While segmenting individual similar lumbar vertebrae, I want to copy a completed single lumbar segmentation and move it to the next one and edit anything that didn’t match just to see how that compares to my current work flow.
The vertebrae all need to be separated. Straight threshold or grow from seeds requires a fair amount of post editing cleanup. I’m experimenting with a few work flows using threshold masking, flood fill, closing and median smoothing for each one to achieve an accurate and faster process.
This idea may work but might not be ideal, due to:
Fixing an imperfect a segmentation usually takes more time than creating it from scratch
If you start from a similar segmentation then that may end up with less accurate segmentation, as you are more likely to leave small segmentation errors unfixed; and if you make small fixes then they editing may leave artifacts.
If you try it then at least don’t do a simple rigid transform but do landmark-based warping. You can use SlicerIGT extension’s Fiducial Registration Wizard module for this; or you may try automatic image-based registration using SlicerElastix.
Also, it may be better to register a set of vertebrae at once instead of a single vertebra. It may save you time, it may make automatic intensity-based registration more robust, and vertebrae at the same level may be more similar to each other (even if they are from different patients).
Grow from seeds with intensity-based masking (see this segmentation recipe for an example) is always better than threshold and flood fill. You can put a different seed in each vertebra, which gives you automatic separation between vertebrae.
Basic smoothing operations, such as closing and median filtering can fill small holes but if “Grow from seeds” cannot fill cancellous internals of vertebral bodies then you can use “Wrap Solidify” effect (provided by SurfaceWrapSolidify extension).
Keep us updated about what you find working for you or if you need more details about any of the above techniques.
Thank you for all of the options included in your response.
The included screenshot is L-1 (red) transformed onto to L-2 (pink), which had the best results for me. The conclusion was best said in your first and second point about creating each lumbar individually. It was a cleaner and faster process, but I’m very happy to learn and interchange the many options available.
