I was just repeating something that Tim or
Saad had said about using the waytotal. ;) I think this is the main issue with
the approach: you will lose the differences in pathway strength to some degree.
Thus, the approach is great for thresholding pathways so that their spatial
distributions are similar (and makes a lot of sense if you have confidence the
pathway exists). To make comparisons of pathway strength, you might compare
your pathway of interest to some control pathway. Also, within a single brain
you can use the waytotals themselves as a measure of pathway strength, and you
could use the ratio of a waytotal of interest and a control waytotal to compare
across subjects, should you have reason to believe your control pathway will
not vary significantly.
I believe the waytotals for a single ROI
tractography are the total number of samples sent out. This means that
waytotals / # voxels in the brain mask will give you a number that you already
know without performing the lengthy analysis: the number of samples you are sending
out from each seed voxel.
Peace,
Matt.
From: FSL - FMRIB's
Software Library [mailto:[log in to unmask]] On
Behalf Of Markus Gschwind
Sent: Friday, December 05, 2008
10:42 AM
To: [log in to unmask]
Subject: Re: [FSL] question on old
and new waytotals
Hello!
With respect to Matts proposal of thresholding at a fractional proportion of
the waytotal:
Doing this it will geographically concentrate the path. However how to derive
an adapted mesure of connectivity, as the waytotals will not change.
Would it make sense to threshold the waytotals? How would this be done?
As to a global correction factor for a subjects tractability I've started
tracking from the whole nodif_brain_mask, the resulting waytotals to be devided
by the # voxels in the brain mask.
I'll give news when terminated...
Cheers,
Markus
2008/12/5
cathyliu <[log in to unmask]>
Hi,
Good morning!
As you mention that normalization is
less useful when you are making comparisons within a brain, to
find the asymmetries of the same pathway between two hemispheres. What can we
do for thresholding this kind of case? It happens quite often in the
pathological cases.
Regards,
Yan
> Date: Fri, 5 Dec 2008 01:04:55 -0500
> From: [log in to unmask]
> Subject: Re: [FSL] question on old and new waytotals
> To: [log in to unmask]
>
> This has been a great discussion and it will be interesting to see
> how tractography analysis techniques evolve in the future.
>
> We have been testing the reproducibility of one method that is
> similar to the earlier idea of normalizing to the maximum paths,
> which has flaws as Matt has pointed out, and I was hoping to get some
> feedback from the list. Our approach is aimed at making a qualitative
> comparison of voxels within the seed mask to test hypotheses across a
> group and between groups. We start by performing seed->waypoints/
> termination tractography, then extracting all the streamline values
> in the seed mask, without any thresholding applied to the image, then
> normalize to the maximum value in the seed. This gives us the
> relative "importance" of a seed voxel in connecting to the
target. We
> can then add these images across the group and again normalize to the
> maximum. Our group result then illustrates foci of important voxels
> within the seed on the group level. What we would now like to do is
> perform a simple ttest or regression for hypothesis-testing. Have
> others tried similar approaches and have any suggestions?
>
> I'll also add that we too have tried multiple ways of establishing a
> set of metrics by which to standardize tractography results across
> subjects with little luck.
>
> Any comments would be appreciated.
>
> ted
>
>
> On Dec 4, 2008, at 10:07 PM, Matt Glasser wrote:
>
> > My understanding is that dividing by the waytotal is the way to
> > normalize
> > across subjects as best as you can. Then you could threshold at some
> > fractional proportion of the waytotal and binarize. This should
> > give you
> > the most similar spatial distributions across subjects. However, I
> > think
> > that this method is less useful when you are making comparisons
> > within a
> > brain, for example between the same pathway in the two hemispheres,
> > when you
> > are interested in pathway asymmetries for example.
> >
> > Peace,
> >
> > Matt.
> > -----Original Message-----
> > From: FSL - FMRIB's Software Library [mailto:[log in to unmask]] On
> > Behalf
> > Of David Gutman
> > Sent: Thursday, December 04, 2008 8:46 PM
> > To: [log in to unmask]
> > Subject: Re: [FSL] question on old and new waytotals
> >
> > I'v talked with Matt about this issue in the past. Other than
> > controlling for the volum of the seed you use (which since I run most
> > things in a standard space isn't an issu) just dividing by the total
#
> > of threads, which should be the same across subjects, would just
scale
> > the images, but not actually "normalize" between subjects.
> >
> > Generally it seems the "trackability" really does
significantly vary
> > between subjects. The variance between subjects is very unlikely due
> > to some real biological difference, since the variance is sometimes
> > on the order of 1-2 magnitudes between subjects, in terms of the
"raw"
> > number of fibers that may make it from seedmask A to targets B and C.
> > I'll see numbers vary from the 100's in subject 1 to the thousands or
> > even 10s of thousands between subjects....
> >
> >
> > Ive played around in the past, as I said, with some internal standard
> > "non" interesting tract but was never super happy with the
results.
> > Maybe Tim/Saad/Steve or some of the more physics minded people have
> > some thoughts about this-- in particular maybe looking at the
variance
> > in one of the FA derived metrics (and I won't pretend to fathom which
> > one) that could be used as some sort of pseudo-objective standard of
> > DTI image quality across some large/homogenous region of cortex (if
> > such a thing exists....)
> >
> >
> > DG
> >
> >
> >
> > On Thu, Dec 4, 2008 at 8:42 PM, Cherif Sahyoun <[log in to unmask]> wrote:
> >> Hi David,
> >>
> >> I like the internal standard idea. As you pointed it is hard to
> >> implement/justify though.
> >> Perhaps something based on the eigenvalues could be used to
quantify
> >> Oo the "trackability" of each subject...
> >>
> >> The percentage idea is basically what I had in mind when talking
> >> about
> >> dividing by the max (or 98th percentile), but see also Matt's
> >> concerns.
> >>
> >> Thanks,
> >> Cherif
> >>
> >>
> >> On Thu, Dec 4, 2008 at 8:08 PM, David Gutman <[log in to unmask]>
> >> wrote:
> >>> Cherif I have noted similar absolute number differences
between
> >>> subjects, like a scaling difference, where the tracts
themselves
> >>> look
> >>> extremely similar. My current thoughts are that the
> >>> "trackability" of
> >>> an image varies significantly from subject to subject
depending on
> >>> things like you mentioned, scan quality, motion, etc...
> >>>
> >>>
> >>> One thing I was thinking of, almost analogous to other
methodologies
> >>> (like when you try and amplify DNA/RNA you may choose a gene
that
> >>> "shouldn't be different" to serve as a loading
control and use
> >>> that to
> >>> scale all of your values... myosin or actin or other
messenger genes
> >>> are often used in this sort of analysis, and the relative
> >>> mRNA/whatever expression is correlated using some sort of
internal
更多热辣资讯尽在新版MSN首页！ 立刻访问！
更多热辣资讯尽在新版MSN首页！ 立刻访问！
--
Dr. med. Markus Gschwind, M.D.
Laboratory for Neurology and Imaging of Cognition
Dept of Neurosciences
1 Michel-Servet - 1211
Tel 0041 (0) 22 379 5324
Fax 0041 (0) 22 379 5402
email: [log in to unmask]
http://labnic.unige.ch