Hi Christian,
the platform is indeed 32bit Intel and I tried to run the process on a
FreeBSD machine (compiled sources) and on a Fedora 5 machine (binaries),
both with 2 GB of RAM. I guess it would help if the data were stored in
shorts (with a scale factor) instead of floats, but can that be done?
The dataset unzipped is 156618368 bytes (~150 Mb) and is 54x64x50 voxels by
453 "timepoints". I uploaded the file with the access number 308806; I also
included some snapshots of the "IC maps of the IC maps" that are found by
melodic, which I thought you may find interesting since they match so well
with the results reported in your paper (there are also some artifact maps,
but only a few).
On a related issue, I gather that the associated "timecourse" of each
second-level IC map represent the relative contribution of each individual
(first-level) IC and can therefore be used to identify the most relevant IC
for each subject for a particular group-wise resting state network. Is this
correct?
Thank you very much for all your input, it has been tremendously helpful.
cheers,
giuseppe
|