Here are my two cents :-)
A small sample size per se does not invalidate Random Field Theory
(RFT), but so many things could potentially go wrong. A small sample
size DOES make the FWE-corrected threshold (by RFT) very stringent. In
such cases, SPM calculates the FWE-corrected threshold based on
Bonferroni correction, because it is actually less stringent compared to
RFT. As for cluster size p-values, SPM may underestimate the expected
number of voxels per cluster, possibly <1 voxel per cluster
(counterintuitive, isn't it?). If that happens, a cluster with just a
few voxels could have a very small corrected p-value.
So, I would recommend using SnPM in this case. It's been shown to work
well under low df. You have 6 images in your data set, that means there
are 2^6=64 possible permutations. So it probably won't take you too much
CPU time and memory to calculate. Corrected p-values are accurate, both
for voxel-level and cluster-level. It does not require the data to be
Gaussian, although it helps to smooth the data somewhat for the sake of
matched-filter theory.
-Satoru
Satoru Hayasaka PhD ----------
Assistant Professor, Public Health Sciences & Radiology
Wake Forest University School of Medicine
(ph) +1-336-716-8504 / (fax) +1-336-716-0798
(email) shayasak _at_ wfubmc _dot_ edu
|