Dear Natalie,
given the peak coordinate (independently of the number of subjects within that study) the Anatomy toolbox assigns some probability for being BA x, BA y, ... a low probability might thus result from activations "too far away" but might also reflect limitations of the probabilistic maps (seem to be based on 10 subjects only; different normalisation routines during preprocessing might lead to deviations). As the authors provide MNI coordinates and show brain slices, I would rely on these information to assess whether the label makes sense or not.
I guess the reason for an HPF of 400 s was that one sequence of the block-design (as displayed in fig. 1) is 200 s, which should also correspond to the average interval between two subsequent blocks of the same type (there were three repetitions within a session). According to Nyquist theorem they used 2x 200 s, thus 400 s. Personally I regard it as useful, but note that usage of the Nyquist theorem in this context is disputed (see older mails on that issue).
Just to make sure, in a standard design with short blocks/events and short intervals between trials of the same type the expected time course does not contain much information within the low frequency range. However, there is scanner noise within this range. With an HPF of 128 s (which is derived from such a standard design, and the exact value 128 seems to have to be chosen since it's 2^7) you remove the slow frequencies (the noise) from both the expected and the measured time course.
If you have a non-standard design with long blocks / long intervals, then there might be quite some meaningful signal within the low frequency range. When using an HPF with 128 you will still remove the noise effectively, but also relevant signal. When adjusting the HPF, you might keep the relevant signal, but include the noise at the same time. I think in designs like this one it might be better to go with the latter and "hope" that the noise does not correlate with the expected time courses. Unfortunately there's not much literature about HPFs for long blocks / long intervals. Instead of an HPF one could also try and add some regressors which change linearly, quadratically, ... over time to account for temporal drifts.
Best,
Helmut
|