Tradeoffs in pushing the spatial resolution of fMRI for the 7T Human Connectome Project

Abstract

Whole-brain functional magnetic resonance imaging (fMRI), in conjunction with multiband acceleration, has played an important role in mapping the functional connectivity throughout the entire brain with both high temporal and spatial resolution. Ultrahigh magnetic field strengths (7T and above) allow functional imaging with even higher functional contrast-to-noise ratios for improved spatial resolution and specificity compared to traditional field strengths (1.5T and 3T). High-resolution 7T fMRI, however, has primarily been constrained to smaller brain regions given the amount of time it takes to acquire the number of slices necessary for high resolution whole brain imaging. Here we evaluate a range of whole-brain high-resolution resting state fMRI protocols (0.9, 1.25, 1.5, 1.6 and 2mm isotropic voxels) at 7T, obtained with both in-plane and slice acceleration parallel imaging techniques to maintain the temporal resolution and brain coverage typically acquired at 3T. Using the processing pipeline developed by the Human Connectome Project, we demonstrate that high resolution images acquired at 7T provide increased functional contrast to noise ratios with significantly less partial volume effects and more distinct spatial features, potentially allowing for robust individual subject parcellations and descriptions of fine-scaled patterns, such as visuotopic organization.

Keywords: 7T; Connectome; High-resolution; Resting state; Whole brain; fMRI.

Figure 1

Representative GE EPI cross sections from a single subject. A) Axial slices. Top row: 3 T 2 mm isotropic (left), 7 T 1.6 mm isotropic (middle) and 7 T 1.25 mm isotropic (right). Bottom row: 7 T 2 mm isotropic (left), 7 T 1.5 mm isotropic (middle), 7 T 0.9 mm isotropic (right). B) Sagittal slices. Formatting same as A. Note the striping artifact in the brain stem and cerebellum regions of the 1.5 mm and higher data (red arrow). These are indicative of the enhanced motion sensitivity in the iPAT3 reference scans.

Figure 1

Representative GE EPI cross sections from a single subject. A) Axial slices. Top row: 3 T 2 mm isotropic (left), 7 T 1.6 mm isotropic (middle) and 7 T 1.25 mm isotropic (right). Bottom row: 7 T 2 mm isotropic (left), 7 T 1.5 mm isotropic (middle), 7 T 0.9 mm isotropic (right). B) Sagittal slices. Formatting same as A. Note the striping artifact in the brain stem and cerebellum regions of the 1.5 mm and higher data (red arrow). These are indicative of the enhanced motion sensitivity in the iPAT3 reference scans.

Figure 2

Representative seed based connectivity map from a single subject for both the 3 T and 7 T HCP protocols. The seed voxel is placed in primary visual cortex (denoted by the crosshairs). The 7 T data results in similar if not stronger correlation values than the 3 T data, even though the voxel volume is roughly half that of the 3T data. Benefiting from the reduced partial voluming, the 7 T data reveals broader connectivity maps, which extends into sulcal regions where the cortex is relatively thin. Note: to control for the different number of timepoints, only the top 50 PCA components were used when calculating the correlation values for both 3 T and 7 T.

Figure 3

Cross-sulcal to within-sulcal correlation ratio as a function of field strength and resolution. A) Box plots calculated across the four resting state scans per resolution for pilot subject 1. As expected, the metric decreases from 2 mm to 1.5 mm. Improvements beyond 1.5 mm may in part be limited due to blurring in the phase encoding direction associated with increased PF and echo train lengths required for higher resolution acquisitions. B) Same as A, but for pilot subject 2. C) Comparison of correlation ratio metric between 3 T and 7 T HCP protocols averaged across 24 HCP subjects. Error bars are STDEV across subjects.

Figure 4

Number of resting state signal components as a function of field strength and resolution. Reducing the iPAT factor to 2, was most efficient in terms of TR and SNR per unit time which translated to larger number of detectable signal components. While retaining optimal TE and echo train blurring a resolution of 1.6 mm was attainable.

Figure 5

Resting state component composition: 3 T vs 7 T, on average across 24 subjects. Left) Total number of ICA components (7 T > 3 T; p < 0.05). Middle) Number of noise ICA components. Right) Number of BOLD signal ICA components (7 T > 3 T; p < 0.05). Errorbars are STDEV across subjects.

Figure 6

Resting state variance breakdown: 3 T vs 7 T, on average across 24 subjects. Left) Unstructured Noise divided by original variance; where unstructured noise is the residual variance remaining not contributed by the ICA components classified as either “noise” or “signal” (7 T < 3 T; p < 0.05). Middle) BOLD variance divided by original variance; where BOLD variance is the variance contributed by the ICA components classified as “signal” (7 T > 3 T; p < 0.05). Right) fCNR as defined as square root of the BOLD variance divided by unstructured noise variance (7 T > 3 T; p < 0.05). Dashed red line indicated fCNR after compensating for difference in number of timepoints sampled in the 7 T and 3 T HCP protocols. Errorbars are STDEV across subjects.

Figure 7

Resting state fCNR: 7 T vs 3 T, on average across 24 subjects. The results are displayed in CIFTI grayordinate space and show the advantage of the 7 T protocol across the brain.

Figure 8

Exemplary dense Connectome for a seed placed in posterior parietal cortex. The 7 T data shows enhanced connectivity/correlation throughout the brain.

Figure 9

Exemplary dense Connectome for a seed placed in the subcortical nucleus, pulvinar (red arrow). Due to the improvement in partial voluming, the 7 T data is able to show connectivity/correlation throughout the brain while the 3T data is not.

Figure 10

Dense Connectome connectivity gradients. The 7 T data show stronger, more well defined gradients.