Zusammenfassung
Introduction:
The BOLD signal arises predominately from susceptibility changes in and around the vessels. This effect varies with each vessel's orientation relative to the B0 axis. The cortical vasculature is structured with large pial veins lying tangentially on the cortical surface; intracortical veins perpendicular to the surface, and randomly oriented capillaries(Fig. 1a). Together this imparts a cortical orientation dependence on the signal. Since large veins dominate the gradient-echo (GRE) BOLD signal, its amplitude is maximized when pials are perpendicular to B0, i.e. when the cortical surface normal is parallel to B0, as demonstrated in a 3T hypercapnia study and in simulations[1,2]. The effect should increase with higher resolution and field strength. Here, we assess the effect in resting-state (rs) BOLD fMRI tSNR at 3T in the Human Connectome Project data and at 7T using our own data. We also test if orientation can explain tSNR bias and variance in population studies, using a mean cortical orientation atlas and an orientation variability atlas. We compare mean tSNR and tSNR variability across subjects with these two atlases, and consider these measures in 7 major functional connectivity (FC) networks[3].
Methods:
5 subjects were scanned on a 7T scanner (Magnetom, Siemens, Germany). Each scan comprised 6 rs-fMRI runs with GRE BOLD SMS-EPI: 1.1mm isotropic (iso) resolution,TR/TE=1700/26ms,SMS/GRAPPA=3/4,matrix=174×174×87,160 volumes; and a T1-w MEMPRAGE (0.75mm iso resolution). FreeSurfer was used to generate cortical surfaces from the T1-w image, including the pial (depth=1), the white (depth=0) and intracortical surfaces at 0.1 depth increments for cortical depth analyses(Fig. 1a/b). Surfaces were then registered to the subject's head position during the EPI runs and each mesh vertex' orientation angle θB0 between the surface normal and the B0 axis (Fig.1b) was calculated. Each rs-fMRI voxel was assigned the angle of the vertex closest to its centroid and tSNR was computed from the timeseries. To study orientation effects in a modern rs-fMRI dataset, we selected at random 118 subjects from the 3T HCP cohort[4]. rs-fMRI (4 runs of GRE BOLD SMS-EPI: 2mm iso resolution,TR/TE=720/33ms,SMS=8) and the T1-w MPRAGE were downloaded. The HCP minimal pre-processing pipeline[5] was adapted to avoid MNI registration and keep the data in the subject's native head position. Cortical surface reconstruction, orientation and tSNR calculation was performed as above. Mean orientation and orientation variability was calculated in fsaverage with circular statistics.
Results:
tSNR differed up to +35% between parallel and perpendicular orientation at the pial surface in the 7T data(Fig.1c). The bias declines with cortical depth, but never vanishes, implying a pial vein contribution on the signal at all depths. The orientation effect is also present in the 3T HCP data with a maximum tSNR bias of about 20%(Fig.1d). Fig. 2 shows the mean orientation and orientation variability over 118 subjects. Certain areas exhibit more consistent orientation across the population than others(Fig. 2a) and tSNR variability across subjects σ(tSNR) increases with orientation variability by up to 30%(Fig.2b). FC networks with higher orientation angle tend to higher tSNR(Fig.2c) and networks with higher orientation variability tend to higher tSNR variance.
Conclusions:
The orientation bias is apparent in high-resolution 7T and moderate-resolution 3T data. The effect induces within-subject tSNR bias: tSNR is higher for cortex with a surface normal which is more perpendicular to B0. tSNR across subjects is more variable in areas with more orientation variability.
Besides being a nuisance, orientation may help to distinguish vascular contributions in fMRI techniques since the signal bias is expected to be attributable to pial veins. The orientation effect may characterize the microvascular contribution in spin-echo fMRI that is more specific to the randomly oriented capillaries.
