Dear Catherine,
You are right that the correct method of entering the motion parameters is by entering a separate rp*.txt file per session (motion parameters are typically estimated separately for each session). The 'degrees of freedom' issue that you are finding arises because CONN will use a rather conservative method for removing physiological and other confounds, and the default settings of this 'preprocessing' step are not appropriate for your data (which contains very limited number of scans per session; only 20). The default procedure regresses out 10 'noise' parameters (noise characterization, estimated from white matter and CSF areas), another 12 'motion' parameters (estimated subject motion plus their first derivatives), and another 2 'session' parameters (compensating for any small transient effects at the beginning of the session). This default procedure is not able to be run on your data because the sessions are not sufficiently long. I would recommend to reduce the dimensionality of these confouding effects to more appropriate settings. For example, select the 'White matter' and 'CSF' confounds and set their 'dimensions' field to 2, select the 'realignment' confound and set its 'derivatives order' field to 0, and also remove the 'Effects of Session' effect from the confounds list. This will reduce the number of required degrees of freedom to 10 which should be fine on your data (if the distributions of voxel-to-voxel connectivity values still look strange, I would try reducing even further the dimensionality of the 'White matter' and 'CSF' confounds, or trying the option below).
Another alternative would be to concatenate all of your data into a single session. This is typically not preferred because it assumes similar movement effects across sessions, as well as a false continuity of your timeseries across sessions (e.g. when band-pass filtering). If this nevertheless turns out to be the only viable route, make sure to, in addition to the concatenated motion parameters, also define a new set of 'session' first-level covariates that would at least remove the average session effects from your timeseries. You can do this, for example using the following matlab lines:
R=kron(eye(8),ones(20,1));
save SessionsCovariates.mat R;
and then entering the resulting 'SessionsCovariates.mat' file as an additional first-level covariate (in the Setup->Covariates->First-level gui).
Hope this helps
Alfonso
ps. you may want to visit http://www.nitrc.org/projects/conn/ and use the forum there to post questions specific to the CONN toolbox in the future (and you could also download the latest version of the toolbox there, currently conn v.13i)
|