Introduction

Simultaneous multislice EPI (SMS-EPI) can reduce volume acquisition times in brain MRI and is widely used in fMRI and diffusion MRI.¹ However, the gain in scanning efficiency is often used to acquire more complex scanning protocols (such as multiple b-values, more diffusion directions), which may lead to artifacts due to patient movement. Of particular concern are larger motions, which may change coil sensitivity maps. Therefore, we investigated real-time motion correction (MoCo) enabled SMS-EPI with split slice-GRAPPA (SSG) and SENSE reconstruction ^2-4 with coil sensitivity interpolation.

Materials and Methods

The study was performed on a 3 T whole-body MR imaging system (Siemens Prisma, Erlangen, Germany) using a 20-channel head coil. A blipped-CAIPIRINHA SMS sequence was used.⁴ Unlike prior implementations, ^4,5 the blip scheme of the sequence has no prewinding lobe; see Fig. 1 for two simultaneous slices (bottom slice at isocenter). The phase variation at top slice edges is $$$\theta(z',N_{y})=\sum_{n_{y}=1}^{N_{y}}rA_{blip}(n_{y})z'$$$, with slice offset z' and blip area along phase direction $$$A_{blip}(n_{y})$$$. The interslice image shift is $$$\frac{FOV}{4}$$$, and the phase variation at slice edges is $$$\frac{\pi}{2}\pm\theta$$$ (Fig. 1a). The $$$G_{z}$$$ blip cycle is 1:1:(-3):1 (Fig. 1b). When MoCo alters slice positions, an additional phase is imposed by the $$$G_{z}$$$ blips (Fig. 1c). Accordingly, the receiver phases need to be altered; this was implemented in real-time by updating the phases of ADC events (lines along $$$k_{x}$$$).
To test the accuracy of the real-time phase compensation scheme, we injected 10mm slice offsets while acquiring a two-slice single-shot SMS-EPI time series in a stationary phantom. Single-band reference images were acquired with EPI parameters matched to the SMS acquisition. SSG was used to separate the simultaneously acquired slices.³
Next, motion experiments were performed in a phantom with an internal structure using tracking markers. The phantom was scanned inside the head coil with the SMS-EPI sequence (thickness 5 mm, SMS-factor 2, 10 mm slice-to-slice distance, TR/TE 3000/37 ms, FOV 280 × 280 mm², matrix 64 × 64, bandwidth 2170 Hz/pixel, 20 repetitions). Phantom motion started at the 10th repetition, primarily x-rotations (up to 20°) and z-translations (up to 50mm). Prospective motion updates (slice offsets and rotation matrix) were applied, and the receiver phase was additionally compensated in real-time as noted.
For SENSE and SSG, raw data were reconstructed twice, using:
1. original coil sensitivity maps (oCSM) or GRAPPA weights from single-band reference images, and
2. updated coil sensitivity maps (uCSM) calculated by interpolation from oCSM for altered slice positions, or updated GRAPPA weights, calculated for altered slice positions from single-band reference images with uCSM.
Additionally, raw data from the first repetition (stationary phantom) was used to simulate and quantify motion in SSG reconstructions, from 1-50mm translations and 1-50° rotations. With no-motion images (I_standard) as the gold standard, motion-induced errors were quantified for each reconstructed image (I_estimated) using the cumulative normalized mean squared error (cNMSE), ⁶ $$cNMSE=100\times\sum_{s=1}^S\frac{\sum_r{\mid{\mid{I_{estimated}^s}(r)}\mid-\mid{I_{standard}^s}(r)\mid\mid}^{2}}{\sum_r\mid{I_{standard}^s}(r)\mid^{2}},$$ where r denotes the voxel position, and S is the slice number. Next, the movement was reversed (i.e., motion correction was performed) and cNMSE recalculated.

Results

Fig. 2 shows the SMS-EPI images with and without real-time receiver phase compensation. The left column is the aliased image of two slices with $$$\frac{FOV}{4}$$$ blipped-CAIPI interslice shift, and the right two columns show the SSG reconstructions of the two slices acquired simultaneously. Shifting the slices by 10mm without compensating the receiver phase caused substantial aliasing artifact in the individual images (Fig. 2a). Conversely, ghost artifacts were eliminated when the receiver phase was updated in real-time (Fig.2b).
Fig. 3a shows the SMS reconstruction of SSG and SENSE methods with oCSM and uCSM, while the motion was at its maximum (motion parameters within TR period in Fig.3b). The reconstructed images using oCSM show residual aliasing artifacts that obscure the inner structures and outer boundaries. Conversely, images reconstructed with uCSM show improved delineations of inner structure and boundaries as well as much less aliasing artifacts.
Fig. 4a shows the results of the two reconstruction methods towards the end of the scan when the phantom was stationary (x/y/z-translation -2.4/6.4/21.5mm and x/y/z-rotation 10.5°/5.2°/-0.5°). Again, residual slice-aliasing artifacts appear in most slices of the SSG with oCSM (solid arrows). Conversely, the images reconstructed by SSG with uCSM show much less artifact and clearer phantom inner structure and boundaries. SENSE with oCSM shows aliasing artifact (red boxes), but SENSE with uCSM shows clear slice structures, and there is no slice-aliasing artifact.
Fig. 5 shows the cNMSEs comparisons between images with oCSM and uCSM for SSG reconstructions (for translations 1-50mm and rotations 1-50°). The cNMSE (black curves) with original kernels rise supra-linearly with translation and rotation increases (black curves). When coil sensitivity maps are corrected, the cNMSE (red curves) show minimal dependence on motion parameters.

Discussion

For larger movements, SSG and SENSE with dynamically updated sensitivity maps (uCSM) show marked improvements in aliasing artifacts and image quality compared to reconstructions using the original sensitivity maps (oCSM). By interpolating spatial sensitivity profiles, the two techniques have the ability to separate slices from collapsed multislice images even in the presence of severe motion. Of note, a larger number of slices, thinner slices and smaller slice gaps will most likely result in smoother uCSMs and ultimately better reconstructions.