All about Registration

Hey, I am embarking a project that involves human brain/head registration. I have 100 patient data and would like to register them without losing the inter-subject variability and shape and size of different subjects(morphometricity).
What would be the best registration algorithm to use? For example: affine, deformable/non-rigid registration.
I guess deformable registration will distort different subject head shapes.

Another thing is how to visualize the registration results? Is there a metric or so to compare and monitor the 100 registration process?
Though it is slicer3D forum but I am also familiar with ITK.

Hi -

You can certainly use Slicer for this, but when working on batch processing of neuroimages you may be better off with one of the neuro-specific packages like freesurfer, FSL or SPM.

For slicer registration you could start by looking at examples here:

https://www.slicer.org/wiki/Documentation/4.10/Registration/RegistrationLibrary

1 Like

I have used FSL before, how about ITK?
And I need intersubject size and shape variability to experiment on different heads. I mean anatomical or structural information is top priority to differentiate between subjects(These registered images will also be segmented to create different tissue labels of head)
Which algorithm should I go for like rigid-nonrigid, affine-bspline or deformable?
Thanks for the link it looks helpful.

If you are looking for head-shape differences then maybe SlicerMorph or SlicerSALT would have interesting tools for you. Also if you do non-rigid registration then the resulting transformation could be thought of as a way to define the variability of the shapes.

1 Like

Interesting thought.
I need the head shapes intact so was thinking about rigid registration or affine(considering shear and scaling also change the head shapes).

But if I perform non-rigid with respect to either MNI305 or MNI152 or any reference to get the transformation matrix, what non-rigid method you think would be best?

Also to keep in mind that I will segment these registered images to get the different tissue label maps.
I will prefer running a script for 120 3D images rather than registering one by one.

TIA

If you segment the images then you don’t need registration. You can compute displacements and shape statistics based on those segmented structures.

1 Like

If you just statistic the shape you can use the segment statistic module or if you try to get more information you could perform the radiomic analysis.

Hey, thanks!
I have never used it before. Could you provide some tutorials or links to learn more about it?
Looks very helpful.

You can search in the forum and you could get what you want.
For example:
1、segment the ROI
2、select which statistic method in ‘segment statistic’ moduel and then apply.
3、The results and the information should be fould just as:

I was thinking about MultiAtlas segmentation for subcortical brain regions like putamen, substantia-nigra and all. Do you think I won’t need registration(at least rigid/affine) for that reason?
I have used FSL FAST for segmenting peripheral regions such as WM, GM, CSF.

To assign different electromagnetic properties to different tissues for a simulation I will need segmentation labels.
Previous work pipeline included Affine Registration and then MultiAtlas segmentation(MAS).
For MAS, is it mandatory to have registration?
Moreover the head shape and size also has importance the study I am doing.