Hi Tugan, glad that sorted things out for you.
You're right - it would be nice if memory issues were never a problem,
likewise it would also be nice if the programs checked that there was
enough disk space to complete the analysis - they don't currently do that
either, sorry! One day maybe....but on the other hand, the cost of (for
example) a decent linux workstation with plenty of RAM, swap and hard disk
is very small compared with general costs of neuroimaging ;-)
In fact, in your case, FILM contains spatial calculations as part of its
prewhitening estimation, so you can't trivially just "chunk up" the data
without changing the results. In general people find all of the pre- and
post-stats in FEAT much faster (and easier to setup) than with SPM, but
yes the stats itself (FILM) is slower, because it uses the much more
accurate robust prewhitening method, which makes a lot more calculations
per voxel. There is always the option to turn off "FILM prewhitening" in
the GUI if you would rather get faster rather than the most accurate
results, for example, if you are just wanting a quick look at the data
after running a scan.
With respect to multi-session analyses, yes, the memory and cpu-time
requirements for the group-level analysis (FLAME) are even bigger/slower
than at first-level, because of the time-consuming MCMC process which is
needed to get the best possible estimates of group-level variance given
the low degrees-of-freedom involved. Again, there is an option to turn
this "best" estimation off and get faster results (and in the version to
be released shortly, there is a third intermediate option which is nearly
as fast as OLS and nearly as accurate as full FLAME). Also, in the new
version, to save memory for large group studies, the script does do
exactly as you suggested and process the data a slice at a time, which is
ok here, as FLAME does not use spatial processes.
Hope this all makes sense :)
On Thu, 1 May 2003, T Muftuler wrote:
> Hello again,
>
>
> Yes, increasing the swap size did resolve this problem.
>
>
> If the creators of this program don't mind, I would consider this a
> weakness of the program. In my opinion, the program should have checked the
> available memory size (both physical and swap) and load the data in chunks,
> if necessary, instead of trying to load everything at once into the memory.
>
>
> Then, this brings the next question: is this going to be a problem with
> multi-session, multi-subject analysis, too? Is it going to try loading all
> subjects/sessions into the memory?
>
>
> One other thing I have noticed is that FSL is considerably slower compared
> to SPM99 in running the same data set, with similar attributes/parameters
> on the same machine (8-9 times slower). This could be due to inefficient
> use of hard drive as swap memory. Continuous hard drive access may slow it
> down.
>
>
> I certainly liked many features of this software but speed is a major
> concern.
>
>
> Thanks again for taking your time to reply my inquiries.
>
> Best regards,
>
> Tugan
>
>
> Hi - it looks like this is just a memory problem, not hard disk then. You
> probably just need to increase your swap space (to maybe 1GB or more) and
> all should be well.
>
>
> Thanks, Steve.
>
>
>
> On Fri, 25 Apr 2003, Tugan Muftuler wrote:
>
>
> Hi,
>
>
> an additional observation:
>
>
> I had two data sets with 256x256x15 image size, 70 and 90 volumes,
> respectively. None of them made it to a successful FEAT analysis. On
> the other hand, I tried another data with 128x128x16 image size, 124
> volumes and I got success. So, there is certainly something related
> to file size but it is not disk space or quota. Does this program
> have any memory, file size or array/pointer size limitations (in the
> Linux distribution)?
>
>
> Thanks again,
>
>
> Tugan
>
>
> >Hi - it looks like the output from ip was not successful so film_gls
> >wasn't happy - this is most likely that you ran out of disk space or
> >quota?
> >
> >Can you send us the result of running in the FEAT directory:
> >
> >ls -l
> >ls -l stats
> >
> >thanks :)
> >
> > >
> > >
>
>
> --
> ****************************************
>
>
> L. Tugan Muftuler, Ph.D.
>
>
> Center for Functional Onco-Imaging
> Irvine Hall 164
> University of California
> Irvine, CA, 92697-5020
>
>
> Tel: (949) 824-6290
> fax: (949) 824-3481
> E-mail: [log in to unmask]
> ****************************************
>
Stephen M. Smith MA DPhil CEng MIEE
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
|