医学图像配准中的相似性测试研究

【作者】 陈芳；

【导师】 张建秋；

【作者基本信息】 复旦大学 ， 电路与系统， 2009， 硕士

【摘要】 医学成像技术的蓬勃发展为临床提供了大量实用的解剖及功能方面的影像数据。临床上常常需要配准并融合同模态或者不同模态的医疗图像,以获取更全面的病患信息,指导临床诊断和治疗。医学图像配准技术就成了目前医学图像处理和分析中基础及关键技术,具有重要的临床应用价值,而配准的相似性测度直接决定了配准的鲁棒性和有效性。本文对刚体图像、序列刚体图像以及非刚体图像现有的配准方法中的相似性测度进行了研究,针对其存在的不足,提出了适用于各种配准应用的新方法。本文的主要内容包括:针对刚体医学图像配准的相似性测度进行了研究。为提高其配准鲁棒性,本文提出了一种新的自适应指数加权的互信息(Adaptive Exponential WeightedMutual Information,AEWMI)测度,该测度通过对互信息测度进行指数加权可以提高测度曲线的峰值尖锐性和平滑性;通过评估待配准图像的质量和分辨率大小来自适应确定指数权值,进一步提高了配准测度的鲁棒性和普适性。仿真实验结果表明,基于本文所提测度的配准方案,对图像噪声、分辨率差异等有较高的鲁棒性,且可有效地提高配准的成功率。针对序列医学图像配准方法中的相似性测度进行了创新。传统的序列图像配准方法大多基于两幅图像的配准方法,并且在低信噪比的情况下面临着鲁棒性问题。为解决此问题,本文提出了一种全新的适用于序列图像配准的相似性测度--归一化高阶互累积量系数(Normalized Higher-order Cross-cumulant Coefficient,NHCC)。分析表明:本文提出的NHCC测度不仅可以同时敏感多幅图像之间的相关程度,而且能免疫加性高斯噪声对配准的影响。仿真和实验结果验证了该测度的有效性和优越性。针对非刚性空间变换的数字减影图像(Digital Subtraction Angiography,DSA)配准,本文提出了一种新的基于块图像内容自适应选择相似性测度的匹配方法。它根据DSA图像的血流特性对块图像进行分类,对不同内容的块图像采用不同的相似性测度,分析表明该方法可有效地减小血管成分和图像模糊等对配准精度造成的影响,提高图像整体配准的鲁棒性。结合边缘控制点提取以及薄板样条(Thin Plate Spline,TPS)非线性变换,我们对医院采集到的DSA图像进行配准实验,实验结果验证了本文提出匹配方法的有效性。更多还原

【Abstract】 Along with the fast development of the medical imaging technology,more and more anatomic and/or functional medical images could be obtained in clinic.It is always required to register mono-modality and multimodality medical images in order to obtain patients’ all-round information,which is helpful in clinical diagnosis and therapy.Medical image registration is a key technology of medical image processing and analysis and has significant meanings in clinical analysis.This dissertation mainly focuses on the registration methods for rigid images, rigid image sequences and non-rigid images.In view of their disadvantages,this dissertation proposes several new and robust registration methods respectively.The main contents of this dissertation are as follows:In order to improve the robustness for rigid images registration,a new similarity measure for medical image registration,called Adaptive Exponential Weighted Mutual Information(AEWMI),is proposed.AEWMI could improve the smoothness and peak keenness performance of similarity measurement curve via adaptively weighting MI according to image quality and resolution.The experimental results,demonstrating the analyzing result,show that our method is robust to image noise or resolution difference and improves the registration robustness effectively.This dissertation innovate the registration method for medical image sequences. A novel registration measure,called as Normalized Higher-order Cross-cumulant Coefficient(NHCC) is proposed.The analytical results show that the proposed NHCC cannot only easily capture the correlation information among the multiple variables simultaneously,but also suppress the additive Gaussian noise influence on the image registration results.While simulation results verify the effectiveness and robustness of the proposed measure,better experimental results of the digital subtraction angiography image registration are also obtained.Under the non-rigid image registration frame,the dissertation focuses on studying an accurate and effective processing method for digital subtraction angiography(DSA) image registration.An image registration method was proposed to be based on the similarity measurement of block image content with self-adaptive selection.In the process of digital subtraction angiography image registration,block image content was classified with blood flow character and for different block image content different similarity measurement was selected with self-adaptation.The analyses and experimental results demonstrated that our method could decrease the influences of the image blurring and vessels on the accuracy of the DSA image registration.The proposed method can effectively improve the robustness and accuracy of DSA image registration.更多还原