Susceptibility distortion correction for fMRI
Gina Joue1, Tobias Sommer1, and Siawoosh Mohammadi1

1Department of Systems Neuroscience, University Medical Center Hamburg-Eppendorf, Hamburg, Germany


Multiband echo planar imaging (EPI) offers increased temporospatial resolution and statistical power for functional magnetic resonance imaging (fMRI) but the higher spatial resolution comes at the cost of higher susceptibility-related spatial distortions. In diffusion MRI (dMRI), studies have shown that distortion correction is better when using blip-reversed EPI data (known under the term blip-up/down images) as compared to the standard fieldmap approach. This has motivated fMRI studies to acquire their data with blip-up/down directions and to use these to reduce susceptibility distortion. Here, we qualitatively illustrate why this can lead to erroneous results and quantify this error across 10 subjects.


Multiband echo planar imaging (MB-EPI) is a recently developed protocol used in functional magnetic resonance imaging (fMRI)1 that offers increased temporal/spatial resolution with increased statistical power2,3 but the higher spatial resolution comes at a price of considerable spatial distortion in the phase-encoding direction caused by susceptibility-related field inhomogeneities. With the rising popularity of multiband imaging methods and increasingly short TRs, susceptibility-induced distortions (SD) are becoming a common issue for fMRI that still needs to be addressed. Diffusion MRI (dMRI) studies suggest that SD is better corrected when using two images with reversed phase-encoding directions (blip-up/down images) compared to acquiring a field map4,5, as is standard for fMRI. However, when extending the blip-up/down approach to fMRI by acquiring one additional gradient-echo (GE) EPI scan in a phase-encoding direction opposite to the standard functional images and combining the data to correct SD in the GE/fMRI data, an error can be introduced as illustrated in Figs.2 and 3. Since, this problem has frequently been reappearing in prominent neuroimaging software forums (e.g. FSL or SPM), we decided to quantify the induced error when using GE-EPI as compared to SE-EPI for blip-up/down correction.


Pairs of SE- and GE-EPIs with opposite phase-encoding directions were acquired on a Siemens Magnetom Prisma 3T scanner from 10 healthy participants (aged 19–29 years, 5 females), with a multiband acceleration factor of 4 and no in-plane acceleration, resulting in substantial distortions. Phase was encoded anterior-to-posterior. Slice resolution was isotropic 2-mm in-plane. GE-EPIs had TE=40ms, TR=1.54s, and flip angle=60$$$^\circ$$$. For each phase-encoding direction, ten volumes of GE-EPIs were acquired and averaged before analyses in order to smooth over signal dropouts. SE-EPIs had TE=200ms, TR=5s, flip angle=90$$$^\circ$$$, and refocus flip angle=180$$$^\circ$$$. Single volumes of SE-EPIs were acquired. T1-weighted MPRAGE structural images were also acquired with isotropic 1-mm voxels in-plane; no slice gap in coronal slices, TE=2.98ms, TR=2.3s, and flip angle=9$$$^\circ$$$. We estimated the field inhomogeneities using two open-source tools: the Computational Morphometry Toolkit6 and HySCOv.27. HySCO is part of the ACID toolbox integrated into the commonly used fMRI analysis software SPM8. CMTK uses nonrigid transformations generated by multilevel B-splines, whereas HySCO adopts a diffeomorphic transformation algorithm. To normalize image intensity fluctuations due to signal drop-outs in GE-EPIs, the averaged GE-EPIs were first bias-corrected before estimating the deformation field using HySCO or CMTK.Corrections were performed on the SE-EPIs, rather than the GE-EPIs, in order to evaluate corrections unconfounded by signal dropouts. To facilitate evaluation, the average of the corrected/unwarped blip-up/down of SE-EPIs were coregistered to the individual’s MPRAGE, and the transformation applied to all other images. Ideally, corrections should result in the blip-up/down SE-EPIs being more similar to each other and to the MPRAGE, here used as the anatomical reference. We quantified goodness in terms of the differences in the root mean squares (ΔRMS) of image intensity (1) between the corrected blip-up/down SE-EPIs, and (2) the corrected EPIs vs. MPRAGE, where the lower the ΔRMS, the more similar the images are and hence the better the correction. We also statistically tested ΔRMS differences between SE-EPI- vs. GE-EPI-based corrections with paired Wilcoxon signed rank tests, where α<0.017 indicates significant differences, corresponding to a (Bonferonni) corrected threshold of α<0.05.


Blip-up/down SE-corrected EPIs visually differed less from the MPRAGE (Figs.2-3). The lower ΔRMS between blip-up/down and MPRAGE when corrected using SE-EPIs vs. GE-EPIs (dashed blue/dotted green lines, respectively, in Fig.4) confirmed this, though statistically there were no differences in images corrected by CMTK. GE-corrected pairs were overall less different than SE-corrected pairs. ΔRMS between blip-up/down SE-EPIs were also more similar for SE-corrected images than GE-corrected images for HySCO but not for CMTK (see Fig.4 for details).


Individual visual inspection of air-tissue boundaries and between tissue types of corrected SE-EPIs compared to MPRAGE as an anatomical reference showed that field inhomogeneities estimated from blip-up/blip-down SE-EPI pairs lead to better spatial correction overall than when estimated from GE-EPIs. This is confirmed when quantitatively comparing the intensity differences (ΔRMS) between corrected blip-up and blip-down images and between corrected images and MPRAGE.


In order to correct for susceptibility-induced distortion correction in fMRI (GE-EPI), a pair of additional SE-EPIs should be acquired with the same sequence parameters in opposite phase-encoding directions. The additional ca. 1-minute scan time results in better SD corrections.


This research was partially funded by the German Research Foundation (DFG-Antrag SO 952/8-1). SM received funding from the European Union’s Horizon 2020 research and 654 innovation program under the Marie Sklodowska-Curie grant agreement No 658589, and from the BMBF (01EW1711A and B) in the framework of 656 ERA-NET NEURON.


1Feinberg DA, Setsompop, K. Ultra-fast MRI of the human brain with simultaneousmulti-slice imaging. J. Mar. Res. 2013; 229: 90–100.

2Chen L, Vu AT, Xu J, Moeller S, Ugurbil K, Yacoub E, Feinberg D. Evaluationof highly accelerated simultaneous multi-slice EPI for fMRI. NeuroImage. 2015; 104: 452–459.

3Todd N, Moeller S, Auerbach EJ, Yacoub E, Flandin G, Weiskopf N. Evaluationof 2D multiband EPI imaging for high-resolution, whole-brain, task-based fMRI studies at 3T: Sensitivity and slice leakage artifacts. NeuroImage. 2016; 124: 32–42.

4Esteban O, Daducci A, Caruyer E, O'Brien K, Ledesma-Carbayo MJ, Bach-Cuadra M, Santos A. Simulation-based evaluation of susceptibility distortion correction methods in diffusion MRI for connectivity analysis. In 2014 IEEE 11th International Symposium on Biomedical Imaging (ISBI), April 29 2014-May 2 2014. 2014; 738–741.

5Graham MS, Drobnjak I, Jenkinson M, and Zhang H. Quantitative assessment ofthe susceptibility artefact and its interaction with motion in diffusion MRI. PLoS ONE1. 2017; 2:e0185647.

6Rohlfing T. Image similarity and tissue overlaps as surrogates for image registration accuracy: widely used but unreliable. IEEE Trans Med Imaging. 2012; 31: 153–163.

7Ruthotto L, Kugel H, Olesch J, Fischer B, Modersitzki J, Burger M, Wolters CH. Diffeomorphic susceptibility artefact correction of diffusion-weighted magnetic resonanceimages. Phys Med Biol. 2012; 57: 5715–5731.

8Friston KJ, Ashburner JT, Kiebel S, Nichols T, Penny WD, eds. Statistical Parametric Mapping: The Analysis of Functional Brain Images. Oxford: Elsevier; 2007.


Illustration of what can go wrong, when using GE blip-up (↑) and down (↓) images to reduce susceptibility-induced distortions. The cross-hair highlights a region with high spatial distortions, where distortion-induced signal variation in the SE-EPIs (right column) are only due to spatial compression and stretching (i.e. can be described by the Jacobian of the deformation field), but in the GE-EPIs, spatial distortion is accompanied by signal dropout. Consequently, incorrect regions of the blip-up/down images are registered to each other, namely those areas adjacent to the signal dropout. The T1-weighted MPRAGE (top row, left panel) is shown for anatomical reference.

Edges of the MPRAGE overlaid on the corrected SE-EPIs of an individual to illustrate that images corrected from field inhomogeneity estimations based on GE-EPIs (middle) differed more from MPRAGE (left and overlaid red contours) than when based on SE-EPIs (right). This can be seen in the curvature and borders of the lateral ventricles. Blip-up (↑) and blip-down (↓) images are shown separately.

Edges of white matter segmented from the MPRAGE overlaid on white matter segmentedfrom corrected SE-EPIs to illustrate unwarping success for white matter (and hence gray matter) tissue types for an individual. Gyrations are relatively maintained in medial regions, but as can be seen in the shape of the brain stem, images corrected from distortion estimations based on GE-EPIs (middle) differed more from MPRAGE (left and overlaid red contours) than when based on SE-EPIs (right). Blip-up (↑) and blip-down (↓) images are shown separately.

Differences between corrected blip-up (↓) and blip-down (↑) SE-EPIs (solid red lines) and between corrected SE-EPIs and MPRAGE (dotted green and dashed blue lines), when estimated from GE-EPIs or SE-EPIs. Large RMS differences indicate worse correction. SE-corrected is statistically better than GE-corrected for HySCO, whether blip-up/down pairs vs. MPRAGE(upper right; V=55, p<0.0019 for blip-down, dotted green lines; V=47.5, p<0.046 for blip-up,dashed blue lines), or between blip-up/-down pairs themselves (lower right; V=55, p<0.0059).These were not statistically significant for CMTK (left; all p’s>0.03). Points mark the mean across 10 participants. Error bars mark standard deviations of the differences.

Proc. Intl. Soc. Mag. Reson. Med. 26 (2018)