多模医学图像配准及融合方法研究
|
摘要
医学图像配准是近年来医学图像领域的研究热点,本文的主要目的是研究基于最大化互信息的多模态医学图像配准,主要是脑图像CT-MRI之间的配准。文中使用的标准实验数据来自开源的医学图像配准分割库ITK,使用的真实病人数据来自吉林大学第二医院CT室提供的有病变的CT、MRI图像,另外还有一部分实验数据来自RIRE项目。
本文首先介绍了课题的选题背景以及研究意义,图像配准的概念以及当前的研究现状。然后给出一个通用的医学图像配准框架,并根据该框架对医学图像的配准过程进行了详细的分析与说明,该框架主要由四个模块组成:几何变换模型、图像插值、相似性测度、优化函数。并重点介绍了互信息的基础理论,以及用于估计互信息的Parzen窗和Historgram估计。
接着介绍了互信息的计算过程,给出了互信息计算的图示和伪代码流程,并用一个最简单的示例演示了互信息的计算。因为在整个配准过程中,互信息的计算是最耗时的,所以针对两个主要参数像素样本数量和直方图bins的数量进行了实验,为了避免图像中组织器官细节的影响,使用的是ITK提供的标准实验图像。首先选取像素样本为固定图像总像素的50%,并改变直方图bins数量,实验结果表明bins取32或50左右的表现都较好。
然后固定bins为32和50,像素样本选取从固定图像总像素的5%~100%,结果表明当样本超过总像素的20%时都表现非常好,即没有必要使用太多的样本。之后,我们又针对真实病人的图像进行了实验,使用的参数值就是前面的实验得出的结果,几何变换模型是仿射变换,插值策略是简单的3阶B-样条插值。并给出了互信息值以及X轴随着迭代次数的变化曲线图,实验的结果也验证了前面使用标准图像得到的参数是正确的。
利用冗余Contourlet变换的多尺度、方向性、各向异性和平移不变性等优点以及基于区域能量的融合规则在选取融合系数上的优势,提出了一种基于冗余Contourlet变换和区域能量融合规则的图像融合算法,并引入粒子群优化算法求解阈值的全局最优解.实验结果表明:该算法是一种行之有效的图像融合算法。
今后的研究方向主要在:
1.针对腹部等易变形的组织器官进行研究,使用B-样条、偏微分方程等弹性模型进行实验研究。
2.由于局部优化函数容易出现局部极值而导致错误的结果,研究复杂的全局优化函数,并使用真实的病人数据进行实验。
3.研究3D医学图像的配准,以及使用GPU进行辅助加速。
4.由于最大化互信息配准测度丢失了图像的空间信息,从而影响了配准的精度,所以要结合图像的空间等信息进行研究,以更大程度地提高配准的精度,以及适用范围。
5.医学图像配准评估,由于没有统一的评价标准,所以对多模态医学图像配准的评价非常困难,评估模型尤其是针对多模态医学图像的评估模型显得非常有必要。
Medical image registration has been a heated research area in recent years, the main purpose of this paper is to study the maximization of mutual information based on multi-modal medical image registration, mainly brain CT-MRI image registration. The standard text used in the experimental data from the open source segmentation of medical image registration database ITK, using real patient data from the CT room of the Second Hospital of Jilin University, provided the diseased CT, MRI images, in addition to some experimental data from the RIRE project.
This paper introduces background and significance of the research, the concept of image registration and the current research status. Besides, it gives a general framework for medical image registration. The framework mainly consists of four modules: geometric transformation model, image interpolation, similarity measure, optimized function. Also, highlight the basic theory of mutual information, as well as Parzen window and Historgram estimates which used to estimate the mutual information.
Then introduced the mutual information calculation process, given the mutual information calculation process icon and pseudo-code, and use one of the simplest examples to illustrate the calculation of mutual information. Because in the registration process, the calculation of mutual information is the most time-consuming part, in order to avoid the image detail of tissues and organs, we use ITK provides standard test images. Besides, carry out experiments to two main parameters for the pixel number of samples and the number of histogram bins. First select a fixed image pixel sample 50% of the total pixels, and change the number of histogram bins, experimental results show that the performance of bins which obtained 32 or 50 are perform much better.
Then fixed bins of 32 and 50, selected from a fixed image pixel sample 5% of the total pixels to 100%, the results show that when the sample of more than 20% of the total pixels are doing very well, that is, there is no need to use too many samples.
Then we did experiments based on real patient's images, using the parameter value based on previous experimental results. The geometric transformation model is affine transformation; interpolation strategy is a simple 3-order B-spline interpolation. Given values of mutual information and X-axis changes with the number of iterations curve, experimental results also verify the previous parameters which obtained using standard image is correct.
We use the redundant Contourlet transform the multi-scale, orientation, anisotropy and the advantages of translation invariance and the fusion rules which based on regional energy integration in the selection coefficient of superiority, a transformation based on redundancy and regional energy integration Contourlet Rules Image fusion algorithms and particle swarm optimization algorithm for the introduction of thresholds for the global optimal solution.
The experimental results show that: this algorithm is a well-established image fusion algorithm.
本文首先介绍了课题的选题背景以及研究意义,图像配准的概念以及当前的研究现状。然后给出一个通用的医学图像配准框架,并根据该框架对医学图像的配准过程进行了详细的分析与说明,该框架主要由四个模块组成:几何变换模型、图像插值、相似性测度、优化函数。并重点介绍了互信息的基础理论,以及用于估计互信息的Parzen窗和Historgram估计。
接着介绍了互信息的计算过程,给出了互信息计算的图示和伪代码流程,并用一个最简单的示例演示了互信息的计算。因为在整个配准过程中,互信息的计算是最耗时的,所以针对两个主要参数像素样本数量和直方图bins的数量进行了实验,为了避免图像中组织器官细节的影响,使用的是ITK提供的标准实验图像。首先选取像素样本为固定图像总像素的50%,并改变直方图bins数量,实验结果表明bins取32或50左右的表现都较好。
然后固定bins为32和50,像素样本选取从固定图像总像素的5%~100%,结果表明当样本超过总像素的20%时都表现非常好,即没有必要使用太多的样本。之后,我们又针对真实病人的图像进行了实验,使用的参数值就是前面的实验得出的结果,几何变换模型是仿射变换,插值策略是简单的3阶B-样条插值。并给出了互信息值以及X轴随着迭代次数的变化曲线图,实验的结果也验证了前面使用标准图像得到的参数是正确的。
利用冗余Contourlet变换的多尺度、方向性、各向异性和平移不变性等优点以及基于区域能量的融合规则在选取融合系数上的优势,提出了一种基于冗余Contourlet变换和区域能量融合规则的图像融合算法,并引入粒子群优化算法求解阈值的全局最优解.实验结果表明:该算法是一种行之有效的图像融合算法。
今后的研究方向主要在:
1.针对腹部等易变形的组织器官进行研究,使用B-样条、偏微分方程等弹性模型进行实验研究。
2.由于局部优化函数容易出现局部极值而导致错误的结果,研究复杂的全局优化函数,并使用真实的病人数据进行实验。
3.研究3D医学图像的配准,以及使用GPU进行辅助加速。
4.由于最大化互信息配准测度丢失了图像的空间信息,从而影响了配准的精度,所以要结合图像的空间等信息进行研究,以更大程度地提高配准的精度,以及适用范围。
5.医学图像配准评估,由于没有统一的评价标准,所以对多模态医学图像配准的评价非常困难,评估模型尤其是针对多模态医学图像的评估模型显得非常有必要。
Medical image registration has been a heated research area in recent years, the main purpose of this paper is to study the maximization of mutual information based on multi-modal medical image registration, mainly brain CT-MRI image registration. The standard text used in the experimental data from the open source segmentation of medical image registration database ITK, using real patient data from the CT room of the Second Hospital of Jilin University, provided the diseased CT, MRI images, in addition to some experimental data from the RIRE project.
This paper introduces background and significance of the research, the concept of image registration and the current research status. Besides, it gives a general framework for medical image registration. The framework mainly consists of four modules: geometric transformation model, image interpolation, similarity measure, optimized function. Also, highlight the basic theory of mutual information, as well as Parzen window and Historgram estimates which used to estimate the mutual information.
Then introduced the mutual information calculation process, given the mutual information calculation process icon and pseudo-code, and use one of the simplest examples to illustrate the calculation of mutual information. Because in the registration process, the calculation of mutual information is the most time-consuming part, in order to avoid the image detail of tissues and organs, we use ITK provides standard test images. Besides, carry out experiments to two main parameters for the pixel number of samples and the number of histogram bins. First select a fixed image pixel sample 50% of the total pixels, and change the number of histogram bins, experimental results show that the performance of bins which obtained 32 or 50 are perform much better.
Then fixed bins of 32 and 50, selected from a fixed image pixel sample 5% of the total pixels to 100%, the results show that when the sample of more than 20% of the total pixels are doing very well, that is, there is no need to use too many samples.
Then we did experiments based on real patient's images, using the parameter value based on previous experimental results. The geometric transformation model is affine transformation; interpolation strategy is a simple 3-order B-spline interpolation. Given values of mutual information and X-axis changes with the number of iterations curve, experimental results also verify the previous parameters which obtained using standard image is correct.
We use the redundant Contourlet transform the multi-scale, orientation, anisotropy and the advantages of translation invariance and the fusion rules which based on regional energy integration in the selection coefficient of superiority, a transformation based on redundancy and regional energy integration Contourlet Rules Image fusion algorithms and particle swarm optimization algorithm for the introduction of thresholds for the global optimal solution.
The experimental results show that: this algorithm is a well-established image fusion algorithm.
引文
[1]陈明.三维医学图像刚性配准新算法研究[D].第一军医大学. 2003.
[2]胡顺波.凹函数测度和医学图像配准技术研究[D].山东大学. 2008.
[3]陈露斯,胡学锋,石锦平. MR I/CT融合与增强CT对鼻咽癌靶区勾画的比较[J].实用癌症杂志. 2010.25 (3): 172-174
[4] L ester H,A rridge SR. A survey of hierarchical non-linear medical image registration [J]. Pattern Recognition. 1999. 32 (2): 129-149.
[5] A Rangarajan, Chui Hail, Duncan James. Rigid point feature registration using mutual information [J]. Medical Image Analysis. 1999.3 (4):425-440.
[6] Antoine Maintz , Max A. Viergever. A survey of medical image registration [J]. Medical Image Analysis. 1998. 2 (1) : 1-36.
[7]张红颖.医学图像配准算法研究[D].天津大学. 2007.
[8]张见威,韩国强,张见东.医学解剖图像与功能图像配准技术综述[J].医疗设备信息. 2006.21 (6):90~94.
[9]李雄飞,张存利,李鸿鹏,减雪柏.医学图像配准技术进展[J].计算机科学,2010, 37(7): 27-33.
[10] Elsen P, Pol E, Viergever M. Medical image matching–a review with classifica- tion[J]. IEEE Engineering in medicine and Biology Magazine, 1993,12(1):26–39.
[11]卢振泰.医学图像配准算法研究[D].南方医科大学. 2008.
[12] Maintz JBA, Viergever MA. A survey of medical image registration[J], Medical Image Analysis,1998,2 (1):1-36.
[13] RIRE[EB/OL]: http://www.insight-journal.org/rire/
[14] Barbara Zitova′*, Jan Flusser,Image registration methods: a survey. Image and Vision Computing 21 (2003) 977–1000.
[15] Crum WR, Griffin LD, Hill DL, Hawkes DJ. Zen and the art of medical image registration:correspondence,homology,and quality[J]. Neuroimage.2003, 20(3): 1425-1437.
[16]田捷,赵明昌,何辉光.集成化医学影像算法平台理论与实践[M].北京:清华大学出版社,2004:211~212.
[17]王海南,郝重阳,雷方元,张先勇.非刚性医学图像配准研究综述[J].计算机工程与应用2005, 41(11).
[18] Lester H, Arridge SR. A survey of hierarchical non-linear medical image registration[J], Pattern Recognition,1999, 32(1): 129-149.
[19] Bookstein FL. Principal warps:thin-plate splines and the decomposition of deformations[J]. IEEE Trans on Pattern Analysis and Machine Intelligence,
[20] B. Kim et al, Mutual Information for Automated Unwarping of rat brain autoradiographs[J], Neuroimage,1997,5(1),31-40.
[21] Do M N,Vetterli M Contourlets:a directional muhiresolution image representation[C]//Proceedings of International Conference on Image Processing,2002,1:357-360
[22] Nunez J,Otazu X.Multiresolution-based image fusion with additive wavelet decomposition[J].IEEE Trans.Geoseience and Remote Sensing,1999,37(3):1204-1211
[23] Wang H,Peng J,Wu W .Fusion algorithm for multisensor images based on the discrete multiwavelet transform [C]//IEE Proceedings of Visual image signal processing,2002,149(5):1243-1246
[24] P. Hong and M. J. T. Smith,“An octave-band family of non-redundant directional filter banks,”in IEEE proc. ICASSP, vol. 2, pp. 1165-1168, 2002.
[25] Do M, VetterliM. The Contourlet Transform: An efficient directional multiresolution image representation [ J ] , IEEE Transactions on Im age Processing, 2003, 14 ( 12 ) : 2091-2106.
[26] da Cunha AL, Zhou J, DoMN. The nonsubsamp led contourlet transform: theory, design, and app lications[ J ].IEEE Trans Image Processing, 2006, 15 ( 10 ) : 3089- 3101.
[27] del Valle Y,Venayagamoorthy G K,Mohagheghi S,et al. Particle Swarm Optimization: Basic Concepts, Variants and Applications in Power Systems [J]. IEEE Transactions on Evolutionary Computation(S1089-778X),2008,12(2):171-195.
[28] Wang Zhou,Bovik A C.A universal image quality index.IEEE Signal Processing Letters,2002,9(3):81- 84
[29] Wang Zhou,Bovik A C,Sheikh H R,et a1.Image quality assessment l From error visibility to structural similarity.IEEE Transactions on Image Processing,2004, 13(4):600-612
[2]胡顺波.凹函数测度和医学图像配准技术研究[D].山东大学. 2008.
[3]陈露斯,胡学锋,石锦平. MR I/CT融合与增强CT对鼻咽癌靶区勾画的比较[J].实用癌症杂志. 2010.25 (3): 172-174
[4] L ester H,A rridge SR. A survey of hierarchical non-linear medical image registration [J]. Pattern Recognition. 1999. 32 (2): 129-149.
[5] A Rangarajan, Chui Hail, Duncan James. Rigid point feature registration using mutual information [J]. Medical Image Analysis. 1999.3 (4):425-440.
[6] Antoine Maintz , Max A. Viergever. A survey of medical image registration [J]. Medical Image Analysis. 1998. 2 (1) : 1-36.
[7]张红颖.医学图像配准算法研究[D].天津大学. 2007.
[8]张见威,韩国强,张见东.医学解剖图像与功能图像配准技术综述[J].医疗设备信息. 2006.21 (6):90~94.
[9]李雄飞,张存利,李鸿鹏,减雪柏.医学图像配准技术进展[J].计算机科学,2010, 37(7): 27-33.
[10] Elsen P, Pol E, Viergever M. Medical image matching–a review with classifica- tion[J]. IEEE Engineering in medicine and Biology Magazine, 1993,12(1):26–39.
[11]卢振泰.医学图像配准算法研究[D].南方医科大学. 2008.
[12] Maintz JBA, Viergever MA. A survey of medical image registration[J], Medical Image Analysis,1998,2 (1):1-36.
[13] RIRE[EB/OL]: http://www.insight-journal.org/rire/
[14] Barbara Zitova′*, Jan Flusser,Image registration methods: a survey. Image and Vision Computing 21 (2003) 977–1000.
[15] Crum WR, Griffin LD, Hill DL, Hawkes DJ. Zen and the art of medical image registration:correspondence,homology,and quality[J]. Neuroimage.2003, 20(3): 1425-1437.
[16]田捷,赵明昌,何辉光.集成化医学影像算法平台理论与实践[M].北京:清华大学出版社,2004:211~212.
[17]王海南,郝重阳,雷方元,张先勇.非刚性医学图像配准研究综述[J].计算机工程与应用2005, 41(11).
[18] Lester H, Arridge SR. A survey of hierarchical non-linear medical image registration[J], Pattern Recognition,1999, 32(1): 129-149.
[19] Bookstein FL. Principal warps:thin-plate splines and the decomposition of deformations[J]. IEEE Trans on Pattern Analysis and Machine Intelligence,
[20] B. Kim et al, Mutual Information for Automated Unwarping of rat brain autoradiographs[J], Neuroimage,1997,5(1),31-40.
[21] Do M N,Vetterli M Contourlets:a directional muhiresolution image representation[C]//Proceedings of International Conference on Image Processing,2002,1:357-360
[22] Nunez J,Otazu X.Multiresolution-based image fusion with additive wavelet decomposition[J].IEEE Trans.Geoseience and Remote Sensing,1999,37(3):1204-1211
[23] Wang H,Peng J,Wu W .Fusion algorithm for multisensor images based on the discrete multiwavelet transform [C]//IEE Proceedings of Visual image signal processing,2002,149(5):1243-1246
[24] P. Hong and M. J. T. Smith,“An octave-band family of non-redundant directional filter banks,”in IEEE proc. ICASSP, vol. 2, pp. 1165-1168, 2002.
[25] Do M, VetterliM. The Contourlet Transform: An efficient directional multiresolution image representation [ J ] , IEEE Transactions on Im age Processing, 2003, 14 ( 12 ) : 2091-2106.
[26] da Cunha AL, Zhou J, DoMN. The nonsubsamp led contourlet transform: theory, design, and app lications[ J ].IEEE Trans Image Processing, 2006, 15 ( 10 ) : 3089- 3101.
[27] del Valle Y,Venayagamoorthy G K,Mohagheghi S,et al. Particle Swarm Optimization: Basic Concepts, Variants and Applications in Power Systems [J]. IEEE Transactions on Evolutionary Computation(S1089-778X),2008,12(2):171-195.
[28] Wang Zhou,Bovik A C.A universal image quality index.IEEE Signal Processing Letters,2002,9(3):81- 84
[29] Wang Zhou,Bovik A C,Sheikh H R,et a1.Image quality assessment l From error visibility to structural similarity.IEEE Transactions on Image Processing,2004, 13(4):600-612