[VoxBo] Question about FIR analysis for event-related designs
Charan Ranganath, Ph.D.
cranganath at ucdavis.edu
Sun Jun 8 18:19:26 EDT 2008
> As a follow up, it would seem from your comments that not generating an
> FIR set for conditions that one is not directly interested in would not
> create an accurate model, which makes sense.
Well if the conditions that you are not interested in are still sources
of task-related neural activity, then you could model those by
convolving the covariates with an HRF instead of using the FIR set for
those conditions.
> My understanding of your description, then, is that I should be
> generating different FIR sets depending on the duration of the events in
> my experiment. To make this more concrete, the TR for the entire study
> is 3s; the experiment includes 4 different types of events with various
> durations: condition A = 12s (4TRs), condition B=12s (4 TRs), perceptual
> baseline=6s (2 TRs), and null events=3s (1 TR). Is there a "right" way
> to generate the appropriate numbers for each of the FIR sets? For
> instance, would 19-20 points for my 12s events, 15 points for my 6s
> events, and 10 points for my 3s events be correct?
I don't think you should need to use an FIR for the null events, unless
these actually represent an active condition that needs to be modeled
separately from activity during the ITI.
As far as how many points to include, here is a rule of thumb: try
convolving the event with an HRF and see how many TRs it takes for the
convolved event to return to baseline. That should be a good starting
point, and you can then add on 1-2 extra TRs if you want to play it
safe. Given that you have a 3s TR, I am guessing that you can go with
less than 19-20 points for conditions A and B and less than 15 points
for the perceptual baseline. For example, you can probably do 8 or 9
points for the 6s events (this should cover 24-27s which should be
plenty to cover a BOLD response to a 6s neural event).
> Also, if the above is correct, how can one set up contrasts between
> conditions that do not have the same number of points (e.g., an
> experimental condition with 20 points, vs. a perceptual baseline with 15
> points)? Should the results be transformed somehow or should the points
> match?
Making inferences with basis sets is always tricky. One thing you can do
is simply contrast the time point that would be expected to represent
the peak of the HRF for each condition. Another is to take a set of
points from condition A and contrast them against a set from condition
B. So lets say you have 20 points for one condition and 15 points with
another. Then you could, say weight timepoints 4-8 as +3 and then weight
timepoints 4-6 for the baseline as -5.
I must admit that we haven't done that much with FIR sets, so this is
all theoretical. You might want to check with someone who has more
practical experience with them. But I hope this helps.
CR
>
> I'm sorry for all the additional questions, but I just want to make sure
> I'm doing this right.
>
> Thanks so much again for your help-
>
> Best,
>
> ~Lila
>
> On Fri, Jun 6, 2008 at 6:35 PM, Charan Ranganath <cranganath at ucdavis.edu
> <mailto:cranganath at ucdavis.edu>> wrote:
>
> Lila, I'm not sure I understand your question because I don't know
> the duration of your TR and what you're exactly going for. But my
> initial hunch is that this isn't the way that you want to do this.
> In a traditional analysis, you have a model specifying the onset and
> duration of NEURAL activity that is convolved with a hemodynamic
> response. The FIR set is designed to model the HEMODYNAMIC response
> to an event, based on the onset of a neural event, but without
> strict assumptions about the duration of neural activity.
>
> A typical HRF lasts 16-20s. So let's say that you have a 2s TR, an
> event that is short (~2s) could be modeled with an FIR set with 8-10
> points. For a longer event (~8s), you'd want to use more points (say
> 15?).
>
> Regarding whether you can include other covariates that are not
> modified w/an FIR set, that is fine. But if the other covariates are
> not modeled in a realistic way, they might either be useless or suck
> up variance that they shouldn't.
>
> Re your last question, I believe voxbo sets up a 0 covariate to
> model the onset of the event (t=0), whereas the other covariates
> model subsequent TRs.
>
> Hope this helps. CR
>
> Lila Chrysikou wrote:
>
> Hello everyone,
>
> I am trying to analyze data from an event-related design using
> an FIR analysis. I am interested in examining differences
> between two experimental conditions relative to a perceptual
> baseline task. My experimental conditions have a duration of
> 4TRs, whereas my Baseline has a duration of 2TRs.
> While setting up my GLMs, a couple of questions came up:
>
> 1) After introducing the diagonal set as a condition of
> interest, I have selected each experimental condition and the
> perceptual baseline separately and subsequently modified each
> selection with the appropriate TR for each covariate type (i.e.,
> 4 TRs for each experimental condition, 2TRs for the baseline).
> Is that correct?
>
> 2) If my model includes other covariates for which I am not
> interested regarding the FIR, can I leave them as is?
>
> 3) I noticed that voxbo automatically generates an additional
> covariate (0); what does this covariate correspond to? Should
> one instead enter the FIR TRs as n-1 to account for this?
>
> I haven't been able to find much information on FIR analysis on
> the mailing list; if anyone has additional
> details/advice/recommended readings that wouldn't mind sharing,
> it would be greatly appreciated.
>
> Many thanks in advance,
>
> ~Lila
>
> --
> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> Evangelia G. Chrysikou, Ph.D.
> Post Doctoral Research Fellow
> Thompson-Schill Lab
>
> Center for Cognitive Neuroscience
> University of Pennsylvania
> Office address: 3810 Walnut St.,Room 307, Philadelphia, PA 19104
> 215. 573.6726 (phone)
> 215. 898.1982 (fax)
> E-mail: evangelg at psych.upenn.edu
> <mailto:evangelg at psych.upenn.edu>
> <mailto:evangelg at psych.upenn.edu <mailto:evangelg at psych.upenn.edu>>
>
>
> --
> Charan Ranganath, Ph.D.
> Associate Professor
> Center for Neuroscience and Dept. of Psychology
> University of California at Davis
> 1544 Newton Ct.
> Davis, CA 95618
>
> phone: 530-757-8750
> fax: 530-757-8640
> http://entorhinal.ucdavis.edu/~evan/index.php
> <http://entorhinal.ucdavis.edu/%7Eevan/index.php>
>
> [PLEASE NOTE THAT OUR NORMAL LAB URL (DynamicMemoryLab.org) IS NOT
> WORKING]
>
>
>
>
> --
> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> Evangelia G. Chrysikou, Ph.D.
> Post Doctoral Research Fellow
> Thompson-Schill Lab
>
> Center for Cognitive Neuroscience
> University of Pennsylvania
> Office address: 3810 Walnut St.,Room 307, Philadelphia, PA 19104
> 215. 573.6726 (phone)
> 215. 898.1982 (fax)
> E-mail: evangelg at psych.upenn.edu <mailto:evangelg at psych.upenn.edu>
More information about the voxbo-general
mailing list