Slice-accelerated (SliceAcc) multi-slice techniques (i.e. “multiband”) utilize RF pulses which excite multiple 2D slices simultaneously [1-4]. As all of k-space is measured, SliceAcc has the benefit of acquiring more slices per unit time without the usual SNR penalty. One use of SliceAcc is high spatial resolution EPI applications where the minimum TR would otherwise be higher than needed (i.e. high-resolution fMRI or DTI), allowing shorter scan time. Another use is high temporal resolution EPI where the maximum number of slices per TR would otherwise be lower than desired (i.e. resting state fMRI or DSC perfusion).

Parametric T₁ mapping using inversion recovery (IR) has competing requirements for speed, SNR, spatial resolution, anatomical coverage, and adequate sampling of the longitudinal magnetization recovery. One approach involves a non-selective IR pulse followed by 2D multi-slice echo-planar imaging (EPI), which is repeated with the EPI slice acquisition order permuted each time such that each slice experiences a different effective TI within each repetition [5,6] (here called shuffled-IR-EPI). Previous work described the application of SliceAcc to quantitative T₁ mapping, providing increased slices and/or TI points acquired in a given measurement time [6-7]. Here, we highlight new aspects in which SliceAcc is beneficial for high spatial resolution applications at 7T.

Slice-accelerated shuffled-IR-EPI was implemented as previously described [10]. A non-selective IR pulse inverts the magnetization, and is followed by the SliceAcc EPI imaging kernel which is repeated N_TI times. An optional recovery period (T_relax) can be used before another IR pulse is applied. The slice ordering is permuted for each repetition to change the effective TI for each slice. The updated implementation permitted N_TI <= (N_slice/SliceAcc), where N_slice is the total number of slices, and SliceAcc is the slice acceleration factor. When, N_TI < Nslice, the TIs are distributed evenly throughout the IR for each slice, but the TIs are slightly shifted for adjacent slice groups.

Potential sources of slice-to-slice variability were investigated, including slice cross talk due to imperfect slice profiles, and the range of TIs used to fit each slice. To investigate the effect of the range of TIs on the parametric maps and to differentiate it from slice cross talk effects, a dataset with N_TI = (N_slice/SliceAcc) was fit with different subsets of TIs as would occur if N_TI=0.5*N_slice/SliceAcc, and the resulting T₁ values were compared to the fully sampled case. To investigate the effect of slice cross talk, T₁ maps were compared for protocols with T_relax=2sec and T_relax=0. In the case of T_relax>0, the effective TR for the first slice interleave equals TR, while the effective TR for the second slice interleave (for any portion of the slice that was excited by the other interleave due to imperfect slice profiles) is (TR-T_relax)/2.

Measurements were performed on a Siemens MAGNETOM 7T using a Nova Medical 24-channel head coil. The protocols used were: TE/TR=24/5000-7000ms, T_relax=0-2sec, (1.2mm)³ voxels, GRAPPA=3, SliceAcc=2, 102 slices, N_TI=25 & 51, TI 32-5000ms, interleaved slice ordering, scan time 4:14-6:00. The series of magnitude images were fit voxel-wise to the 3-parameter model S(TI) = M₀*|1-A*exp(-TI/T₁)|, for maps of T₁, M₀, and A, where A=2 in the case of ideal inversion.

T₁ differences of 5-10% were seen between adjacent slices when the range of TIs were not identical, even though they only differed by ~100ms. This is seen in the multiplanar reformatted images in Figure 1. The signal differences were reduced when T_relax=0, indicating a small amount of slice cross talk effects. However, the T₁ differences were still evident. When both T_relax=0 and the TI range were the same for all slices, these differences were no longer significant (Figure 2). Slice-to-slice variations were also seen when the dataset in Figure 2 was subsampled to simulate the data in Figure 1, to demonstrate it was not purely an SNR issue.

Slice acceleration improves imaging efficiency for protocols with thin slices (i.e. N_slice>80), by allowing thin slices without longer TRs and longer scan time. It can also be used as an alternative way to reduce the number of repetitions (slice permutations) needed to provide the same TI range to all slices, thereby reducing scan time. Although this can introduce dead time between slices if the TR is not reduced, it avoids slice-to-slice variations in the T₁ maps.

In conclusion, we demonstrated that T₁ relaxometry benefits significantly from SliceAcc in not only scan time and slice coverage, but in its ability to reduce quantitative errors by allowing the same TI sampling across all slices.