Hello,
we have observed some rather odd effects of slice timing and corresponding match of microtime onset recently. It seems that in this case, the slice timing as such has a negligible effect, but changing the microtime onset has a drastic one, independently of slice timing as such.
Acquisitons parameters were TR = 2 s, TE = 30ms, 34 sices, acquisition order ascending interleaved, starting with even numbers. The data set was originally preprocessed in SPM5 with slice timing and the temporally middle slice (1) serving as reference. During model specification (epoch related design), microtime resolution was left at 16 to avoid oversampling, but microtime onset was set to 8 reflecting the temporal middle time bin.
The data sets later served as a control group in another study. Preprocessing was redone completely, this time in SPM8 without slice timing (and during model specification, standard microtime onset 1). Interestingly, results in the group analysis of the contrast of interest looked very different.
We therefore reran the preprocessing with slice timing in SPM8 to exclude effects due to different SPM versions, but differences remained. We then further examined the data. Our finding is that
a) Slice timing + microtime onset 8 leads to the same results as no slice timing + microtime onset 8
b) No slice timing + microtime onset 1 leads to the same results as slice timing + microtime onset 1
c) Results of a) and b) are different, that is some significant clusters in a) are gone in b) (not only a little smaller and subthreshold by chance).
Now the idea of changing the microtime onset is to account for effects of slice timing. But in our case, effects of modified microtime onset are independent of the slice timing process. This means that, at least in this example, changing the microtime onset is not justified.
Has anyone noted similar problems? What exactly happens when changing the microtime onset?
Thanks for comments,
Helmut
|