 |
 |
Dear Cesar Sorry for the late reply. You are right that it is a simple thing to do, what makes it slightly complicated is that is not possible to set this via the otherwise fantastic batch interface. But maybe you can even make your script as a processing step in the batch :-) Best Torben Den Uge:27 04/07/2011 kl. 17.29 skrev Cesar Caballero: > Dear Torben, > > Thanks very for your answer. I am actually more interested in whether there would be an effect in the statistics at the second-level random effect analysis. > My understanding is that the the contrast images used at the second level are not biased and therefore no problem, is that right? > > Furthermore, SPM.xX.iG is the field indicating the columns corresponding to the covariates of no-interest. As you indicate, this field is by default set as empty in the SPM structure defined at the first level. Thus, these columns are considered to build the F-contrast that selects the relevant voxels for the non-sphericity estimates (see spm_spm from line 445). I guess it is possible to modify the SPM structure before estimation and indicate these columns in the xX.iG field. Before creating an script... I am wondering whether this is possible with the GUI? > > Thanks very much, > Cesar > > > On Jul 4, 2011, at 12:02 PM, Torben Ellegaard Lund wrote: > >> Dear Cesar >> >> If the AR model makes sense it makes sense also in the case of nulling regressors. The situation does however get complicated by the fact that the AR model is estimated collectively for a group of voxels located in a mask where there was a significant effect of your regressors of interest. Unfortunately SPM by default considers all regressors except for the mean to be of interest. When you have a nulling regressor you are very likely to get the whole brain in that mask and not just the voxels where your activation is present. This will bias the p-values at the single subject level. The solution to this is to define the nulling regressors as covariates of no interest (see previous posts about this). >> >> Best >> Torben >> >> >> Torben Ellegaard Lund >> Associate Professor, PhD >> The Danish National Research Foundation's Center of Functionally Integrative Neuroscience (CFIN) >> Aarhus University >> Aarhus University Hospital >> Building 10G, 5th floor, room 31 >> Noerrebrogade 44 >> 8000 Aarhus C >> Denmark >> Phone: +4589494380 >> Fax: +4589494400 >> http://www.cfin.au.dk >> [log in to unmask] >> >> >> >> >> >> >> >> Den Uge:25 23/06/2011 kl. 00.20 skrev Cesar Caballero: >> >>> Dear SPMer's >>> >>> I am dealing with some datasets with numerous scans corrupted by motion and artefacts. >>> I have decided to null these scans by using specific 1s regressors (nulling regressors) instead of remove these scans from the analysis. >>> In that case, does it make sense to use an autoregressive process ( AR(1) ) for the serial correlations of the residuals? >>> >>> Thanks in advance for your recommendations, >>> Cesar >>> >>> >>> Cesar Caballero Gaudes >>> Hôpitaux Universitaires de Genève >>> CIBM / Radiologie >>> Rue Gabrielle-Perret-Gentil 4 >>> (ex-rue Micheli-du-Crest 24) >>> CH-1211 Genève 14 >>> Switzerland >>> >>> [log in to unmask] >>> Phone: (+41) 0223725217 >> >> >> >> >> >> > > Cesar Caballero Gaudes > Hôpitaux Universitaires de Genève > CIBM / Radiologie > Rue Gabrielle-Perret-Gentil 4 > (ex-rue Micheli-du-Crest 24) > CH-1211 Genève 14 > Switzerland > > [log in to unmask] > Phone: (+41) 0223725217 > >