User Manual Privacy Policy Disclaimer Contact us
  Advanced SearchBrowse





Partial Fourier imaging anisotropically reduces spatial independence of BOLD signal time courses

There are no MPG-Authors available

(Any fulltext)

Fulltext (public)
There are no public fulltexts available
Supplementary Material (public)
There is no public supplementary material available

Zaretskaya, N., & Polimeni, J. (2016). Partial Fourier imaging anisotropically reduces spatial independence of BOLD signal time courses. Poster presented at 22nd Annual Meeting of the Organization for Human Brain Mapping (OHBM 2016), Geneva, Switzerland.

Cite as: http://hdl.handle.net/21.11116/0000-0000-7B6A-0
Introduction: Partial Fourier (pF) imaging is a common method used to accelerate image acquisition or shorten the minimum TE in 2D EPI. The technique takes advantage of conjugate symmetry in k-space, which allows acquiring a subset (at least half) of k-space and estimates the skipped data from the acquired data [1], [2]. Despite its wide use for BOLD fMRI and the theoretically predicted blurring effects, the actual impact of pF on spatial independence of the BOLD signal time courses-and hence on the spatial resolution of fMRI experiments-is not known. In this study we quantify the effect of pF on temporal correlation of resting-state BOLD signal time courses along each of the three image encoding dimensions. We find that with increasing pF spatial correlations between neighboring voxels increase anisotropically, with most blurring occurring along the phase-encoding direction. Methods: Resting-state BOLD fMRI data were acquired from four healthy volunteers on a Siemens TIM Trio 3T MR scanner using a 32-channel head coil and either a GE (TR=3.3 s, TE=30 ms, FA=90°, 2.5 mm3 resolution, 48 axial slices with A–P phase encoding, 120 volumes per measurement) or SE (same parameters as GE but 30 slices and TE=70ms) single-shot 2D EPI sequence. The partial Fourier factor was varied from run to run between 8/8 (i.e. no pF) to 5/8 in a pseudorandom order to avoid confounding effects of fatigue, motion, etc. EPI data were reconstructed using zero-padding algorithm [3]. After standard preprocessing we employed an analysis procedure illustrated in Fig. 1. This procedure generated three maps representing the amount of local temporal correlation along phase-encoding, readout, and slice dimensions. The resulting values were then averaged across all voxels within the brain to yield one global value per run. To rule out effects of brain tissue asymmetry or physiological noise correlations along different dimensions we conducted separate measurements where the EPI matrix was rotated by ±30° relative to the A–P axis. To exclude the effect of a specific image reconstruction algorithm used, we compared the results obtained using default zero-filling reconstruction with those obtained using the POCS algorithm [4]. Results: We observed the expected tSNR increase with increasing pF due to implicit spatial smoothing caused by omitting the high spatial frequencies and interpolation during image reconstruction (Fig. 2). The only exception was observed for pF 5/8 in GE sequence, presumably because the signal loss due to strong dephasing (as the echo becomes closer to the edge of the acquisition window) outweighs the smoothing effect. Importantly, we also observed an anisotropic increase in temporal signal correlation in neighboring voxels, with strongest increase occurring along the phase-encoding dimension (again with an exception of pF 5/8; see Fig. 3A). The observed effect of pF is similar to applying a 1D Gaussian smoothing kernel along the same dimension (Fig. 3B), where pF 6/8 is roughly equivalent to a smoothing of 1.1 mm FWHM (as determined by linear interpolation of the results in Fig. 3B). Additional control experiments show that the correlation anisotropy is unlikely to be due to differences in tissue or noise properties along different dimensions (Fig. 4A) or the specific reconstruction algorithm used (4B). Conclusions: Using partial Fourier for fMRI introduces an asymmetric loss of resolution along the phase-encoding dimension comparable to the effect of smoothing single volumes with a 1D Gaussian kernel during standard preprocessing. This can not only cause bias in high-resolution fMRI analyses where explicit smoothing is avoided [5], but also can complicate interpretation of resting-state functional connectivity including promising new measures of local connectivity anisotropy [6]. This effect is expected to increase with field strength since increases in B0 inhomogeneity cause assumptions of conjugate symmetry and smoothly-varying phase in the image domain to be less justified.