Introduction

EPI can provide whole-brain coverage rapidly, but does not lend itself to high in-plane resolution imaging due to severe geometric distortion, voxel pile-ups and T2*- or T2-related blurring. This has impaired its utility in high-resolution structural imaging and limited its application to functional or diffusion (dMRI) at lower in-plane resolutions of 1.5-2 mm. Sensitivity encoding allows for up to R_inplane=4 acceleration to mitigate, but not eliminate these artifacts. Recent developments in navigator-free multi-shot EPI reconstruction has allowed for R_inplane8 (2,3), but this requires several shots of data for adequate image quality and fails to eliminate distortion.
An alternative strategy is to collect multiple shots with alternating phase-encoding polarities, and reconstruct these jointly using field map information in the forward model as in Hybrid-SENSE (4). BUDA incorporates Hankel structured low-rank regularization into Hybrid-SENSE to eliminate the need for phase navigation (1), and can provide distortion-free dMRI at high in-plane acceleration with minimum T2-blurring.
With dSAGE, we propose to utilize the empty sequence time during the b=0 image acquisition. We add two additional EPI readouts to exploit the dead time created by the absence of diffusion gradients. This provides a T2*- and a T2’-weighted contrast image without additional scan time. In essence, we replace the simple b=0 acquisition with a spin- and gradient-echo (SAGE) sequence (5). Combining this with 4-shot BUDA encoding at high R_inplane allows us to create a distortion-free, 1-minute clinical scan at high in-plane resolution (1x1x4 mm³) with gradient-, spin-, and mixed-echo images as well as 3-direction dMRI.

Methods

Fig1 shows the proposed prototype sequence diagram where two additional readouts, with gradient and mixed gradient- spin echo contrasts, are added in the unused sequence time during the b=0 acquisition. Four shots of EPI are acquired for each diffusion weighted or b=0 image, where the alternating shots are sampled with opposite phase-encoding polarities. This way, two blip-up and two -down shots are collected for BUDA reconstruction.

Acquisition: Data were collected on a 3T Siemens Prisma (MAGNETOM Prisma, Siemens Healthcare, Erlangen, Germany) using 32-channel reception with FOV = 220×220×128mm. 1×1×4mm³ resolution b=1000s/mm² diffusion data were acquired at R_inplane=6 with 4-shots using TE/TR=68/3500ms. The diffusion gradients were applied along the x-, y- and z-directions. b=0 acquisition was made at R_inplane=9 and partial Fourier=6/8, which provided two additional echoes. 4-shots were acquired at the TEs = [14 48 68]ms.

Reconstruction: For spin-echo contrast where shot-to-shot phase variations are minimal due to the refocusing pulse, we first sum the k-space data for the two blip-up and the two blip-down shots and perform SENSE reconstruction to generate interim blip-up and blip-down images. We then estimate a field map using topup (6) and incorporate this in the joint BUDA reconstruction of the 4-shots in the other contrasts:
$$min_x \sum_{t=1}^{N_{s}}{ \left \| {F_{t}W_{t}Cx_{x}-d_{t}} \right\|}_2^2 + \lambda \left \| {H(x)} \right\|_*$$
Where F_t is the undersampled Fourier operator in shot t, W_t is distortion operator in shot t, C is sensitivity map from ESPIRIT [2], and dt is the k-space data of each shot. ||H(x)||_* is Hankel structured low-rank constraints. The 3-direction diffusion averaged DWI and ADC maps were calculated after the joint BUDA reconstruction.

Results

Figure 3 shows distortion-free high-resolution multi-contrast images including T2*-, T2- and T2-weighted images.
Figure 4(a) shows distortion-free dMRI obtained with diffusion gradients applied along the x-, y- and z-directions. The averaged dwi images and ADC maps are shown in Figure4(b).
The total acquisition time for all contrasts was 56s, plus a 2s calibration scan for coil sensitivities.

Discussion & Conclusion

We demonstrated distortion-free multi-contrast clinical images and dMRI at high in-plane resolution from a single, 1-minute scan. High geometric fidelity of the images can be appreciated in the lower slices in Figs 3 and 4. As the shot-to-shot variations are higher in dMRI, we limited the acceleration to R_inplane=6 to facilitate high quality reconstruction. Structural images supported yet higher (R_inplane=9) acceleration with 6/8 partial Fourier, which minimized their readout duration and allowed them to fit in the unused sequence time. As in single-shot SAGE imaging, we anticipate that these echoes will lend themselves to T2 and T2* parameter mapping, further improving the utility of this sequence.

Acknowledgements

This work is supported in part by the National Natural Science Foundation of China (No: U1809204, 61525106, 61427807, 61701436), by the National Key Technology Research and Development Program of China (No: 2017YFE0104000, 2016YFC1300302), and by Shenzhen Innovation Funding (No: JCYJ20170818164343304, JCYJ20170816172431715), and by:NIBIB Award Number: P41 EB015896, R01 EB017337, R01 EB019437, R01 EB020613 and U01 EB025162;NIMH, Award Number: R01 MH116173 and R24 MH106096;Shared instrumentation grants S10-RR023401 and S10- RR023043; andNVIDIA GPU grants. Zijing Zhang is supportedby the China Scholarship Council for 2-years study at Massachusetts General Hospital.