Introduction

Diffusion weighted imaging (DWI) is widely used to probe the displacement of water molecules within the tissue. DWI is usually acquired using a single-shot spin-echo echo pIanar imaging (SS-EPI) with diffusion encoding gradients placed symmetrically on either side of the 180° refocusing pulse. In this traditional implementation, DWI has been robustly applied to several organs.

Despite, its speed and efficiency, SS-EPI images remain severely affected by B₀ inhomogeneities, which create image distortion artifacts in the phase encoding direction. One way to reduce this B₀ sensitivity is to shorten the length of the EPI echo train (T_Epi) by reducing the number of k-space lines acquired in the EPI readout. High readout bandwidth, partial Fourier, and parallel imaging all reduce T_Epi, but at the expense of a lower signal-to-noise ratio (SNR).

Multi-shot spin-echo EPI (MS-EPI), in which k-space is acquired segmentally, results in a lower number of k-space lines acquired per shot, hence a shorter T_Epi. However, MS-EPI suffers from shot-to-shot phase variations due to the interaction between physiologic motion and the diffusion encoding gradients^1,2. These phase variations generate strong image aliasing artifacts. Several retrospective methods have been proposed^3,4, such as low-rank constrained approaches, However these retrospectives strategies remain impractical as they require long, offline reconstruction processes and are subject to extensive ad hoc tuning.

In this work, we propose prospective strategies to reduce shot-to-shot phase variation for interleaved phase-segmented MS-EPI. First (M₀M₁) and second order (M₀M₁M₂) motion-compensated gradient waveforms have been used in cardiac diffusion imaging to limit the impact of cardiac motion on the diffusion signal^5,6. The first objective of this work was to investigate motion-compensated waveforms to reduce shot-to-shot phase variation compared to traditional monopolar encoding (M₀). The second objective was to evaluate a new phase-segmented EPI k-space sampling strategy, in which the center of k-space is fully sampled with each shot. This sampling pattern involves two speeds of EPI phase encoding and few extra k-space lines. In silico, in vitro, and in vivo validations on the brain, liver and heart are proposed.

Methods

All analyses proposed in this work were based on a protype DWI sequence (FOV=200x200mm², matrix=128x128, Partial Fourier=6/8 slice thickness=5mm, Bandwidth=2056Hz/px TR=2000ms). One b-value=0 image and six diffusion-weighted directions were used with b-value=800s/mm² for brain and in vitro experiment, b-value=500s/mm² for liver and b-value=350s/mm² for heart. All comparison were done at a constant acquisition time (2min) by adjusting the number of averages (N_avg=8,4,2,1) for the corresponding number of shot (N_shot=1,2,4,8). TE was minimized for a given N_shot and a given diffusion encoding waveform.

The dual-speed EPI readout was built by changing the phase encoding blip amplitude and by adding extra lines at the center of the k-space as shown in Figure 1. The number of extra lines was +8 for N_shot=2 and +16 for N_shot=4,8. T_Epi was then longer for dual-speed EPI than single speed EPI at equivalent N_shot.

In silico simulations using a Shepp-logan phantom were performed to study shot-to-shot phase variation and B₀ inhomogeneities. Single and dual-speed EPI with N_shot=1,2,4,8 were compared.

In vitro and in vivo experiments were performed on a 3T scanner (MAGNETOM Prisma, Siemens Healthcare, Erlangen, Germany). First B₀ inhomogeneities were studied by using a phantom composed of seven tubes with different diffusion coefficients and a tube with trapped air. Then, three healthy volunteers (N=3) were imaged using single and dual-speed EPI with N_shot=4. Diffusion encoding waveform M_0, M₀M₁ and M₀M₁M₂ were compared in the brain and liver while only M₀M₁M₂ was acquired on the heart. All reconstructions were performed offline. GRAPPA with 24 external reference lines was used as an unaliasing technique on each individual shot.

Results

In silico analysis (Figure 2) shows image aliasing due to shot-to-shot phase variation. Image aliasing increases as the number of N_shot increases. GRAPPA reconstruction can unalias images at N_shot=2, but cannot recover highly aliased images at N_shot=4,8. Dual-speed EPI mitigates aliasing and enables unwrapped images with GRAPPA even for N_shot=4,8. Lower root-mean-square-error (RMSE) were found for dual-speed EPI compared to single-speed.

As shown in Figure 3, in silico and in vitro B₀ inhomogeneity experiments show a decrease of image distortion as the number k-space lines/shots decrease. Since its T_epi is longer, dual-speed EPI yields more image distortion than single-speed EPI for equivalent N_shot. Nonetheless, lower RMSE was achieved with dual-speed EPI compared to SS-EPI.

Examples of in vivo acquisitions for brain, liver and heart are shown in Figure 4. M₀M₁M₂ motion-compensated diffusion encoding waveforms are sufficient to mitigate shot-to-shot phase variation in brain even using single-speed EPI. For the liver and heart acquisitions, only dual-speed EPI led to non-aliased images. Equivalent MD values were found between single-speed and dual-speed EPI across organs (Figure 5). Dual-speed EPI with M₀M₁M₂ leads to lower FA in gray matter and a higher FA in white matter than single-speed EPI. Dual-speed EPI also significantly increases SNR in the brain and liver.

Discussion and conclusion

In this work, M₀M₁M₂ waveforms were shown to mitigate shot-to-shot phase variation for MS-EPI diffusion imaging in the brain. Dual-speed EPI offers a tradeoff between B₀ inhomogeneity and aliasing artifacts mitigation. Dual-speed EPI enabled MS acquisitions in moving organs like in the liver and the heart.

Acknowledgements

No acknowledgement found.