Hi,
thanks for the response. Just one further question, though, do you think that the change in the estimation accuracy could have such drastic effects on data even if it was not close to threshold? In at least one of the datasets, the old findings were very robust and really nowhere near threshold at all, and now the activation is almost entirely gone. I'm curious whether, since you said that the change with the new version is usually minimal but we had quite extensive changes on two separate analyses, if you know of something we could have done wrong in the second analysis to make this happen or if it is at all possible that there is something else wrong with the way our FSL is running? I just wanted to double check before we have to completely reconsider the state of our data...
thanks very much,
Katie
> ----------
> From: Stephen Smith
> Reply To: FSL - FMRIB's Software Library
> Sent: Thursday, October 30, 2003 12:14 PM
> To: [log in to unmask]
> Subject: Re: [FSL] group analysis on 3.1 vs 3.0
>
> Hi Katie,
>
> Yes, this is possible. In the "what's new" notes with the 3.1 release it
> says: "FLAME..updated to give increased estimation accuracy at the MCMC
> stage" - basically we improved some of the estimation maths and also
> increased the number of MCMC samples we use to get improved accuracy
> compared with the previous version (with a small hit on computational
> time). In most datsets the change is minimal, but with some the
> "improvement" can change the results, particularly if the activation was
> close to threshold before. Yes, the log files will look quite different in
> the two versions.
>
> If asked to give a bottom line thought on how to interpret the change: I'm
> afraid you should trust the new results more than the old!
>
> Thanks, Steve.
>
>
>
> On Wed, 29 Oct 2003, Karlsgodt, Katherine wrote:
>
> > > I have a question about running group analyses on FSL 3.1. Our lab
> > > has 2 data sets that were run previously on version 3.0 (one on Mac
> > > OSX and one on Solaris 8) that produced reasonable looking group data.
> > > When we reran the same data on FSL 3.1 (once on OSX and once on Redhat
> > > 9, linux kernel 2.4.19) the results were drastically different. The
> > > single subjects registration and activation looked essentially the
> > > same, but in the group analyses all of the activation was gone in one
> > > of our datasets and the other was left with only a few active voxels.
> > > In trying to figure out why, we looked in the log files and found that
> > > they look quite different from each other, even with analyses that we
> > > thought were set up to be the same.
>
> Stephen M. Smith DPhil
> Associate Director, FMRIB and Analysis Research Coordinator
>
> Oxford University Centre for Functional MRI of the Brain
> John Radcliffe Hospital, Headington, Oxford OX3 9DU, UK
> +44 (0) 1865 222726 (fax 222717)
>
> [log in to unmask] http://www.fmrib.ox.ac.uk/~steve
>
>
|