Combination of Compressed Sensing and Parallel Imaging for Highly-Accelerated 3 D First-Pass Cardiac Perfusion MRI-猫眼课题宝

DOI：

发表时间：

2009

期刊：

影响因子：

通讯作者：

中科院分区：

其他

文献类型：

作者：

R. Otazo;J. Xu;D. Kim;L. Axel;D. Sodickson

引言：对于首过心脏灌注MRI研究，每次心跳的全心脏覆盖是理想的。3D成像为多切片2D技术的SNR和体积覆盖限制提供了一种特别有吸引力的替代方案[1-2]。然而，3D方法由于其较长的采集时间而更容易受到心脏运动伪影的影响。使用多元件线圈阵列的并行成像可用于大幅加速3D采集。在去年的ISMRM会议上，报告了使用具有8倍加速的32元件阵列的每次心跳的全心脏覆盖[3]。然而，需要进一步加速，以达到足够的空间和时间分辨率的临床研究。灌注MRI的另一种加速技术是压缩感知（CS）[4]。CS利用空间和时间相关性，导致图像序列内容的稀疏性，以实现高水平的欠采样而不丢失图像信息。我们最近提出了一种CS和并行成像（PI）的组合，通过利用联合多线圈稀疏的特性来提高2D灌注MRI的加速率[5]。在这项工作中，我们将我们以前开发的CS和PI组合扩展到3D灌注MRI，以使用净加速因子为16的32元件线圈阵列实现每次心跳的全心脏覆盖，并提高空间和时间分辨率。方法：对2名健康志愿者进行首过3D心脏灌注MRI检查，使用0.1mmol/kg的Gd-DTPA（Magnevist）。修改了3D饱和恢复TurboFLASH脉冲序列，以包括用户定义的相位编码、分区编码和时间（ky-kz-t）采样模式（图1），并在配备32元件心脏线圈阵列（In Vivo）的1.5T扫描仪（Siemens，Avanto）上实施该脉冲序列。轴向采集是在中腔室进行的，以降低对心脏运动的敏感性。相关成像参数包括：FOV = 340×340×100 mm，图像矩阵= 128×128×16，空间分辨率= 2.65 ×2.65×6.25 mm，翻转角= 10，TE/TR = 0.9/2.3 ms，时间分辨率= 294 ms（完整容积），重复次数= 40。在动态成像的第一次心跳期间采集低空间分辨率线圈灵敏度数据，翻转角= 5且没有饱和脉冲。使用ky-kz-t随机欠采样来完成加速，其中对每个时间体积使用沿沿着ky-kz的不同可变密度欠采样模式，以在稀疏x-y-z-f域中产生所需的不相干伪影。使用压缩感知和SENSE [5]的组合离线执行图像重建，其中在多线圈组合上而不是单独在每个线圈上强制执行联合稀疏性，以利用沿着线圈维度的过采样和不相干性。每个线圈的采集模型由下式给出，i x y z i = y F S d，其中yi是欠采样数据，i x y z F是沿沿着x、y和z的傅立叶变换，Si表示线圈灵敏度，d是要重建的动态图像。多线圈采集模型通过将各个模型连接成= y艾德来公式化。傅立叶变换

INTRODUCTION: Whole-heart coverage per heartbeat is desirable for first-pass cardiac perfusion MRI studies. 3D imaging offers a particularly appealing alternative to the SNR and volumetric coverage limitations of multi-slice 2D techniques [1-2]. However, 3D approaches are more susceptible to cardiac motion artifacts due to their longer acquisition time. Parallel imaging using many-element coil arrays can be used to substantially accelerate 3D acquisitions. At last year’s ISMRM meeting, whole-heart coverage per heartbeat was reported using a 32-element array with 8-fold acceleration [3]. However, further acceleration is required to achieve an adequate spatial and temporal resolution for clinical studies. An alternative acceleration technique for perfusion MRI is compressed sensing (CS) [4]. CS exploits spatial and temporal correlations, resulting in sparsity of image series content, to achieve high levels of undersampling without loss of image information. We have recently presented a combination of CS and parallel imaging (PI) to increase the acceleration rate for 2D perfusion MRI by exploiting the properties of joint multicoil sparsity [5]. In this work, we extend our previously developed combination of CS and PI to 3D perfusion MRI to achieve whole-heart coverage per heartbeat with increased spatial and temporal resolution using a 32-element coil array with a net acceleration factor of 16. METHODS: Fist-pass 3D cardiac perfusion MRI was performed on two healthy volunteers with 0.1 mmol/kg of Gd-DTPA (Magnevist). A 3D saturation-recovery TurboFLASH pulse sequence was modifed to include user defined phase-encoding, partition-encoding and time (ky-kz-t) sampling pattern (Fig. 1), and this pulse sequence was implemented on a 1.5T scanner (Siemens, Avanto) equipped with a 32element cardiac coil array (In Vivo). An axial acquisition was performed in mid-diastole to reduce sensitivity to cardiac motion. The relevant imaging parameters include: FOV = 340×340×100 mm, image matrix = 128×128×16, spatial resolution = 2.65×2.65×6.25 mm, flip angle = 10, TE/TR = 0.9/2.3 ms, temporal resolution = 294 ms (complete volume), repetitions = 40. Low spatial resolution coil sensitivity data were acquired during the first heartbeat of the dynamic imaging, with flip angle = 5 and without the saturation pulse. Acceleration was accomplished using ky-kz-t random undersampling in which a different variable density undersampling pattern along ky-kz was used for each temporal volume to produce the required incoherent artifacts in the sparse x-y-z-f domain. Image reconstruction was performed offline using a combination of compressed sensing and SENSE [5], where joint sparsity is enforced on the multicoil combination rather than on each coil separately, in order to exploit oversampling and incoherence along the coil dimension. The acquisition model for each coil is given by , , i x y z i = y F S d , where yi is the undersampled data, , , x y z F is the Fourier transform along x, y and z, Si represents the coil sensitivities and d is the dynamic image to be reconstructed. The multicoil acquisition model is formulated by concatenating the individual models into = y Ed . A Fourier transform

Combination of Compressed Sensing and Parallel Imaging for Highly-Accelerated 3 D First-Pass Cardiac Perfusion MRI

文献摘要

被引文献

相似文献