Most fMRI echo-planar imaging (EPI) techniques acquire a single brain volume of data per radio-frequency (RF) excitation pulse. The receiving elements at the scanner record signals during a readout window centered on the echo time (TE); which is the moment when the recoverable signal of interest is maximized. This TE, which is setup by the experimenter, depends on factors such as field strength (e.g., 1.5T, 3T, 7T), target tissue (e.g., GM, WM, CSF), and target imaging parameter (e.g., T1, T2, T2*). For Blood Oxygen Level Dependent (BOLD) fMRI, which is interested in T2* signal fluctuations, it is desirable to setup the TE as close as possible to the T2* of grey matter (GM). Because data is acquired at only one time-delay (the TE) from each RF pulse, it is common to refer to this data acquisition scheme as single-echo fMRI.
In practice, nothing precludes researchers from having additional readout windows centered at TEs (delays from the RF pulse) other than the optimal TE discussed above. Such acquisition schemes are commonly referred to as multi-echo fMRI (ME-fMRI). The question then is: it is worth acquiring these additional data at suboptimal TEs? If so, why? At the SFIM, we believe it is extremely useful. Functional MRI time-series contain fluctuations of many different origins. In addition to neuronally induced fluctuations (the ones of interest for most experiments), the fMRI signal also contains nuisance signal fluctuations due to hardware instabilities, progressive heating, subject motion, respiration, cardiac function, inflow related signals, etc. Fortunately, many of these nuisance fluctuations are non-TE dependent, while neuronally induced fluctuations are heavily dependent on the selected TE. At the SFIM we are developing a series of algorithms that exploit this TE/non-TE dependence to automatically separate BOLD (such as neuronally induced signal changes) from non-BOLD fluctuations (e.g., subject motion, hardware instabilities, inflow effects, etc.). This results in a significant boost in temporal signal-to-noise ratio (TSNR) that permits the detection of interesting phenomena commonly buried below noise levels in single-echo fMRI. Our portfolio of ME-fMRI research is very diverse and includes: programming of new ME-fMRI sequences, optimization of scanning parameters for ME-fMRI, and the development of a toolbox for analyzing and automatic denoising of ME-fMRI data.
1. Differentiating BOLD and non-BOLD signals in fMRI using Multi-Echo
A central challenge in the fMRI-based study of functional connectivity is distinguishing neuronally related signal fluctuations from the effects of motion, physiology, and other nuisance sources. Conventional techniques for removing nuisance effects include modeling of noise time courses based on external measurements followed by temporal filtering. These techniques have limited effectiveness. Previous studies have shown using multi-echo fMRI that neuronally related fluctuations are BOLD signals that can be characterized in terms of changes in R2* and initial signal intensity (S0) based on the analysis of echo-time (TE) dependence. We hypothesized that if TE-dependence could be used to differentiate BOLD and non-BOLD signals, non-BOLD signal could be removed to denoise data without conventional noise modeling. To test this hypothesis, whole brain multi-echo data were acquired at 3 TEs and decomposed with Independent Components Analysis (ICA) after spatially concatenating data from different TEs across space and TE. Components were analyzed for the degree to which their signal changes fit models for R2* and S0 change, and summary scores were developed to characterize each component as BOLD-like or not BOLD-like. These scores clearly differentiated BOLD-like "functional network" components from non BOLD-like components related to motion, pulsatility, and other nuisance effects. Using non BOLD-like component time courses as noise regressors dramatically improved seed-based correlation mapping by reducing the effects of high and low frequency non-BOLD fluctuations. A comparison with seed-based correlation mapping using conventional noise regressors demonstrated the superiority of the proposed technique for both individual and group level seed-based connectivity analysis, especially in mapping subcortical-cortical connectivity. The differentiation of BOLD and non-BOLD components based on TE-dependence was highly robust, which allowed for the identification of BOLD-like components and the removal of non BOLD-like components to be implemented as a fully automated procedure. [Kundu et al. 2012, NeuroImage]
2. Integrated Strategy for improving Functional Connectivity Mapping using Multi-Echo fMRI
Functional connectivity analysis of resting state BOLD functional MRI is widely used for noninvasively studying brain functional networks. However, recent findings have indicated that even small (≤1 mm) amounts of head movement during scanning can disproportionately bias connectivity estimates, despite various preprocessing efforts. Further complications for interregional connectivity estimation from time domain signals include the unaccounted reduction in BOLD degrees of freedom related to sensitivity losses from high subject motion. To address these issues, we describe an integrated strategy for data acquisition, denoising, and connectivity estimation. This strategy builds on our previously published technique combining data acquisition with multi-echo (ME) echo planar imaging and analysis with spatial independent component analysis (ICA), called ME-ICA, which distinguishes BOLD (including neuronal) and non-BOLD (artifactual) components based on linear echo-time dependence of signals—a characteristic property of BOLD signal changes. Here we show for 32 control subjects that this method provides a physically principled and nearly operator-independent way of removing complex artifacts such as motion from resting state data. We then describe a robust estimator of functional connectivity based on interregional correlation of BOLD-independent component coefficients. This estimator, called independent components regression, considerably simplifies statistical inference for functional connectivity because degrees of freedom roughly equals the number of independent components. Compared with traditional connectivity estimation methods, the proposed strategy results in fourfold improvements in signal-to-noise ratio, functional connectivity analysis with improved specificity, and valid statistical inference with nominal control of type 1 error in contrasts of connectivity between groups with different levels of subject motion. [Kundu et al. 2013, PNAS]