Characterization of ultra-high resolution Gradient Echo and Spin Echo BOLD fMRI in the human visual cortex at 7 Tesla
Catarina Rua1,2, Mauro Costagli2,3, Mark R Symms4, Laura Biagi3, Mirco Cosottini2,5, Alberto Del Guerra1, and Michela Tosetti2,3

1Department of Physics, University of Pisa, Pisa, Italy, 2Imago7 Research Center, Pisa, Italy, 3IRCCS Stella Maris, Pisa, Italy, 4GE Healthcare, Pisa, Italy, 5Department of Translational Research and New Technologies in Medicine and Surgery, University of Pisa, Pisa, Italy


This study compared GRE-EPI and SE-EPI sequences with different spatial resolutions at 7T for fMRI in the visual cortex. It is demonstrated that SE-EPI yields higher specificity than GRE-EPI. However, the decreasing temporal signal-to-noise at submillimeter acquisitions affected significantly the extension of the activated volume in a SE acquisition. At this level, GRE-based functional maps showed significant increased specificity compared to standard resolutions, and a preserved cluster volume. The reduction of partial volume effects allowed the selection of an activation sub-cluster excluding the highest z-scores, which co-localized preferably in non-gray matter, potentially increasing the performance of UHF high-resolution fMRI.


High field MRI allows the acquisition of ultra-high resolution functional images of the human brain using the blood oxygenation level dependent (BOLD) contrast1. Debate exists whether studies should still rely on the higher sensitivity to signal changes of Gradient Echo (GRE) acquisitions or instead use T2-weighted Spin Echo (SE) sequences which, with higher intrinsic SNR given by the larger B0, could potentially show greater advantages in correctly identifying the site of activation2-7.


Given that the characteristics of the BOLD signal and noise vary non-linearly with resolution8,9, in this study a quantitative evaluation of the sequences is made, comparing acquisitions at sub-millimeter to conventional resolutions at 7T. Signal dynamics, extent, and spatial distribution of the functional statistical maps are measured in the early visual cortex.


Four healthy volunteers were scanned on a 7T MRI scanner (GE Healthcare, USA) with optimized GRE and SE-EPI sequences for targeted fMRI. Ten slices were prescribed obliquely, parallel to the calcarine sulcus and covering most of the visual cortex, including V1 and V5. All fMRI scans used a 2-shot EPI approach with effective TR=3s (shot-TR=1.5s), rBW=250kHz, Fat Suppression, ASSET factor=2, phase encoding=Anterior-to-Posterior, slice thickness=1.4mm, slice spacing=0.1mm. In SE-EPI, TE was set to 45ms and in GRE-EPI it was set to 23ms, with FA=62º. SE and GRE scans were acquired at three different in-plane resolutions: 1.5x1.5mm2, 1.0x1.0xmm2 and 0.75x0.75mm2. The functional task consisted of a dynamic display of coherent motion (15s blocks of dots moving coherently, followed by 15s blocks of gray screen) with total duration of 3'00'' per scan. 12 of dummy scans were added at the start to achieve steady state. At each resolution, single-volume GRE and SE-EPIs with whole brain coverage were acquired with identical parameters as the functional scans but with effective TR=8s to allow an almost complete coverage of the brain (~50 slices). A standard T1-weighted sequence was also acquired for anatomical co-registration of V1 and V5 masks from MNI space. Automated segmentation of gray matter (GM) was achieved with FAST (FSL v5.0.5, applied to the long-TR GRE-EPI scans; tissue not included in the GM mask was labeled as "non-GM".

Functional data were motion-corrected in AFNI and high-pass temporal filtered with Hcut-off=60s and pre-whitened using FEAT. A general linear model analysis was used to extract statistically significant signal changes elicited by the visual stimulus. The output z-score maps were cluster thresholded at z>2.3. These maps, designated "z-full",were further upper-thresholded at different percentages (10%, 30%, 50%, 70%, and 90%) to create 5 additional z-maps: voxels above the z-score percent threshold were set to zero. Average values of temporal SNR, percent signal change (ΔS/S) and cluster volume were extracted from the z-statistical maps in GM of V1 and V5. Specificity of the activation was defined as: Spec=TN/(TN+FP), where TN (true negatives) = number of non-active voxels in non-GM and FP (false positives) = number of active voxels in non-GM10.


At all resolutions tSNR in GRE was higher (47±13)% than in SE, and this difference was statistically significant (χ2=9.72; p-value=0.0015), while the percent signal change was similar in SE and GRE (χ2=1.33; p-value=0.25). For GRE, in V1 we observed lower ΔS/S in lower z-score voxels (purple/blue curves in Figure 1) but this effect diminished in V5. As expected the cluster size was bigger in GRE than in SE. However, only in the latter we observed a significant decrease in activation volume when acquiring at small voxel sizes (Figure 2). Also, the contribution of lower-z voxels to the active volume in GRE was higher in V5 than in V1. Specificity was consistently higher in SE; however, specificity in GRE increased significantly with increasing resolution, and increased in the lower z-scored voxels (Figure 3).


At 7T, SE BOLD signal at standard spatial resolutions reflected meaningful activations and had high spatial specificity to active microvasculature. Nonetheless, the decreasing temporal signal-to-noise ratio at high spatial resolutions reduced the extent of the detected activation. This occurred to a lesser degree in GRE where, at submillimeter acquisitions, the reduced partial volume effects, the increased specificity and ΔS/S allowed the differentiation between very high percent signal changes that corresponded mostly to non-grey matter regions (vasculature), and smaller yet statistically significant activations located predominantly in the gray-matter.


We conclude that properly masked GRE-based functional maps should be preferred for fMRI applications at high field using high spatial resolutions.


This work was supported by the Initial Training Network, HiMR, funded by the FP7 Marie Curie Actions of the European Commission (FP7-PEOPLE-2012-ITN-316716).


1. Van der Zwaag W, Francis S, Head K, et al. fMRI at 1.5, 3 and 7 T: Characterizing BOLD signal changes. NeuroImage. 2005;47:1425-1434.

2. Norris D G. Spin-echo fMRI: the poor relation? NeuroImage. 2012;62:1109–1115.

3. Pflugfelder D, Vahedipour K, Uludag K, et al. On the numerically predicted spatial BOLD fMRI specificity for spin echo sequences. Magn. Reson. Imaging. 2011;29:1195-1204.

4. Budde J, Shajan G, Zaitsev M, et al. Functional MRI in human subjects with gradient-echo and spin-echo EPI at 9.4 T. Magn. Reson. Med. 2014;71:209–218.

5. Boyacioglu R, Schulz J, Mueller N, et al. Whole brain, high resolution multiband spin-echo EPI fMRI at 7 T: a comparison with gradient-echo EPI using a color-word stroop task. NeuroImage 2004;97:142–150.

6. Sanchez-Panchuelo R, Schluppeck D, Harmer J, et al. Assessing the spatial precision of SE and GE-BOLD contrast at 7 Tesla. Brain Topogr. 2015;28:62-65.

7. Chiacchiaretta P, Romani G L, Ferretti A. Sensitivity of BOLD response to increasing visual contrast: Spin echo and gradient Echo EPI. NeuroImage. 2013;82:35-43.

8. Triantafyllou C, Polimeni J R, Elschot M, Wald L L. Physiological noise in gradient echo and spin echo EPI at 3 T and 7 T. Proc. Intl. Soc. Mag. Reson. Med. 2009;122.

9. Yacoub E, Van De Moortele P F, Shmuel A, Ugurbil K. Signal and noise characteristics of Hahn SE and GE BOLD fMRI at 7 T in humans. NeuroImage. 2005;24:738–750.

10. Klauschen F, Goldman A, Barra V, Meyer-Lindenberg A, Lundervold A. Evaluation of automated brain MR image segmentation and volumetry methods. Hum. Brain Mapp. 2009;30:1310-1327.


Figure 1: Quantitative measurements of the percent signal change (ΔS/S) as a function of voxel volume in significantly active voxels in V1 and V5 (gray matter) for GRE and SE. The color-coded curves correspond to the ROIs obtained from the regional upper-thresholded maps.

Figure 2: Active volume as a function of voxel size (mm3): (A) plot of "z-full" activated volume for GRE and SE in V1 and V5; (B) Contribution of z-score sub-levels to the total active volume as a function of voxel size (normalized to the total volume of the 3.6mm3 acquisition).

Figure 3: Specificity of GRE and SE as a function of voxel size in V1 and V5. The color-coded bars represent measurements extracted from the regional upper-thresholded maps in V1 or V5 accordingly. Inter-subject variability is displayed with error-bars on the "z-full" measurement.

Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)