I thought it might have implications for both resting and task-based data given the possibility that the AR(1) noise correction isn't capturing everything sufficiently enough to have a truly random baseline. But, I could be wrong...
Another thing that was weird in the paper was how they changed the default thresholds from 0.001 to 0.05, which in my mind would only increase their chance of finding a problem when maybe there isn't one (within 0.001 tolerance, at least)?
Jeff
On May 13, 2012, at 4:31 PM, Watson, Christopher wrote:
> They only looked at resting-state data, and applied those designs when specifying the 1st-level design in SPM. I might be missing something big about the paper, but I don't think it's worth a panic; it's just showing that there are voxels that significantly vary at similar frequencies to the block designs specified. I'm not sure I would call those false positives, as they could be voxels that are part of the default mode network, which are known to have slowly-varying BOLD signal. So even though the subjects didn't perform a task, there could still be voxels that truly do vary in BOLD signal along with the "task frequency" that the authors entered into their design.
>
> If I am misunderstanding what this paper shows, I look forward to some more discussion on this.
>
> Chris
> ________________________________________
> From: SPM (Statistical Parametric Mapping) [[log in to unmask]] on behalf of Jeff Browndyke [[log in to unmask]]
> Sent: Sunday, May 13, 2012 4:23 PM
> To: [log in to unmask]
> Subject: [SPM] Eklund et al. Panic - What to do?
>
> Eklund, A., Andersson, M., Josephson, C., Johannesson, M., and Knutsson, H. (2012). Does parametric fMRI analysis with SPM yield valid results?—An empirical study of 1484 rest datasets NeuroImage DOI: 10.1016/j.neuroimage.2012.03.093<http://dx.doi.org/10.1016/j.neuroimage.2012.03.093>
>
> My take from this enlightening article is that long block, short TR data is problematic if analyzed in SPM8, but what if run on SPM5 or earlier iterations? I suspect the same problems given the argument is that it is the noise correction AR(1) scheme that isn't capturing things well under long block, short TR conditions (i.e., what is assumed to be noise actually includes meaningful brain activity?)
>
> Also, what if someone ran a block design that did not include rest conditions (i.e., alternating N-back with varying levels of N)? Do the Eklund et al. SPM problems apply to purely task-based data?
>
> Jeff
>
> -----------------------------------------------------------------------
> Jeff Browndyke, Ph.D.
> Clinical & Research Neuropsychologist
> Durham VA & Duke University Medical Centers
>
> [log in to unmask]<mailto:[log in to unmask]> / [log in to unmask]<mailto:[log in to unmask]>
> office: (919) 286-0411 ext. 4656
> cell: (336) 264-4222
-----------------------------------------------------------------------
Jeff Browndyke, Ph.D.
Clinical & Research Neuropsychologist
Durham VA & Duke University Medical Centers
[log in to unmask] / [log in to unmask]
office: (919) 286-0411 ext. 4656
cell: (336) 264-4222
|