数据。
The state-of-the art for solving the nonlinear material decomposition problem in spectral computed tomography is based on variational methods, but these are computationally slow and critically depend on the particular choice of the regularization functional. Convolutional neural networks have been proposed for addressing these issues. However, learning algorithms require large amounts of experimental data sets. We propose a deep learning strategy for solving the material decomposition problem based on a U-Net architecture and a Sim2Real transfer learning approach where the knowledge that we learn from synthetic data is transferred to a real-world scenario. In order for this approach to work, synthetic data must be realistic and representative of the experimental data. For this purpose, numerical phantoms are generated from human CT volumes of the KiTS19 Challenge dataset, segmented into specific materials (soft tissue and bone). These volumes are projected into sinogram space in order to simulate photon counting data, taking into account the energy response of the scanner. We compared projection- and image-based decomposition approaches where the network is trained to decompose the materials either in the projection or in the image domain. The proposed Sim2Real transfer strategies are compared to a regularized Gauss-Newton (RGN) method on synthetic data, experimental phantom data and human thorax data.