Partial Surface Registration Tutorial

Hi devs.

I have made a tutorial for Partial Surface Registration, something similar to weight-painting ICP that is done in Blender.

Here you can access it:

If there are no suggestions for corrections I’ll later upload it to Youtube.

Mauro

3 Likes

I think this looks good. I was able to follow the process and fully understand it after watching it twice.

A question I have is what is the primary motivation for partial surface registration (vs other registration options)? I can imagine a few possible use cases:

  • Maybe partial surface registration is much faster to when full model to full model registration is prohibitively slow?
  • Maybe you are trying to align a particular region in order to look at differences in another region, so a full registration which tried to align everything would be counterproductive
  • Maybe regions you are trying to align do not have clear landmarks (like the smooth section of bone you chose)

Any other reasons someone should think of partial surface registration when trying to solve a problem?

Are there any caveats to the method or things people should keep in mind when choosing regions? Perhaps a trade-off in size of chosen region vs registration time? Are disconnected regions OK or must the regions to be registered be a single connected region?

Many Slicer tutorial videos do not address these broader questions (I assume to focus on brevity), so I don’t think you need to either, but these are a couple of things I wondered about after watching and which aren’t addressed.

Thanks for making the video demonstration tutorial, these kinds of videos are very helpful in learning how to do things in Slicer!!

Hi Mike.

Thanks for the feedback.

A question I have is what is the primary motivation for partial surface registration (vs other registration options)?

I think this is the main reason:

Maybe you are trying to align a particular region in order to look at differences in another region, so a full registration which tried to align everything would be counterproductive

Are there any caveats to the method or things people should keep in mind when choosing regions?

I would choose a little bit of a well-matched region to work as pivot, and then the rest of the selection would be badly matched parts. The idea in my mind is trying to set up a context that would push or rotate the moving-model to the correct place considering points of both models will try to be as close as possible as they can. For a deeper idea of this, I think you should study how the algorithm works, I think I just have a general idea of it.

Are disconnected regions OK or must the regions to be registered be a single connected region?

I think there is no constraint that obligates you to use a single connected region.

1 Like

This videotutorial show how to make a rigid registration of two models corresponding to the same anatomy on different subjects so that the selected region of one model matchs a similar region on the other model (no entire model matching is done, only region of interest is accounted for registration)

First an approximate registration is done with PointToPoint registration.
Then the selected part of one of the models is used for ICP resulting on similar regions matching/overlapping.

Workflow developed ad-honorem by Mauro I. Dominguez

3 Likes