 |
 |
Thank you very much for your response and important precision regarding the concepts. In fact you are right. If I take a lower test threshold for the VBM analysis with the SPM-DARTEL procedure, the results displayed are more similar to the results obtained with CAT12 procedure. So I assume the differences come from the sensitivity. Best Michaël -----Original Message----- From: SPM (Statistical Parametric Mapping) <[log in to unmask]> On Behalf Of Christian Gaser Sent: vendredi, 22 février 2019 22:46 To: [log in to unmask] Subject: Re: [SPM] Comparison CAT12 and classical SPM12-DARTEL VBM pipeline Dear Michaël, the different handling of the Jacobian scaling as John suspected was only used in the older VBM toolboxes (in short I only used the non-linear part of the Jacobian which allowed to skip the additional TIV correction in the statistics. This is like scaling the segmentations with a TIV-like measure). Ian Malone convinced me after some discussions that the use of TIV as nuisance parameter in the stat. model has several advantages and is more robust. Therefore, in CAT12 I use the same strategy as in SPM12 where modulation is using the whole Jacobian and TIV is used as nuisance parameter or as globals scaling (if TIV is correlating with your parameter of interest), so this might not cause the different effects you have noticed. There are many differences in the preprocessing approaches between SPM12 and CAT12 and as so often different methods give you different results (even between the CAT12 versions there are differences in the results). In CAT12 I additionally apply de-noising, local adaptive segmentation and other methods that can influence the results. The only hint I can give is to choose the approach that better fits to your hypothesis. It might also help to lower the thresholds to get a closer look at the differences. Best, Christian On Thu, 21 Feb 2019 11:40:16 +0000, Mouthon Michaël <[log in to unmask]> wrote: >Dear Christian and John, >I am conducting a correlation study between Grey Matter volume and a behaviour variable with 30 healthy participant. With the same data (and confounds), I have conducted the VBM processing through the CAT12 pipeline (as explain in the manual) and with the classical SPM12-DARTEL pipeline as explain in the John Ashburner tutorial (https://www.fil.ion.ucl.ac.uk/~john/misc/VBMclass15.pdf). > >The results of these two pipelines give me a different result (significant TFCE corrected cluster in the frontal pole of the CAT12 processing and no significant cluster at all in classical SPM-DARTEL pipeline). > >I know that the segmentation procedure as slightly different but is it the only reasons why the two results are different? >Can you help me to know which result is the most trustful regarding to my initial question (does the change of my behavioural variable correlate with GM volume)? >Thank you very much in advance > >Note: I have used CAT12.5 (r1264), SPM12 (v6906), TFCE toolbox (174).