 |
 |
Hi Arun. I've got even more basic concerns than Joe and Johannes. The main one is about the normalisation that you carried out; EV1 looks at the change in BOLD response level and EV4 looks at the change in PET signal - but whether these two things make any sense to compare in this subtraction depends on how the images have been (intensity) normalised - unless you're very careful with this, this subtraction is likely to be meaningless. Secondly, you aren't feeding variances up into this higher-level anlaysis, at least for the PET data, so you may as well just use the OLS option, and the groupings won't get used, so you don't need to worry about this. Hope this makes sense! Cheers, Steve. On Fri, 13 Feb 2004, Arun Bokde wrote: > Hello, > > I have a question about modeling interaction effects. I believe that I > have it right but I want to double check. > > I would like to look at a Group x Modality interaction effect. I have > subjects with measurements from 2 different imaging modalities. I have > the fixed effects analysis (fmri data) from each subject in 2 groups: > Normals and Patients. In addition, I have a PET image from each > subject. The assumption from this point on is that there is a linear > relationship between the PET data and my fmri results on a voxel by > voxel level. In the preprocessing, I also took care that the final > smoothing of the data is the same in both modalities (i.e. the fmri data > was smoothed more than the pet data). > > > Setting up the RFX analysis (I have about 20 subjects/group - using only > 3 here for simplication). > > The group membership would NOT indicate Normals and Patients, but would > differentiate between fmri and pet data. > So Group 1 membership would indicate fmri data. Group 2 membership > would indicate pet data. > > Under the Group column I also indicate which data comes from normals > (-nor) and which are from patients (-pat) > > > Group EV1 EV2 EV3 EV4 > 1(fmri-nor) 1 1 0 0 > 1(fmri-nor) 1 1 0 0 > 1 (fmri-nor) 1 1 0 0 > 1(fmri-pat) -1 1 0 0 > 1(fmri-pat) -1 1 0 0 > 1 (fmri-pat) -1 1 0 0 > 2 (pet-nor) 0 0 1 1 > 2 (pet-nor) 0 0 1 1 > 2 (pet-nor) 0 0 1 1 > 2 (pet-pat) 0 0 1 -1 > 2 (pet-pat) 0 0 1 -1 > 2 (pet-pat) 0 0 1 -1 > > EV1 would indicate a difference between normals and patients (fmri data) > EV2 - the average of the fmri data > EV3 - the average of the pet data > EV4 - would indicate a difference between normals and patients (pet data) > > The contrasts would be: > 1 0 0 0 = Simple main effect fmri-nor > fmri-pat (that is, a t-test) > -1 0 0 0 = Simple main effect fmri-pat > fmri-nor > 0 0 0 1 = Simple main effect pet-nor > pet-pat > 0 0 0 1 = Simple main effect pet-pat > pet-nor > 1 0 0 -1 = Interaction (greater activation in fmri (nor-pat) = greater > "deact" in pet (nor-pat) > -1 0 0 1 = Interaction (greater activation of fmri (pat-nor) = greater > "deact" in pet (pat-nor) > > Does the above sound reasonable ? I would appreciate any advice. > > Another questions, is that I am unsure which contrast would show that > difference between normals and patients in the fmri set is linearly > related to the difference between normals and patients in the pet data set. > > > > Thanks in advance. > > > Cheers, Arun > Stephen M. Smith DPhil Associate Director, FMRIB and Analysis Research Coordinator Oxford University Centre for Functional MRI of the Brain John Radcliffe Hospital, Headington, Oxford OX3 9DU, UK +44 (0) 1865 222726 (fax 222717) [log in to unmask] http://www.fmrib.ox.ac.uk/~steve