Hi,
The ROI way is still a valid method to analyse TBSS data, but you are
right that
it is not without problems. The principle is that the average over
the ROI has a
greatly reduced noise (through averaging) and that the signal will be
less affected.
However, the latter is only true if there is substantial amounts of
similar signal change
within the ROI. If, as you say, there is very inhomogeneous signal
which averages
out, or if most of the ROI has no change, then the results of the ROI
analysis
are likely to be null. Only when the ROI contains a relatively large
amount of
similar change will this be a better method. Hence, doing the whole
skeleton
is probably not be very sensitive, but targeting a smaller portion
that you
hypothesize is changing, might do better.
Your masking approach sounds reasonable and the mean is the best summary
to start with.
All the best,
Mark
On 8 Jan 2008, at 02:33, David Gutman wrote:
> I saw some old posts (see attached)1, but wanted to get some
> feedback. I am interested in group differences within connectivity
> patterns within a given ROI (i.e. ROI based TBSS).
>
> I was thinking I could/should run TBSS as normal (I would imagine
> trying to register just an ROI would not work very well). Then
> mask the tbss_4_prestats output by the ROI and run the stats from
> there.
>
> From the old post-- Dr. Smith suggested>
> > If you already have a tract ROI defined, you might get more
> sensitive
> > results by averaging the TBSS-preprocessed data within the ROI and
> > then doing stats on the single summary number across subjects - but
> > what you're doing is fine too. 1) Is this still valid with the
> newest incarnation of TBSS (this post is like 1.5 years old)
>
> 2) Isn't it quite possible that two groups could have the same
> mean value but actually have quite large differences in their
> structure? Of course if the simple summary statistic yields a
> result, it doesn't matter. But in simple terms, I imagine a tract
> with the same value (say 0.5) across the entire tract vs a tract
> with really low values on one side and really high values on the
> other (constructed in such a way the average value along the tract
> = 0.5)
>
> 3) If I did have an individual subject's FA map and masked out the
> specific ROI I was interested in, what summary statistics would be
> useful to characterize the tract? Just get the mean intensity of
> the skeletonized mask for all non-zero voxels?
>
> Hope this makes sense,
>
>
> DAG
>
> --
> David A Gutman, M.D. Ph.D.
> Department of Psychiatry & Behavioral Sciences
> Emory University School of Medicine
>
>
>
>
>
>
> i,
> ok, thanks, I'll try out averaging the TBSS data within the ROI and
> see
> how doing stats on that goes.
> I still want to be sure I'm clear on my original process, though.
> So, to
> demean I would just do avwsplit on all_FA_skeletonised, merge each
> group
> into its own 4d file, do avwmaths -Tmean, split it again so that I can
> subtract the appropriate mean from each file, and then merge it all
> together
> into one big file? (or is there an easier way?).
> I think maybe I am a bit confused about some of the differences
> between
> randomise and Feat. If in addition to my existing EVs with demeaned
> behavioral data, I put in two more EVs that each code for a group
> (1's and
> 0's), would that also model the group mean FA (and make the manual
> demeaning
> unnecessary), or is randomise different with regards to this? I know
> covariates of non-interest have to be modeled in a separate design
> matrix,
> but am not sure whether that also means that this would not work?
> thanks,
> Katie
> ___________________________________
> Katie Karlsgodt
> Dept of Psychology/Cognitive Neuroscience
> University of California, Los Angeles
>
> [log in to unmask]
> phone: (310) 794-9673
> fax: (310) 794-9740
>
>
>
> > From: Steve Smith < [log in to unmask]>
> > Reply-To: FSL - FMRIB's Software Library < [log in to unmask]>
> > Date: Tue, 30 May 2006 07:48:21 +0100
> > To: <[log in to unmask]>
> > Subject: Re: [FSL] DTI ROI analysis
> >
> > Hi Katie,
> >
> > Sounds good. If you're using randomise (as opposed to FLAME) there's
> > no point setting the group membership IDs in the Glm GUI - randomise
> > doesn't use this information.
> >
> > If you already have a tract ROI defined, you might get more
> sensitive
> > results by averaging the TBSS-preprocessed data within the ROI and
> > then doing stats on the single summary number across subjects - but
> > what you're doing is fine too.
> >
> > It sounds like you are not modelling the group mean FA though! If
> you
> > just have 2 EVs, containing demeaned behavioural data, then you also
> > need to demean the data - i.e. demean each groupt separately before
> > re-concatenating to all subjects' 4D file and running randomise -
> > otherwise your model-fitting in randomise won't make sense...
> >
> > Cheers.
> >
> >
> > On 29 May 2006, at 20:13, Katie Karlsgodt wrote:
> >
> >> Hi,
> >> I'm trying to run an ROI based analysis in TBSS/randomise and
> >> would like
> >> to check whether what I'm doing is appropriate. I have a tract of
> >> interest,
> >> which I defined on the regular FA map and drew a mask of. I then
> >> masked out
> >> the portion of that tract that was included in the TBSS-skeleton
> using
> >> avwmaths. I ran randomise, using the skeleton masked tract as the
> >> mask in
> >> the -m option. I would like to see differences between groups in
> >> how FA of
> >> the tract relates to a behavioral measure. I created 2 groups in
> >> the group
> >> column, and entered 2 EV's, one with demeaned behavioral data for
> >> one group,
> >> and one with demeaned behavioral data for the other group. My
> >> contrasts then
> >> compare them (1 -1 and -1 1). I can't find anything on the list
> >> quite about
> >> this sort of thing, so just wanted to make sure this whole
> process is
> >> kosher? Thanks very much.
> >> Katie
> >> ___________________________________
> >> Katie Karlsgodt
> >> Dept of Psychology/Cognitive Neuroscience
> >> University of California, Los Angeles
> >>
> >> [log in to unmask]
> >> phone: (310) 794-9673
>
> --
> David A Gutman, M.D. Ph.D.
> Department of Psychiatry & Behavioral Sciences
> Emory University School of Medicine
|