基于变电压图像序列盲分离的X射线多谱CT成像

发布时间：2018-05-20 04:46

  本文选题：X射线图像分解 + 多能谱CT　； 参考：《中北大学》2016年博士论文

【摘要】：X射线多谱CT成像相对于传统单能假设的CT成像,其能谱信息更丰富,可依据能谱与衰减系数的多谱对应性,实现检测对象组分的有效区分,满足新型材料、矿石深加工以及现代医学中组分微观结构等定量化功能成像需求。现有的双能CT、光子计数型CT、能谱滤波分离多谱CT在时间和能谱分辨率上存在一定的局限性,而单色性较好的同步辐射CT为国家大科学装置,共享面较宽,机时有限,限制了样品测试实验效率和规模。对此,论文在不改变常规CT成像系统物理组成的基础上,研究连续谱X射线投影图像能谱分离方法,期望分解所得投影的重建图像能达到窄能谱CT图像效果,从而在常规CT成像系统上实现组分有效区分,为微观结构的定量表征提供支撑。论文在研究X射线多谱衰减特性、多谱成像特点和单能假设CT重建的基础上,分析了连续谱X射线投影序列分解的可行性,提出了基于盲源分离的多谱序列分离的CT成像方法。在此基础上,采用非负矩阵分解方法,以误差平方和最小为优化准则,构建X射线图像序列分解模型,并推导了求解算法。同时针对分解算法的局部收敛性问题,引入具有全局收敛的遗传算法,提出了基于遗传算法的X射线图像序列非负矩阵分解方法,实现了多谱序列的有效分离,获取的投影重建后图像有窄能谱图像特征。针对CT成像过程中散射因素的影响,利用分解残差的局部方差和作为散射信号低频特性的描述,以分解残差的局部方差和最小为优化准则,改进了X射线图像序列非负矩阵分解模型,提高了多谱序列的分离精度。针对盲源分离结果的无序性和窄谱投影的能量不确定性,利用光电效应和康普顿效应分解代替分解模型中的衰减系数,提出了基于衰减系数分解的能谱校准方法,实现了窄谱投影的能量指向;同时,考虑CT图像序列表征问题,耦合多谱成像的物理先验,利用基于DCM模型的窄谱CT序列融合算法,实现了组分的定量表征。论文在理论研究同时,进行实验对比分析。实验过程中,以硅铝材质构成的圆柱体为对象,通过仿真与实验相结合的方法,验证了X射线投影序列分离的可行性,以及所提出的基于遗传算法的X射线图像序列非负矩阵分解方法、改进方法、以及能谱校准和窄谱CT序列融合算法。结果表明,论文提出的基于盲源分离的多谱CT成像方法,可在不改变系统物理组成的基础上,通过扫描模式和数据处理方法的创新,实现材料组分定量表征。这对提高实验效率,降低成本,促进X射线多谱反演与定量检测技术的发展,具有重要理论意义和应用价值。
[Abstract]:Compared with conventional single-energy CT imaging, X-ray multispectral CT imaging has more information of energy spectrum. It can effectively distinguish the components of detecting objects according to the multi-spectral correspondence between energy spectrum and attenuation coefficient, and meet the requirements of new materials. Quantitative functional imaging needs such as ore deep processing and the microstructure of components in modern medicine. The existing dual-energy CTs, photon counting CTs and spectral filtering multi-spectral CT have some limitations in time and spectral resolution, while the monochromatic synchrotron radiation CT is a large national scientific device with a wide shared area and limited computer time. The efficiency and scale of the sample test are limited. Therefore, on the basis of not changing the physical composition of conventional CT imaging system, this paper studies the method of separating the energy spectrum of continuous spectrum X-ray projection image. It is expected that the reconstructed image can achieve the effect of narrow spectrum CT image. Thus, components can be effectively distinguished in conventional CT imaging system, which provides support for quantitative characterization of microstructure. On the basis of studying the characteristics of X-ray multispectral attenuation, multispectral imaging and single-energy assumption CT reconstruction, the feasibility of continuous spectral X-ray projection sequence decomposition is analyzed, and the CT imaging method based on blind source separation is proposed. On this basis, the non-negative matrix decomposition method is used to construct the X ray image sequence decomposition model with the minimum sum of error square as the optimization criterion, and the solution algorithm is derived. At the same time, aiming at the problem of local convergence of decomposition algorithm, a genetic algorithm with global convergence is introduced, and a non-negative matrix decomposition method of X-ray image sequence based on genetic algorithm is proposed, which realizes the effective separation of multi-spectral sequences. The reconstructed images are characterized by narrow-spectrum images. In view of the influence of scattering factors in CT imaging, the local variance of decomposition residuals is used as the description of the low frequency characteristics of scattering signals, and the local variance and minimum of decomposition residuals are taken as the optimization criteria. The nonnegative matrix decomposition model of X-ray image sequence is improved and the separation accuracy of multi-spectral sequence is improved. Aiming at the disorder of blind source separation results and the energy uncertainty of narrow spectral projection, a calibration method for energy spectrum based on attenuation coefficient decomposition is proposed, which is based on the decomposition of photovoltaic effect and Compton effect instead of the attenuation coefficient in the decomposition model. At the same time, considering the representation of CT image sequence, coupled with the physical priori of multispectral imaging, the fusion algorithm of narrow-spectrum CT sequence based on DCM model is used to realize the quantitative representation of components. At the same time, the thesis carries on the experiment contrast analysis in the theory research. In the course of experiment, the feasibility of separating X-ray projection sequence is verified by combining simulation and experiment with the cylinder made of silicon and aluminum. The non-negative matrix decomposition method of X-ray image sequence based on genetic algorithm, the improved method, and the fusion algorithm of spectral calibration and narrow-spectrum CT sequence are also presented. The results show that the proposed multispectral CT imaging method based on blind source separation can realize quantitative characterization of material components without changing the physical composition of the system and through the innovation of scanning mode and data processing methods. It is of great theoretical significance and application value to improve the experimental efficiency, reduce the cost and promote the development of X-ray multispectral inversion and quantitative detection technology.
【学位授予单位】：中北大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：TP391.41;O434.1
，

本文编号：1913265