I am getting unexpectected results from smoothing.
I have 2 fMRI datasets (pre&post surgery) acquired 3 months apart.
Both datasets were acquired using the same paradigm and the same pulse sequence and the same timing, etc. They were acquired identically.
It's a block design. TR is 3.5 sec. The voxel size is 3.59mmx3.59mmx3.9mm. There are 120 scans consisting of 40 A, 40 B and 40 Control as follows:
10A 10 B 10Control 10A 10 B 10Control 10A 10 B 10Control 10A 10 B 10Control
When I analyze the unsmoothed data, SPM99 reports similar estimated smoothness for both sets of data:
5.1mmx5.0mmx5.3mm and 4.5mmx4.5mmx4.7mm
If I apply a filter kernel whose FWHM is 2 times the voxel size (7.2mmx7.2mmx7.8mm) and analyze the data, the estimated smootness reported by SPM99 is very different:
33.4mmx33.1mmx31.0mm versus 12.8mmx12.7mmx12.0mm
How could such a modest amount of smoothing (2 times the voxel size) cause such a large estimed smoothess in the data (9 times the voxel size) in one set of data, but much less in the other?
How does such a large disparity in smoothness affect my attempt to compare the pre&post surgery data?
Jerry Allison, Ph.D.
Medical College of Georgia
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|