Dear SPMers,
My questions are only related to SPM, but I am sure some of you have
encountered (and hopefully solved) similar problems, too, when using SPM
- so, here we go:
I am writing a Matlab routine to do principal component analysis on
SPECT scans of epileptic patients. Now, when I read in the scans of lets
say 30 subjects, each scan consisting of 128x128x40 voxels and then
calculate the correlation matrix...the thing just gets HUGE and my
computers out of memory. So, I went through the literature and found a
nice algorithm created by Dr. Friston (J Cereb Blood Flow and Metab,
1993), but if I implement it, Matlab exceeds the allowed recursion depth
and crashes.
I am working with a AMD K7 600 MHz machine (128 MByte RAM), and a
MacIntosh G3 333 MHz (350 MByte RAM) ...and both cannot handle the job.
Does anybody know a way to implement a memory saving pca algorithm...or
how much RAM I would need to handle such high-dimensional matrices?
I would warmly welcome any hints and advices.
Sincerely,
Kaspar Schindler, MD, PhD
|