Efficient T₂ mapping with Blip-up/down EPI and gSlider-SMS (T₂-BUDA-gSlider)

Abstract

Purpose:

To rapidly obtain high isotropic-resolution T₂ maps with whole-brain coverage and high geometric fidelity.

Methods:

A T₂ blip-up/down echo planar imaging (EPI) acquisition with generalized Slice-dithered enhanced resolution (T₂-BUDA-gSlider) is proposed. A radiofrequency (RF)-encoded multi-slab spin-echo EPI acquisition with multiple echo times (TEs) was developed to obtain high SNR efficiency with reduced repetition time (TR). This was combined with an interleaved 2-shot EPI acquisition using blip-up/down phase encoding. An estimated field map was incorporated into the joint multi-shot EPI reconstruction with a structured low rank constraint to achieve distortion-free and robust reconstruction for each slab without navigation. A Bloch simulated subspace model was integrated into gSlider reconstruction and utilized for T₂ quantification.

Results:

In vivo results demonstrated that the T₂ values estimated by the proposed method were consistent with gold standard spin-echo acquisition. Compared to the reference 3D fast spin echo (FSE) images, distortion caused by off-resonance and eddy current effects were effectively mitigated.

Conclusion:

BUDA-gSlider SE-EPI acquisition and gSlider-subspace joint reconstruction enabled distortion-free whole-brain T₂ mapping in 2 min at ~1 mm³ isotropic resolution, which could bring significant benefits to related clinical and neuroscience applications.

Introduction

High resolution T₂ mapping has shown great potential in clinical and neuroscience applications, including but not limited to epilepsy(1,2), glioma(3), monitoring tumor progression(4), brain maturation(5), femoroacetabular impingement(6) and Parkinson disease(7). However, the use of conventional spin-echo acquisitions at multiple TEs(8), which is considered the gold standard approach for T₂ mapping, requires lengthy acquisition time (e.g. about 60 min for 10 TEs acquired at 1×1×5 mm³ resolution with limited slice coverage) and prohibits its clinical application. Other approaches including DESPOT₂(9), MR fingerprinting(10,11), 3D turbo spin-echo (TSE) T₂ mapping(12,13), multi-echo acquisition(14) and model-based acceleration(15–17) have been developed to enable faster T₂ parameter estimation. These techniques can reduce the scan time and improve clinical feasibility, but still require several minutes to provide whole-brain high-resolution T₂ maps, for example, 17 min for DESPOT₂(9), 11–16 min for model-based T₂ mapping(18) and 5–8 min for 3D MR fingerprinting (while also provides T₁ mapping)(11,19).

Another potential solution is using spin-echo (SE) echo planar imaging (EPI) to reduce the scan time. To achieve whole-brain coverage, 2D SE-EPI efficiently utilizes the idle time during the lengthy TR required for T₁ recovery to acquire data from multiple slices. While this enables faster scans, it is limited to thick slice acquisition as 2D encoding fails to provide sufficient SNR for high resolution quantitative imaging. Recently, a slab encoding technique named generalized slice-dithered enhanced resolution (gSlider)(20) was proposed, which utilizes RF pulses with different excitation profiles to encode individual thin slabs. The gSlider technique fully utilizes the long TR by acquiring data from all slabs sequentially while enjoying higher SNR due to its volumetric encoding, which is crucial for whole-brain, high-resolution T₂ mapping.

Despite their high efficiency k-space encoding, drawbacks common to all EPI-based readout strategies are T₂- or T₂*-related voxel blurring, geometric distortion and voxel pileups stemming from B₀ inhomogeneity. These preclude high in-plane resolution imaging with a single-shot EPI readout. A common practice is to use parallel imaging techniques(21–24) to reduce the effective echo spacing, but these are usually limited to R_in-plane ≤ 4 in-plane acceleration using modern receive arrays. Multi-shot EPI approaches segment and acquire a plane of k-space data across multiple TRs, thereby reducing the geometric distortion and blurring and permitting higher in-plane resolution imaging. Shot-to-shot phase variations due to physiological noise in multi-shot EPI need to be accounted for to enable successful image reconstruction. A navigator echo can help estimate these variations(25), but this comes at the cost of reduced acquisition efficiency since navigation further prolongs the acquisition time. Navigator-free methods such as Hankel structured low-rank constrained parallel imaging techniques(26–28) have been introduced to address this problem. However, such advanced reconstruction techniques(28) require a larger number of shots at very high in-plane acceleration (e.g. 4-shots acquisition with R_in-plane = 8) to achieve good image quality and low distortion, which again reduces the acquisition efficiency.

To mitigate the geometric distortion, FSL topup(29,30) is commonly used, where two EPI acquisitions with reverse phase-encoding polarities, one with blip-up and another with blip-down, are acquired separately to estimate a field map and perform distortion correction. In this way, distortion-free images can be obtained with only 2-shots of data, which helps reduce the acquisition time compared to standard multi-shot EPI using only blip-up encoding. Based on this approach, a hybrid-space SENSE method(31,32) was proposed to jointly reconstruct the blip-up/down shots with estimated field maps and incorporated the phase differences into the forward model. This method was able to reduce the g-factor penalty and improve the SNR over single-shot EPI. However, since the hybrid-space SENSE requires explicit knowledge of phase variations between the blip-up and -down shots, their inaccurate estimation may introduce reconstruction artifacts and noise amplification(33).

In this work, we propose to combine gSlider encoding and blip-up/down acquisition (BUDA(34)) to achieve high-resolution and distortion-free T₂ mapping with whole-brain coverage. First, we incorporate Hankel structured low-rank constraint into our BUDA reconstruction to recover distortion-free images from blip-up/down shots without navigation. To utilize the similarity among different TEs, we introduce a model-based gSlider-subspace joint reconstruction to recover high-resolution thin-slice images. This projects time-domain images into temporal coefficient maps with a temporal basis(35,36) and then the reconstructed coefficients are used to obtain quantitative T₂ maps. The proposed method enables distortion-free high-quality whole-brain T₂ mapping with ~1 mm isotropic resolution in 2 minutes. This work is an extension of our earlier work which was reported in abstract of ISMRM 2020(37).

Method

BUDA-gSlider acquisition

Figure 1A shows the sequence diagram of proposed multi-slab BUDA-gSlider acquisition. Ten different TEs are acquired to obtain different T₂ weighted contrasts. Consecutive TE volumes are acquired with alternating blip-up or -down polarity. For example, TE₁ data are encoded with 5 gSlider shots using blip-up polarity, whereas the 5 RF encoding shots in the following TE₂ acquisition use blip-down phase encoding. The entire acquisition thus requires a total of 50 shots (N_TE×N_RF = 10×5, where N_TE is the number of different TEs used for generating different T₂ weightings, and N_RF is the number of RF pulses used for encoding one slab).

Figure 1.

(A) The sequence diagram of the BUDA-gSlider acquisition with multi-TE and spin-echo EPI readout.

(B) The acquisition trajectory of the blip-up/down 2-shots EPI.

(C) The flowchart of k/t BUDA reconstruction. The reconstructed SENSE images by respectively using blip-up and -down shots are jointly used to estimate the field map using FSL topup. Then the field map is incorporated into the Hankel structured low-rank constrained forward model to jointly reconstruct distortion-free images.

To obtain whole-brain coverage efficiently, we utilize the idle time of each TR to acquire data from other slabs. With the 5-mm slab thickness, acquiring 26 slabs will correspond to a 130 mm FOV in the slice direction, thus providing the desired whole-brain coverage. In this way, the entire acquisition time could be described as T_acq=TR× N_TE×N_RF. To further accelerate the acquisition, we also introduced blipped-CAIPI encoding for simultaneous multi-slab acquisition(38) to reduce the TR to 2400 ms with multi-band (MB) factor 2. With this, we have obtained T_acq=2.4×10×5=120 sec.

For gSlider slab-encodings, the 90° excitation pulses are designed to achieve a highly independent encoding. The five 90° excitation pulses are named as RF₁ to RF₅ sequentially and used to encode the same slab, which is 5 times as thick as the desired slice thickness. In this work, the targeted slice resolution is 1 mm, and the corresponding thickness of each individual slab is 5 mm. Since each of these RF pulses excites a slab rather than a single slice, there is volumetric SNR gain for each acquisition(20).

k/t BUDA reconstruction

With acquired interleaved blip-up and blip-down shots, distortion-free images of each RF encoding can be jointly reconstructed using the pipeline shown in Figure 1C, which includes the following:

1. The blip-up EPI data (yellow lines in Figure 1B) and the blip-down EPI data (blue lines in Figure 1B) were separately reconstructed using SENSE to obtain distorted images with ten different TEs, SENSE blip-up and SENSE blip-down shown in Figure 1C. The SENSE blip-up/down images have different geometric distortion due to opposite phase-encoding direction as the arrow indicates.

2. All of the ten SENSE blip-up/down images were jointly processed in FSL topup (http://fsl.fmrib.ox.ac.uk/fsl) to estimate field maps (29). The advantage by using multiple SENSE blip-up/down images for field estimation over one single pair will be shown later in Results section.

3. The estimated field maps were incorporated into the Hankel structured low-rank constrained joint reconstruction blip-up/down data of all TEs (33). This can be described as:

   $$mi{n}_b{\sum }_{t=1}^{t=10}\Vert {\text{F}}_t{\text{E}}_t\text{C}{b}_{t,r}-d_{t,r}\Vert {}_2^2+\Vert \mathcal{H}{\left(b\right)\Vert }_{\text{*}}$$ [1]

where t is the index of TE and r is the index of RF encoding. F_t is the undersampled Fourier operator in t^th TE’s shot, E_t is the estimated off-resonance information, C are the ESPIRiT (23) coil sensitivity maps estimated from distortion-free gradient-echo calibration data, b_t,r is the distortion-free image and d_t,r are the t^th TE’s and r^th RF-encoding k-space data. The constraint $\Vert \mathcal{H}{\left(b\right)\Vert }_{\text{*}}$ enforces low-rank prior on the block-Hankel representation of the blip-up and blip-down data, b_t&t+1,r, which is implemented by consecutively selecting 9×9 neighborhood points in k-space from each shot and then concatenating them in the column dimension. This reconstruction is implemented by using a projection onto convex sets (POCS) like(39,40) iteration with the tolerance of 0.01% RMSE between two successive iterations. Please note that the reconstructed images b_t,r are still RF-encoded slab images. After using this approach to reconstruct five RF encodings (r=1,2…5), a total of 50 distortion-free k/t BUDA images were obtained.

In addition, to justify if the TE-specific contrast is confounded by the Hankel constraint, a comparison of BUDA reconstruction results by using different TEs’ data (proposed k/t BUDA reconstruction) and using same TE’s data was implemented and shown in Results section.

gSlider-subspace joint reconstruction

A model-based gSlider-subspace joint reconstruction was proposed and shown in Figure 2:

1. A dictionary comprising signal evolution curves of T₂ from 1 to 1000 ms with a step size of 1 ms was built. With principal component analysis, the first three principal components were selected as the temporal basis Φ, shown in Figure 2A. With this basis, the desired high-resolution thin-slice images x_t,s (where s=1,2…5 is the slice index within a RF-encoded slab) could be expressed as Φc_e,s(where e=1,2,3 is the index of subspace components), where c_e,s is the temporal coefficient map.

2. To jointly reconstruct all the 50 shots of different TEs and RF encodings, all the BUDA reconstructed images b_t,r(t=1,2…10, r=1,2…5) were expanded in the column dimension as b_exp, resulting in a 50×1 vector for each voxel with contributions from all TEs and RF encodings. To meet the form of b_exp, c_t,s is also expanded but along the TE-Slice dimension as c_exp.

3. Figure 2B shows the gSlider-subspace joint reconstruction, which can also be described as:

   $$mi{n}_{c_{\text{exp}}}\Vert A_{\text{exp}}{\Phi }_{\text{exp}}{c}_{\text{exp}}-b_{\text{exp}}\Vert {}_2^2+{\lambda }_{\text{Tik}}\Vert c_{\text{exp}}\Vert {}_2^2$$ [2]

   where A_exp , a matrix with size of 50×50, contains ten RF-encoding matrices A(This matrix contains the excitation profiles of all RF-encoding pulses and can be calculated using Bloch simulation). Φ_exp contains five temporal bases Φ corresponding to all the five RF encodings. $\Vert .\Vert {}_2^2$ denotes Tikhonov regularization and λ_Tik is the gSlider Tikhonov subspace regularization parameter and was set to 0.3 to achieve a high SNR gain and sharp partition resolution based on a retrospective experiment, which will be shown in Results section. To validate that three-component PCA is fair for subspace reconstruction, Figure 3 shows exemplary ten coefficient maps from a single slice, where the first three coefficient components capture ~95% of whole signal intensity while the rest seven weak coefficient maps are noise-like.

4. The temporal coefficient maps were projected back to time domain to recover the high-resolution thin-slice images with different T₂ contrasts by ${\Phi }_{\text{exp}}^T{c}_{\text{exp}}$. Then, the recovered images were used to get the T₂ maps using template matching with the pre-calculated dictionary voxel-by-voxel.

Figure 2.

The gSlider-subspace reconstruction process including:

(A) Generating T₂ dictionary by using EPG algorithm and corresponding temporal basis by principal component analysis.

(B) The gSlider-subspace joint reconstruction which is used to generate the high-resolution thin-slice images.

(C) Template match the high-resolution thin-slice images with a pre-calculated T₂ dictionary to obtain the final T₂ maps.

Figure 3.

The coefficient maps projected from the high-resolution thin-slice images by using a 10-components temporal basis.

In addition, this model could enable further acceleration, for example, by reducing the number of acquired TEs per RF-encoding from 10 to 6 with the sub-sampling pattern shown in Supporting Information Table S1, resulting in a 40% reduction of acquisition time from 120s to 72s. Then, by utilizing the high SNR of RF-encoded slab images, the missing points are fitted using an exponential fitting as initialized inputs of the shuffling model. This approach enables faster acquisition but compromised the image quality with more noise as shown in Supporting Information Figure S1. Therefore, we stick to 120s version for the comparison with gold-standard method.

In-vivo validation

To validate our proposed method, three healthy volunteers were scanned with the approval of Institutional Review Board.

The proposed T₂-BUDA-gSlider data were collected using the following protocol: R_in-plane=4, partial Fourier 6/8, multi-band factor=2, TR=2400 ms, 10 different TEs (from 49ms to 121ms with a gap of 8ms, where the 49ms is the shortest possible TE that EPI readout could achieve and 121ms is selected to cover the T₂ range of common tissues in brain). A phase-encoding shift Δk_y of 2 was set between the blip-up and blip-down shots to improve the k-space coverage and the parallel imaging reconstruction. 26 thin slabs (slab thickness=5mm) were acquired with 5 RF encodings for each slab, resulting in ~1-mm slice thickness (influenced by the selection of λ_Tik, which will be discussed later in Result section) for the final high-resolution thin-slice images. With 1×1 mm² in-plane resolution, the resulting voxel size was 1×1×~1 mm³ with FOV=220×220×130 mm³. For each TE and RF encoding, both blip-up and -down shots were acquired as reference, resulting in a total of 100 shots and the total acquisition time of 240s (T_acq = TR×N_TE×N_RF-encoding×blip-up/down = 2.4s×10×5×2). Then it was subsampled by alternatively selecting blip-up or blip-down shots in the TE dimension to 50 shots, corresponding to an acquisition time of 120s. A FOV-matched low-resolution 2D gradient-echo (GRE) was also acquired to obtain distortion-free sensitivity maps for BUDA reconstruction.

To test the accuracy of the proposed method, multi-TE single-echo spin-echo (SE) data were also collected as the gold standard to estimate T₂ values. The protocol of SE sequence was set the same as the proposed method but with a slice thickness of 5 mm. T₂ values were then fitted voxel-by-voxel using the same dictionary as T₂-BUDA-gSlider to obtain the T₂ maps. In addition, to demonstrate the distortion-free property of the proposed BUDA-gSlider, a 3D fast spin-echo was also acquired as reference. To compare the T₂ values in specific subcortical regions by using proposed method and gold standard SE, a T₁-weighted MPRAGE sequence was acquired as the input for Freeserfer’s subcortical segmentation(41).

To validate the robustness of the regularization parameter selection, additional two subjects were scanned.

All studies were performed on a 3 Tesla (T) MAGNETOM Prisma scanner (Siemens Healthcare, Erlangen, Germany) with a 64-channel head receiver coil. Computations were performed on a Linux (Red Hat Enterprise) server (with Core i7 Intel Xeon 2.8 GHz CPUs and 64GB RAM) using MATLAB R2014a (The MathWorks, Inc., Natick, MA).

Results

Figure 4 shows the RF-encoded slab images of RF1/TE1 (i.e. 49 ms) by using hybrid-space SENSE and BUDA reconstruction, respectively. As indicated by the blue arrows, the individual SENSE reconstructed images for blip-up and blip-down shots exhibited significant geometric distortions compared to the 3D FSE reference, while the results from BUDA are consistent with the reference. As indicated by the red arrows, the BUDA results had reduced noise amplification and artifacts compared to hybrid-space SENSE. Figure 4B shows the RF-encoded slab images of different TEs using BUDA reconstruction, which provide the different T₂ contrasts for subsequent T₂ mapping.

Figure 4.

(A) Reconstructed RF-encoded slab images of RF1/TE1 and reference 3D-FSE images. The individual SENSE reconstructed images show significant distortion (blue arrows). The images using BUDA reconstruction has reduced artifacts compared to hybrid-space SENSE (red arrow) and high geometric fidelity consistent with the reference images. Minor distortion has remained in regions where B0 field map may not be estimated perfectly (yellow arrow).

(B) RF-encoded slab images of different TEs by using BUDA reconstruction.

Figure 5 shows T₂ maps obtained by using different λ_Tik values for Tikhonov regularization of gSlider-subspace joint reconstruction. The impulse responses of the central slice within an RF-encoded slab were shown in the bottom. Here we replaced b_exp with a gSlider RF-encoded delta function of the central slice and pass it through the proposed method to obtain the impulse response. The reference results were obtained by using the fully-sampled 100-shots acquisition with λ_Tik of 0.1. Compared with the reference, when λ_Tik was selected between 0.1 to 0.3, a high SNR gain and sharp partition resolution could be achieved. For λ_Tik of 1, significant blurring in partition direction could be observed.

Figure 5.

The comparison of T₂ maps based on different lambda for Tikhonov regularization from 0.01 to 1. The reference is from the full-sampled 100-shots acquisition.

Figure 6 shows four exemplary slices of T₂ maps by using the proposed method with λ_Tik =0.3 (top). To directly compare the gold standard method and the proposed method regardless of resolution difference, reformatted T₂ maps (averaged across 5 adjacent slices) from BUDA-gSlider were also shown in the second row. T₂ maps by using the gold standard SE method (third row) and the tenfold scaled difference maps (fourth row) were also shown. T₂ values from six specific regions (including white matter, gray matter, pallidum, thalamus, caudate and putamen) were evaluated and listed in Table 1 for both methods. Estimated T₂ values were close to the gold standard SE but with faster acquisition (2 min vs 60 min), higher resolution (1×1×1 mm³ vs 1×1×5 mm³), and larger slice coverage (130 mm vs 75 mm) which demonstrate the utility of T₂-BUDA-gSlider. It was found that the T₂ values of thalamus deviated the most from the gold standard acquisition. This phenomenon could be caused by two reasons: i) The segmentation is based on an external MPRAGE acquisition. Since this region is relatively small, a small motion between the acquisitions of gold standard and proposed methods could lead to a significant difference. ii) This region is very close to CSF which could influence the T₂ values due to partial volume effect.

Figure 6.

T₂ maps by using the proposed T₂-BUDA-gSlider method and gold standard spin-echo method as well as absolutive difference maps. The proposed method can achieve higher resolution (1×1×1 mm³ vs 1×1×5 mm³) and wider coverage (130 mm vs 75 mm in the slice direction) within short acquisition time (2 min vs 60 min).

Table 1.

T₂ values of specific regions estimated by using gold standard method and the proposed method.

Unit: ms	WM	GM	Pallidum	Thalamus	Caudate	Putamen
T2-BUDA-gSlider	64.17± 6.59	75.56±11.16	52.75±4.63	58.88±5.70	65.43±5.14	51.80±3.63
Gold Standard	66.23± 6.77	73.69±10.72	53.12±4.81	63.24±5.91	67.42±4.69	51.10±3.91

Figure 7 shows the field maps (first column) estimated by using only one single pair of blip-up/down SENSE reconstruction images (top) and blip-up/down SENSE images of all TEs (bottom), respectively. With more pairs of blip-up/down SENSE images as input for FSL topup, the field map tends to become less noisy and more reasonable, and thus results in better quality for BUDA reconstruction and subsequent T₂ mapping.

Figure 7.

Comparison between using one single pair of SENSE blip-up/down images to do field map estimation with using five pairs. The corresponding BUDA reconstruction results and T₂ maps are also included.

Figure 8 shows the comparison of BUDA reconstruction by using different TEs’ data (k/t BUDA) and same TE’s data (non-k/t BUDA). The noise-like difference maps and small values of RMSE at same TE with k/t and non-k/t BUDA reconstruction suggest that the k/t BUDA didn’t confound the TE-contrast by using a Hankel constraint.

Figure 8.

Non-k/t BUDA reconstruction using blip-up&down shots of only TE1 data (first column as TE1_ref) and of only TE2 data (second column as TE2_ref). The k/t BUDA by using blip-up shot of TE1 and blip-down shot of TE2 jointly was shown in the fourth column. The difference maps between non-k/t BUDA and k/t BUDA were shown in the fifth column and the RMSEs were listed at the right top of each difference map, which are much smaller than the RMSEs between TE1_ref and TE2_ref (non-k/t BUDA, third column).

Figure 9 shows the in-vivo results of two subjects. The similar image quality demonstrated the robustness of regularization parameter selection.

Figure 9.

Two subjects were scanned to validate the regularization robustness by the similar image quality.

Discussion and Conclusion

The proposed T₂-BUDA-gSlider provides a new approach for rapid and distortion-free T₂ mapping with the ability to provide whole-brain coverage at ~1-mm³ isotropic resolution in 2min, with an SNR-efficient RF-encoded SE-EPI acquisition. Compared with current state-of-the-art methods listed in the Introduction section, the proposed method is fast with good image quality. It takes advantages of the following:

1. Distortion-free EPI: By incorporating B₀-correction in the forward model, BUDA acquisition and its corresponding joint reconstruction can provide high-quality distortion-free images while the EPI readout can enable fast acquisition.

2. SNR gain of volumetric gSlider acquisition: Since the individual SENSE blip-up and -down reconstructions are RF-encoded slab images, they benefit from the volumetric SNR gain of using thicker slabs as opposed to conventional 2D acquisitions. The estimated T₂ maps also benefit from the SNR boost.

3. Improved image quality with gSlider-subspace joint reconstruction: We utilized the similarity of images with different TEs to generate better quality T₂ maps through joint reconstruction. This has further capitalized on the SNR gain from gSlider and BUDA techniques, which has improved the image quality of the T₂ maps. In addition, we used the temporal basis to project images of all TEs and RF encodings into a joint reconstruction model and then reduced the noise by eliminating the weak coefficient components.

The T₂ values measured by the proposed method were in accordance with the gold standard SE. We also demonstrated the distortion-free property of the proposed method by comparing the images with 3D FSE data, where BUDA was seen to mitigate the distortion caused by off-resonance and eddy current effects. By comparing the field map estimated by single pair and multiple (i.e. ten) of SENSE blip-up/down images, it could also be found that the latter one enabled better reconstruction results for both BUDA and T₂ mapping.

Since the Tikhonov regularization used in the gSlider-subspace joint reconstruction would induce side lobes, the partition resolution might degrade when λ_Tik is big. Therefore the λ_Tik should be selected within the range from 0.1 to 0.3, corresponding to a total side-lobe proportion from 7.08% to 16.64%. Thus a sharp partition resolution as well as high SNR gain could be achieved. And λ_Tik=0.3 was selected in Figure 6 based on the consideration of better image quality.

One limitation is the relatively narrow range of acquired TEs (i.e. from 49ms to 121ms, where the 49ms is the shortest possible TE based on current acquisition parameters). We have employed these values to target gray and white matter in the brain, but this range may not lend itself to targeting short T₂ species (e.g. cartilage and myelin) or to capture very long T₂ species (e.g. CSF). Another potential fact that could influence the BUDA reconstruction is the imperfectly estimated B0 field map especially in regions with less signal, resulting in uncorrected distortion. To address this issue, a potential approach is to use an external acquisition to estimate B0 field map precisely. To further improve the accuracy of T₂ mapping, a potential approach is to compensate magnetization transfer effect by using an EPG model with magnetization transfer consideration(42). In addition, for translating this technique to other body parts, for example abdominal imaging, the challenges may come due to shot-to-shot differences caused by motion and the difficulty for estimating an accurate B0 map.

Based on the BUDA-gSlider SE-EPI acquisition strategy and gSlider-subspace reconstruction, a rapid, distortion-free, high-resolution, whole-brain T₂ mapping approach is proposed. The proposed method could obtain whole-brain distortion-free T₂ maps with 1-mm³ isotropic resolution in 2 min, which provides a favorable balance between acquisition speed, image quality and the accuracy of T₂ estimation.

Supplementary Material

tS1&fS1

Supporting Information Table S1. Undersampling pattern. Only six out of ten TEs per RF-encoding were acquired.

Supporting Information Figure S1. T₂ maps by using different numbers of shots. Top row, for each TE and RF-encoding, blip-up and blip-down shots were both acquired. Middle row, for each TE and RF-encoding, only one blip-up or blip-down shot was acquired in an interleaved way. Bottom row, for each RF, only six out of ten TEs’ shots were acquired.