Hi Benjamin,
I know very little about AFNI, but I'll have a quick effort, hopefully
someone else will be more thorough, but if not, I hope this helps...
> In addition to the
> preprocessed images (I do not have intermediate steps) I have the
> original analyze files from the DICOM conversion/ extraction.
[...]
> 1. Would it be suitable to convert AFNI files to the analyze file type
> for the final preprocessed images, or does image processing need to be
> re-done in SPM to yield proper results in subsequent analyses?
I think so. I assume you have AFNI BRIK files, in which case AFNI
contains several functions like 3dAFNIto<something> where <something>
includes ANALYZE and NIFTI. The latter would be better for SPM5, if
you are using that. In fact, I believe the current version of AFNI is
happy working with NIfTI files, so it should be possible to use both
this and SPM5 on the same data.
A few things to watch out for when switching between programs:
- orientation is a very thorny issue, 3dAFNItoANALYZE has some help
info on this. Take a subject with an obvious left/right asymmetry and
check that the afni viewer and SPM's Display function have left/right
consistent.
- intensity scaling can be rather non-standard, for the non-float
data-types. Again, check that viewers agree on the intensity of
corresponding voxels. If you have problems, AFNI's 3dcalc could be
used to convert AFNI data to floating point (this might not be the
best way of doing this...)
> Can I
> port the preprocessed AFNI files into SPM and expect to get the same /
> similar results in t and F contrasts in SPM?
If the orientation and intensities are correct, then I would expect
t/F values to be similar. However, AFNI and SPM have different
methods/options for non-sphericity correction / variance component
estimation, and probably a few other subtleties, so they will probably
not be the same.
Also, even with identical t or F values, FWE/RFT statistical
thresholds will probably differ (hopefully not by too much, but I
haven't looked) due to differences in the way the two programs
estimate smoothness and related calculations. FDR should be more
likely to be the same, as far as I can tell.
Note though that AFNI and SPM pre-processing (such as motion
correction and intersubject normalisation if used) might differ quite
a lot, so if you did do separate analyses from scratch in the two
programs, the results could differ a fair bit -- but hopefully not too
drastically, if they are both analysing the same brain activity ;-)
> 2. Does anyone have experience working with AFNI and SPM and know what
> the fidelity of SPM maps between program is? Being able to validate my
> work against the AFNI results would be invaluable.
I think they will differ, but it will probably be due to different
modelling choices (e.g. in HRF parametrisation or error
autocorrelation structure) rather than greater or less degrees of
"fidelity/validity" so it won't be easy to say which is better/right.
Sorry I don't have any more direct experience to relate to you.
Best,
Ged.
|