Hi,
It sounds like you need to fix the bias field in these EPI images.
To do that you can either use FAST with 2 or 3 classes (depending on the resolution/contrast/quality of your EPI images) or if the bias field is quite strong (and it sounds like it is) then it is possible that a smoothing approach might work (it is less good than using FAST, but a reasonable alternative if FAST does not work well). To use the smoothing approach you need to do the following:
1 - make a brain mask (you need the binary brain mask and the original image masked by this mask)
2 - smooth the masked image by a large amount (e.g. fslmaths with -s 20 or even higher)
3 - smooth the binary brain mask by the same amount
4 - divide the smoothed masked image (2) by the smoothed brain mask (3) to get a new smoothed image
5 - divide the original masked image (1) by the result of (4)
This should hopefully get rid of the big bias field enough to allow the registrations to work better.
All the best,
Mark
On 24 May 2013, at 11:03, jamy <[log in to unmask]> wrote:
> Dear FSL experts,
> Some EPI images have been acquired with prescan normalize OFF, which resulted in a gradient of intensities in some EPI data.
> After running registration using FNIRT (T1 to standard T1) then FLIRT BBR (EPI to T1), we can see some parts of the EPI data exceeding the boundaries of the standard T1, which seems to be due to this gradient of intensities (The part of the EPI that have high intensities seem to exceed the boundaries of the standard T1).
> I was wondering if you have a solution to suggest to correct this problem?
> I know fslmaths can rescale image intensities, but I am not sure it is the best option.
> Any help will be greatly appreciated,
> Best regards
>
> J.
|