It seems that TR/2 should be subtracted, not added.
If EPI slice timing is shifted to TR/2 (more specifically to nslices*(nslices-1)*TR/2 in the case of alt+z, for example), then for use in AFNI’s 3dDeconvolve or 3dREMLfit, TR/2 should be subtracted from the event times. If the EPI timing happens later, then relatively, the stimulus events would happen earlier.
For example, suppose TR=2s and the first stimulus event happens at time t=1s. After 3dTshift, it would be as if the original volume were acquired at time t=1s, corresponding with that initial event. So from the perspective of 3dDeconvolve, that event should be adjusted to time t=0 to still correspond with the initial EPI volume.
Thanks,
– rick
rick reynolds9 months
Aaaaand I mixed up 2 things in my mind when writing the average, which should divide by nslices, not multiply.
(nslices-1)*TR / (2*nslices)
Sorry to further any confusion.
– rick
Jeanette Mumford9 months
Good call, thanks! Will repair to reflect this and also the code error that I made in the last plot.
It seems that TR/2 should be subtracted, not added.
If EPI slice timing is shifted to TR/2 (more specifically to nslices*(nslices-1)*TR/2 in the case of alt+z, for example), then for use in AFNI’s 3dDeconvolve or 3dREMLfit, TR/2 should be subtracted from the event times. If the EPI timing happens later, then relatively, the stimulus events would happen earlier.
For example, suppose TR=2s and the first stimulus event happens at time t=1s. After 3dTshift, it would be as if the original volume were acquired at time t=1s, corresponding with that initial event. So from the perspective of 3dDeconvolve, that event should be adjusted to time t=0 to still correspond with the initial EPI volume.
Thanks,
– rick
Aaaaand I mixed up 2 things in my mind when writing the average, which should divide by nslices, not multiply.
(nslices-1)*TR / (2*nslices)
Sorry to further any confusion.
– rick
Good call, thanks! Will repair to reflect this and also the code error that I made in the last plot.