Dear Mark,
Thank you for your prompt and comprehensive response.
The fieldmaps in rad/sec are fairly good quality with some random noise
which is easily removed with a median filter. The fieldmaps in rad/sec
do sometimes show the lateral ventricles as dark compared to surrounding
voxels. From what you have said there should be no contrast based on
structures in the fieldmap and as such they should be smoothed out?
The b=0 images do have some intensity outside of the brain and I use a
weak bet to remove that. So even after removing the non-brain material
from the fieldmap I still have some speckled voxels at the boundary.
Would it be a better option to erode the fieldmap or despike it using
fugue?
Thank you for your input on the registration. My concern at the moment
is applying the correct or most appropriate regularisation to the
fieldmap so that the forward warped magnitude image linearly matches the
B=0 image, since non-linear deformations in the B=0 not accounted for in
the warping process will mean that the field map does not reflect the
true distortion in the B=0 image and therefore will not unwarp
correctly.
Thank you.
Thanks for your suggestion to use FEAT. One question I have with the
FEAT GUI. It asks for the EPI TE, whereas FUGUE requires the EPI Dwell
Time. How is the dwell time computed from the TE if that is what it is
doing?
Thank you again for your response.
Chris Adamson.
On Thu, 2011-08-18 at 00:12 +0100, Mark Jenkinson wrote:
> Dear Chris,
>
> If you have good quality fieldmaps then it may not be necessary
> to do any regularisation at all. Often the problems are just around
> the edge of the brain and performing tight skull-stripping (even if
> it removes a few brain voxels from the fieldmap) is normally enough
> to get rid of these noisy voxels and give you good results. However,
> if you can see structure in the fieldmap - in the rad/s or phase part -
> such as anatomical structures or slice-by-slice variations, then some
> smoothing might be useful. I would not use the cost function to
> determine the quality as I have found this to be extremely unreliable
> in the past. Just judge it by eye.
>
> If your b=0 images do not show much non-brain material then I
> would skull-strip your fieldmap magnitude image prior to registration.
> Always do the skull stripping of the magnitude image for masking the
> phase (rad/s) image though, for the reasons given above. It doesn't
> matter if you lose brain voxels in the fieldmap, as it extrapolates beyond
> this and normally does a much better job than you get if any partial
> volume edge voxels are included.
>
> As for registration, creating the forward warped magnitude image
> and registering that to the b=0 image is what you should do. If it
> is better to use one or other as the reference then that's fine, just
> take the appropriate transformation inverse. But make sure you
> are using the b=0 image and the forward warped fieldmap magnitude
> and not any other pair of images.
>
> Apart from what I've said above, the steps you are going through
> sound good to me.
>
> You might also find it useful to try things through the FEAT GUI
> (as it does a range of pre-processing steps to the fieldmap phase).
> To do this just use your b=0 image as the 4D feat input and deselect
> everything except the B0 unwarping in the Pre-Stats tab. The
> outputs will then be in the unwarp subdirectory, which you could compare
> to your pipeline.
>
> All the best,
> Mark
>
>
> On 17 Aug 2011, at 23:58, Chris Adamson wrote:
>
> > Everyone,
> >
> > I am trying to register a diffusion image to a T1-structural image. I
> > have found non-linear distortions in the DWI images and wish to correct
> > these using FUGUE and then register the B=0 image to the T1 structural.
> > I was wondering about what the best required steps are for using FUGUE
> > and I had some questions about the regularisation parameters. I am using
> > FUGUE to operate on diffusion-weighted EPI images.
> >
> > Firstly, I just wanted to ask about the best sequence of steps used to
> > register the field map to the warped data. The procedure that I have
> > been using so far is:
> >
> > 1. Median filter and regularise field map (FM-Undistorted).
> > 2. Forward warp field map magnitude image to produce (FMMAG-Distorted)
> > 3. Register FMMAG-Distorted to B=0 diffusion image (BZero-Distorted),
> > and produce T(FMMAG-Distorted, BZero-Distorted).
> > 4. Apply T(FMMAG-Distorted, BZero-Distorted) to the field map
> > (FM-Undistorted) to register the field map to the BZero image.
> > 5. Unwarp the BZero image using transformed field map in undistorted
> > space producing (BZero-Undistorted).
> > 6. Register (BZero-Undistorted) to the T1-structural image using linear
> > registration.
> >
> > Another option for 2 and 3 is to register the B=0 to the fieldmap
> > magnitude and then use the inverse transformation to register the
> > fieldmap to the B=0 image.
> >
> > Is this the best sequence of steps to use?
> >
> > My other questions are:
> > Should you skull strip the fieldmap image (skull stripping the magnitude
> > image and and applying the brain mask) prior to regularisation?
> >
> > The other questions I have relate to regularisation of the fieldmap.
> >
> > Thus far I have used two datasets and I have been using the alignment of
> > the corpus callosum as a qualitative measure of success. For one dataset
> > no or little regularisation seems to work best, whereas for the other
> > dataset I cannot find a parameter set that gives good alignment of the
> > corpus callosum.
> >
> > My question is: I have been using the FLIRT cost function output to
> > gauge which regularisation produces the best registration of the
> > fieldmap magnitude image with the B=0 image. Is there any other way of
> > determining the "best" regularisation parameters to use?
> >
> > Thank you in advance,
> >
> > Chris Adamson.
> >
> > --
> > Dr Christopher Adamson, PhD (Melb.), B Software Engineering (Hons.,
> > Monash)
> > Research Officer
> > Developmental Imaging, Critical Care and Neurosciences
> >
> > Murdoch Childrens Research Institute
> > The Royal Children’s Hospital
> > Flemington Road Parkville Victoria 3052 Australia
> > T 9345 4306
> > M XXXX XXX XXX
> > E [log in to unmask]
> > www.mcri.edu.au
> >
> > ______________________________________________________________________
> > This email has been scanned by the MessageLabs Email Security System.
> > For more information please visit http://www.messagelabs.com/email
> > ______________________________________________________________________
> >
>
> ______________________________________________________________________
> This email has been scanned by the MessageLabs Email Security System.
> For more information please visit http://www.messagelabs.com/email
>
> If you have any question, please contact MCRI IT Helpdesk for further assistance.
> ______________________________________________________________________
--
Dr Christopher Adamson, PhD (Melb.), B Software Engineering (Hons.,
Monash)
Research Officer
Developmental Imaging, Critical Care and Neurosciences
Murdoch Childrens Research Institute
The Royal Children’s Hospital
Flemington Road Parkville Victoria 3052 Australia
T 9345 4306
M XXXX XXX XXX
E [log in to unmask]
www.mcri.edu.au
______________________________________________________________________
This email has been scanned by the MessageLabs Email Security System.
For more information please visit http://www.messagelabs.com/email
______________________________________________________________________
|
