Hi Regina,
if the mask is in standard space and you want to use your filtered_func_data from the 1st level, then yes: you will need to transform it into native subject space first. Make sure to either use nearest neighbour interpolation or to threshold and binarise the mask appropriately.
Using fslmeants makes primartily sense if you want to look at your preprocessed data itself, which may or may not give you a better idea what is going on (depending on your design).
Hope that helps,
cheers-
Andreas
________________________________________
Von: FSL - FMRIB's Software Library [[log in to unmask]] im Auftrag von Regina Lapate [[log in to unmask]]
Gesendet: Freitag, 24. April 2009 18:21
An: [log in to unmask]
Betreff: Re: [FSL] AW: [FSL] Plotting time series (verification of sensitivity of BOLD modeling)
Thanks so much for your reply and sorry about the confusion (convolve vs.
deconvolve).
One thing that I am not very clear on is whether I need to do any sort of
standard space transformation to each participants' filtered_func_data when
extracting the time-courses from individual subjects using fslmeants within
a mask that is on standardized space. Since the mask I would like to use to
extract the time series is a thresholded_zstat from the group analyzes
level, can I still use the filtered_func_data from each subject without
transforming them first?
A second question I have is whether fslmeants would have any advantage over
Featquery for the purposes of having a look at the mean time series within
the mask of interest. If I understand it correctly, Featquery is able to
detect the space the mask is in and do the appropriate transformation back
to native space- so, if I were able to use it instead of fslmeants, perhaps
the question in the paragraph above would be irrelevant?
Thanks a lot!
Regina
On Tue, 21 Apr 2009 10:32:24 +0200, Andreas Bartsch
<[log in to unmask]> wrote:
>Hi,
>
>you are not really "de-"convolving anything in a GLM-type of analysis. You
do convolve you EV, e.g. with the gamma HRF, to optain your model and then
fit you data to it.
>However, you may extract time-courses within a mask from your
filtered_func_data using fslmeants and collapse this across epochs and
subjects. Alternatively, you may use the PEs of a FLOBS model to
"reconstruct" the associated fit or go for MELODIC.
>Hope that helps, cheers-
>Andreas
>
>________________________________________
>Von: FSL - FMRIB's Software Library [[log in to unmask]] im Auftrag von
Regina [[log in to unmask]]
>Gesendet: Dienstag, 21. April 2009 07:21
>An: [log in to unmask]
>Betreff: [FSL] Plotting time series prior to deconvolution (verification of
sensitivity of BOLD modeling)
>
>Dear all,
>
>I am interested in verifying the degree to which I am effectively modeling
>the hemodynamic response to a certain event (original EV). In other words,
>given it is a new paradigm, I would like to verify whether or not I might be
>over or under-capturing the lenght of an event of interest.
>
>The data I am currently analyzing were originally deconvolved using a Gamma
>function. Upon running group level analyzes, I found a cluster that reached
>significance. What I would like to do, as a means to see the original shape
>of the typical response in an area that in principle is sensitive to this
>original EV, is to plot the averaged time series for this cluster (group
>average), prior to it being deconvolved with the canonical HRF (Gamma). I
>believe that by doing so I would be able to get a sense of whether the Gamma
>function in association with the event duration I am currently modeling are
>a good option to most effectively model these responses (or whether I should
>try something different, e.g., different function and/or event duration).
>
>The way that I am currently planning on doing this is by using Featquery:
>1)extracting the average time series within the cluster (binary) mask
>mentioned above from my filtered_func_data for each participant (lower level
>copes); 2) "manually" extracting the epochs of interest from these averaged
>time series (i.e., when the event occurs); 3)averaging across participants
>and epochs of interest to get the averaged time series in this area, prior
>to deconvolution.
>
>My question is: is there a more practical way of achieving this? Or is what
>I describe above a good way to go about it?
>
>Thank you very much in advance for any feedback!
>
>Regina
|