Inadequate fat suppression for diffusion imaging

Diffusion imaging is often included as a component of functional neuroimaging protocols these days. While fMRI examines functional changes on the timescale of seconds to minutes, diffusion imaging is able to detect changes over weeks to years. Furthermore, there may be complimentary information from the white matter connectivity obtainable from diffusion imaging – that is, from tractography - and the functional connectivity of gray matter regions that can be derived from resting state or task-based fMRI experiments.

I was recently made aware of some artifacts on diffusion-weighted EPI scans acquired on a colleagues’ scanner. When I was able to replicate the issue on my own scanner, and even make the problem worse, it was time to do a serious investigation. The origin of the problem was finally confirmed after exhaustive checks involving the assistance of several engineers and scientists from Siemens. The conclusion isn't exactly a major surprise: fat suppression for diffusion-weighted imaging of brain is often insufficient. And it seems that although the need for good fat suppression is well known amongst physics types, it’s not common knowledge in the neuroscience community. What’s more, the definition of “sufficient” may vary from experiment to experiment and it may well be that numerous centers are unaware that they may have a problem.

Let’s start out by assessing a bad example of the problem. The diffusion-weighted images you’re about to see were acquired from a typical volunteer on a Siemens TIM/Trio using a 32-channel receive-only head coil, with b=3000 s/mm² (see Note 1), 2 mm isotropic voxels, and GRAPPA with twofold (R=2) acceleration. These are three successive axial slices:

(Click to enlarge.)

The blue arrows mark hypointense artifacts whereas the orange arrow picks out a hyperintense artifact. Even my knowledge of neuroanatomy is sufficient to recognize that these crescents are not brain structures. They are actually fat signals, shifted up in the image plane from the scalp tissue at the back of the head. (If you look carefully you may be able to trace the entire outline of the scalp, including fat from around the eye sockets, all displaced anterior by a fixed amount.) I’ll discuss the mechanism later on, but at this point I’ll note that the two principal concerns are the b value (of 3000 s/mm²) and the use of a 32-channel array coil. GRAPPA isn’t a prime suspect for once!

Now, part of the problem is that the intensity of the artifacts – but not their location - changes as the direction of the diffusion-weighting gradients changes. In the following video you see the diffusion-weighted images as the diffusion gradient orientation is changed through thirty-two directions (see Note 2):

The signal from white matter fibers changes as the diffusion gradient direction changes. That’s what you want to happen. But the displaced fat artifacts also change intensity with diffusion gradient direction, meaning that the artifact is erroneously encoded as regions of anisotropic diffusion. Thus, when one computes the final diffusion model, the brain regions contaminated by fat artifacts end up looking like white matter tracts. In the next figure the data shown above was fit to a simple tensor model, from which a color-coded anisotropy map can be obtained:

The white arrow picks out the false “tract” corresponding to the artifact signal crescent we saw on the raw diffusion-weighted images. I suppose it’s remotely possible that this is the iTract, a new fasciculus that has evolved to connect the subject’s ear to his smart phone, but my money is on the fat artifact explanation.

Clearly, in the above image there is no easy way to distinguish the artifact from real white matter tracts by eye, except by using your prior anatomical knowledge. And it's likely to confuse tractographic methods, too, because it has very similar geometric properties to those that tractographic methods attempt to trace. So let's take a look at the origin of the problem and then we can get into what you want: solutions. 

Review: Differentiating BOLD from non-BOLD signals in fMRI time series using multi-echo EPI

Disclaimer: I'm afraid I haven't done a very good job reviewing the entirety of this paper because the stats/processing part was pretty much opaque to me. I've done my best to glean what I can out of it, and then I've focused as much as I can on the acquisition, since that is one part where I can penetrate the text and offer some useful commentary. Perhaps someone with better knowledge of stats/ICA/processing will review those sections elsewhere.


The last paper I reviewed used a bias field map to attempt to correct for some of the effects of subject motion in time series EPI. A different approach is taken by Prantik Kundu et al. in another recently published study. In their paper, Differentiating BOLD from non-BOLD signals in fMRI time series using multi-echo EPI, Kundu et al. set out to differentiate between signal changes that have a plausible neurally-driven BOLD origin from those that are likely to have been modulated by something other than neuronal activity. In the latter category we have cardiac and respiratory fluctuations and, of course, subject motion.

The method involves sorting BOLD-like from spurious changes using an independent component analysis (ICA) and to then "de-noise" the time series before applying connectivity analysis. For resting state fMRI in particular, the lack of any sort of ground truth and an absence of independent knowledge that one has with task-based fMRI makes disambiguating neurally driven signal changes from artifacts a major problem. Kundu et al. use a relatively simple philosophical approach to the separation:

"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 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."

And, noting again the caveat that there is an absence of ground truth, the approach seems to work:

"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."