Hi - good questions:
On Thu, 15 Jan 2004, Vikas Rao wrote:
> Hello everyone, I have a few questions about extracting trial-averaged
> timeseries data using FSL. Basically, I am trying to extract HRFs from
> subjects using a checkerboard task.
>
> 1) I was just wondering if the timeseries output of Featquery
> (mean_mask_ts.txt) takes into account the filtering and prewhitening
> that occurs in the feat glm.
Featquery just extracts time series from filtered_func_data which is the
input to the FILM GLM stats and therefore has not been prewhitened at that
stage. (All other filtering such as highpass temporal filtering _has_
already been applied in the creation of filtered_func_data.)
> 2) Should I be using tsplot instead, would it take these issues into
> account?
It's up to you really, as to whether you want to extract filtered or
filtered&whitened time series for your later analysis. You could probably
argue that you really want just filtered time series as that is the
_data_. tsplot uses whitened data for the plotting largely because if it
didn't, the "fitted time series" wouldn't appear to fit the "data".
> 3) How can I use tsplot over a whole roi instead of a single voxel?
> Why would I want to weight my extracted timeseries by z-value (as
> tsplot does by default)? That seems like it would clean up the
> timecourse alot, but unnaturally so and eliminate my ability to compare
> HRFs of different subjects.
You may well not want to weight them in this way - I like this in tsplot
because it reduces the impact of arbitrary thresholding in the definition
of the clusters used and seems sensible - but note you can easily turn
this off on the command line call to tsplot (-n).
> 4) I have been trying to find activity in my task by using a set of 3
> sin and 3 cos waves (harmonics, the lowest frequency one being 18s).
> If I use all 3 harmonics of the 18s sin, I get the linear combination
> error with a ratio on the order of 10^-2. My understanding is that a
> ratio at that level is ok, and that I can safely continue. Is there a
> better way to do this, perhaps using a gamma basis set ?
Yes, this ratio is fine. (There are complicated reasons why we don't just
set the threshold more liberal.) The choice of basis function shapes is up
to you, but in general people seem to use gammas when it is the HRF that
they want to allow to vary, and sines when it is the (longer-term)
response over a "block" that they want to allow to vary (often in visual
mapping experiments).
Cheers, Steve.
Stephen M. Smith DPhil
Associate Director, FMRIB and Analysis Research Coordinator
Oxford University Centre for Functional MRI of the Brain
John Radcliffe Hospital, Headington, Oxford OX3 9DU, UK
+44 (0) 1865 222726 (fax 222717)
[log in to unmask] http://www.fmrib.ox.ac.uk/~steve
|