Simultaneous Multislice Acquisition Technique on Auto-Regularized Parallel Imaging for Fast MR Imaging

自磁核造影技術發展以來已廣泛應用在人體各重要組織；縮短掃描時間亦成為廣泛討論的課題，隨著軟硬體技術進步，現今已發展出多種快速成相方式；諸如：EPI（回波平面成像），快速螺旋陣列成像技術等等。隨著時代演進，快速平行影像技術亦因應而生，藉由減少取樣數目以縮短掃描時間及增加時間及空間的解析度，並利用陣列線圈其非均勻的磁場敏感度特性作為影像重建依據以還原原始影像。這類技術亦已陸續發表，例如：SMASH、SENSE等等。現已廣泛利用於解剖影像設以期得到較短的掃描時間及較高的影像解析度及訊雜比。 平行加速影像其最大加速倍率決定於其陣列現圈的數目，本文將使用多截面激發技術(Simultaneous Multislice Acquisition簡稱SIMA)同時激發多截面來得到更佳的空間解析度及訊雜比，並大幅縮短影像掃描時間。在多重截面激發研究當中，分別利用Hadamard transform及陣列現圈非均勻的敏感度以分離多截面積發後的混合影像。在本文中，多截面激發技術也成功應用於仿體及動物的擴散張量影像(Diffusion Tensor Imaging，DTI)，並且得以驗證。本文更結合多截面激發技術與快速平行影像，在動物及仿體的實驗當中；成功地同時激發兩個截面，並且在取樣時間上分別分別做到2、3、4倍的加速，並且應用Tikhonov正規化演算法於處理平行影像重建及多截面激發混合訊號分離，已得到較佳的影像訊雜比。 此外，由於加速因子及空間解析度的提升，訊雜比常成為犧牲代價；此歸究於快速平行影像演算多涉及ill-condition問題，使得部份雜訊放大導致訊雜比下降。為改善此問題，本文中將引入Tikhonov 規範化(regularization)演算法，利用L-curve及GCV兩種方式尋找其最佳正規化參數。這個概念已成功履行在本實驗室3T的BRUKER硬體上並且應用於仿體及動物實驗，並且應用在2、3、4倍的加速序列。藉由得到較小的對稱參數(g factor)，可證明本方法提供較佳的訊雜比於影像重建結果上。藉由多截面激發的方式TR的時間可以縮短，並結合多通道陣列現圈於快速平行影像的加速方式，於本系統現行的四通道陣列現圈中，可達到超越四倍的加速倍率，其最大加速因子可達到8倍，這個嶄新的概念也成功應用在仿體及動物實驗上於現行硬體系統上。Since the invention of MRI, major improvements in imaging speed have been conceived and implemented, like echo planar, turbo, and spiral acquisitions, and even combinations of these methods. A very different approach to spatial encoding has been known of old, i.e. using the sensitivity profiles of a set of receiver coil elements to localize the signal’s source [1]. This so-called parallel imaging or sensitivity encoding methodology has recently gained renewed interest by the introduction of SMASH [2], and especially SENSE [3]. Parallel imaging was generally implemented today to achieve speed scanning, SNR enhancement, improvement spatial and temporal resolution. Furthermore, the maximum of acceleration always depends on the numbers of elements in phase array coil. In order to speed scanning and enhance SNR, we will combine Simultaneous Multislice Acquisition (SIMA) [7-12] with parallel imaging for higher spatiotemporal resolution and SNR in anatomical image. In this work, we excited two parallel slice profiles simultaneously and segregated intermixed slices by Hadamard transform and utilizing non-homogeneous sensitivity profile of phase array coil [13]. DTI (diffusion tensor image) was well put into practice with SIMA technique in phantom and in vivo experiment. Moreover, we excited two slices and speeded up scan time in 2, 3 and 4 accelerations both in phantom and animal study. Utilizing Tikhonov regularization to parallel image reconstruction and mingled image segregation, the approaches were also faultlessly practiced in phantom and in vivo experiment. From this novel idea, the acceleration ratio will transcend the amount of phase array coil. Furthermore, by increasing accelerations and spatiotemporal resolution, sensitivity encoding always involves ill-condition problem [4] which induce serious noise amplification in anatomical image reconstruction. It’s also the price for the increased spatiotemporal resolution in parallel MRI. In this thesis, we propose a reconstruction based on Tikhonov regularization [5] that reduces SNR loss due to geometric correlations in the spatial information from the array coil elements [6]. Reference scans are utilized as a priori information about the final reconstructed image to provide regularized estimates for the reconstruction using the L-curve technique. In Bruker 3T MRI system, we could achieve more than 4 accelerations by exploiting SIMA and auto-regularized SENSE with a 4-channel rat head array coil for higher SNR and spatiotemporal resolution.Content 1. Preface……………………………………………………………..…………1 1.1 Background………………………………………………………………… 1 1.2 Motivations………………………………………………………………….2 1.3 Framework…………………………………………………………………..3 2. Introduction……………………………………………………………………..4 2.1 Introduction to magnetic resonance imaging………………………...………..4 2.1.1 RF pulse excitation………...…………………………………………..7 2.1.2 Slice selection…………………………………...……………………..9 2.2 Simultaneous Multiple-slice Acquisition…………………………………...11 2.2.1 Background…………………………………………………………11 2.3 Parallel Imaging…………………………………………………………….15 2.3.1 Background ………………………………………………………...15 2.3.2 Ill-Posed Problem on Parallel Imaging……………………………..19 3. SIMA (Simultaneous Multislie Excitation)…………………………..21 3.1 Theory………………………………………………………………….….....21 3.1.1 RF pulse Design……………………………………………………..21 3.1.2 Hadamard Transform on Intermixed Signal Segregation…………....22 3.1.3 Intermixed Signal Separation by array coil………………………….23 3.2 Material and Method………………………………………………………....24 3.3 Result………………………………………………………………….……..28 3.3.1 Hadamard Encoding for Image Segregation…………………………28 3.3.2 SIMA with Hadamard Encoding for DTI…………….………………35 3.3.3 Utilizing Sensitivity Profile to Signal Separation……………………38 4. Auto-Regularized SENSE………………………………………….…….43 4.1 Theory………………………………………………………………………..43 4.1.1 Tikhonov Regularization for Ill-Condition on Parallel Imaging……..46 4.1.2 Tikhonov Regularization……………………………………………..48 4.1.3 Automatic Choosing for Optimal Regularization Parameter……...…50 4.1.4 Auto-Regularized SENSE Operation………………………………...54 4.2 Experiment Result……………………………………………………..……57 4.2.1 Auto-Regularized SENSE on Phantom Study…………………..……57 4.2.2 Auto-Regularized SENSE on Animal Study……………….…….…..60 5. SIMA with Auto-Regularized SENSE…………………………..……...65 5.1 Material and Method………………………………………………………..65 5.2 Phantom Study…………………………………………………..………….67 5.3 Animal Study…………………………………………………………...…..69 6. Discussions……………………………………………..…………………..79 7. Conclusions and future works…………………………………………..87 7.1 Conclusions………………………………………………………………..…87 7.2 Future works………………………………………………………………….88 Appendix……………………………………………………………………….….89 Matlab Code A. Auto-Regularized SENSE………………………………………………...….89 B. Resolving image from intermixed slices……………………………….…….93 C. Tikhonov Regularization……………………………………………..………96 D. L-curve……………………………………………………………………….99 E. GCV………………………………………………………………..……….102 Reference……………………………………………………….……………….10