I am currently working on a statistical shape mode of cardiac biventricular meshes. I am performing a multidomain optimization of the left ventricle and right ventricle and it works just fine, however, I am wondering if there is a way to add a constraint on the septum so that the particles lying in that shared boundary are shared between the two shapes.
I am not sure if there is a built-in way to do this but any advice on how to represent the shared boundary could be helpful.
Thank you for you reply!
I tried implementing the shape model similarly to the example and I’m getting some good result for the single domains, however in the reconstructed mean shape the surfaces are not matching at the boundaries and they intersect (I’m attaching a screenshot of the mean shape).
I suppose that there is an underlining issue with the orientation and alignment of the shape, but I am not sure what is the best approach to fix this. At the moment I’m doing ICP alignment during the grooming step and Procrustes alignment during the optimization step but maybe I need to do some an extra alignment step the pre-processing too? Do you have any advice how to fix this?
For our biventricle models, we aligned the ventricle meshes before generating the shared boundary, we then didn’t used any further alignment (e.g. 4 domain alignment or procrustes).
I have a further question regarding shared boundary implementation. I am following the example on the GitHub, but I am having quite jagged edges in the meshes after shared boundary extraction. I am currently using .smoothSinc(30, 0.5) but the result is still not ideal. Is there anything else that can help getting better results? Perhaps some specific pre-processing of the original meshes.
I managed to obtain a satisfactory smoothing of the boundaries playing around with smoothing factors, however now I am having some troubles during the optimization.
While the particles on the other domains are in good correspondence, the ones on the boundary contour seem to be completely off and the first modes are mostly picking up on that (see image - this is mode 2).
