Hi, just a follow up.
The longitudinal siena seems to give ok results. Almost all the subject show a PBVC smaller than 1%. So this is good.
However, the test retest reliability for the SienaX at time point one and two is not very good, being .62.
It seems to me that there is something intrinsically wrong with my images, but looking at them and looking at the results of the segmentation I can not really point that out.
Is there anyone who have used FIRST in children that can give me any tip ?
De : FSL - FMRIB's Software Library <[log in to unmask]> de la part de Niels Bergsland <[log in to unmask]>
Envoyé : jeudi 1 octobre 2015 11:11
À : [log in to unmask]
Objet : Re: [FSL] bad test retest reliability specifically for right accumbens
Hi - I haven't used FIRST in a pediatric sample but those values for
the test-retest reliability is quite low, even for the caudates. We
have looked at this ourselves in a paper where we used FIRST as part
of a processing pipeline. We found excellent intraclass correlation
coefficients for all structures in sample of 6 multiple sclerosis
patients and 6 healthy controls, scanned twice (one week apart). (0.92
and 0.95 for the caudate and accumbens, respectively.) We looked at
total volume rather than left/right separately but still, yours are
So your results are a bit surprising. It's perhaps an obvious
question, but has anything changed in the scanning protocol or on the
scanner? Have you looked at the reproducibility of an other measures?
Even if it's not part of your study, you could consider looking at
some other parameters to see if you evidence of some systematic
change. Perhaps comparing SIENAX values at both time points, as well
as running the longitudinal SIENA pipeline. These are of course very
different measures than what FIRST provides you, but it might help (or
not!) get some additional insight into what's going on. Also, I guess
not since you are specifically interested in the test-retest
reliability but is anything happening with the subjects during these
eight weeks? I can't imagine it would have this large of an effect,
but something like less motion at the second timepoint could have some
impact although not to the degree you're seeing, provided the data
BTW, how many subjects are you using for your test-retest comparisons?
On Thu, Oct 1, 2015 at 10:22 AM, Alain Imaging
<[log in to unmask]> wrote:
> Hi everybody,
> I was doing a check on the segmentation of caudate, putamen and accumbens
> obtained from FIRST in a sample of children (6 yo).
> I had these subjects scanned twice, with a gap of around 8 weeks.
> I was testing a sort of test-retest reliability of the segmentation in this
> sample (even accepting the idea that some noticeable change in the
> structures of interest could occur in 8 weeks, I would not expect this
> change to be dramatic).
> Overall, the test-retest reliability is not awesome, being between .6 and .7
> for left and right caudate and left accumbens. The funny thing is that it
> drop specifically for the right accumbens (.2).
> Could it be any reason why this is so bad specifically in the right
> accumbens ? I know that segmentation of smaller structures is worse than
> that of bigger structure, but why the hemisphere has any influence is beyond
> my understanding.
> If anyone have had this experience or have any idea about the reason why I
> am observing this, I am willing to listen
Buffalo Neuroimaging Analysis Center
100 High St. Buffalo NY 14203
[log in to unmask]