Multiband (MB) imaging, which involves the simultaneous excitation of multiple slices, has been developed to speed up 2D EPI acquisitions^1,2. In order to solve the slice-unaliasing problem of MB imaging, a fully sampled calibration dataset is needed. Typically, a separate reference scan with a MB-factor=1 (MB1) is collected. However, collecting the reference scan in this way can lead to an increased risk of motion between the reference scan and data acquisition, and long TRs resulting in longer acquisitions with varied T1 weighting. At 7T, increased susceptibility effects necessitate the addition of in-plane acceleration to reduce distortion and T2* related signal loss. To address these issues, a multiband scan with multishot (MS) autocalibration was developed and prepended to a MB-EPI acquisition. Furthermore, an application of Hadamard unaliasing with an acceleration factor lower than the rank of the Hadamard encoding matrix was implemented to unalias the slices³, yielding a fully sampled calibration scan unreliant upon parallel imaging techniques.

Autocalibration repetitions were Hadamard encoded through the addition of a temporally modulated slice-wise phase tag incorporated into the excitation pulse. Phase tags are selected so that a linear combination of the different aliased repetitions results in the signal being cancelled in all slices but one³. Initial testing was performed on a phantom on a 7T GE Healthcare Discovery MR950 system using a 32-channel Nova receive-only head coil and quadrature transmit coil. Further evaluation was conducted on three healthy volunteers under informed consent. Each subject underwent a resting state sagittal MB gradient-echo EPI scan. All data was collected with the following parameters: matrix=128x128, resolution=1.6mm isotropic, TR/TE=1000/23ms, FA=45°, MB5, in-plane R=2, excitations per repetition=17. This resulted in 85 total slices (17x5) and full brain coverage. Controlled aliasing was performed using the Blipped-CAIPI approach with a FOV shift of 3⁴.

At the beginning of the MB acquisition, 32 interleaved autocalibration volumes were acquired using a Hadamard-encoded multi-slice excitation³. To Hadamard unalias five slices, eight repetitions are needed to form a standard Hadamard encoding matrix in which there are effectively three encoded null images. The k-space volumes from the odd and even interleaves were combined prior to Fourier reconstruction and Hadamard unaliasing resulting in a fully sampled, unaliased calibration volume. The time series data following the calibration repetitions was reconstructed using the autocalibration scan as the reference and a slice-GRAPPA algorithm⁴. Navigator echoes were obtained at the beginning of each shot and used to correct for ghosting on a slice-by-slice and repetition-by-repetition basis. This algorithm is shown in Figure 1.

The task-free fMRI data was analyzed using independent component analysis (ICA) and FSL’s (www.fmrib.ox.ac.uk/fsl) MELODIC⁵ software with 25 components. 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 default mode network was manually extracted. For each timeseries, the tSNR was also computed.

The Hadamard unaliased, multishot autocalibration volumes were suitable for use as reference scan for the slice and in-plane unaliasing of the subsequent MB-EPI volumes. Autocalibration images from a volunteer are shown in Figure 2 alongside reconstructed MB-EPI volumes and tSNR maps. The mean tSNR in the whole brain across subjects was 19.0+/-2.8, however tSNR in gray matter, where brain activation is detected, was consistently greater than 30 (Figure 2C). Reliable functional connectivity of the default mode network was found for all subjects using ICA (Figure 3).One advantage of Hadamard unaliasing is the effective averaging inherent in the unaliasing process. The unaliasing of five Hadamard encoded slices is essentially equivalent to eight averages since eight repetitions are used to unalias the five slices³. Thus, a sufficient calibration volume could be obtained in 16s (8 Hadamard repetitions x 2 interleaves x TR=1s). This usage of eight repetitions to Hadamard unalias five slices and three null slices is novel. This allows the utility of Hadamard unaliasing while retaining a MB acceleration factor which is within hardware RF amplitude limits and coil sensitivity limits for subsequent GRAPPA based unaliasing.A Hadamard-encoded multiband, multishot autocalibration scan can be used to reliably unalias MB-EPI data. This reference scan was prepended to the EPI acquisition, reducing the chance of motion between the reference scan and EPI acquisition. This work further shows an application of Hadamard unaliasing with an acceleration factor lower than the rank of the Hadamard encoding matrix through the inclusion of the mathematical construct of null slices. Finally, the averaging inherent in the Hadamard unaliasing process, and the fact that k-space is fully sampled, yields high SNR calibration volumes in a short amount of time.