Hi Stephen,
I'm afraid we don't have any clear rules on optimal -l and -i options
(and of course these interact with each other). I would just try a
range of each, as I guess you've been doing.
Note that in the -oba <n> option, you _don't_ need to optimise the
<n> - this ("iterations" as output by the usage here isn't quite the
right word) just tells you how much to dilate the estimated bias - so
just set this high and leave it high (100 should be enough?)
The "wait" when you start FAST could either be the fast wrapper
converting the input data to Analyze (FAST is the only FSL program
that we've not yet converted to reading NIFTI, so the wrapper
converts to anlayze) or running FLIRT if you've turned on the priors
options - in neither case though should the CPU be sitting around
idle, except when it is IO-bound - so this suggests that your NFS
(etc) file IO is running very slowly.
Good luck, Steve.
On 20 Mar 2006, at 17:52, Stephen Towler wrote:
> Greetings all. I'm trying to eliminate major inhomogeneity from a
> run of
> 70 T1 structural (Allegra 3T head coil) on which we will be doing
> regional gm/wm/csf analysis.
>
> To that end I'm multiplying (avwmaths_32R -mult) the original
> volumes by
> their individual "fast -oba" output to try to eliminate the bias field
> (a la http://www.aston.ac.uk/lhs/research/facilities/mri/biascrt.jsp).
>
> My question: is there documentation describing the practical ranges of
> fast's -l, -i, and -oba options for this purpose? If not I'd welcome
> input on the effective range of each option. I'm shooting for decent
> inhomogeneity correction and *reasonable* CPU time (an hour per brain
> would be nice).
>
> My approach thus far: The field inhomogeneity originally manifest
> itself
> on the un-bias-field-corrected CSF volumes (_pve_0), which show a high
> probability of CSF in places they shouldn't (e.g. dorsal third of
> frontal lobes). To remedy I've been trying different combinations of
> fast's -l -i and -oba values and then using avwstats -V to compare the
> number of non-zero voxels in the resultant segmented CSF volumes.
> Example: fast -n -l 100 -i 8 -oba 100 produces far more CSF voxels
> than
> fast -n -l 125 -i 16 -oba 100.
>
> I've also noticed that we experience many minutes with little
> processor
> utilization before the first Main Iteration starts (and maybe even
> before each subsequent iteration). Any idea why we're not getting full
> CPU usage during this period? Trying to cut down on the current 30+
> hrs/brain (Pentium IV's 2.8+).
>
> Any help greatly appreciated!
>
> --
> Stephen Towler
> [log in to unmask]
> 352-294-0048 office
> 352-258-6409 mobile
> 352-392-8347 fax
>
> Leonard Lab
> Department of Neuroscience
> PO Box 100244
> University of Florida HSC
> Gainesville, FL 32610
------------------------------------------------------------------------
---
Stephen M. Smith, Professor of Biomedical Engineering
Associate Director, Oxford University FMRIB Centre
FMRIB, JR Hospital, Headington, Oxford OX3 9DU, UK
+44 (0) 1865 222726 (fax 222717)
[log in to unmask] http://www.fmrib.ox.ac.uk/~steve
------------------------------------------------------------------------
---
|