Point Correspondence Issue

Hi everyone,

I am working with a data set of about 447 atrial fibrillation patients. I am looking at the left atrium and am having some trouble with my point correspondence. I was wondering if anyone had some advice on what parameters I should adjust to improve this correspondence? I have attached some screenshots of what my shapes are looking like and my current parameters.



I would first recommend targeting a lower number of particles and ensuring that you have a reasonable model there. I would suggest something like 256 particles, or even 128 to start. If there is not a good model at this level, further splits will just make things worse.

Besides that, I would recommend an initial_relative_weighting of 0.05 and a relative_weighting of 1.0 to start. I think you can reduce the number of iterations, probably 1000 per split and I wouldn’t worry too much about the optimization iterations until you confirm that you have a consistent correspondence, so I would set it to 500 iterations or so.

Also, did you try on a smaller dataset first, maybe 50 shapes instead of the full 447?

Thank you for these suggestions. I took them into account and there are still correspondence issues. These are what my shapes are looking like.

Do you have any other parameter suggestions?

These are the reconstructions? Can you show what it looks like on the groomed models?

Can you point me to this data (on SCI filesystem)?

I am unsure how to view the groomed models as I ran this from the terminal and not the studio originally. They are on glados in CIBC.

I am running them again in studio right now to visualize the groomed models.