 |
 |
Hi Torben and Helmut, Thank you for the info. I re-ran the analysis 3 different ways. The following reports are for 1 subject. The results for this 1 subject are consistent across all subjects. The first thing I note is that the results largely look the same across all 3 implementations (phew!). However, there are some voxels that pass threshold for a particular implementation and don't in another implementation. Interestingly, in order of most to least statistical power, it seems like: Changed_T0 (which was what I did initially), default, and then default_Torben. The power is not just stronger for the most ventral slices. There are also more voxels that pass threshold in the dorsal slices. I'm not sure what to make of this. Any ideas? My study design is 36 slices in ascending order, acquired axially. TR=2. Changed_T0.ps: T=36, T0=1, Regular Timing File Default.ps: T=36, T0=18, Regular Timing File Defaults_Torben.ps: T=36, T0=18, Timing File-= TR/2. Much thanks, Rita On Tue, Jun 16, 2015 at 8:46 AM, H. Nebl <[log in to unmask]> wrote: > Dear Rita, > > In SPM8 and earlier versions the default settings during model > specification were microtime onset 1 and microtime resolution 16, meaning > the predictors based on the actual onsets (= no shift applied) would > coincide with the beginning of the TRs. This works very well for the first > slice acquired, but the last slice acquired would be mismodeled by (almost) > one TR. Even if you do not rely on slice timing the temporal misfit across > all the slices would be smaller if you set the microtime onset to the > middle time bin = microtime reolution * 1/2, e.g. 8. The predictor would > then be optimal for the slice acquired in the middle of the TR, and max. > misfit would be +/- 1/2 TR. Accordingly, the default microtime onset in > SPM12 has been adjusted to reflect the middle time bin of the default 16 > time bins. > > Thus the default settings in SPM12 should be perfect for no slice timing > and for data that was slice-time corrected onto the temporally middle > slice. If you want the predictor to reflect the timing of a certain > different slice it's probably better to work with microtime onsets and > resolutions instead of shifting the trial onsets. Shifting the trial onsets > is equivalent to choosing a certain time bin and should result in very > similar results as long as you use the correct settings, but shifting is > more prone to errors (shifting into the wrong direction). > > > the former analyses actually gives me substantially more signal > Do you mean larger beta estimates or more significant findings? In any > case, microtime resolution 1 would indeed be a better fit for voxels in the > first quarter of slices acquired, while it would be worse for the orther > three quarters. With descending/ascending acquisition schemes it might > indeed be the case that some regions show up more consistently if they were > covered during the first quarter of the TR (although data is interpolated > between different slices several times during the preprocessing). However > and especially if you obtain "more results" across the whole brain, then > it's a global effect, pointing to your regions reaching their peak earlier > or later than under the usual assumptions. > > Best > > Helmut > >