3D US image registration

Hi all,

With regards to registering 3D US volumes, which approach + optimizer is the most accurate for this application in ImFusion? Is it NCC with the “ImFusion Nelder-mead algorithm”?


Hi Sewify,

The most accurate approach can only be measured by comparison to ground truth registrations on data sufficiently representing the target data distribution. Other than that, the similarity function and the tolerance mostly govern accuracy, the optimizer is just how you get there (potentially failing).

If I understand you correctly, you want to perform mono-modal registration. For this, NCC should be sufficient. As for the optimizer, I would recommend BOBYQA.

Best regards,