Dear Freesurfer community:
I have some (perhaps silly, but I think fundamental) questions about contrasts in functional analyses:
1. When wanting to see what regions are activated by exposure to a certain stimulus (blocked-design), what is the exact stats that is happening voxel-wise:
(i) correlation between the time series of the HRF-convoluted BOLD and that of the behavioural stimulus that was used, or (ii) t-test between the BOLD signal during the stimulus and the BOLD signal during a resting baseline
If ii) is the answer, then is "baseline" taken as an average of all non-stimulus (rest) periods in the scan, i.e. the "zero" periods in the block-design boxcar diagram?
2. When wanting, on the other hand, to see what regions are active in condition A but not in condition B (so a comparison this time), is it the case that i) is no longer a choice and a t-test always has to be used between condition A and condition B? If so, is the stimulus-vs-baseline situation from the previous question just a special case of this, where condition B = (passive) baseline?
3. Is it the case that a contrast probing the difference between two groups is using independent samples t-tests, whereas one probing the within-subject difference between two time-points is using paired-samples t-tests?
Thank you for your help :-)
--Francesco
On 03/15/2014 06:35 PM, Francesco Puccettone wrote:
Dear Freesurfer community:
I have some (perhaps silly, but I think fundamental) questions about contrasts in functional analyses:
- When wanting to see what regions are activated by exposure to a
certain stimulus (blocked-design), what is the exact stats that is happening voxel-wise:
(i) correlation between the time series of the HRF-convoluted BOLD and that of the behavioural stimulus that was used, or (ii) t-test between the BOLD signal during the stimulus and the BOLD signal during a resting baseline
If ii) is the answer, then is "baseline" taken as an average of all non-stimulus (rest) periods in the scan, i.e. the "zero" periods in the block-design boxcar diagram?
Neither. A GLM is set up with regressors for you conditions and nuisance variables. The GLM is fit to the time course to compute a regression coefficient for each regressor. This is a multivariate operation and does not reduce to a simple correlation. For a task regressor, the regression coefficient indicates the amplitude of the HRF. This amplitude is tested with a t-test
- When wanting, on the other hand, to see what regions are active in
condition A but not in condition B (so a comparison this time), is it the case that i) is no longer a choice and a t-test always has to be used between condition A and condition B? If so, is the stimulus-vs-baseline situation from the previous question just a special case of this, where condition B = (passive) baseline?
This would be a conjunction analysis that does not reduce to a simple contrast.
- Is it the case that a contrast probing the difference between two
groups is using independent samples t-tests, whereas one probing the within-subject difference between two time-points is using paired-samples t-tests?
Between groups is independent two-sample. Within subject uses a GLM and does not reduce to a simple paired formula (but that is basically the right way to think about it) doug
Thank you for your help:-)
--Francesco
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
freesurfer@nmr.mgh.harvard.edu
-
Douglas N Greve -
Francesco Puccettone