Dear Caret Users,
I wanted to share with you a report and a matlab functions to do statistics
for functional data on a caret surface. These functions basically do a
SPM-style analysis of surface data, including calculating the statistical
parameter map (SPM), estimating the smoothness, correcting the p-values for
the search-region, finding clusters of activation and calculating
cluster-wise p-values. The motivation for this was the following:
For my current experiments I do a surface-reconstruction for every single
subject. Then I map the non-smoothed functional data onto the individual
surface and then normalize these to David's new PALS Atlas, using the 6
core landmarks, and finally smooth the data in 2D.
Although this is a LOT of work, the results seem to be clearer, the areas
better defined, than a standard SPM-normalization (with subsequent mapping
to a caret surface), most certainly for areas that are well described by
anatomical landmarks (for example the motor and somatosensory-cortices).
I'd be curious to hear what other groups' experiences are in this respect.
Now that I am looking at the results on a surface representation, I wanted
of course to do my statistics on the surface as well, as I am used to from
SPM, that is, find corrected p-values, as well as cluster-wise p-values.
So, what you see here is really a transfer of the SPM-style statistics to 2
dimensions and the caret data format.
I hope some of you might find this useful. If we look at our results on a
caret-surface, we are really constricting our search region from a volume
to a sheet of cortex and I think our statistics should reflect this.
Otherwise caret will always remain "just" a visualization tool.
A document describing the functions and the approach can be found under:
http://www.bme.jhu.edu/~jdiedric/download/caret/Caret_surface_statistics_020205.pdf
The functions themselves, including a sample data set are in:
http://www.bme.jhu.edu/~jdiedric/download/caret/
As this is constitutes only a first stab at things and I would of course
appreciate any comments, questions, or bug reports you might have. I hope
that, if many people find this useful, there is a potential of integrating
these functions into caret itself.
Thanks,
Joern Diedrichsen
----------------------------------------------------------------------------------------
Jörn Diedrichsen
Postdoctoral Fellow
Department of Biomedical Engineering
Johns Hopkins University
720 Rutland Ave, 419 Traylor Building
Baltimore, MD 21205-2195
email: [EMAIL PROTECTED]
web: http://www.bme.jhu.edu/~jdiedric/
fax: 410 614-9890
phone: 410 614-8266
