Dear Kelly,
I dare to share some of my thoughts since I have met similar difficulties in
a project.
I think the approach which you mention (FIR) should be suitable.
Alternatively (maybe closer to the HRF), you could move your recorded onsets
back [/forward - depending on which way you look at it] in time (e.g. shift
by -6 seconds) and then use a/the Fourier basis set to model the data. You
can specify the time period you would like to span with your regressor, i.e.
model as a single condition, and then e.g. 'bin' to your epochs of interest
at the inference stage -
which is where the mess starts: because of the mentioned variability of the
'HRF' across the brain, it is difficult to draw conclusions based on the
response shape (or time when it raises) etc. Also, classifying your bins is
a bit like ERP research - when does the response start to rise etc.... Also,
I was told the hemodynamic modelling (see button in SPM) should be preferred
over the 'simple' post stimulus time histograms... If your design allows,
the best approach would be DCM as I assume from the timing information you
might want to conclude which brain regions causes activity in another?
So long.
Hope this helps a bit,
Helmut
----- Original Message -----
From: "kelly Diederen" <[log in to unmask]>
Sent: Monday, December 18, 2006 6:19 PM
Subject: How to analyse brain activity prior to the onset of a modelled
event?
Dear all,
At present, we're running an experiment in which subjects indicate the on
and offset of a mental event by means of button presses. We're interested
in the brain activity during the event itself as well as a few seconds
prior to the onset. Preferably we would like to split the interval prior to
the onset into different blocks, for example -6 till -3 and -3 till 0.
We're interested in the way in which different brain regions activate
consecutively. This raises a problem concerning the fact that different
regions display different HRF's (to our knowledge up to 3 seconds).
This leads to two main questions:
1 Is anyone aware of a good overview of the differences in HRF's between
different regios?
2 Secondly, we wonder how researchers running this kind of analysis have
tried to overcome this problem.
In our opinion, this can be done in several ways: First, by making seperate
blocks for each time period prior to onset in the range of 4 - 7 seconds ,
that is long enough to distinguish between differences in HRF's between
different regions and changes in the signal due to the experimental
condition.
Secondly, by using temporal derivatives or finally by doing a FIR analysis.
The disadvantage of the first option is that you can only distinguish
between activity in the range of about 4 -7 seconds.
Do people have experience with these kinds of analyses or other types of
analyses suitable for our design or could someone perhaps give us some
useful tips?
Thanks in advance and kind regards,
Kelly Diederen
|