Feasibility of Diffusion Tensor Imaging at 1.5T Using Multi-Band Echo Planar Acquisition

We report that diffusion tensor imaging (DTI) and tractography (DTT) of the pyramidal tracts using multi-band (MB) EPI could be a useful tool with a 1.5T MRI. We compared images using single-band EPI (SB-EPI) and MB-EPI. MB-EPI could reduce the scanning time by about 40%. We demonstrated that it is comparable between image qualities of SB-EPI and MB-EPI using tract-specific analysis and dice coefficients. Therefore, MB-EPI can promote high-speed DTI and DTT in clinical applications.


Introduction
Recently, microstructure analysis of the brain (structural, fun ctional, diffusion, etc.) by means of magnetic resonance imaging (MRI) has been actively investigated. 1 In such anal yses, highresolution isotropic images are very useful. How ever, numerous slices are necessary to cover the whole brain, which poses a major drawback in that the imaging time is increased. To address this problem, simultaneous imaging of multiple slices is useful.
The simultaneous imaging technique was originally dem onstrated by Larkman et al. using a gradientrecalled echo sequence. 2 Those authors reported that simultaneous excited slices could be separated by means of coil-sensitivity profiles. However, it is difficult to separate the simultaneous excited slices when coil-sensitivity profiles are similar among these slices. In the multiband (MB) technique, several slices are concurrently excited using a single radiofrequency (RF) pulse tailored for selective excitation at multiple frequencies. These slices are then subsequently separated based on the coil-sensitivity profiles and the phase-shift technique, which uses the phase of the RF pulses to shift the position of adja cent slices in such a way as to reduce overlap between aliased slices in the image space. This facilitates separation of aliased slices where the coil-sensitivity profiles are similar. [3][4][5] The MB technique was applied to the echo planar imaging (EPI) sequence in a number of studies. 5-7 MB can acquire more highresolution images by increasing the number of slices without extending the repetition time (TR) and acquisi tion time, and can reduce acquisition time while maintaining image resolution.
MRI using diffusion phenomena is useful for various analyses, including the commonly used diffusionweighted imaging (DWI), qspace imaging (QSI), and diffusion spec trum imaging (DSI). 8,9 As other examples of diffusion MRI, diffusion tensor imaging (DTI) and diffusion tensor tractography (DTT) have been proposed as noninvasive techniques for identifying white matter tracts in vivo, and the clinical utility of these techniques has been already reported. [10][11][12][13][14][15][16] There are, however, no reports quantitatively evaluating DTI and DTT using MBEPI. Furthermore, most previous reports on the MB sequence have used highspec devices (e.g., high field [≥ 3T] and multi-channel coils [≥ 32-channels]). In this study, we demonstrate that DTI and DTT using MBEPI could be a useful tool even with a 1.5T MRI, which is com monly used clinically.

MR imaging
Subjects were 10 male healthy volunteers (mean age: 26.6 years). This study was approved by the institutional ethics committee of the University of Tokyo, and only subjects who gave written informed consent were included in the study.
All MR images were acquired using a 1.5T clinical scanner (MAGNETOM Avanto; Siemens Healthcare, Erlangen, Ger many; 45 mT/m maximum gradient strength, 200 mT·m −1 ·s −1 maximum slew rate) and a commercial 12channel matrix M. Mitsuda et al.
where X represents a voxel of DTT from SBEPI data, Y represents a voxel of DTT from MBEPI data, and V represents the volume of the relevant voxel. DSC < 0.40, 0.40 ≤ DSC < 0.60, 0.60 ≤ DSC < 0.75, and 0.75 ≤ DSC indicate poor, fair, good, and excellent similarity, respectively. 20 Next, we implemented tract-specific analysis (TSA) to evaluate the apparent diffusion coefficient (ADC) and FA, to compare each MBf of MBEPI with SBEPI. 21 ROIs of TSA were set at the both right and left pyramidal tracts on DTT of SBEPI.

Statistical analysis
We compared ADCs and FAs from TSA, for each MBf of MBEPI and SBEPI using Dunnett's test implemented in commercially available software (SPSS v.20; IBM, Tokyo, Japan). P < 0.01 was considered significant.

Results
Using MBEPI, scan time was reduced by approximately 40% as compared with SBEPI. The DTT of each condition is shown in Fig. 1. The imaging capacity of DTT using MBf3 and MBf4 was markedly lower than that using SBEPI. DSC were 0.67 ± 0.03 (average ± standard error [SE]) and 0.69 ± 0.02 in the right and leftpyramidal tracts of MBf2; 0.40 ± 0.05 and 0.46 ± 0.06 in right and leftpyramidal tracts of MBf3; and 0.14 ± 0.07, 0.12 ± 0.07 in the right and left pyramidal tracts of MBf4, respectively. The DSC was consid ered good in MBf2, fair in MBf3, and poor in MBf4 (Fig. 2).
The average ± SD of FAs were 0.45 ± 0.03 and 0.49 ± 0.03 in the right and leftpyramidal tracts for SBEPI; 0.44 ± 0.03 and 0.47 ± 0.03 in the right and leftpyramidal tracts for MBf2; 0.38 ± 0.04 and 0.42 ± 0.03 in the right and leftpyramidal tracts for MBf3; and 0.35 ± 0.03 and 0.39 ± 0.03 in the right and leftpyramidal tracts for MBf4. The FA for MBf3 and MBf4 were significantly lower than that for SB-EPI (Figs. 5, 6).

Discussion
Using MBf2 of MBEPI could reduce the scan time by approximately 40% as compared with SBEPI. In this study, we used fixed TR (4200 ms) in MB-EPI. Higher MBf could use a shorter TR (e.g., minimum TR = 2500 ms in MBf3), but it is not practical because of T 1 saturation effects. It seems head coil. The MBEPI sequence used in this study was Release R011a for VB17A (https://www.cmrr.umn.edu/ multiband/#refs).
The imaging parameters for conventional singleband EPI

Data processing
The distortion caused by eddy currents was corrected for all DWI data using a software from the Oxford Centre for Func tional MRI of the Brain (FSL v.5.0.7; FMRIB Software Library, http://www.fmrib. ox.ac.uk/fsl/). 17 b0images and T 1 weighted images were coregistered using MATLAB2012b (Mathworks; Natick, NA) and Statistical Parametric Mapping 8 (SPM8, www.fil.ion.ucl.ac.uk/spm/software/spm8/), and coregistered T 1 weighted images were eliminated, except for the cerebral parenchyma, using MRIcron (http://www.mccaus landcenter.sc.edu/mricro/mricro/index.html). Diffusion toolkit was used for tensor analysis of corrected DWI. 18 The parame ters on diffusion toolkit were set the initial value. The target region of interest (ROI) was set at the primary motor cortex on T 1 weighted images, using VOLUMEONE and dTV II.FZR (developed by Yoshitaka Masutani; Hiroshima City Univer sity). Seed ROIs were set at the cerebral peduncle on a frac tional anisotropy (FA) map, using TrackVis. 18 Tractography of the pyramidal tracts was performed using these ROIs and TrackVis. The Interpolated Streamline method was used for the tractography algorithm. 19

Image analysis
We evaluated the DTT similarity of SBEPI and MBEPI using the Dice similarity coefficients (DSC). The DSC was calculated using the following equation: that the amount of reduction in scan time that can be gained by using MBEPI is limited. Additionally, MBEPI (SEtype) has other limitations, due to the peak power of the RF pulse and/or specific absorption rate. Recently, however, it has been reported that these drawbacks are beginning to be resolved, 22,23 and further improvement is to be expected. During evaluation of TSA, the ADC and FA of MBf2 was comparable to SBEPI. The DSC of MBf2 was also comparable to that found in a previous study (DSC = ca. 0.7) that investigated the repeatability of DTT. 20 We believe that MBEPI using MBf2 is superior to SBEPI in terms of the balance between scan time and image quality, because MBEPI using MBf2 can reduce scan time and maintain imaging quality. More specifically, in terms of clinical utility, MBEPI is particularly useful for patients who cannot main tain the same posture for an extended period.

M. Mitsuda et al.
On the other hand, the ADCs and FAs were significantly different, and the DSC were lower, in MBf3 and 4. The leakage factor, which is a factor related to residual aliasing among simultaneously excited and acquired slices, increases when MBf is set higher, and this results in a reduction in image quality. 6 In preliminary phantom examination, the relative leakage factors to SBEPI have increased by a factor of 1.1 in MBf2, 2.3 in MBf3 and 3.2 in MBf4 (data not shown). Thus, it appeared that the results of MBf3 and 4 are caused by an increased leakage factor in our study.
Recently, an approach that can decrease the leakagefactor by optimizing the GRAPPA technique for slice direction, was reported by Cauley et al. 24 This technique could decrease the leakage factor, while maintaining image quality with high MBf.
In addition, we confirmed another fiber bundle, which is the corpus callosum. The result image is shown in Fig. 7. The imaging capacity tended to be similar to the DTT of the pyramidal tracts. It is likely that MBEPI is useful for the DTT of pyramidal tracts as well as that of other fiber tracts. This study demonstrated that MBEPI using 1.5T MRI and a 12 channel coil, which are widely used in clinical prac tice, is useful for DTI and DTT. DTT is used for imaging nerve fibers, but it is difficult to image crossing and adjoining fibers within an individual voxel by DTT. 25 To resolve this issue, some methods (e.g., diffusion spectrum imaging, qball imaging, and highangular resolution diffusion imaging) have been proposed. 9,26,27 We plan to apply MBEPI with these techniques, in an attempt to obtain high precision trac tography images using a short imaging time.

Conclusion
We have demonstrated that MBEPI, using MBf2 is a useful tool for DTI and DTT employing 1.5T MRI. The technique may facilitate the use of DTI and DTT in clinical scenarios.