通道化Hotelling观察器在PET系统图像质量评估中的应用研究

发布时间：2018-04-11 19:19

本文选题：正电子发射断层扫描技术 + 主观评价法　；参考：《杭州电子科技大学》2017年硕士论文

【摘要】：正电子发射断层扫描技术(Positron Emission Tomography,PET)是核医学领域中一种先进的临床检查影像技术,可以准确地反映机体内部的生理、生化变化,进入21世纪后更是迅猛发展,已成为目前癌症的早期诊断和脑功能成像中非常重要的检查手段。PET图像质量评估是PET成像中一个极为重要的方面,也是对PET成像系统参数进行优化的关键步骤。主观评价法(Human Observer)是目前评估图像质量的主要手段,但并不适用于系统参数优化的研究,因为需要花费医生大量的时间和精力,为了结果的客观性需要多名医生共同参与。数字观察器(Numerical Observers)可以很好地克服这些缺点,被越来越多的应用于图像质量评估中。在众多数字观察器中,考虑到应用的普遍性,本文选择通道化Hotelling观察器(Channelized Hotelling Observer,CHO)作为PET图像质量评估的方法。论文第一部分首先介绍了目前常用的图像质量评估方法包括本文主要用到的数字观察器CHO及其研究现状、PET系统成像原理和图像重建过程、以及影响重建图像质量的因素;然后介绍本文数据的获取方式GATE(Geant4 Application for Tomographic Emission)仿真方法。论文第二部分应用CHO评估PET系统参数对图像质量的影响。影响PET图像质量的因素包括硬件参数和软件参数,其中硬件参数主要有晶体厚度和轴向视野(Axial Field of View,AFOV),软件参数主要有滤波器和迭代次数。硬件参数晶体厚度、轴向视野(AFOV)对图像质量的影响:晶体厚度和轴向视野(AFOV)是硬件中两个主要因素,厚度和AFOV的增加能增加系统灵敏度并改善图像质量,但是成本也会随之提高,扫描速度会下降。经过对灵敏度,扫描速度以及成本问题的综合考虑,本文将轴向视野分别设置为201.6mm和403.2mm,晶体厚度分别设置为15mm和17.5mm。用CHO评估它们对图像质量的影响,得到的结果显示晶体厚度一样时,轴向视野为403.2mm的系统生成的PET图像比轴向视野为201.6mm系统的清晰,轴向视野一样时,晶体厚度为17.5mm的系统生成的PET图像比晶体厚度为15mm系统的清晰。软件参数滤波器、迭代次数对图像质量的影响:滤波器和迭代次数是经常研究的两个软件参数,它们的取值对图像质量的影响非常大,通常图像质量会随着它们的参数值的增加先变好后变差,因此为了得到最好的图像质量,确定它们的最佳参数非常重要,CHO可以对图像质量进行量化从而选择最佳参数。首先通过对图像质量、仿真时间等因素的综合考虑选择3min为最佳扫描时间。其次本文用到的滤波器为高斯滤波器和非局部均值(Non-Local Mean,NLM)滤波器,并将高斯参数σ分别设置为0.5、0.8、1.1、1.4、2.0、3.0、4.0,NLM参数f分别设置为0.5、0.7、0.9、1.2、2.0、3.0、4.0。通过CHO对它们进行优化,并评估两种滤波器在最佳参数下的滤波效果。结果表明σ=1.4和f=0.5为高斯滤波器和NLM滤波器的最佳参数,而且NLM滤波器的滤波效果比高斯滤波器好。最后本文将迭代次数分别设置为1、2、3、4、6、8、10次,当迭代次数变大,图像变清晰,但是由于图像中的噪声也被放大,当超过某个值后,图像因噪声影响变模糊。为了得到最佳迭代次数,本文用CHO对上述迭代次数对应的图像进行评估,结果表明迭代次数为4次时,图像质量最好。仿真结果说明CHO可以用于评估PET系统参数对图像质量的影响,其对PET研制以及PET图像的辅助诊断具有重要意义。本文的创新之处在于研究者们通过CHO计算AUC值可以定量的确定参数,并且这是国内首先开展采用CHO方法评估PET系统参数对图像质量影响的研究。
[Abstract]:Positron emission tomography (Positron Emission Tomography, PET) is an advanced clinical examination of nuclear medicine imaging technology, can accurately reflect the physiological and biochemical changes within the organism, after entering twenty-first Century is the rapid development, has become the current inspection methods of.PET image quality assessment is very important for early diagnosis and functional imaging of brain cancer which is a very important aspect in PET imaging, is also a key step for the optimization of PET imaging system parameters. The subjective evaluation method (Human Observer) is the main means of evaluating image quality, but is not suitable for the parameter optimization of the system, because the doctor needs to spend a lot of time and energy, in order to the objectivity of the results takes more than doctors to participate. The digital observer (Numerical Observers) can overcome these shortcomings, has been applied to more and more Image quality assessment. In many digital viewer, considering the universality of application, this paper selects the channel of Hotelling observer (Channelized Hotelling Observer, CHO PET) as a method of image quality assessment. The first part introduces the research status of CHO and the observer digital image quality assessment methods are used in this paper include. The principle of PET imaging and image reconstruction process, and the factors affecting the quality of the reconstructed image; then describes the data acquisition mode (GATE Geant4 Application for Tomographic Emission) simulation method. The second part CHO was used to evaluate the effect of PET system parameters on the image quality. The factors affecting the quality of PET image including hardware parameters and software parameters, including the hardware parameters include the crystal thickness and axial view (Axial Field of View, AFOV), the main parameters of the software The filter and the number of iterations. The hardware parameters of crystal thickness, axial view (AFOV) effect on image quality: crystal thickness and axial view (AFOV) are the two main factors in hardware, increase the thickness and AFOV can increase the sensitivity of the system and improve the image quality, but the cost will be increased, the scanning speed will drop after. The sensitivity, considering the scanning speed and cost problems, the axial field of view were set to 201.6mm and 403.2mm respectively, the crystal thickness is set to 15mm and 17.5mm. using CHO to evaluate their impact on image quality, the results showed the same crystal thickness, axial view PET 403.2mm image generation system than axial view the 201.6mm system is clear, the axial view as when the crystal thickness is PET 17.5mm image generation system than the crystal thickness of 15mm system clear. Software filter parameters, Diego Effect of number of image quality: filter and the number of iterations is two software parameters frequently studied, their influence on the image quality value is very large, usually the image quality will be increased with the increase of their parameter values becomes better then worse, so in order to get the best image quality, to determine the optimal parameters of them very important, CHO can quantify the image quality so as to choose the optimal parameters. Based on the image quality, considering the simulation time and other factors the choice of 3min is the best scan time. Then filter used in this paper for the Gauss filter and the non local means (Non-Local, Mean, NLM) filter, and Gauss parameters were set up 0.5,0.8,1.1,1.4,2.0,3.0,4.0, NLM parameters of F were set to 0.5,0.7,0.9,1.2,2.0,3.0,4.0. by CHO to optimize them, and evaluate the two kinds of filters under the best parameters of the filter Results. The results show that the best parameters =1.4 and f=0.5 for the Gauss filter and NLM filter, and the filtering effect of NLM filter Gauss filter. Than the number of iterations were set for 1,2,3,4,6,8,10 times, when the number of iterations becomes larger, the image becomes clear, but because of the noise in the image is magnified, when more than one value, because the influence of noise image blur. In order to get the optimal number of iterations, the image of the number of iterations corresponding with the CHO evaluation results show that the iterative number is 4 times, the best image quality. The simulation results indicate that CHO can be used to evaluate the effect on image quality parameters of PET system, which is of great significance to PET development and diagnosis PET images. The innovation of this paper lies in the researchers through the CHO AUC calculation parameters can be determined quantitatively, and this is the first to carry out using CHO Methods to evaluate the influence of PET system parameters on image quality.

【学位授予单位】：杭州电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41;R817

【参考文献】