Synopsis
Multi-band EPI is becoming a standard acquisition scheme for functional MRI. Here we will evaluate how multi-band acceleration benefits fMRI. We will address how accelerating to increase temporal resolution leads to de-aliasing of nuisance physiological signals and supports resolving complex BOLD activity. In accelerating to increase spatial resolution, we will consider temporal signal-to-noise ratio losses due to high resolution, and how multi-band compensates through imaging more volumes. We will assess multi-band multi-echo fMRI, which incorporates T2* relaxometric techniques for susceptibility artifact compensation and thermal noise reduction. Lastly, we will review the multi-band EPI configuration of the Human Connectome Project and clinical translatability of its multi-band approach.Overview and Acquisition
Multi-band EPI, in the blipped-CAIPI scheme,
addresses critical limitations to fMRI resolution and fidelity from
conventional EPI [1] by increasing imaging speed. In this educational talk we
will review methods, benefits, costs, and variations of the multi-band
approach. Multi-band EPI images multiple slices simultaneously by multiplexing
excitations across interspaced frequency bands corresponding to target slices.
The blipped approach adds alternating phase offsets to these slices from center
in either direction by some fraction of field of view (FOV). Leveraging coil
sensitivity maps, a multi-channel reconstruction with a GRAPPA or SENSE kernel
then yields well-separated slice images. This strategy leads to a factor-N
speed-up of volume acquisition [2]. Critically, partially overlapping slices are quite unique from each
other, k-space lines are not skipped, and thus coils richly sample separable signals. Resultantly,
reconstructed images are associated with low noise penalty (G-factor much lower
than factor-N), enabling high factor-N accelerations [3]. In contrast,
conventional fMRI in-plane acceleration of single-shot/slice/echo EPI, while
also using SENSE/GRAPPA reconstruction, accelerates by skipping alternate
k-space lines, and high-density receiver arrays sample signals of only one
slice. This in-plane acceleration leads to approximately $$$\sqrt{N}$$$ noise penalty for
less than factor-N speed-up, with N>2 leading to substantial losses. Thus,
on its own, multi-band represents a critical innovation for accelerating fMRI
and is a major advance over in-plane acceleration.
Temporal Resolution Considerations
We will review how increased temporal resolution
is a key benefit of multi-band EPI. Conventionally BOLD signal is considered low
temporal frequency, but greater temporal resolution is argued to better
dealias/disentangle overlapping signals from cardiac and respiratory sources
that are of higher frequency (than conventional EPI acquisition Nyquist) [4].
In the context of task fMRI however, the specific shape of hemodynamic response
is known to be informative of pathological or pharmacological states, which is
explained in higher frequency temporal components, and so faster sampling can
be directly beneficial for characterizing BOLD [5]. In the context of resting
state, with the increasing focus on dynamics [6], it is also important to
realize that the capture of non-stationary transitions (spikes, sharp waves,
etc.) also require high temporal frequencies for characterization, highlighting
another benefit for temporally accelerated BOLD imaging.
Spatial Resolution Considerations
We will review spatial resolution
considerations of multi-band EPI. For a fixed experiment duration, there are
fewer total readouts, and more time can be spent per readout (i.e. slice-set),
enabling higher spatial resolution. In this way multi-band EPI nominally
enables high spatial resolution for fMRI, but important considerations are
required on experiment design and analysis towards determining the optimal
increase in spatial resolution. This is because fMRI sensitivity, calculated as
temporal signal-to-noise ratio (tSNR), scales linearly with voxel size. As especially
relevant to SPM-style univariate analyses, multi-band EPI does not mitigate the
increase in thermal noise associated with higher resolution (by sampling signal
of a given voxel location less densely), and image smoothing to attenuate
thermal noise obviates high-resolution acquisition and penalizes false
discovery rate estimation. However, in multivariate decomposition, typified
by independent component analysis (ICA), tSNR's fixed-point sensitivity measure
is considered less relevant, and aggregate power tSNRx$$$\sqrt{N}$$$ for resolving signal
components by (sparsity-weighted) statistical contrast is increasingly
considered more relevant [7]. We will also consider how in the context of
population studies, anatomical variation of the neural correlates of function
sets an upper bound to the benefits of spatial resolution increases.
Multi-Echo Multi-Band fMRI
We will then review the more recent
development of multi-echo multi-band (MEMB) fMRI, and analysis with multi-echo
independent components analysis, in its own right and in relation to
single-echo multi-band EPI. Multi-echo fMRI enables relaxometric assessments of
BOLD signals, with benefits ranging from susceptibility artifact compensation,
to thermal noise reduction without smoothing, to non-BOLD component removal,
and time series output in dT2* units [8]. We will discuss how MEMB combines
these benefits with additional ones: specifically determining the statistical
dimensionality of total and BOLD spaces, isolation of thermal noise, and signal
denoising (including slice leakage, cardiac/respiratory, GRAPPA artifacts) without
template-matching algorithms such as FSL FIX, at 3T and 7T [9-11].
Human Connetome Project and Clinical Alternatives
Finally, we will review the multi-band
acquisition strategy of the Human Connectome Project (HCP). After summarizing
acquisition parameters and the current workflow, we will consider implications
of the wider NIH objective of harmonization of neuroimaging with HCP-like
protocols. In particular, it will be emphasized that HCP multi-band EPI
parameters are designed with a noise tolerance based on 1-hour resting state acquisition
[12]. In clinical or neurodevelopmental contexts, however, this strategy may be disagreeable, so alternative multi-band configurations will be
considered.
Acknowledgements
Dr. Kundu is supported by the Brain Imaging Center (BIC) and Translational and Molecular Imaging Institute at Mount Sinai. Dr. Lara Marcuse of the Mount Sinai Epilepsy Program, Dr. Priti Balchandani of the 7T Program are also acknowledged here. Dr. Ben Poser and Dr. Essa Yacoub and the Center for Magnetic Resonance Research (Minnesota) are acknowledge for their efforts on multi-echo multi-band imaging. References
[1]
Setsompop, K., B. A. Gagoski, J. Polimeni, T. Witzel, V. J. Wedeen, and L. L.
Wald. "Blipped CAIPIRHINA for simultaneous multi-slice EPI with reduced
g-factor penalty." In Proceedings of the 18th annual meeting of ISMRM,
Stockholm, Sweden, p. 551. 2010.
[2]
Mansfield, P. "Real-time echo-planar imaging by NMR." British
medical bulletin 40, no. 2 (1984): 187-190.
[3]
Setsompop, Kawin, Borjan A. Gagoski, Jonathan R. Polimeni, Thomas Witzel, Van
J. Wedeen, and Lawrence L. Wald. "Blipped-controlled aliasing in parallel imaging for simultaneous
multislice echo planar imaging with reduced g-factor penalty." Magnetic Resonance in Medicine 67,
no. 5 (2012): 1210-1224.
[4]
Xu, Junqian, Steen Moeller, Edward J. Auerbach, John Strupp, Stephen M. Smith,
David A. Feinberg, Essa Yacoub, and Kâmil Ugurbil. "Evaluation of slice
accelerations using multiband echo planar imaging at 3T." Neuroimage
83 (2013): 991-1001.
[5]
Luchtmann, M., K. Jachau, C. Tempelmann, and J. Bernarding. "Alcohol
induced region-dependent alterations of hemodynamic response: implications for
the statistical interpretation of pharmacological fMRI studies." Experimental
brain research 204, no. 1 (2010): 1-10.
[6]
Allen, Elena A., Eswar Damaraju, Sergey M. Plis, Erik B. Erhardt, Tom Eichele,
and Vince D. Calhoun. "Tracking whole-brain connectivity dynamics in the
resting state." Cerebral cortex (2012): bhs352.
[7]
Smith, Stephen M., Christian F. Beckmann, Jesper Andersson, Edward J. Auerbach,
Janine Bijsterbosch, Gwenaëlle Douaud, Eugene Duff et al. "Resting-state
fMRI in the human connectome project." Neuroimage 80 (2013):
144-168.
[8]
Kundu, Prantik, Souheil J. Inati, Jennifer W. Evans, Wen-Ming Luh, and Peter A.
Bandettini. "Differentiating BOLD and non-BOLD signals in fMRI time series
using multi-echo EPI." Neuroimage 60, no. 3 (2012): 1759-1770.
[9]
Olafsson, Valur, Prantik Kundu, Eric C. Wong, Peter A. Bandettini, and Thomas
T. Liu. "Enhanced identification of BOLD-like components with multi-echo
simultaneous multi-slice (MESMS) fMRI and multi-echo ICA." NeuroImage
112 (2015): 43-51.
[10]
Salimi-Khorshidi, Gholamreza, Gwenaëlle Douaud, Christian F. Beckmann, Matthew
F. Glasser, Ludovica Griffanti, and Stephen M. Smith. "Automatic denoising
of functional MRI data: combining independent component analysis and
hierarchical fusion of classifiers." Neuroimage 90 (2014): 449-468.
[11]
Boyacioglu, Rasim, Jenni Schulz, Nils CJ Müller, Peter J. Koopmans, Markus
Barth, and David G. Norris. "Whole brain, high resolution multiband
spin-echo EPI fMRI at 7T: a comparison with gradient-echo EPI using a
color-word Stroop task." Neuroimage 97 (2014): 142-150.
[12]
Van Essen, David C., Kamil Ugurbil, E. Auerbach, D. Barch, T. E. J. Behrens, R.
Bucholz, A. Chang et al. "The Human Connectome Project: a data acquisition
perspective." Neuroimage 62, no. 4 (2012): 2222-2231.