Hi Doug,
We've been running a regression analysis on our subjects, and the results are quite puzzling as they don't seem to sync up with the individual pieces of the regression. The data is taken from the Sternberg memory paradigm, with subjects memorizing groups of 1, 3, 5 and 7 letters. We have looked at imaging data of 1vFix, 3vFix, 5vFix, 7vFix, and then our regression analysis, in theory, we should see areas of significant linear pattern when moving from the activation between 1, 3, 5, and 7 letters. However, in our latest analyses, the regression maps have not been complimenting some of the individual subjects' 1,3,etc vs Fix data. In the attached PDF, there are 2 subjects with their 1vFix, 3vFix, etc maps shown, as well as the regression map in the middle, and as you can see, particularly for the 2nd subject, the regression map just doesn't make sense considering the maps from which it was derived.
The third sheet of the excel file plots the ROI values generated from the regression analysis against a calculated slope derived from the funcroi values from 1, 3, 5, and 7 vs Fix. The two outliers in the top left are the subjects we further analyzed in the attached powerpoint (which includes ROI derived values).
Another strange occurance is that when making regression maps for sub-groups of approx. 8 subjects, the maps don't seems to match up well with the numbers, but the regression map of all of our subjects together looks great, yet the average ROI derived value is quite low.
We've been searching for a few days for a "stupid mistake" that could have caused this, but nothing is popping out. I'm attached the analysis info for the regression analysis as well as an example paradigm file for that analysis (slopepar). We just are not sure why the regression maps and individual subject ROI values are so low consider the rather strong regression effect implied from simply eyeballing the data. I understand that the maps may not always be as strong as anticipated, but they should certainly not show dark negative blue when the pattern is positive.
Examples of group commands:
isxconcat-sess -sf /cluster/roffman/users/Stable5_PerRun/Subject_Files/9_ActiveFolate_Post_5T -d /cluster/roffman/users/Stable5_PerRun -analysis SIRP_LoadRegression_Stable5 -c Cond2vFix -o /cluster/roffman/users/Stable5_PerRun/August_2011/9_ActiveFolate_Post_5T_TEST
"mri_glmfit --y /cluster/roffman/users/Stable5_PerRun/August_2011/9_ActiveFolate_Post_5T_TEST/SIRP_LoadRegression_Stable5/Cond2vFix/lh.ces.nii --fsgd /cluster/roffman/users/fsgd/9_ActiveFolate_Post_5T.fsgd --osgm --glmdir /cluster/roffman/users/Stable5_PerRun/August_2011/9_ActiveFolate_Post_5T_TEST/SIRP_LoadRegression_Stable5/osgm/LH2vFix --surf fsaverage lh --fwhm 4.6
Adam Nitenson, B.S. Brain Genomics Lab Massachusetts General Hospital nitenson@nmr.mgh.harvard.edu Phone: 617-643-3215