dmBLOCK basis function and percent signal change

AFNI Message Board
1

Dear AFNI experts,

Is it correct to interpret parameter estimates calculated using dmBLOCK (not normalized to a peak of 1 as peak is based on second parameter e.g., reaction time) as %signal change?

Thank you,
Ankit Parekh

2

With ‘dmUBLOCK(-X)’, you have the total control of deciding to which value ‘X’ you would like to center the modulation parameter. In other words, with the setting you want to interpret the first effect as the percent signal change associated with the parameter (e.g. reaction time) value of X.

3

Thank you Gang!
So if were to center the modulation parameter such that the maximum reaction time has a peak of 1, for each subject, I am in essence doing a normalization based on the reaction time to then interpret the beta weight as the %signal change?

4

if were to center the modulation parameter such that the maximum reaction time has a peak of 1, for each subject

I’m not so sure why you prefer to use the maximum reaction time as your benchmark for brain response interpretation. Also, you need to take into consideration the group analysis step: do you want to keep the same or different reaction time benchmark across subjects (which may have consequence at the group level)?

5

I understand now. Thank you Gang!
Last question - Say you have a mixed block-event design and the contrast is [1 -1]. For an individual subjects, are the parameter estimates for the difference (A-B) actually %signal change regardless of the scaling as long as its the same for the two conditions (dmBLOCK(0) or dmBLOCK(-X))? Of course, to be able to take this to group level, it would be necessary to have the same scaling factor across subjects right?

6

For an individual subjects, are the parameter estimates for the difference (A-B) actually %signal change regardless of the scaling
as long as its the same for the two conditions (dmBLOCK(0) or dmBLOCK(-X))? Of course, to be able to take this to group level,
it would be necessary to have the same scaling factor across subjects right?

For the events, use BLOCK(d) while trying dmUBLOCK(-X) for the varying blocks (not dmBLOCK(-X)). You have to make the decision as to which duration X of the blocks can be meaningfully compared to the events BLOCK(d). This is more a scientific (or contextual) issue than a modeling (or statistical) question.