 |
 |
> Q1 > > I wondered if for DARTEL also output files of the unified segmentation > (e. g. SPM5) could be used as starting point to start the process? > > If I understood it correctly, the import step produces grey matter > segments that only contain the native => affine transformation. Could > this be replaced by native => unified segmentation/cleaning? Not easily as it would need the code for doing the DARTEL importing to be quite heavily modified. Ideally, the tissue probability maps that are used as priors would be generated such that they did not include the various bits of non-brain tissue in the GM prior. With slightly different priors, I find the segmentation to be quite a lot cleaner. > > (The main reason I ask is rather practical in that the cleaning step > of the unified segmentation was quite helpful to remove misclassified, > and this step is not accessible easily now within DARTEL) You could run DARTEL with the usual imported images, but apply the DARTEL warps to c1* files directly generated from the segmentation. This should give you results that are slightly closer to what you are after. > > Q2 > > Do images that contain an affine transformation in the header / the nii > file generally need to be resliced before smoothing/testing? All NIFTI images have an affine transform (two actually) in the headers. The thing to ensure is that these transforms are identical for all images. > > Q3 > > To perform VBM in 1x1x1 resolution, would you recommend to run the whole > DARTEL > template generation at that resolution from scratch? Is reslicing from 1.5 > x 1.5 x 1.5 > > > 1 mm3 introducing a tolerable error (I know, difficult to say; however, I > > wondered > why the defaults in DARTEL are 1.5 mm and not 1.0 mm as a "typical" VBM > resolution) They are set to 1.5mm because DARTEL would require far too much memory to run, and would be quite a lot slower if the data were 1mm resolution. Because VBM normally involves blurring by quite a lot, I figured that having 1.5mm resolution spatially normalised images should not be too much of an issue. > > Q4 > > A two-scanner question: for an image pool B with slightly different T1-raw > data, > should we run DARTEL by itself, independently from pool A, and just covary > for it at the stats level? Or would it be okay to warp onto templates of > pool A? If the GLM you plan to use is going to involve factoring out sequence effects (which in this case include different pre-processing effects), then a slightly different pre-processing strategy should be OK. Note that you'll need fairly balanced data between the two different sequences/scanners (ie definately not patients collected with one sequence compared with control data collected with the other). Best regards, -John