I have fixed this bug and found and fixed another bug having to do with pre-stim removal in FIR designs. I have also made some changes to assure that the GUI and the command-line versions of mkanalysis-sess give the same results.
For those in the Martinos Center, the changes are in dev and will be in stable on the next update (probably tomorrow).
For those outside of the Martinos Center, you can download a patch here:
ftp://surfer.nmr.mgh.harvard.edu/transfer/outgoing/flat/greve/fsfast.stable.patch.2009-04-17.tar.gz
To install:
cd $FREESURFER_HOME mv fsfast fsfast.backup tar xvfz fsfast.stable.patch.2009-04-17.tar.gz
There is a README file (included below) that describes the changes and how to fix your analyses.
Once again, sorry for the inconvenience. Let me know if you have any questions or other concerns.
doug
----------------------------------------------------------------------------------------------------------------------------------------------
This version was created on 4/17/09 as a patch to FreeSurfer stable 4.3.0. It fixes the FSFAST GUI bug as well as the RmPreStim bug (see below), and makes a few more changes to make the command-line and GUI compatible.
It can be obtained from:
ftp://surfer.nmr.mgh.harvard.edu/transfer/outgoing/flat/greve/fsfast.stable.patch.2009-04-17.tar.gz
To install:
cd $FREESURFER_HOME mv fsfast fsfast.backup tar xvfz fsfast.stable.patch.2009-04-17.tar.gz
CHANGES:
---------------------------------------------------------------- Below are changes to the mkanalyses-sess GUI:
Fixes the FSFAST GUI BUG in which unbalanced contrasts create in the mkanalysis-sess GUI were summed instead of averaged.
Changes the default intensity normalization target in analyses created with the GUI. The GUI previously used 100 (meaning that all the voxels were re-scaled so that the in-brain mean was 100). Now it uses 1000, which is the command-line (mkanalysis-sess) always used. This will only scale the values and will have no effect on significances.
Changes the default GUI TER when gammafit or spmhrf are selected. Previously, it set the TER=TR. Now it uses TR/20, which is the same as mkanalysis-sess. If the stimulus presentations are synched to the TR, then this has no effect. If the stimuli appear off-TR, then this will have an effect. This has no effect on FIR designs.
The GUI now has an "AutoStimDur" button, which is off by default. When AutoStimDur is used, it tells the analysis to automatically determine the duration of the stimulus based on the time from the stimulus onset to the onset of the next stimulus. In the prior version of the GUI, AutoStimDur was the default. Note that this has no effect when you use a 3- or 4- column paradigm file in which you explicitly specify the stimulus duration, nor does it have an effect on FIR designs.
To apply the above fixes, you can do one of two things depending upon whether you want to keep your old GUI-based analyses or not (eg, for comparison):
1) If you do NOT want to keep them, you can simpl:y
mkanalysis-sess -gui -analysis youranalysis
This will bring up your analysis. Just hit the "Save" button, then exit. Then re-run selxavg3-sess on all of your data.
2) If you DO want to keep them, you can simply:
mkanalysis-sess -gui -clone youroldanalysis -analysis yournewanalysis
This will bring up your analysis. Just hit the "Save" button, then exit. If you want to analyze youroldanalysis, you must set
setenv FSF_PROC_GUI_BUG 1
---------------------------------------------------------------- Below are changes to selxavg3-sess:
Fixes the "RmPreStim" bug in which a contrast is configured to subtract the pre-stimulus mean from the post-stimulus FIR values. This bug was present in both the GUI and command-line analyses. The bug was that RmPreStim had no effect. This fix takes effect when selxavg3-sess is run.
Douglas N Greve wrote:
IMPORTANT! PLEASE READ!
I am writing to let you know I have recently uncovered a bug in FSFAST. While I don't expect the bug to have widespread effects, the places where it does have an effect could be serious.
I have set up the following web page to document the bug and help track what is being done to asses its impact and to fix it:
surfer.nmr.mgh.harvard.edu/fswiki/FsFastGuiBug
The bug has to do with contrasts configured with the mkanalysis Graphical User Interface (GUI). The bug is particularly bad for unbalanced contrasts, eg, comparing A to the AVERAGE of B and C. The contrast actually would compare A to the SUM of B and C.
Please read the above wiki page for more details.
If you have not used the GUI, then you are unaffected.
I am currently developing tools to detect and fix these errors.
If you are (or think you might) affected by this bug, please email me as I'd like to know exactly how widespread the problems really are.
Finally, I apologize for the inconvenience that this will cause.
doug
ps. Thanks to Amitai Shenhav and Moshe Bar for bringing this to my attention.