I am trying to automatically register tracked US images to a CT scan (with 3D model). I already developed a way to automatically segment bone from US images in real-time. The segmented binary image can be loaded in slicer as scalar volume or labelmap, which looks like this:
Now, I am trying to find a way to register these segmentations to the CT scan, but I am a bit struggling which method is most efficient. I tried exporting each 2D segmentation to a 3D volume, but then you end up with a lot of different segments in the slicer scene. I also tried the volume reconstruction module, but this results in a 3D segmentation with a thickness of multiple millimeters, while I only want the bone surface. I was also thinking: maybe it is possible to convert each 2D US segmentation to a 3D point cloud? Then the point cloud can be registered to the CT bone model with the fiducials-model registration module.
Which method do you think is most efficient? Or do you have other suggestions?