Multi-echo fMRI: A review of applications in fMRI denoising and analysis of BOLD signals
Introduction
Blood oxygenation level dependent (BOLD) functional MRI (fMRI) is widely used to study brain activity based on hemodynamic signals (Bandettini et al., 1992, Bullmore et al., 1996, Buxton et al., 1998, Friston et al., 1995, Ogawa et al., 1990). However, recent studies show that fMRI data can be severely affected by artifacts (Power et al., 2012). These artifacts relate to subject head motion, cardiac and respiratory effects, and hardware (Glover et al., 2000, Jo et al., 2010). Studies on the effects of fMRI artifacts have brought into question many of the compelling findings on brain function based on fMRI, for example, relating to human brain development (Fair et al., 2009). fMRI artifacts also reduce the statistical power of fMRI studies and lead to spurious findings, which has been associated with a crisis of confidence in fMRI research (Button et al., 2013). Thus, while in recent years the field of fMRI research has both enjoyed advanced technology and expanded use, it also deals with a deep discomfort related to many known and unknown effects of artifact. In this review, we discuss an emerging fMRI approach, called multi-echo (ME)-fMRI, which focuses on improving the fidelity of fMRI signals through a physically-driven determination of the origins of fMRI signals as arising from either BOLD contrast or artifact. We also discuss how ME methods can be combined with emerging multi-band acceleration methods to create fMRI strategies with both high resolution and fidelity. Altogether, it is shown that ME-fMRI involves a few relatively small changes from standard fMRI acquisition technique, but enables analyses that support major improvements in fMRI data quality.
To date, the challenges in controlling fMRI artifacts have been met with generic time series signal processing methods such as regression and frequency-restricting bandpass filters (Satterthwaite et al., 2012). Head motion artifacts, modeled from shifts in brain images over time, are regressed out of fMRI time series. Recordings of physiology during MRI are used to model cardiac and respiratory artifact regressors (Glover et al., 2000). Spontaneous activity is band pass filtered to retain a narrow range of frequencies in order to exclude hardware related signal drifts and high-frequency noise (Carp, 2013). Time series models of noise and usually their temporal derivatives are linearly regressed out of data. Despite the significant reduction in information in data after applying these steps, it is now clear that much artifact remains. Of late, simple deletion of volumes from fMRI datasets has been suggested (Power et al., 2012). This altogether means that artifact signals are not characterized well enough by modeling artifact time courses and regressing them out of data. This reality also suggests that artifacts are themselves various and complex, and may interact in unpredictable ways. At the heart of the issue, however, is that standard fMRI approaches do not have a strong and general ground truth to precisely relate signals to biophysical signal mechanisms versus artifacts. Given information on how signals scale across the echo images of an ME-fMRI experiment, however, valuable insight on the origins of fMRI signals in BOLD contrast or artifact can be gained.
Section snippets
ME-fMRI and fMRI relaxometry
After excitation, standard fMRI uses 2-D echo planar imaging to acquire slice images at a single TE, one slice at a time. At 3 T, this TE is usually 30 ms. ME-fMRI uses a slightly different approach. After a normal excitation pulse, a slice image is acquired at the earliest TE possible. Without exciting again, readout of another image of the same slice is then acquired immediately afterward, at a longer TE, and so forth up to the desired number of images and TEs. This happens for each slice of
TE-dependence and TE-independence of signal changes in ME-fMRI data
Managing the contribution of various artifacts such as from subject head motion to fMRI time series is the central problem of fMRI denoising and analysis. When using single-echo fMRI, the challenge is in distinguishing neurally-related activity from artifacts. However, the scaling of fMRI signals across TEs, based on ME-fMRI data, can be analyzed to infer origins in neurally-related versus artifactual signal processes.
The amplitude of a signal fluctuation from standard single-echo fMRI data is
Classifying statistical components of fMRI datasets as problem and opportunity
TE-dependence and TE-independence models can be applied to classify signal components of fMRI that are found using spatial independent component analysis (ICA) (Kundu et al., 2012). ICA is a powerful method that can factor fMRI image time series into a set of linearly separable component maps and their corresponding time courses, expressed aswhere X is a voxel-by-time matrix of time series, A is a voxel-by-component matrix expressing spatial maps, and W is a
A general model for fMRI time series
ME-ICA showed how brain network components could be grouped apart from artifacts of many kinds without spatial or temporal templates of expected functional effects. While ME-ICA was shown first for human resting-state fMRI data from 3 T MRI, it is in principle more broadly applicable. To denoise fMRI, TE-independent component time series could be regressed out of fMRI time series (McKeown et al., 2005). This approach can be used to denoise ME-fMRI of both task and resting-state activity since
Motion artifacts
While motion artifacts have long been known to affect fMRI, only recently has the severity of their impact on studies of connectivity become widely accepted (Power et al., 2012, Satterthwaite et al., 2012). There remains a vital need for an fMRI denoising method that can remove these artifacts in a principled way. Ideally, such a denoising method would lead to improvements in both subject and group levels of study. At the subject level, motion artifacts can either inflate or reduce
ME-fMRI using MB EPI and 7 T MRI
A perceived cost of the ME approach relates to the investment of imaging time to acquire more echoes, instead of using imaging time to acquire at high resolution. On the one hand, high spatial resolution is important for localizing brain function precisely. On the other hand, smaller voxel size leads to lower per-voxel signal to noise ratio, lowering the odds of detecting weaker signals, such as in deep brain areas, despite high spatial resolution. In regards to temporal resolution, acquiring
ME-fMRI of non-normative neuroanatomy
The use of ME-ICA at 7 T could make high-resolution fMRI studies possible for anatomies that are non-standard since ME-ICA needs no comparisons to standard anatomy to identify networks and artifacts. Fig. 17 demonstrates MEMB-fMRI with 2.5 mm isotropic voxel size with 4 TEs using 7 T MRI acquired from controls and epilepsy patients, analyzed with ME-ICA. For a 10-minute resting state dataset of a normal volunteer, representative BOLD networks are shown. Subcortical activity readily sub-divides
Making ME-fMRI practical
It might be surprising that with all the information to be gained from ME-fMRI over standard fMRI, ME-fMRI is still not standard. For example, the spiral and echo planar (SPEP) sequence by Wong et al., available for several years for the GE platform, could flexibly implement ME-fMRI as well as other techniques such as echo relaxation imaging. The main practical problem has been that gradient echo image readout takes too much time without in-plane acceleration so that by even a third TE most
Conclusion
Over the history of the development of fMRI methods, ME-fMRI has been important in the process of validating BOLD-related origins of novel fMRI signal observations, such as resting state time series correlation. The need for validation of novel BOLD phenomena is due to the fact that only a fraction of variance captured in a typical fMRI experiment is related to BOLD contrast. Bandettini et al. and Peltier and Noll, separated by a few years, used highly targeted ME-fMRI applications to conduct
Acknowledgements
PK and PB are supported by the Icahn School of Medicine Capital Campaign, the Translational and Molecular Imaging Institute, Brain Imaging Center and the Department of Radiology at the Icahn School of Medicine. PB is supported by funding from NIH-NCI (R01 CA202911) and NIH-NIMH (R01 MH109544) and Siemens Healthcare. VV is a Wellcome Trust Intermediate Fellow in Clinical Neurosciences (093705/Z/10/Z). We would like to thank Drs. Lara V. Marcuse and Madeline Fields at the Mount Sinai Epilepsy
References (67)
- et al.
The development and future of perfusion fMRI for dynamic imaging of human brain activity
NeuroImage
(2012) - et al.
Improved sensitivity and specificity for resting state and task fMRI with multiband multi-echo EPI compared to multi-echo EPI at 7 T
NeuroImage
(2015) - 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
(2014) - et al.
Is fMRI “noise” really noise? Resting state nuisance regressors remove variance with network structure
NeuroImage
(2015) - et al.
Removing motion and physiological artifacts from intrinsic BOLD fluctuations using short echo data
NeuroImage
(2013) - et al.
Wavelets and functional magnetic resonance imaging of the human brain
NeuroImage
(2004) Optimizing the order of operations for movement scrubbing: comment on Power et al
NeuroImage
(2013)- et al.
BOLD fractional contribution to resting-state functional connectivity above 0.1 Hz
NeuroImage
(2015) - et al.
Performance of blind source separation algorithms for fMRI analysis using a group ICA method
Magn. Reson. Imaging
(2007) - et al.
Compensation of susceptibility-Induced BOLD sensitivity losses in echo-planar fMRI imaging
NeuroImage
(2002)