 |
 |
I have another question related to this thread. In our case, the design is not paired and comparing 2 groups, but on a first TBSS run our patient group shows a very diffuse reduction of FA. So, we thought we would use the mean FA over the whole skeleton for each individual as a covariate in order to detect further underlying regional patterns. There are different ways to do this and I need some clarification of what the different options mean. 1) one could divide each individual skeleton image by the mean FA for that subject (proportional scaling in PET lingo) 2) one could use mean FA as a regressor after demeaning. Here there are 4 options. a. Demean based on the average FA across all subjects, and use only one column for the regressor. b. Demean based on the average FA across all subjects, but use two regressors, with zeros for one group in one and zeros for the other group in the other c. Demean based on the mean within each group, and use only one column for the regressor. d. Demean based on the mean within each group, but use two regressors, with zeros for one group in one and zeros for the other group in the other. I am oriented to using option1 (the simplest) or 2d which I think provides most flexibility to analyze the data. If there are not differences in the relationship between mean FA and any given voxel between the groups, these methods should give equivalent results. Is this correct? Any feedback? Stefano Marenco ________________________________ From: Stephen Smith [mailto:[log in to unmask]] Sent: Wednesday, October 12, 2011 7:05 AM To: [log in to unmask] Subject: Re: [FSL] using mean FA as a confound regressor in TBSS / setting up a paired-design with this confound regressor Hi - technically this seems OK - you just have to very carefully change how you interpret the results from doing this...... Cheers. On 11 Oct 2011, at 19:15, Diederick Stoffers wrote: Hi all, I am analyzing dti data from a group of subjects using tbss and want to use a paired design; I set up regressors to model the means within each pair. From my initial analysis it is obvious that mean FA is much higher in one of the groups of scans relative to the other group in the paired design. Two questions; 1. Does it make sense in TBSS to use the mean FA within the skeleton as a "confound" regressor (analogous to using native GM volume in a VBM analysis?). If so, how should I interpret differences between the analysis with and without mean FA as a regressor? 2. If I were to use mean FA as a confound regressor, should I first demean over the whole group and subsequently within pairs before I copy it into my matrix? Cheers Diederick --------------------------------------------------------------------------- Stephen M. Smith, Professor of Biomedical Engineering Associate Director, Oxford University FMRIB Centre FMRIB, JR Hospital, Headington, Oxford OX3 9DU, UK +44 (0) 1865 222726 (fax 222717) [log in to unmask]<mailto:[log in to unmask]> http://www.fmrib.ox.ac.uk/~steve ---------------------------------------------------------------------------