Hi Doug,
As you suggested, I ran mc simulation data with a threshold taken from fwhm.dat so that the smoothing on the simulation matches the smoothing on the real data. What I was wondering is, if I use the same simulation data for several contrasts (they have the same fsgd, so this seems like a valid thing to do), it becomes less clear how to choose the smoothing. Or maybe it is not possible to use the same mc simulation data, because the smoothing turns out to be different. Should I instead use an average of the smoothing values, or maybe use the lowest one? Conceptually it seems like this might make the results less accurate, but in practice is there a certain level of slack in smoothing without sacrificing accuracy? If not, it makes a big difference, because the full simulation will take 6 days instead of 2. The z simulation, which at at least with the level of smoothing we have applied, showed results being much less significant, which is why I'm trying the full mc simulation.
Thanks, Rob
Just to give you an idea of the variance in the smoothing reported in fwhm.dat for the three contrasts: lh.ASvfix_8s_BwGroupError5.fsfast.glmdir/fwhm.dat 17.128043 rh.ASvfix_8s_BwGroupError5.fsfast.glmdir/fwhm.dat 16.675573
lh.ASevASc_8s_BwGroupError5.fsfast.glmdir/fwhm.dat 15.202638 rh.ASevASc_8s_BwGroupError5.fsfast.glmdir/fwhm.dat 15.771982
lh.ASevfix_8s_BwGroupError5.fsfast.glmdir/fwhm.dat 14.438915 rh.ASevfix_8s_BwGroupError5.fsfast.glmdir/fwhm.dat 15.766636
If the simulation command-line would be the same across the different contrasts and the contrast matrices have the same number of rows, then you do not have to redo the simulation. Why is the smoothing changing if you are using the same fsgd file?
Robert Levy wrote:
Hi Doug,
As you suggested, I ran mc simulation data with a threshold taken from fwhm.dat so that the smoothing on the simulation matches the smoothing on the real data. What I was wondering is, if I use the same simulation data for several contrasts (they have the same fsgd, so this seems like a valid thing to do), it becomes less clear how to choose the smoothing. Or maybe it is not possible to use the same mc simulation data, because the smoothing turns out to be different. Should I instead use an average of the smoothing values, or maybe use the lowest one? Conceptually it seems like this might make the results less accurate, but in practice is there a certain level of slack in smoothing without sacrificing accuracy? If not, it makes a big difference, because the full simulation will take 6 days instead of 2. The z simulation, which at at least with the level of smoothing we have applied, showed results being much less significant, which is why I'm trying the full mc simulation.
Thanks, Rob
Just to give you an idea of the variance in the smoothing reported in fwhm.dat for the three contrasts: lh.ASvfix_8s_BwGroupError5.fsfast.glmdir/fwhm.dat 17.128043 rh.ASvfix_8s_BwGroupError5.fsfast.glmdir/fwhm.dat 16.675573
lh.ASevASc_8s_BwGroupError5.fsfast.glmdir/fwhm.dat 15.202638 rh.ASevASc_8s_BwGroupError5.fsfast.glmdir/fwhm.dat 15.771982
lh.ASevfix_8s_BwGroupError5.fsfast.glmdir/fwhm.dat 14.438915 rh.ASevfix_8s_BwGroupError5.fsfast.glmdir/fwhm.dat 15.766636
Hi,
That's a good question. I have no idea why changing only the contrast (and changing nothing anatomical) should result in the smoothing being different. I also learned that previous research assistants in our lab encountered the same exact issue but they didn't solve it. My best guess is it is that some randomness that results from average7 being incompatible with an operation along the stream, either in stxgrinder, mris_preproc, or mri_glmfit.
Thanks, Rob
Doug Greve wrote:
If the simulation command-line would be the same across the different contrasts and the contrast matrices have the same number of rows, then you do not have to redo the simulation. Why is the smoothing changing if you are using the same fsgd file?
Robert Levy wrote:
Hi Doug,
As you suggested, I ran mc simulation data with a threshold taken from fwhm.dat so that the smoothing on the simulation matches the smoothing on the real data. What I was wondering is, if I use the same simulation data for several contrasts (they have the same fsgd, so this seems like a valid thing to do), it becomes less clear how to choose the smoothing. Or maybe it is not possible to use the same mc simulation data, because the smoothing turns out to be different. Should I instead use an average of the smoothing values, or maybe use the lowest one? Conceptually it seems like this might make the results less accurate, but in practice is there a certain level of slack in smoothing without sacrificing accuracy? If not, it makes a big difference, because the full simulation will take 6 days instead of 2. The z simulation, which at at least with the level of smoothing we have applied, showed results being much less significant, which is why I'm trying the full mc simulation.
Thanks, Rob
Just to give you an idea of the variance in the smoothing reported in fwhm.dat for the three contrasts: lh.ASvfix_8s_BwGroupError5.fsfast.glmdir/fwhm.dat 17.128043 rh.ASvfix_8s_BwGroupError5.fsfast.glmdir/fwhm.dat 16.675573
lh.ASevASc_8s_BwGroupError5.fsfast.glmdir/fwhm.dat 15.202638 rh.ASevASc_8s_BwGroupError5.fsfast.glmdir/fwhm.dat 15.771982
lh.ASevfix_8s_BwGroupError5.fsfast.glmdir/fwhm.dat 14.438915 rh.ASevfix_8s_BwGroupError5.fsfast.glmdir/fwhm.dat 15.766636
Oh, sorry, you mean the lower-level contrast? Yes, that can definitely cause difference in smoothness at the higher level because it is a totally different input.
Robert Levy wrote:
Hi,
That's a good question. I have no idea why changing only the contrast (and changing nothing anatomical) should result in the smoothing being different. I also learned that previous research assistants in our lab encountered the same exact issue but they didn't solve it. My best guess is it is that some randomness that results from average7 being incompatible with an operation along the stream, either in stxgrinder, mris_preproc, or mri_glmfit. Thanks, Rob
Doug Greve wrote:
If the simulation command-line would be the same across the different contrasts and the contrast matrices have the same number of rows, then you do not have to redo the simulation. Why is the smoothing changing if you are using the same fsgd file?
Robert Levy wrote:
Hi Doug,
As you suggested, I ran mc simulation data with a threshold taken from fwhm.dat so that the smoothing on the simulation matches the smoothing on the real data. What I was wondering is, if I use the same simulation data for several contrasts (they have the same fsgd, so this seems like a valid thing to do), it becomes less clear how to choose the smoothing. Or maybe it is not possible to use the same mc simulation data, because the smoothing turns out to be different. Should I instead use an average of the smoothing values, or maybe use the lowest one? Conceptually it seems like this might make the results less accurate, but in practice is there a certain level of slack in smoothing without sacrificing accuracy? If not, it makes a big difference, because the full simulation will take 6 days instead of 2. The z simulation, which at at least with the level of smoothing we have applied, showed results being much less significant, which is why I'm trying the full mc simulation.
Thanks, Rob
Just to give you an idea of the variance in the smoothing reported in fwhm.dat for the three contrasts: lh.ASvfix_8s_BwGroupError5.fsfast.glmdir/fwhm.dat 17.128043 rh.ASvfix_8s_BwGroupError5.fsfast.glmdir/fwhm.dat 16.675573
lh.ASevASc_8s_BwGroupError5.fsfast.glmdir/fwhm.dat 15.202638 rh.ASevASc_8s_BwGroupError5.fsfast.glmdir/fwhm.dat 15.771982
lh.ASevfix_8s_BwGroupError5.fsfast.glmdir/fwhm.dat 14.438915 rh.ASevfix_8s_BwGroupError5.fsfast.glmdir/fwhm.dat 15.766636
freesurfer@nmr.mgh.harvard.edu
-
Doug Greve -
Robert Levy