Hello all,
I am having some difficulty understanding the relationship between the fstatX and zfstatX (and zflame1lowerfstatX) files produced by flameo (running flame1), and I was wondering if someone could help me out.
After looking at the source of flameo and gsmanager.cc, I can see that zfstatX is the same as zflame1lowerfstatX, which is consistent with my results. What confuses me, though, is that these 2 files are not the same as if I performed:
ftoz -zout zfstatX_f2zredo fstatX fdof1_fX fdof2_fX
despite the fact that my look at the source code seemed to indicate that the z statistic should be created with very similar internal calls across ftoz and flameo (in Gsmanager::f_ols_contrast within flameo). Because the z-converted files are not the same, I'm not sure which file I should trust to use for inference. The fstat file (and the zfstat file produced by an explicit call to ftoz) looks like what I would expect, in that the image histogram looks continuous, and the spatial distribution of those statistics across the brain looks exactly like I would expect. The zfstat file produced directly by flameo looks like a thresholded image in which the absolute value of all voxels is greater than approximately 5.75. I have attached a picture of the three image histograms, with the fstat image on the left, the zfstat image produced by flameo in the middle, and the zfstat file produced by a call at the command line to ftoz.
To help with interpretation, here is some more information regarding my design. In the first-level analysis, I am attempting to use FIR basis functions to estimate a timecourse in response to various stimulus conditions. I am using a set of 10 delta functions that span the 20 s following stimulus onset. I am then using a second-level model on the copes reflecting each of the delta functions, such that I have 10 copes for each of the 18 subjects (180 input files total). In the second-level model, I am using an intercept (1 regressor) and dummy coded factors for subject (17 regressors, each corresponding to the presence of a different subject) and timepoint (9 regressors, each corresponding to the presence of a different time point). The f statistic of interest is the main effect of timepoint, which should have a first degree of freedom of 9, and a second degree of freedom of 153 (the fdof1_fX and fdof2_fX files do reflect this).
Any help would be much appreciated. Thanks in advance.
-Jeremy
Jeremy Reynolds, Ph.D.
Assistant Professor
Department of Psychology,
University of Denver
2155 S. Race St
Denver, CO 80208
Phone: 303.871.4622
Fax: 303.871.4747
[log in to unmask]
|