【作者】 梁晓云；

【导师】 罗立民；

【作者基本信息】 东南大学 ， 生物医学工程， 2005， 博士

【摘要】 磁共振成像技术利用核磁共振原理对人体或生物体的某部分进行断面成像或立体成像,以获得组织和器官的解剖结构、功能结构和病变状况。磁共振成像技术是一种无损体外探测技术,是二十一世纪生物学和医学研究的重要工具。但是磁共振成像重建技术仍然存在重建图像质量不高的问题,临床上期待给出更有效的方法。功能磁共振成像(fMRI)是90年代初磁共振成像技术的一项新发展。该技术基于血氧含量水平依赖对比度机制(blood oxygenation level dependent,BOLD),能够实时地对大脑皮层神经功能活动进行成像,为临床MR诊断从单一形态学的研究到形态与功能信息相结合的系统研究开辟了一条崭新的道路。fMRI以较高的空间分辨率、时间分辨率及完全无损的特性已成为进行脑功能研究的重要工具。较之一般成像技术,回波平面成像技术(EPI)由于采样时间短,能够在几秒之内得到整个脑部的图像,这已成为fMRI广泛采用的一种技术。通过对fMRI中的任务相关的信号改变的分析检测可以确定对应的功能活动。但是,在1.5Tesla的磁场中,任务相关的信号改变范围通常在1～2%之间。为了获得对功能反应的较高检测率,需要高度稳定的信号。在图像采集过程中,有很多因素导致伪影。所有这些干扰因素都对功能激活区域的准确检测及定位造成不同程度的影响,虽然通过改善成像设备的硬件性能、优化扫描参数等措施可以减小这些干扰效应,但去除效果不是太理想。目前主要采用的为k空间校正方法,后处理是一种值得探讨的方法。生理伪影是功能信号检测的主要干扰因素,主要包括呼吸伪影和心脏运动伪影。这些伪影的出现严重影响了脑功能信号的有效检出,必须研究有效的去除生理伪影的方法。针对以上问题,本课题的主要目的有两个方面:一是研究新的磁共振图像重建方法,能重建更高质量的图像,以及对于序列图像重建,如何使时间分辨率和空间分辨率二者之间均衡;二是研究功能磁共振图像中生理伪影的去除方法,使功能信号检测更有效。本文主要的研究成果包括以下几个方面:1)提出了一种非笛卡尔采样的磁共振图像重建方法。对于非均匀采样数据,用非均匀FFT方法对其重建,大大提高了重建图像的速度。同时该方法的加权系数能以多种满足条件的函数形式给出,而已有方法的加权系数是给定的,所以相比之下该方法具有更大的灵活性,精度更高。最终结果表明该方法具有较好的重建图像质量。2)提出了一种基于信息相关性的动态磁共振图像重建方法。该方法利用已获取的两幅高空间分辨率参考图像作为先验信息,通过减少编码方式获取序列中其它数据,采集数据时间减少,重建可得到较高空间分辨率的图像。同时考虑到需要求解的系统为病态的,采用改进的TSVD方法进行正则化,重建序列图像。文中还给出利用其它几种方法实现的结果。结果表明,该方法具有较好的重建效果。3)提出了一种基于功率谱相减去除生理伪影的方法。生理伪影是由具体的周期性生理活动引起的,所以总是对应一定的频率,在功率谱上有较明显的特征。由于脑脊液中体素不包含任何功能激发信号,只含有生理伪影和随机噪声,所以首先选择脑脊液中体素,得到时间序列,估计出噪声的功率谱。然后取感兴趣体素,其功率谱减去脑脊液的平均功率谱,可得到去除生理伪影后的功率谱。实验结果表明该方法能有效地去除生理伪影。4)提出了一种基于空间独立成分分析去除生理伪影的方法。利用空间独立成分分析方法对功能磁共振数据分解,对独立成分对应的时间序列求功率谱,判断是否含有生理伪影,然后将含有生理伪影的独立成分去除并重建去除生理伪影后的数据。实验结果表明该方法能有效去除生理伪影。5)提出了一种基于图像空间数据的欠采样生理伪影去除方法。在利用长重复时间(repetition time,TR)采集的多层数据研究中,采用典型的成像参数(秒数量级),则每层不足以对生理伪影严格采样,得到的时间序列被高频生理伪影引起的混叠谱分量所污染。我们将原来按层排列的时间序列按时间顺序重排列,并对每幅图像取均值,得到一个时间序列,求功率谱,估计生理频率。然后根据频率混叠性质,判断混叠位置,采用滤波器去除呼吸生理伪影。实验结果表明该方法能较好地解决混叠问题,从而较好地去除生理伪影。论文最后给出工作展望:如何重建高质量的磁共振图像,以及如何有效地去除包括生理伪影在内的多种伪影仍然是我们以后的研究工作重点。更多还原

【Abstract】 Magnetic Resonance Imaging (MRI) techniques are used to obtain anatomical, functional, and pathological information of certain part of human bodies or animals based on the principles of nuclear magnetic resonance. MRI is a non-invasive technique, and especially an important tool for medical research. The problem, that the reconstructed images using MRI have low quality, still exists, so more efficient methods are expected.Functional MRI is a new development of MRI technology in the early 1990s. Cerebral cortex functional activation can be imaged real-time based on the contrast mechanisms of blood oxygenation level dependent (BOLD). And the advent of functional MRI leads to a new method for the clinical MR diagnosis, which can transform single morphological research to systematic research that combines morphology with functional information. So functional MRI has become a significant tool to study brain function with high spatial resolution, high temporal resolution, and absolutely non-invasive characteristics.Compared with other imaging techniques, EPI has been an extensively used method. Due to its short acquiring time of EPI, the whole brain can be imaged in several seconds, so functional imaging is of high spatial resolution. Functional activity can be determined by analyzing the task-related signal change. But in 1.5 Tesla magnetic filed, the change of functional signal is only 1～2%. To obtain high functional signal detecting rate, high steady signal is needed. During image acquisition, lots of factors can lead to artifacts and therefore interfere with the functional activated area. Though the problem can be mitigated by improving the hardware property, we can’t removal them satisfactorily. K-space based correction methods are used mostly, and post-processing strategy is a considerable method.Physiological artifact is an important interference of functional signal detection, including respiratory and cardiac artifact. The advent of these artifacts interfere with the detection offunctional signal, so efficient methods are expected to be presented.Aiming at problems mentioned above, there are two purposes in this thesis. Firstly, we want to reconstruct high quality image and make trade-off between time resolution and spatial resolution by presenting new MRI reconstruction methods. Secondly, we want to present efficient methods to remove physiological artifacts, thus functional signal can be detected more efficiently.The research achievements include five following methods:1) For non-cartesian acquisition, previous methods use regridding method and then obtain reconstruction results using FFT mostly. Based on non-uniform FFT (NUFFT) method, a new method for non-cartesian acquisition reconstruction is presented. This method gives weighted coefficients through many suitable functions, but previous methods give the weighted coefficients directly. The results show that our method can give high quality reconstruction images.2) For dynamic sequence images, a new high time and spatial resolution image reconstruction method is presented using information relativity. Using prior information from two reference image, high temporal and spatial resolution images can be reconstructed by reducing encoding. In the method, an ill-conditioned system is solved using modified TSVD method. At the same time, L-curve method is used to determine optimal parameter k. Some results using previous methods are given to be compared with our results. According to the comparison, our method is superior to others.3) Physiological artifacts are induced by concrete physiological activity and correspond to certain frequency, so they show obvious character. A method using power spectrum subtraction to removal physiological artifacts is presented. We estimate noise power spectrum from time series with selected voxel because CSF contains physiological noise and random noise without any activated signal. Using power spectrum of interested voxel, spectrum with physiological artifact removal can be obtained by subtracting estimated noise power spectrum. Experimental results show that the method can remove physiological artifacts efficiently.4) A spatial ICA based physiological artifacts removal method is presented. Using spatial ICA method, fMRI data can be decomposed and independent components are obtained. Power spectrum can be calculated using time series corresponding to independent components. we decide that which independent components contain physiological artifacts and reconstruct data by removal these independent components. Results show that the method can remove physiological artifacts efficiently.5) An image-space data based physiological artifacts removal method is given. In acquired multi-slices data research, if typical imaging parameter is used, physiological artifactscan’t be critically sampling for each slice. Resulting time series artifacts are contaminated by aliased spectral components from the high-frequency physiological artifacts, and thus previous methods are not efficient anymore. We reorder the data from original slice ordering to time ordering and obtain time series by calculating mean value of every image. Thus physiological frequency is estimated by calculating power spectrum of time series. According to the property of aliasing, we can decide the aliasing position and then remove physiological artifacts using digital filter. Experimental results illustrate that the method can remove physiological artifacts preferably.At the end of the paper, prospects of future research are given. The topics on how to reconstruct high quality MR image and how to remove various artifacts efficiently are our future research keystone.更多还原