The maximum parallel imaging acceleration factor generally depends on coil geometry, k-space trajectory and slice positioning. Unfolding data becomes increasingly difficult with higher acceleration factors, resulting in a g-factor penalty in (temporal) SNR. In functional MRI acquisitions with 3D-EPI, acceleration in both phase-encoding directions, but especially the slab-selection direction, leads to shorter volume TRs (1,2) and can be highly beneficial in increasing BOLD sensitivity.

In this study, we show that, using a 32-channel coil and 3D-EPI-CAIPI acquisition (3), a cylindrical excitation can be used to achieve high undersampling factors (4x3), while maintaining high image SNR as well as tSNR and hence, improved BOLD signal detection.

3D-EPI-CAIPI was combined with a 2D-RF pulse (4) to selectively excite a cylinder. Placing the cylinder along the read-out axis maximizes time benefits. Two protocols were compared in 4 volunteers at 7T, both cylindrical excitation and a standard slab selection:

- a 2-mm isotropic acquisition (FOV=200*200*120mm, TR/TR_vol/TE/α=55ms /1.1s/27ms/18^o, GRAPPA=3x2, ΔCAIPI=1, readout gradient and cylinder along the left-right axis, cylinder radius 25mm, alias 22cm)

- 0.9*0.9*2.0 mm acquisition (FOV=200*200*120, TR/TR_vol/TE/α=55ms /1.1s/27ms/18^o, GRAPPA = 4x3, ΔCAIPI=1; same cylinder).

Cardiac and respiratory data were collected for physiological noise removal using RETROICOR. An auditory stimulus (5s natural sounds, 15s silence alternated for 5 minutes) was used to test BOLD sensitivity in the primary auditory cortex.

Image SNR was estimated by dividing the mean signal in a large mid-cylinder ROI (Figure 1, blue box) by the standard deviation of the noise in an extra-cerebral ROI not affected by image artefacts.

Image SNR in the 2-mm isotropic data was 30±4 % (mean ± stderr) higher in the cylindrical excitation data than in the equivalent slab selection data. This improvement was further emphasized on the 0.9mm data - 80±30 %. The difference in image SNR is clearly visible in Figure 1, where both images are scaled to give equal appearance to the noise in the background.

Higher image SNR translated also into increases in tSNR values, of 35±4 % in the 2mm data and 23±2 % in the 0.9mm data. Again, the tSNR is visibly improved in the tSNR maps in Figure 2.

The high tSNR led to highly significant BOLD responses in the cylindrical EPI data, an example of which is shown in Figure 3. Note the high temporal resolution, evident from the shown unfiltered time course of a single voxel. Activation patterns were highly reproducible between subjects and acquisition methods. No significant differences in number of significantly (p<0.05, FWE) voxels were found, possibly due to the much changed number of voxels included in the GLM mask.

One of the main advantages of the 3D-EPI-CAIPI sequence is its flexibility in optimising the acquisition for maximum spatial or temporal resolution (1,2,3,5). The cylindrical excitation pattern could be used to reduce the imaging FOV, thus accelerating the acquisition and allowing higher spatial resolution, however, it is probably more advantageous to increase the GRAPPA undersampling factor, which also leads to significantly shorter volume acquisition times and can be done without incurring the usual undersampling artefacts (g-factor noise amplification), as large parts of the imaging FOV do not contribute with any signal that needs to be unwarped. The higher tSNR in the 0.9mm cylindrical acquisition data is a direct result from this improved unfolding, as all parameters apart from the excitation profile of the pulse remained the same between cylindrical and slab selection acquisitions.Here, we successfully demonstrated that 3D-EPI-CAIPI with cylindrical excitation can be used to acquire fMRI data with submillimetre spatial resolution, 1-second temporal resolution and very high BOLD sensitivity.