 |
 |
Juanjo Lull (UPV) wrote: > Dear Will, SPM, > I have an experiment with two sessions. Each session has four activation > blocks. There is one condition per session. > I have a design matrix that implicitly models the baseline. My matrix > therefore has four columns: first for condition 1, second for condition > 2, third and fourth for global intensity differences. > When I try to observe the activated voxels for cond1, I just ask a t [1 > 0 0 0]' contrast. > I make the same question for condition 2 (t [0 1 0 0]'). But now, say I > want to see differences between cond1 and cond2. If I perform a t [-1 1 > 0 0]' test, I firstly thought I would observe the areas in which cond2 > is greater than cond1, that is, voxels more activated by cond2 than by > cond1. However, as my baseline is implicitly defined, I think that this > is not the case. So when I perform a t [-1 1 0 0] contrast I get the > differences between the conditions regardless of the baseline (i.e. > regardless of whether the voxel activated). Therefore I (suppose I) am > getting voxels that may or may not be active and that they are mostly > different. Regressors 3 and 4 will subtract out the average BOLD level in each session ie. the session baseline. Regressors 1 and 2 are your task-related responses. This seems like a standard epoch design and analysis. So, yes the [-1 1 0 0] t-contrast will show you where the task-related response is greater in the second session than in the first. This is the difference, regardless of the baseline, as you say and is the effect one is usually interested in. But, as far as I understand, you are only interested in those differences for voxels which respond positively. ie. c2 > c1 AND c1 > 0 You don't want to see voxels where c1 is negative and c2 is either less negative or positive. So, you could use the contrast [-1 1 0 0] (ie c2 > c1) but masked with [1 0 0 0] (ie c1 > 0). Or if you don't want to use masking then a conjunction will also do the job. Best wishes, Will. > My questions finally come. > The first one is whether I am right about the problems that assuming an > implicit baseline can generate in my final contrast. > The second one is what approach you think better, in the case that I > should change the way I create the contrasts. I could: > - Redesign my matrix including the baseline (BL) and test: cond1>BL AND > cond2>BL AND cond2>cond1 > or I could > - Ask for a conjunction: cond1 AND cond2 AND cond2>cond1. > Thank you in advance, > Juan J. > PS: please find attached a figure with the design matrix, cond1 and > cond2 seen as a conjunction and a plot with the estimated cond1 and cond2 > > > ------------------------------------------------------------------------ > -- William D. Penny Wellcome Trust Centre for Neuroimaging University College London 12 Queen Square London WC1N 3BG Tel: 020 7833 7475 FAX: 020 7813 1420 Email: [log in to unmask] URL: http://www.fil.ion.ucl.ac.uk/~wpenny/