I’ve tried the registration and found the followings:
- Image size is quite large, therefore I used “Crop volume sequence” module to crop and resample the image (used 1.7x isotropic spacing; resulting in an approximately 100x80x80 volume)
- Segmented frame index=9 using thresholding and then used Sequence registration module (selected same frame index = 9 as fixed frame, “fixed frame to moving frames” direction). Applied output transform to segmentation. Registration result was not accurate enough. Probably the registration is hard because the difference between frames is quite significant (due to motion and difference in contrast distribution).
- Tested “3D CT multi-contrast (cardiac)” preset (set both start frame index and end frame index values to 2 so that I don’t have to wait for registration of 10 frames) and I found that the registration is much better (maybe good enough), it was just very long - took almost 15 minutes. This indicates that the registration results can be improved by tuning registration parameters.
If you are interested in developing a good registration preset that can be used to register such image sequences then I think “3D CT multi-contrast (cardiac)” is a good starting point. Try to simplify (reduce number of steps, etc) and see if you can make the registration faster without making registration results worse. You may ask help on Elastix forum.
If you don’t want to spend much time with automated registration and you just need segmentation for all time points then you may try the more manual, landmark-based registration (for example, using Fiducial registration wizard module in SlicerIGT extension; or maybe Landmark registration in Slicer core).
You might also make segmentation faster by defining seed regions on one frame and then use the same seed regions for “Grow from seeds” in each frame to get complete segmentation.