BTW, here some old message by Torben showing different types of noise for different frequency domains
https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;651c37f7.00
However, in some cases it might be necessary to employ a design with longer blocks (well yes, often it can be avoided), and then the signal of interest might include low frequencies. I'm not sure whether it is better to remove the low frequencies then or rather go with a less aggressive filter (scanner noise should be distributed randomly, so maybe less of an issue for random effects analyses?).
Another interesting message by Torben can be found here
https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=SPM;37a33b0c.1304
The linked PDF shows the results when manipulating the block length, which varies from 8 to 120 s, meaning the cycle "Task Rest" ranges from 16 to 240 s. No high-pass filter is applied. For rather short blocks, the activations are resticted to left motocortex more or less. This is consistent with other papers stating that for standard event-related designs the high-pass filter is not that important in general. With a block length of 120 s lots of noise is evident.
However, what does this actually mean? It could also be an issue due to averaging (15 cycles for blocks of 8 s, 1 cycle for the longest blocks)? Besides, the longer the blocks the less appropriate the expected HRF might be. So what would happen if there are two or more cycles with long blocks (okay, this would also increase the DOF of course)? Some of the noise might be eliminated though. Just wondering :)
|