Hello,
I am trying to calculate the percent signal changes for my monkey data. I did this by running the command selxavg3-sess and looking at the cespct values (or equivalently by using h.nii divided by h-offset.nii). I found in the mailing list that this value should be corrected by a scaling factor scalef if freesurfer-4.5.0 is used (see below).
If I apply this correction factor I got values that at first sight look reasonable but if i calcultate the psc using JIP (the imaging program from Joe Mandeville), i find different values. And looking at the raw data I have to conclude that the value from JIP are the correct ones...
For example if i look at the same voxel in MT if find: with JIP: -14.45% with FS: -8.15 % notice that the values are negative and pretty large due to the fact that we use and iron oxide contrast agent.
If, instead of applying the scaling factor scalef (defined below), I apply the factor max(x.X(:,1)) as a scaling factor (which according to me is the maximum value of the convolved design matrix,or 'the EV' in FSL lingo), then I end up with a value of -15.09%. Which is pretty close to the value obtained with JIP. I got the last idea from mumford.fmripower.org/perchange_guide.pdf
Do you have any clue where the differences come from? or if I am doing something wrong?
Also, one of my colleagues used both the gamma model and FIR model to calculate the percent signal change (psc) for his event-related designed experiment, and the values are also quite different. He applied the scaling factor for the values from the gamma model, and didn't do it for the fir model. What he got is 0.6% signal change if he takes the peak of the time course (which looks quite ok) that fitted from the FIR model and 1.5% signal change from the gamma model. It seems that there is more than 2 times differences between these two values. Is it normal, or we need another scale factor for the fir model?
Thanks for the answers!
regards, Thomas Janssens
ps: My selxavg3-sess analysis was done using the following parameters in the analysis.cfg file:
-gammafit 0 8 -gammaexp 0.3 -timewindow 40.00000000000000000000 -prestim 0 -acfbins 10 -fix-acf -autostimdur -no-fsv3-st2fir -no-fsv3-whiten -polyfit 2 -TER .10000000000000000000 -autowhiten 30 -nskip 0 -fwhm 0 -rescale 1000
----------------------------------------------------------------------------------------------
[Freesurfer] Calculating % Signal Change from roisummary-sess readout
Douglas N Greve Wed, 04 Aug 2010 14:25:52 -0700
That's right, though you may/will run into the problem of very small values in which case you can use this fix:
cd bold/analysis x = load('X'); tirf = x.flac0.ev(3).tirf; Xirf = x.flac0.ev(3).Xirf; scalef = sum(Xirf .* repmat(tirf,[1 size(Xirf,2)]));
scalef will be something like 184. Multiply the percent signal change you get by this number. BTW, this problem is fixed in version 5.
doug
Adam Nitenson wrote: Hi Freesurfers,
I am trying to analyze functional activity in individual subjects.The structural constraint is a dACC label and the functional constraint is positive activation at a specific timepoint (6 sec after stimulus). I was just wondering what the numbers following the "# of estimates per condition" row actually represented, and how I could convert these to % signal change. I saw in an archived email http://www.mail-archive.com/freesurfer@nmr.mgh.harvard.edu/msg00171.html that someone mentioned taking those values, dividing by the baseline offset value (row 4) and multiplying by 100 to get % signal change. Is this a valid method?
Adam Nitenson, B.S. Brain Genomics Lab Massachusetts General Hospital
----------------------------------------------------------------------------------------------
Thomas Janssens, PhD student Laboratorium voor Neuro- en Psychofysiologie K.U.Leuven Medical School Herestraat 49, B-3000 Leuven (Belgium) phone +32 16 33 00 35 cell phone +32 494 115 509