Hi Mark and Andreas,
your suggestions have been very helpful!
I did use Correlation Ratio and -nosearch for the initialized registration of the Angio to the T1. I've now tried the Mutual Information and Normalised Mutual Information, but it only made a very small difference and did not really improve the registration. So I'll stay with the Correlation Ratio, but will have the other options in mind.
Using the scanner info (-usesqform) for the direct registration Angio -> T1 is great, the registration is essentially as good as our previous detour via the EPIs. But so far it only works with a resolution of 1mm iso, the registration fails badly if I use a resolution of 0.5 mm for the T1 and the angio (original resolution: 0.3mmx 0.3mm x 0.5mm). Is the registration algorithm optimized for spatial resolutions above 1mm ?
And I still struggle with the "fine-tuning" of the registration. Since the ToF-Angio is low in contrast, what I see is that the registration fits the outer boundry of the images, which, depending on the quality of the brainstrip, often puts the angio more or less off. Is there an option to just use the scanner coordinates to position the angio to the T1 (-usesqform) without an additional registration ? The ToF-Angio and the high-quality T1 anatomy are measured in the same session using AutoAlignScout, so the sanner coordinates could be sufficient.
Again, any help or comments are welcome,
Thanks,
Renate
Hi,
I'm not sure what these images would look like and so how difficult it
would be to improve BET. Ultimately the last resort is to manually edit
the image, which is relatively easy to do with FSLView.
As for the registration of the Angio to the T1w with an initialization - what
cost function did you use? It might be worth trying Mutual Info or
Normalised Mutual Info if you haven't already. When the images contain
quite different information this can definitely help. And use the -nosearch
option if you haven't done so already.
All the best,
Mark
Hi,
> >When the images contain quite different information this can definitely help.
They do. TOF-MRA is T1-weighted but with very little contrast and mainly flow infomration in the arteries. I would also try to use the scanner info (assuming the images where acquired in one session and the subject did not move much), i.e. the usesqform flag. Cheers- Andreas
On 17 Jan 2011, at 10:18, Renate Schweizer wrote:
> Dear Registration experts,
>
> I have a registration / brainstripping problem with a small volume, high-resolution, double-oblique oriented ToF-MRAngiography, which I want to register to a normally oriented T1-weighted whole brain data-set.
>
> Right now I do a flirt registration of the ToF-MRA to an EPI partialvolume (same orientation, same small volume coverage) which is registered to a "whole brain" EPI (same orientation, covering the whole brain) which is then registered to the normally oriented (ACPC) T1 weighted whole brain volume.
>
> I then take the concatenated transformation matrix from Angio -> partVol EPI -> whole brain EPI -> whole brain T1 to reorient the original angio-volume to the right position on the T1 weighted image. Because of the high spatial resolution of both the T1 and the angio, I do see registration errors. I therefore tried a final registration step in which I used the former transformation matrix as the initialization matrix for the registration of the ToF-MRA to T1, but this worsened the error.
>
> From my various attempts to find out the critical parameters, why this last step fails, I suspect that it is the small volume of the ToF-MRA in combination with the incomplete brain-extraction that excludes a precise registration . Applying bet small Z and -f 0.1 to the ToF-MRA removes the skull but not the CFS/meninges above the cortical surface.
>
> I' d appreciate any suggestions on further options in "bet" to possibly increase the efficiency of brainstripping the angio, or any ideas on alternative routes for registration.
>
> Thanks,
>
> Renate
|