Hi,
There is no good theoretical reason for masking like this - it is a pragmatic choice.
We choose to only analyse voxels where we have valid contributions from all subjects, and hence our mask is setup to exclude voxels where one or more subjects do not have data. If you want to be able to analyse voxels with missing data then you need to setup the model appropriately, and that is somewhat tricky just in terms of the software, the bookkeeping behind the scenes, and making sure that the multiple comparison correction methods are too biased by effective changes in DOFs between voxels. If you really do want to analyse data in these areas then I would either (a) remove the subjects with the bad signal loss from the analysis, or (b) run the analysis in randomise with appropriate voxelwise regressors to account for the missing data (setup_masks can help you create these).
All the best,
Mark
> On 5 Feb 2017, at 19:46, Hallvard Røe Evensmoen <[log in to unmask]> wrote:
>
> When we have been running our fmri analyses in FEAT, we have observed that when we do second-level a group mask seems to be created which is based on where all the subjects have a voxel in the epi image. A challenge with this that we have experienced it is that if for example one of the subjects has an especially large signal loss in the entorhinal cortex due to susceptibility artifacts this will result in seemingly no signal change in that region for the different conditions, even though it indeed was relatively large signal changes in the entorhinal cortex as we observed when we increased the size of the group mask. My question is the reason why the analysis in FEAT is masked the way it is? Any feedback would be highly appreciated.
>
>
> Best regards
>
> Hallvard Røe Evensmoen
> Postdoctoral research fellow
> Department of Neuroscience
> Norwegian University of Science and Technology
|