IRIS has been proposed for diffusion-weighted (DW) image reconstruction with 2D-navigated motion correction from highly aliased DW-MRI data using multishot and regularly subsampled SENSE implementations simultaneously. In this work, it is demonstrated that IRIS framework can be extended for multiband DW image reconstruction for reconstructing motion-free multiband multishot DW images.
Introduction
One of the main challenges in high resolution multishot DW-MRI is to reduce the effect of subject bulk motion while preserving the sensitivity to intravoxel water incoherent motion1,2. Since DW images are usually acquired using strong gradients for sensitizing water molecular diffusion, even a small amount of subject motion can generate significant phase variations and cause large aliasing artifacts, unless the phase variations are properly eliminated3,4. Reconstructing DW images acquired with multishot and parallel imaging is especially challenging when 2D or higher dimensional navigated motion correction is required. Voxel-by-voxel phase correction at an image from each subset of undersampled k-space data is prohibited by significant aliasing due to undersampling by segmented acquisition combined with parallel imaging5,6. IRIS (Image reconstruction using image-space sampling function)7 has been proposed to enable DW image reconstruction of high resolution multishot acquisition with 2D navigated motion correction without resorting to an iterative whole image reconstruction algorithm typical in non-Cartesian acquisitions. Multishot acquisitions are suitable for high resolution DW-MRI, but usually they require much larger scan time than the single-shot counterpart, especially for whole-brain coverage. A recent study on interleaved DW-EPI combined with multiband8 is promising for further reduction of scan time9. In this study, it is demonstrated that IRIS can be easily extended for additional multiband reconstruction.Method
The aliased image voxels for an image column j with cth coil sensitivity and sth shot can be unfolded by using:
gj,c,s = As · Sj,c · Pj,s · fj,
here, As is image-space sampling function for shot s, Sj,c is a coil sensitivity information in column j of coil c, Pj,s is the phase error estimated in column j and sth shot of 2D navigator phase image, fj is column j in an ideal image and gj,c,s is taken from the jth column of the aliased image. To integrate multiband acquisition and reconstruction, equation above can be extended as below,
gj,c,s = As · [ Sj,c,l1 · Pj,s,l1 | Sj,c,l2 · Pj,s,l2 | ··· | Sj,c,lmb · Pj,s,lmb ] · [fj,l1T | fj,l2T | ··· | fj,lmb T]T,
here, T represents transpose, subscripts l1, l2, ..., lmb are the slices simultaneously acquired using multiband excitation and | represents matrix column-wise concatenation. After appending the equation in a matrix column direction for all coils, shots and measurements, image column-wise inversion can provide the entire unfolded image. Multiband acquisition was simulated using existing multishot DW image data with multiband factor of 2 from 6 slices.
Results
The simulated results were compared without and with 2D navigated motion correction (Fig.1). Figure 1 presents without motion correction (top row), with motion correction (the second row), original image (the third row) and g-factor map (bottom row). There are shading artifacts on anterior portion of the brain and regions with relatively high g-factor which may be caused by similarity in the coil sensitivity.
Discussion
In this simulation the use of IRIS was extended for additional multiband acquisition, where in-plane (SENSE factor x number of shots) and through-plane (multiband excitation) reduction can be implemented resulting in the improvement of multishot DW-EPI by SENSE factor x number of shots x multiband factor, and then the highly aliased image were successfully unfolded with motion correction, simultaneously. In this simulation, reconstruction was successful with multiband factor 2, number of shots 4 and SENSE factor 1 (reduction x8) at a given volume of slices. However, the reconstructed images were noisy with SENSE factor 3 (reduction x24) or multiband factor 3 (reduction x12). This is probably due to spatially smooth coil sensitivity at a given imaging volume, which resulted in noisy image reconstruction combined with multiband. Future work will address this issue and include sequence implementation and optimization with multiband IRIS for accelerated high resolution DW imaging.Conclusion
Multiband, multishot and SENSE DW-MRI data were successfully integrated into multiband-IRIS framework and reconstructed with 2D navigated motion correction.