Hi,

On Wed, 17 Jan 2007 16:34:11 +0100, Sebastian Moeller <[log in to unmask]
AACHEN.DE> wrote:

>Dear List,
>
>a short question and some observations regarding the use of FAST for
>bias correction. I am trying to use FAST to correct for the bias field
>of whole brain monkey anatomicals, acquired using a single channel
>surface send and receive coil. As to be expected the intensity
>distribution of the initial scans and their average is highly uneven.
>The scans were acquired at 0.5mm resolution at 256 by 256 by 256
>voxels; the display in fslview is interesting, the three panes are
>flipped and switched (yeah I read the FAQ, but since I want to feed the
>data back into freesurfer I would rather not change what fs considers a
>perfectly fine orientation ;)).
>	Now, will this "wrong" orientation, as fsl regards it, interfere with
>fast doing its job? Especially the 2D option seems to be asking for
>knowledge of the main intensity gradient axis, does it not? Well, the
>2D option's restore result does indeed look strange (it worked in the
>coronal plane, but fsl considered this to be horizontal). Question,
>will correcting the volume to fit fsl's orientation reuirement improve
>the 2D restore? If so I might have to do a there and back transform,
>which I hoped I could avoid...

FAST doesn't use any of the more advanced NIFTI orientation information, so it will treat slices in 
the raw data as it's slices for 2D mode. Hence if you want FAST to work in 2D and to get the 
correct 2D, you will need to re-orient your data with avwswapdim.

>	The archive recommends running bet beforehand (which significantly
>improves the restore), and notes that larger -l values (bias field
>smoothing) will make the biasfield map to be more "low frequent".
>Unlike the documentation states, at least for my inputs, increasing the
>main iterations (-i) does not improve the homogeneity of the restore
>(by visual inspection in fslview) (but this also "improves" by larger
>-l values).

Right- it's hard to predict what would work best. You can vary the smoothness of the bias field 
estimation, and also the number of iterations of the main outer loop - I would play with both.

Also note that if BET has problems due to the bias field you can iterate - namely use FAST to get 
bias field, ask it to dilate that outwards, apply that to the original image, and re-bet.

Cheers, Steve.

>	By the way, many thanks for supplying these powerful tools.
>
>ahoi
>	Sebastian
>--
>Sebastian Moeller
>
>Tel.: 04 21 - 2 18 - 78 38 oder 96 91
>Fax.: 04 21 - 2 18 - 90 04
>GSM:  01 62 - 3 25 45 59
>[log in to unmask]
>
>AG Kreiter / FB 2
>Institut fuer Hirnforschung III
>Abteilung Theoretische Neurobiologie
>Universitaet Bremen
>Biogarten
>Hochschulring 16a
>Postfach 33 04 40
>28359 Bremen
>===========================================================
==============