基于聚类分析提取DSC-MRI脑灌注的动脉输入函数
发布时间:2018-11-15 23:41
【摘要】:使用内源性对比剂的动态敏感对比磁共振成像(Dynamic Susceptibility Contrast-Magnetic Resonance Imaging, DSC-MRI)已经广泛应用于脑灌注加权成像,可用于测量脑血流、脑血容积、平均通过时间等与脑血液动力学参数有关的生理指标,在临床疾病诊断和治疗方案选择方面发挥着重要作用,因此DSC-MRI快速、准确和鲁棒的定量计算对临床实践意义重大。在利用DSC-MRI技术进行脑血液动力学参数定量时,需事先获得动脉输入函数,动脉输入函数的准确性将直接影响最终计算结果。传统情况下,动脉输入函数的提取依赖经验丰富的放射医师手工选取大脑中动脉或颈内动脉的若干像素点来实现,然而该种手动方法耗时较长,且对操作者依赖,导致不同操作者间和同一操作者不同时间点间的结果缺乏可重复性,同时由于DSC-MRI图像的空间分辨率相对较低,基于手动方法提取动脉输入函数的结果也会受到部分容积效应的严重污染。因此,开发能够减少人为干预的自动或半自动动脉输入函数提取算法成为一个迫切需要解决的现实问题。为了解决手动提取动脉输入函数的弊端,本研究评估了不同簇分析算法提取DSC-MRI脑灌注中动脉输入函数的效能。具体步骤包括:采集42位健康志愿者的DSC-MRI脑灌注加权图像;利用离线工作站校正由于呼吸、心跳、被试者难以控制的不自主运动或转动造成各相位容积图像不对齐的情况;通过手工浏览方式选择首幅容积图像中含右水平大脑中动脉的扫描层面;将所选层面图像信号的时间-强度曲线转化为对比剂的时间-浓度曲线;删除曲线下面积较小的曲线、震荡频率较严重的曲线和受部分容积效应污染严重的曲线;最后,将各种簇分析技术应用于剩余曲线,自动提取动脉输入函数,并比较各种簇分析算法在计算动脉输入函数方面的准确性、可重复性及复杂度。由于临床实验缺乏金标准的支持,因此本研究还增添了模拟实验部分,通过估计的动脉输入函数和真实的动脉输入函数的比较,评估各种簇分析算法在检测动脉输入函数方面的可行性。实验结果表明,(1)不可重复的聚类算法:k均值簇分析算法相对手工方法而言能够获得更准确的动脉输入函数,用时也更短;相对模糊c均值算法而言,k均值算法能够获得更准确的计算结果且具有更好的可重复性;(2)可重复的聚类算法:相对快速仿射传播聚类算法而言,归一化分割聚类算法和凝聚层次聚类算法都可以获得更准确的动脉输入函数,但是归一化分割聚类算法比凝聚层次聚类算法的计算复杂度更低,因此具有更好的应用前景。
[Abstract]:Dynamic sensitive contrast magnetic resonance imaging (Dynamic Susceptibility Contrast-Magnetic Resonance Imaging, DSC-MRI) using endogenous contrast agents has been widely used in cerebral perfusion weighted imaging, which can be used to measure cerebral blood flow, cerebral blood volume, The mean pass time and other physiological indexes related to cerebral hemodynamic parameters play an important role in the diagnosis of clinical diseases and the selection of treatment schemes. Therefore, the rapid, accurate and robust quantitative calculation of DSC-MRI is of great significance in clinical practice. When using DSC-MRI technique to quantify the cerebral hemodynamic parameters, the arterial input function should be obtained in advance, and the accuracy of the arterial input function will directly affect the final calculation results. Traditionally, the extraction of arterial input function depends on experienced radiologists to manually select several pixels of the middle cerebral artery or internal carotid artery. However, the manual method is time-consuming and dependent on the operator. The results between different operators and different time points of the same operator are lack of repeatability, and the spatial resolution of DSC-MRI images is relatively low. The result of extracting arterial input function based on manual method will also be seriously polluted by partial volume effect. Therefore, the development of automatic or semi-automatic arterial input function extraction algorithm, which can reduce human intervention, has become an urgent need to solve the practical problem. In order to solve the problem of manually extracting arterial input function, the effectiveness of different cluster analysis algorithms for extracting arterial input function in cerebral perfusion of DSC-MRI was evaluated. The concrete steps include: collecting DSC-MRI perfusion weighted images of 42 healthy volunteers; using off-line workstation to correct the unaligned images of each phase volume caused by involuntary movement or rotation which is difficult for the subjects to control because of breathing, heartbeat and difficulty. The scanning plane with right horizontal middle cerebral artery in the first volume image was selected by manual browsing, and the time-intensity curve of the selected plane image signal was transformed into the time-concentration curve of contrast medium. Delete the curve with smaller area under the curve, the curve with more serious oscillation frequency and the curve polluted seriously by partial volume effect; Finally, various cluster analysis techniques are applied to the residual curve to extract the arterial input function automatically, and the accuracy, repeatability and complexity of the various cluster analysis algorithms in calculating the arterial input function are compared. Due to the lack of gold standard support in clinical trials, this study also adds a simulation experiment to compare the estimated arterial input function with the real arterial input function. To evaluate the feasibility of various cluster analysis algorithms in detecting arterial input functions. The experimental results show that: (1) the non-repeatable clustering algorithm: the k-means cluster analysis algorithm can obtain more accurate arterial input function than manual method, and the time is shorter; Compared with the fuzzy c-means algorithm, the k-means algorithm can obtain more accurate results and has better repeatability. (2) repeatable clustering algorithm: compared with fast affine propagation clustering algorithm, both normalized segmentation clustering algorithm and condensed hierarchical clustering algorithm can obtain more accurate arterial input function. But the computational complexity of the normalized segmentation clustering algorithm is lower than that of the condensed hierarchical clustering algorithm, so it has a better application prospect.
【学位授予单位】:东北大学
【学位级别】:博士
【学位授予年份】:2015
【分类号】:R445.2;R318.04
本文编号:2334743
[Abstract]:Dynamic sensitive contrast magnetic resonance imaging (Dynamic Susceptibility Contrast-Magnetic Resonance Imaging, DSC-MRI) using endogenous contrast agents has been widely used in cerebral perfusion weighted imaging, which can be used to measure cerebral blood flow, cerebral blood volume, The mean pass time and other physiological indexes related to cerebral hemodynamic parameters play an important role in the diagnosis of clinical diseases and the selection of treatment schemes. Therefore, the rapid, accurate and robust quantitative calculation of DSC-MRI is of great significance in clinical practice. When using DSC-MRI technique to quantify the cerebral hemodynamic parameters, the arterial input function should be obtained in advance, and the accuracy of the arterial input function will directly affect the final calculation results. Traditionally, the extraction of arterial input function depends on experienced radiologists to manually select several pixels of the middle cerebral artery or internal carotid artery. However, the manual method is time-consuming and dependent on the operator. The results between different operators and different time points of the same operator are lack of repeatability, and the spatial resolution of DSC-MRI images is relatively low. The result of extracting arterial input function based on manual method will also be seriously polluted by partial volume effect. Therefore, the development of automatic or semi-automatic arterial input function extraction algorithm, which can reduce human intervention, has become an urgent need to solve the practical problem. In order to solve the problem of manually extracting arterial input function, the effectiveness of different cluster analysis algorithms for extracting arterial input function in cerebral perfusion of DSC-MRI was evaluated. The concrete steps include: collecting DSC-MRI perfusion weighted images of 42 healthy volunteers; using off-line workstation to correct the unaligned images of each phase volume caused by involuntary movement or rotation which is difficult for the subjects to control because of breathing, heartbeat and difficulty. The scanning plane with right horizontal middle cerebral artery in the first volume image was selected by manual browsing, and the time-intensity curve of the selected plane image signal was transformed into the time-concentration curve of contrast medium. Delete the curve with smaller area under the curve, the curve with more serious oscillation frequency and the curve polluted seriously by partial volume effect; Finally, various cluster analysis techniques are applied to the residual curve to extract the arterial input function automatically, and the accuracy, repeatability and complexity of the various cluster analysis algorithms in calculating the arterial input function are compared. Due to the lack of gold standard support in clinical trials, this study also adds a simulation experiment to compare the estimated arterial input function with the real arterial input function. To evaluate the feasibility of various cluster analysis algorithms in detecting arterial input functions. The experimental results show that: (1) the non-repeatable clustering algorithm: the k-means cluster analysis algorithm can obtain more accurate arterial input function than manual method, and the time is shorter; Compared with the fuzzy c-means algorithm, the k-means algorithm can obtain more accurate results and has better repeatability. (2) repeatable clustering algorithm: compared with fast affine propagation clustering algorithm, both normalized segmentation clustering algorithm and condensed hierarchical clustering algorithm can obtain more accurate arterial input function. But the computational complexity of the normalized segmentation clustering algorithm is lower than that of the condensed hierarchical clustering algorithm, so it has a better application prospect.
【学位授予单位】:东北大学
【学位级别】:博士
【学位授予年份】:2015
【分类号】:R445.2;R318.04
【参考文献】
相关期刊论文 前7条
1 刘文军;游兴中;;一种改进的凝聚层次聚类法[J];吉首大学学报(自然科学版);2011年04期
2 王羡慧;覃征;张选平;高洪江;;采用仿射传播的聚类集成算法[J];西安交通大学学报;2011年08期
3 王开军;郑捷;;指定类数下仿射传播聚类的快速算法[J];计算机系统应用;2010年07期
4 杨宇鹏;赵卫东;王志成;陈刚;;基于图论的Normalized Cut图像分割方法研究[J];计算机与现代化;2010年01期
5 王开军;张军英;李丹;张新娜;郭涛;;自适应仿射传播聚类[J];自动化学报;2007年12期
6 杨帆;廖庆敏;;基于图论的图像分割算法的分析与研究[J];电视技术;2006年07期
7 闫成新;桑农;张天序;;基于图论的图像分割研究进展[J];计算机工程与应用;2006年05期
,本文编号:2334743
本文链接:https://www.wllwen.com/huliyixuelunwen/2334743.html
最近更新
教材专著