Dear Kwaku,
> For the last half-year, I have been working on auditory fMRI experiments. These experiments require temporal fMRI acquisition schemes that scan a number or "cluster" of fMRI images at a time, with a time delay between clusters facilitation the presentation of auditory stimuli. Perhaps the most advanced of such aquisition schemes is the interleaved silent steady state (ISSS) method [NeuroImage 29 (2006) 774 – 782].
The fundamental challenge is that fMRI analysis packages (including,
but not limited to, SPM) are generally designed to perform analyses on
time-series data. Sparse imaging is not really a problem provided you
set up a model appropriately; however, as you have no doubt
discovered, ISSS presents a special challenge because the data are not
evenly-spaced in time. Although there are not many ISSS papers
published, if I'm not mistaken 1-2 more will likely appear in the
coming year, and may be able to give you some examples to work off of.
> I have been trying to come up with a way of analysing ISSS data in SPM8. I have even posted one or two messages asking for help on this forum, and have seen a few posts about the older "sparse sampling" method (where the cluster has only one image volume). A recurring theme in replies to the sparse sampling posts is that SPM is not really suited to these temporal acquisition schemes.
Analyzing sparse imaging data in SPM shouldn't really pose a problem.
There are a number of posts about this; if you get stuck, please feel
free to let me know and I can try to point you in the right direction.
It's true that it isn't quite as straightforward as continuous
imaging, but once you get the basic idea it's not too much of a
challenge.
Like sparse imaging, the best way to analyze ISSS data probably
depends on the specific parameters of your experiment. For example,
if you have a long gap between data acquisition scans, you may be
comfortable assuming that a stimulus only affects the scans it
immediately precedes, which will probably make your life easier.
Although there may be some modifications to the code that would
facilitate this type of analysis, I suspect it will be more productive
to find ways to analyze data within SPM's existing framework—ISSS is a
rather unique case because it departs from nearly every other study in
the unevenness of data collection.
One option is to use a type of FIR model to estimate the response for
each time bin (scan) following a stimulus. This is probably the most
straightforward approach. [If you just include the ISSS scans in this
model, they are not continuous, which may make the processing steps
that depend on timeseries information (low pass filtering and
autocorrelation) unreliable.]
Another option would be to set up a continuous timeseries by including
a dummy image to fill in the timeseries when no data is collected.
You could use the same image (for example, a mean functional image)
for each dummy image, and then include regressors in the design matrix
to model these out. This can at least help with some of the
assumptions about timeseries continuity. We are currently
experimenting with this approach, and preliminary analyses seem
promising. If all continues to go well I would be happy to share our
approach with you.
Hope this offers at least a bit of encouragement!
Best regards,
Jonathan
--
Dr. Jonathan Peelle
Department of Neurology
University of Pennsylvania
3 West Gates
3400 Spruce Street
Philadelphia, PA 19104
USA
http://jonathanpeelle.net/
|