Functional MRI (fMRI) has been implemented using multiband (MB), multi-shot (MS) EPI at 7T. The development of MB imaging involves the excitation of several slices simultaneously resulting in improved temporal and spatial resolution for 2D EPI acquisitions while maintaining SNR^1,2. EPI-based fMRI at 7T is able to achieve increased blood oxygenation level dependent (BOLD) contrast³ with increased achievable signal-to-noise ratio (SNR) relative to 3T; however these advantages come at the cost of increased image distortion caused by the high effective echo spacing inherent in EPI and increased magnetic susceptibility effects at 7T. One approach to eliminate such distortion while maintaining SNR is to use multishot (MS), segmented imaging techniques. In MS imaging, k-space is collected in multiple segments allowing the effective echo spacing to be reduced. Therefore, in this work we aimed to test a technique combining MB and MS techniques for fMRI at 7T (MBMS EPI).

Adding a temporally modulated slice-wise phase tag to the excited slices in MB imaging allows the slices to be unaliased by a simple addition and subtraction of repetitions using the mathematics of Hadamard encoding⁴. Thus, slices can be unaliased without parallel imaging techniques. Furthermore, the effective averaging inherent in Hadamard unaliasing yields high SNR images. This technique was added to the MBMS sequence.

A standard EPI sequence was modified to replace the excitation pulse with a multiband excitation pulse. The relative phase tags of the excited slices were temporally modulated to allow for unaliasing using Hadamard encoding of the individual shots. The sequence was further modified to collect k-space data in an interleaved multi-shot fashion. Imaging was performed on a 7.0T GE MR950 system with a 32-channel NOVA head coil and head-only quadrature transmit coil. Two volunteers underwent three sagittal MBMS EPI scans with two, three, and four segments respectively, and eight slices excited simultaneously (MB8). Parameters were as follows: matrix=128x128x11, resolution=1.6mm isotropic, FA=45°, and TR=750,1000,1500ms for the two, three, and four segment acquisitions respectively, resulting in an effective TR=3000ms for all scans. This resulted in 88 total slices and full brain coverage. Total imaging time was 6 min. Subjects performed a finger-tapping task, which consisted of 30s of rest followed by 30s of finger-tapping. The reconstruction algorithm is shown in Figure 1. Reconstruction started with Nyquist ghosting correction. To account for shot to shot phase variations inherent with MS imaging, navigator echoes were acquired prior to each shot by sampling the center of k-space in the positive and negative read-out directions⁵. These echoes were used to correct the ghosting on a slice-by-slice and repetition-by-repetition basis. Next, k-space was filled by combining data from each segment. Finally, the data was slice-unaliased using Hadamard encoding and a moving window with a width of 8 repetitions. The fMRI data was analyzed using independent component analysis (ICA) and FSL’s (www.fmrib.ox.ac.uk/fsl) MELODIC software. Data was motion-corrected using MCFLIRT, blurred using a Gaussian kernel with FWHM=3.0mm, and high-pass filtered. For each scan, the component representing the motor network was manually extracted. For each timeseries, tSNR was also computed.

Distortion can be seen in the anterior frontal lobe and occipital lobe and was reduced with increasing number of segments (Figure 2A). Mean tSNR across the whole brain was near 30 for all datasets; however tSNR in gray matter, where brain activation is detected, was consistently greater than 50 (Figure 2B). Activation elicited from the finger-tapping task was reliably found for all datasets using ICA. Activation patterns were very similar between the two, three, and four segment acquisitions (Figure 3). The segmented acquisition has an advantage in that data can be combined to create datasets with effective in-plane accelerations (Reff) up to the number of seqgments by combining all segments, every other segment, or reconstructing each segment individually with GRAPPA. Similarly, the moving window Hadamard encoding can be sub-sampled to yield further MB acceleration which can be unaliased with slice-GRAPPA. Thus, one can titrate the SNR of a reconstructed data set by selecting the shots used for reconstruction. In this scheme, the choice of temporal resolution can be made retrospectively through the selection of the number of shots to be combined. For instance, one can reconstruct a dataset with the low TR, and higher SNR of an R=2 acquisition, but with the reduced distortion of an R=4 acquisition (Figure 4). Hadamard-encoded MBMS imaging is a reliable technique that yields high SNR, reduced distortion EPI images with high temporal and spatial resolution at 7T. Four segment EPI has reduced distortion and provides the ability to choose the effective acceleration.