The infamous dead salmon brain scan study may just have been eclipsed, in the ‘most ghoulish demonstration of a methodological pitfall in fMRI‘ stakes. A new study examines the issue of motion artifacts, a major concern in much neuroimaging research – and it does so by scanning dead people.
The new paper has the unwieldy title of SimPACE: Generating simulated motion corrupted BOLD data with synthetic-navigated acquisition for the development and evaluation of SLOMOCO: A new, highly effective slicewise motion correction.
In this article, authors Erik Beall and Mark Lowe introduce two new tools for dealing with motion. The first, SimPACE, is an fMRI pulse sequence that ‘injects’ motion artifacts directly into the data at the point of acquisition, by means of cleverly (mis)calibrating the gradient pulses. SimPACE can introduce this ‘motion’ on a slice-by-slice basis. Why would you want to create motion artefacts in this way? To find out the best way of correcting for it.
This is where the cadavers come in. Cadavers have no brain activity. They also don’t move. But using SimPACE, Beall and Lowe were able to ‘animate’ the dead – they could acquire fMRI from real human brains, with ‘motion’ that was exactly known (because they injected it). The authors were then able to compare how well various different motion-correction algorithms were able to deal with these artifacts.
Beall and Lowe ran a standard resting-state functional connectivity analysis on the brain ‘activity’ from the cadavers. These analyses are known to show false positives in the presence of motion.
The data revealed that “commonly used motion metrics based on volumetric motion estimation are not robust estimators of actual motion.” In other words, motion correction that operates only at the whole volume level (VOL) – which is the most common approach, used by default in the FSL and SPM software – is inadequate, only removing some 33% of the signal variance introduced by motion.
‘Second-order’ volumetric approaches (SVOL) did rather better. These are already in widespread use, e.g. FSL’s option of including the volumetric motion parameters in the statistical model. Voxel-specific motion correction approaches (SVOX) were also pretty good. These advanced strategies were able to remove about 50-60% of the variance caused by motion.
But the best performance, according to Beall and Lowe, came from using SLOMOCO – which is the second tool they introduce in their paper. SLOMOCO is a novel motion correction algorithm that operates on individual images (slices). Yet even this method, the authors caution, wasn’t perfect. It only removed about 70% of the motion signal.
So 30% of the signal variance introduced by motion survives even the best available mitigation methods, and is indistinguishable from real brain connectivity. That’s pretty worrying: it means that motion is out there, in our data, and we can’t detect all of it.
It’s not clear, though, whether head movement is actually leading neuroimaging researchers astray. The pessimistic view is that head movement, masquerading as brain activity, underlies many exciting claims, especially in neurology and psychiatry (where unwell patients often move more than healthy controls.) However, we don’t know that for sure, at this stage.
Incidentally, this is actually not the first cadaver fMRI paper. There has been one previously that I found, from 1999. Like this paper, it appeared in the journal Necroimage. I mean Neuroimage.
Thanks to regular Neuroskeptic commenter D S for sending me this paper.
Beall EB, & Lowe MJ (2014). SimPACE: Generating simulated motion corrupted BOLD data with synthetic-navigated acquisition for the development and evaluation of SLOMOCO: A new, highly effective slicewise motion correction. NeuroImage PMID: 24969568