Residual aliasing is a well-documented problem for multiband reconstructions, but it can be an important issue with in-plane acceleration methods as well. With GRAPPA in particular, the residually aliased signal can be distributed fairy randomly, making it appear as g-factor noise. We demonstrate that the use of TGAPPA permits not only the elimination of the residually aliased signal but also the determination of L-factor maps, which can be a potentially useful tool in understanding how to minimize residual aliasing.
All experiments were performed on a 3T MAGNETOM Prisma (Siemens Healthcare, Erlangen, Germany), using a Siemens 32-channel head coil. EPI data were acquired from 2 healthy subjects (one female), average age 38 yrs. For each subject two EPI runs were acquired. One run was a standard product EPI sequence with 1.5 mm isotropic resolution, consisting of 10 volumes of 80 transaxial slices, with an FOV of 192x192x120 mm3, iPAT=4, TR=5.5s, and TE=30ms. The second run was multi-shot EPI with the same resolution and coverage as the first run. This run was acquired in 4-shots with a shot TR of 5.5s. Both runs used a bandwidth of 890Hz/pixel. Each shot of the multi-shot dataset was reconstructed offline using a TGRAPPA reconstruction scheme (in-house code), resulting in a 40-volume dataset with a volume TR of 5.5s. In both GRAPPA and TGRAPPA reconstructions, a 4x3 GRAPPA kernel was used.
The last step in the TGRAPPA reconstruction is to apply a temporal notch filter that removes the Nyquist frequency and 0.5*Nyquist frequency. These are the frequencies that the residually aliased signal oscillates at in an iPAT=4 TGRAPPA acquisition. L-factor maps are determined by subtracting a filtered TGRAPPA image from its corresponding unfiltered image and normalizing by the filtered images, as shown in Equation 1. The subscript i denotes voxels in the image volume.
$$L-factor_{i}=\frac{\mid TGRAPPA(unfiltered)_{i}-TGRAPPA(filtered)_{i}\mid}{TGRAPPA(filtered)_{i}}\quad\quad\quad[1]$$
Fig. 1 A&D show example GRAPPA images for Subject A and Fig. 1 B&E show TGRAPPA images. By visual inspection it can be seen that the TGRAPPA images are less noisy, especially toward the middle of the brain and for the lower slices. Fig. 1 C&F show L-factor maps for this subject. From the L-factor maps it can be seen that the residual aliasing can be more that 10% of the filtered signal in many areas. While the regions in the L-factor map of high residual aliasing show some structural features in this dataset, the aliasing is largely incoherent, illustrating why the residually aliased signal in the GRAPPA images can be easily mistaken for g-factor noise.
Fig. 2 shows example results for Subject B. The L-factor maps have similar features to the L-factor maps for subject A, though the locations of high residual aliasing do not appear to coincide well between the two subjects. This suggests that the residual aliasing patterns are not purely a function of coil sensitivities, but also include session specific and/or subject specific factors.
The motivation for the original development of TGRAPPA and its predecessor TSENSE was to improve reconstructions in the presence of subject motion. With normal GRAPPA or SENSE, the misalignment between the calibration data and the data being reconstructed results in both increased and time varying residual aliasing. Thus, the residual aliasing contributes to the deterioration of both image SNR and tSNR. While the TGRAPPA filter removes the residually aliased signal, an important improvement, the L-factor maps can be a useful tool in optimizing the acquisition strategy to minimize the residual aliasing as well.