Posted by Hillary Schaefer on March 13, 2001 at 17:16:38:
Hello all-
we are running 3dDeconvolve to separate out responses to rapidly-presented stimuli in an event-related design. The general design is: a circle or minus "warning" sign predicts a neutral or negative picture, respectively, that is presented a few seconds later. This pattern occurs every 20 seconds and is repeated about 35 times. So there are a total of four impulse responses to estimate, specified by 4 1D files. We estimate the impulse response function for 10 seconds for the "warnings" and for 15 seconds for the pictures. The deconvolution ran, but we saw two issues in the estimated response functions that we were concerned about:
1. The slices were aquired coronally. If you look at an orthogonal slice, the image has a striped appearance, that is, the slices alternate between lighter and darker values. The "tshift" option did not correct this.
2. The estimated impulse response functions have the noise baseline removed, and that puts the values very low- typically less than 100, when the epi values that were inputted are about 15000. Is there any way to figure out what noise baseline was used for each voxel and add it back in? This would be desirable for analyses we plan to do on the data.
Otherwise, the results seem to "make sense" - that is, we get a hemodynamic shape in areas where we would expect it.
Has anyone else seen this "striped" pattern or know how it would come about? And is there any way to add the baseline back in? Any help would be greatly appreciated. Thanks,
Hillary Schaefer