Physicists and neuroscientists interested in fast sampling sequences for fMRI and/or hrf properties.

Temporal resolution in fMRI is typically limited to 2-3s per volume. However, recent developments in acquisition schemes, such as multiband¹ and 3D-EPI-CAIPI² , are overcoming this constraint. The 3D-EPI-CAIPI sequence can provide whole brain coverage including the cerebellum, in 400 milliseconds. This allows a precise hemodynamic response function (HRF) characterization without having to jitter the stimulus onsets.

The purpose of this study was to investigate the temporal properties of the BOLD response, including inter-trial variability, using a high-temporal resolution acquisition at 7T. Several different HRF parameters, such as time of onset, time to peak and undershoot amplitude, as well as inter-trial variability, were measured and compared between six brain regions involved in motor tasks.

Seven participants (3 females; 19-24 yrs) were scanned at 7T (Siemens, Germany). Participants performed a bilateral finger tapping task following a visual cue. Functional ROIs were defined based on a block task localizer (15s ON, 30s OFF alternated). HRF measurements were then taken from a 7-minute event-related run in which the subjects moved their fingers once every 30s.

For the localizer, a 2D EPI (2*2*2mm voxels, 30 coronal-oblique slices, TR/TE: 2500/26ms) was used. For the event-related task, a 3D-EPI-CAIPI (2*2*2mm voxels, 60 coronal-oblique slices, TR/TE:400/27ms, GRAPPA_z:6, Δ_CAIPIRINHA:2, partial Fourier:6/8) was used. Respiratory and cardiac traces were recorded for physiological noise removal using RETROICOR.

Localizer data were GLM analyzed (spm12, p<0.05) to define six bilateral ROIs: primary motor cortex (M1), primary sensory cortex (S1), supplementary motor area (SMA), secondary sensory cortex (S2), cerebellar lobule V (CV) and cerebellar lobule VIII (CVIII). The S1/M1 boundary was drawn manually. The timecourses of the within-ROI voxels were extracted from the realigned 3D-EPI-CAIPI data (CAIPI). Timecourses were also downsampled for comparison (dsCAIPI; five point average; TR=2sec) and the following was performed on both: Voxel timecourses were filtered, physiological noise was regressed out and subsequently, timecourses were averaged within ROIs, then normalized to baseline (two seconds before stimulus onset). Three inverse logit functions^3,4 were fitted to the mean timecourses using the Nelder-Mead Simplex method (Figure 1A,B) to obtain positive peak height, onset time, time-to-peak, full-width at half maximum (FWHM) and undershoot amplitude (Figure 1B). Repeated measures statistics, using a linear mixed model with ROIs as repeated factors, were performed independently on CAIPI and dsCAIPI results, followed by post-hoc analysis (significance level p<.05; FWE ).

The six different ROIs all clearly demonstrate an HRF (Figure 2). However, the variability of the HRF illustrated by the shaded band (i.e. mean standard deviation) is important especially for the cerebellar regions, and is also notably higher in the post-peak period than prior to stimulus onset. For the CAIPI data, significant differences (p<.05) between ROIs were found for peak height, onset time, FWHM and the undershoot amplitude (Figure 3). Cerebral regions showed higher peak amplitude compared to the cerebellar ones, with M1 highest overall. The undershoot amplitude in CV/CVIII and S2 was significantly smaller than in M1, and also somewhat smaller than in S1 and SMA. The absence of a post-stimulus undershoot in the cerebellum is also visible in the time courses (Figure 2). In terms of timing, the HRF started later for the cerebellum, especially CVIII, than most of the cerebral ROIs. Trial-to-trial variability (Figure 2) was highest in CVIII and lowest in SI. Concerning the dsCAIPI, a main effect was observed only for the height and the FWHM (p<.05) and no significant post-hoc differences were found.

Several significant differences in terms of amplitudes and timings were observed between different brain regions involved in motor control when using a TR=400ms 3D-EPI-CAIPI acquisition. This temporal resolution means trial-to-trial fluctuations can also be studied, showing different behavior even between S1 and M1. In contrast, the TR=2 of dsCAIPI did not provide enough information to obtain measurements allowing a good characterization of the HRF.

In the 400ms CAIPI data, the post-stimulus undershoot in cerebellar regions was consistently smaller and even absent in several subjects. The underlying cause could be a vascular phenomenon, such as different vascular compliance⁵ in the cerebellum, or a different post-stimulus neuronal modulation⁶ from the cerebrum. The lack of HRF studies targeting the cerebellum and its very different anatomy from the forebrain still leave both hypotheses open.

We conclude that the HRFs of six different brain regions exhibit between-region differences, such as a smaller cerebellar post-stimulus undershoot and high trial-to-trial stability in the primary sensory area.