Dear David,
Actually there are some event-related fMRI studies that rely on HPF settings like 1/64 Hz, although I can't provide any references right now.
Concerning the HPF, the DCT doesn't know whether it's signal or noise, it will just remove that part. In theory, you would thus determine the physical noise and then adjust the design accordingly so that there's no/hardly any signal within the low frequencies. For some experiments, this is not possible though, which then leads to the trade-off between adjusting the HPF (keeping more of the signal but also introducing more of the noise) or staying with the default settings (possibly losing major parts of the signal). With regard to physiological noise, there are high frequency and low frequency components, the HPF would remove just the low frequencies.
Another aspect, a more "conservative" HPF does not necessarily mean you will get better or more trustworthy results. The detrended predictors might also correlate with some higher-frequency noise within the data, which might have been less of an issue if they had not been detrended.
Best
Helmut
|