Hi all

I have some doubts about the percent signal change normalization procedure of the beta coefficient images in multi-subject studies.

First of all, why do we want to do it?

The most obvious reason for me would be to correct for unpredictable automatic adjustments of the scanner's gain setting. In this case we want to remove the variance associated with it, which has nothing to do with the intrinsic amplitude of the neural response to a stimulus. However, I would assume this gain confound to be spatially uniform across the brain, so that a normalization of the estimated beta coefficients should be done with the whole-brain grand mean (average of all in-brain voxels across all time points of a scanning run), and not performed voxel-wise with the temporal average (or the constant coefficient beta_zero in the fitted model) of that voxel, which would assume that the gain varies across voxels.

If, instead, the main reason for computing voxel-wise percent change is not the scanner's gain but the very spatial variation of the baseline across the brain, it appears to me that we are adopting the VERY strong assumption that the amplitude of the HRF to a stimulus scales linearly with the baseline value. I would be interested to know if there is any published data on that....

Finally I would like to point out that this issue seems all the more critical in studies where the spatial variation of the baseline value across the brain is important (e.g. in studies of the "resting state network"), and may be expected to differ across different subjects populations (e.g., young vs. elders). In this case, you wouldn't want to remove it, if you were testing for differences across the two populations...

Any comments on the above would be deeply appreciated...

thanks!

giuseppe