Hi,

I would not recommend using the FA as a weighting image, as it will put very low weighting on the regions with small values and this effective removes the contrast between the high and low values.  It is generally a bad idea to have a weighting function that removes low or zero values that are part of useful boundaries and structures.

There is also no need to be concerned about different FOVs, as all the registration tools take into account the valid FOV and remove areas outside the FOV from the computations.  So that will automatically happen as long as you did not increase the valid FOV with any zero padding (i.e. that everything inside a given image is part of the valid FOV, and that there are no blank slices/regions inside the image that should be cropped out).

If you are not finding that you are getting good registrations without the weighting, then there is probably another reason for that, and it is unlikely that this form of weighting will generally give you reliable/good results.

All the best,
Mark





On 6 Aug 2014, at 14:52, Ricardo Magalhaes <[log in to unmask]> wrote:

Hello Mark, thanks for the answer, I will try to make everything more clear:

I am trying to use flirt to calculate two transformations with FA maps: from each subject of my dataset too a reference subject in the same dataset and from the reference subject to a FA template. I use no prior initialization. My ideia was to use the same image I use in the -ref option in the -refweight option, so I am using there an image with values between 0 and 1 that is in the same space as the reference, so I have something like:

flirt (flirt search options) -refweight TargetImage -in InputImage -ref TargetImage -omat InputInTarget.mat -out InputInTarget

The first of the registrations, within the dataset, works fine with the weighting option, but the second, from the subject to the template space, gives the results I attached in the previous email. I have also obtained successful registrations between the dataset and the template (without using weighting) before, using both flirt and fnirt, although the results were not necessarily too great. 
I have also tried using the -inweight, with the InputImage, with equal results.

The reason I wanted to try this is because I have in some subjects slightly different FoV's in some subjects, so I wanted a procedure to reduce the importance of the overall shape of the brain and focus on the WM tracts. I am not very sure on how successful this approach will be, but right now I want to try and understand what is happening. 

thanks for the help,


2014-08-06 9:31 GMT+01:00 Mark Jenkinson <[log in to unmask]>:
Hi,

It's hard to know what's going wrong here without knowing a bit more about the exact commands you and running and the nature of your cost weighting image.  Am I correct in understanding that you can get good registrations between individual subjects using the weighting, but not between a subject and the template?  Is you weighting image in the input or reference space (or both)?  Is this only with flirt or with fnirt or with both?  Does you weighting image consist of zeros and ones, or is there more to it?  Do you initialise with a non-weighting registration?  What exact registration cause the result that you attached?

If you can provide answers to these questions, and any other information that you think might be relevant, then we can hopefully help some more.

All the best,
Mark



On 30 Jul 2014, at 15:11, Ricardo Magalhaes <[log in to unmask]> wrote:

Dear FSL experts,

I am working on the registration of rodent FA maps to a version of the recently released Waxholm template, by adapting the TBSS scripts, with success so far. 

My issue came up when I was looking to improve the registration of the main white matter tracts by using the weight arguments (-refweight -inweight) of the the flirt function. The logic is that by giving the main WM tracts (which is what we are interested), as defined in the FA map, a higher weight in the cost function I would have a better registration of them. 

I have been successful in registering images to the template using both flirt and fnirt, and I managed to slightly improve the co-register of different subjects to each other using the -refweight and -inweight options of flirt. But when I try to register one of the subjects to the template using either or both of the options the results are always a distorted blur you can find in the attached file. 

Am I just miss understanding the usage of the weight options or forgetting to consider something?

all the best,

--
Ricardo Magalhães, 
Neurosciences Research Domain (NeRD)
ICVS/3B´s Laboratório Associado
PT Government Associate Laboratory

University of Minho, Campus de Gualtar
4710-057 Braga, Portugal

<flirtrefweight.pdf>




--
Ricardo Magalhães, 
Neurosciences Research Domain (NeRD)
ICVS/3B´s Laboratório Associado
PT Government Associate Laboratory

University of Minho, Campus de Gualtar
4710-057 Braga, Portugal

(+351 253604925)