【作者】 黄鑫；

【导师】 陈武凡；

【作者基本信息】 南方医科大学 ， 生物医学工程， 2008， 博士

【摘要】 磁共振成像被认为是过去100年里世界上最大的应用技术发明之一。与CT相比,磁共振成像具有高组织分辨力、空间分辨力和无放射损伤等优点,同时磁共振检查是一种多参数成像,不仅能够清晰地显示解剖结构,而且还能反映人体的各种功能状态。磁共振成像已经广泛应用于临床,在人类疾病的诊断中发挥着无与伦比的作用。成像速度一直是临床磁共振应用所要考虑的重要因素。因此,自磁共振问世以来,研究者一直致力于提高成像速度。在临床磁共振应用初期,甚至简单的检测都需要花费数小时的扫描时间。发展到90年代,随着场强,梯度硬件,脉冲序列等性能上的增强导致了成像速度的大幅提高。然而,成像速度也极大依赖于磁共振设备的梯度系统的性能,同时硬件的发展也极大制约了速度的进一步提高。其原因在于物理、生理上的限制,以及过快的场梯度切换率给患者神经、肌肉上的刺激。因此,磁共振的成像速度从生理以及技术的角度上看几乎已提高到其内在的极限,也极大限制了磁共振在临床更广泛的应用。在最近10年里,基于多通道采集技术的并行磁共振成像的出现对磁共振成像产生了根本变革,也使得研究者再一次看到了成像速度进一步提高的可行性。这项技术大幅提高成像速度,进而有效减少了运动和磁化伪影的产生,使成像速度提高到超快速的水平,进一步拓宽了临床应用。本文同样也聚焦于此研究热点,致力于研究并解决医学磁共振成像中基于并行成像技术的伪影消除问题,既可以为临床医学诊断提供更高质量的图像,又将形成自主知识产权,为我国自主研发磁共振成像设备提供技术支持。从磁共振成像方式上来讲,成像时间长的一个主要原因是因为在传统的磁共振成像中,相位编码步必须通过逐步转换的磁场梯度来严格按照成像序列顺序执行,这也就限制了数据采集时间。相位编码线的数量同时也决定了图像在相位编码方向上的分辨率。而在并行成像中,空间编码通过使用多线圈部分地代替了通常使用的梯度编码方式,因为格外的一些空间信息可以通过多线圈之间的线圈敏感度差异来获得。简而言之,并行成像技术就是使用相控阵线圈内在包含的空间信息来代替消耗时间的相位编码步。然而,并行成像作为一项新的技术,必然有其不足。例如,使用并行成像技术时信噪比的损失不可避免。但是,在如今的临床应用中,采用降低因子为2甚至更高的情况下,信噪比的损失几乎不会影响诊断的准确性和图像分析。在近十年中,随着多通道并行采集技术的出现,并行成像一直是MR成像领域的研究热点问题之一,配合这项技术产生也下相继提出了许多有效的计算方法(重建算法)来进一步完善和推广这项技术的应用。其中,敏感性编码(SENSitivity Encoding,SENSE)技术是磁共振并行成像技术中开发最早,应用最成熟的一种方法,已经与现有磁共振成像系统完美的结合,如GE公司在EXCITEⅡ磁共振系统中的ASSET(Array Spatial Sensitivity Encoding Technique)技术,Philips公司的SyncraScan扫描技术和SIEMENS公司的iPAT(integrated ParallelAcquisition Technology)技术。然而,伪影和噪声在整个成像过程中不可避免。目前SENSE算法处理破坏数据的做法是建立有效的噪声模型,尚没有对算法进行优化。在本研究中,作者对算法进行深入研究,提出了基于鲁棒估计框架下的优化算法以更好的满足临床实际需求。研究分析得出:在多通道并行成像数据获取过程中线圈数据常常受到运动和噪声的影响而发生异常。在磁共振成像中,伪影和噪声可以看成观测数据集中的异常值。现有的SENSE方法是基于最小二乘估计产生的解,由于破坏数据和异常值的存在往往会造成大残差的情况,而最小二乘解通常对具有大残差的数据点十分敏感,使现有算法并不具有鲁棒性。另外,随着多通道并行成像技术的出现使得快速成像发展到了一个全新的阶段,然而如何把这项全新的技术与现有的成像技术相结合也一直是磁共振研究热点问题之一。部分傅立叶数据重建和多通道并行成像是两种有效的磁共振快速成像技术。由于在部分数据恢复过程中会由于共轭对称性的破坏而产生异常数据并随之带来相位偏移,使得在之后进行的并行重建过程发生错误造成重建图像产生伪影。在本研究中,作者提出了一种新的约束重建算法去消除重建图像中产生的伪影。算法应用AM鲁棒估计的框架对破坏数据进行数据修正,针对上述介绍的两个临床问题给出了理想的解决方案,很好的抑制了异常值对数据集造成的影响。此外,并行成像技术也提供了更加灵活的利用MR信号的方式,以至于使用者可以选择获得更高的图像分辨率或者更快的成像速度。例如,在不增加扫描时间的情况下,通过灵活的线圈组合在完成大视野区域扫描时(如脊椎)可以获得更加清晰的分辨率。然而,如何结合线圈图像得到最终的复合图像是其中的一个关键问题。平方和算法是多线圈采集技术与并行成像中常用的一种图像结合方法,在无法确定线圈敏感度系数的情况下,也被认为是一种最优的结合算法。但是在数据采集过程中,运动常常会使个别位置的线圈数据发生异常,采用平方和算法重建会对最终的复合图像质量产生很大影响,本章对现有的平方和算法做了深入研究,分析其原因在于平方和算法中单个线圈图像数据是采用相等权重的方式结合在一起,这样其中某一个或几个元素发生错误时势必会对产生结果造成巨大影响。本文以图像最大互信息量为判据首先统计出错误数据,并在此基础上构造了一种新的重建算法——加权平方和算法,最大限度的降低破坏数据对最终复合图像造成的影响。总之,本研究基于当前最新的多通道并行成像技术,针对其现有重建算法中存在的缺陷,提出了稳健性与精度都更好的AM-SENSE优化新算法,并使与其原有的快速成像技术进行了有效的结合,在最大限度缩短扫描时间的同时也获得了临床适用的图像质量。另外,基于现有的并行成像中图像结合方法在数据破坏时存在的缺陷,提出了加权平方和算法来最大限度的降低破坏数据对最终复合图像造成的影响,显著提高了存在运动时结合算法的鲁棒性。更多还原

【Abstract】 Magnetic resonance imaging (MRI) has been considered as one of the greatest inventions in the last century. Compared with computer tomography (CT), MRI can provide images with higher soft tissue contrast and spatial resolution and without harmful radiation. In the meantime, MR is imaging technique that provides many parameters to choose for doctors. It not only clearly display human’s anatomic structure, but reflect all kinds of vital functions. Therefore, MRI has been widely applied in clinics, and maybe the most promising non-invasive diagnostic tool in medicine.Speed has always been a critical consideration in MR imaging. Thus, researchers keep devoting themselves to making improvements in acquisition speed over the past decade. Early clinical applications required hours for even the simplest examinations. During the 1990s, advances in field strength, gradient hardware, and pulse sequences brought tremendous improvements in imaging speed. At the same time, it became obvious that further increases in speed along these lines would be progressively difficult to achieve because of physiological limitations. Rapidly switched field gradients produce neuromuscular stimulation, and excessively dense RF pulse trains can lead to unacceptable levels of RF energy deposition and tissue heating. The evolution of MR imaging appeared to have encountered fundamental constraints in terms of speed.In the course of the past decade, a new approach to MR image acquisition has been developed, which can extend or circumvent the speed limits traditionally associated with gradient and RF hardware. This approach, known as "parallel MRI" has taken the world of MRI by storm in the last few years, promising dramatic gains in imaging speed, a reduction in motion and susceptibility artifacts. In this study, we also focused our mind on this new technique. The aim of this work is to resolve the problems of artifacts elimination through which we proposed different strategies based on parallel MRI. We dedicated our study to the service of clinical requirement.From the point of the mode of imaging, the main reason of time consuming is that, in conventional MR imaging, the phase-encoding steps are performed in sequential order by switching a magnetic field gradient step by step, and this determines the speed of the measurement. The number of phase-encoding steps also defines image resolution in the phase-encoding direction (e.g., 256 steps for a 256×256 matrix). In parallel imaging, spatial encoding using multiple coils partially replaces the spatial encoding normally accomplished using gradients, because additional information is obtained from the spatial variation in the coil sensitivity profiles. Simply put, parallel imaging techniques use the spatial information inherent in local coil arrays to replace time-consuming phase encoding steps. Nevertheless, as a new technique, it is inevitable that there are some disadvantages for parallel MRI. For example, the physics of MR dictate that the signal to-noise ratio (SNR) will be reduced when using parallel imaging techniques. However, acceleration factors of 2 or higher can be achieved with virtually no loss in diagnostic accuracy and utility in MR applications today.Great progress in the development of pMRI methods has taken place, thereby producing a multitude of different and somewhat related parallel imaging reconstructtion techniques and strategies. The SENSitivity Encoding (SENSE) is the earliest technique that was developed and was also applied successfully to in vivo imaging based on the parallel MRI. This technique are already implemented and provided as a standard equipment of the MRI scanners with multiple receiver channels. A number of medical companies offer the most robust and versatile parallel imaging solution on the market, for example, GE’s ASSET (Array Spatial Sensitivity Encoding Technique) on EXCITEⅡMR scanner, Philips’s SyncraScan and Siemens’s iPAT (integrated Parallel Acquisition Techniques) et al.However, the appearance of artifacts and noise are inevitable in whole imaging process. In the current version of the algorithm so far, the effect of the noise is removed by use of the appropriate noise model. In this study, we analyze the advantages and disadvantages of present parallel MRI methods in clinical applications and propose the new reconstruction strategy based on robust estimator frame to satisfy clinical requirements.The research showed that the images from component array coil may have strong artifacts due to motion, which in turn have the significant effect on reconstructed composite image. The least squares solution can get arbitrarily wrong when high residual errors are used in the estimation. In this article, these corrupted data are regarded as outliers in observed datum. This paper presents a new algorithm for SENSE which is optimal in presence of artifacts or noise.Additionally, with the appearance of the parallel MRI, fast imaging of MR come into a new stage. How to combine parallel MR with conventional fast imaging methods is a hot subject of the ongoing research. Partial Fourier acquisition and parallel imaging are two effective fast imaging techniques. Both techniques permit decreased acquisition times by reducing the amount of phased encoding needed. The benefits of the combined technique compared with the individual techniques are large reduced imaging time. Unfortunately, during the partial Fourier reconstruction processing, restoration of missing uncollected k-space data rely on corrupt Hermitian conjugate symmetry. This leads to minor phase variations and noise introduced, which in turn corrupt data are amplified in SENSE process. In this study, a constrained reconstruction algorithm is presented to remove artifact in images. We apply robust Annealing M (AM) Estimator scheme to suppress the corrupt data points and make solutions insensitive to the influence caused by outliers.Parallel imaging gives us unprecedented flexibility in exploiting the MR signal, so that you can choose to "trade" speed for improved spatial resolution. In this way, higher resolution images can be obtained in the same amount of time as a conventional exam. Consequently, how to determine the optimum way to combine individual coil images is a key problem during reconstruction. Sum-of-Squares (SOS) is considered a common reconstruction method for phased array magnetic resonance imaging with unknown coil sensitivity. However, if data from one or some coils are corrupt due to motion, SOS results show obvious artifact in combined image. The reason for this is that coil images are combined by equal weights, so that the artifact is not suppressed due to corrupt data. In this work, an alternative technique that is Weighted-Sum-of-Squares (WSOS) method is presented which aim at decreasing the significance of corrupt samples in the data set and hence attenuating artifacts in reconstruction. The WSOS approach is significant valid and robust when there are certain artifacts such as sporadic motion or noise during data acquisition.In conclusion, a novel method for parallel MRI reconstruction was proposed and implemented this is the so-called "AM-SENSE" algorithm. This idea is based on SENSE. The purpose of this work is to deal with the problems of algorithm in present of noise and artifacts. It is shown to be of higher accuracy than the commercial method used in practice. In addition, the proposed algorithm similarly provides acceptable image quality utilizing the idea of applying both speed-up methods simultaneously to provide an almost unlimited reduction in acquisition time. Besides, in this study, we proposed an alternative image combination technique using phased array coils. The WSOS approach is significant valid and robust when there are certain artifacts, e.g. sporadic motion or noise.更多还原