I am more confused than I thought. The MarsBar FAQ on the calculation of
percent signal change (psc) does not mention fitting FIR models. As I
understand it, psc is calculated by:
1) multiplying the basis function for a given event of interest (which is
extracted from the SPM.mat object) by it's corresponding beta value
2) finding the max height of the resulting estimated response to a single
event
3) dividing the max height by the beta for the intercept
4) multiplying the ratio by 100
The basis function and beta are taken from the spm model, which was
specified by convolving the canonical hrf with event onsets of 0 duration in
spm99. As I understand it, this model makes assumptions about the shape of
hrf. How else would it perform the convolution? So, it is not a FIR model.
Given this, why and how would the window size for averaging using FIR models
affect the psc estimate in this case?
They seem unrelated to me. Clearly, this is wrong. Please help me understand
why.
Thanks,
Steve
-----Original Message-----
From: Matthew Brett [mailto:[log in to unmask]]
Sent: Thursday, December 15, 2005 1:23 PM
To: Steven Lacey
Cc: [log in to unmask]
Subject: Re: [SPM] Brett's MarsBar and Poldrack's roi toolbox produce
different results
Hi,
On 12/15/05, Steven Lacey <[log in to unmask]> wrote:
> That's a good suggestion, thanks. However, I want to be clear on what
> is meant by 'default time windows across which the averaging takes
> place'. The window I have been changing unsuccessfully in MarsBar is
> the window used to estimate the height of the estimated response to a
> single event of a given type. MarsBar provides several options for
> this: max, max-min, ect. One option is to specify a window of time in
> which the signal estimates are averaged. In that application it
> doesn't seem to make sense to include any information about the
> response before the onset of the event. Are we talking about the same
> window?
No, I don't think so. The window you are referring to is one way to
calculate percentage signal change from a fitted event time course. For
example, for a window of 4 - 8 seconds, this will give you average signal
betwen 4 and 8 second after event onset:
http://marsbar.sourceforge.net/doc-devel/latest/marsbar/@mardo/event_signal.
html
The window that Daphne and Russ are referring to is the time over which the
FIR time course is calculated - the finite window of Finite
Impulse Response. You can change this in marsbar by changing the bin
length or number of bins in the dialog following the FIR event selection in
the GUI. For example, if your TR is 2, then marsbar will offer you a
default bin length of 2 (the TR) and no of bins = 13, to give you 26 seconds
of FIR time window.
Best,
Matthew
|