I am using TotalSegmentator to segment the spine from CT scans. However, I noticed that certain parts, such as the spinous processes, are not segmented very well. I would like to manually refine the segmentation results based on the original CT data.
I have the following questions:
Is it feasible to manually enhance the TotalSegmentator segmentation results using 3D Slicer?
What is the best workflow in 3D Slicer to compare the TotalSegmentator segmentation results with the original CT data and add missing details to the segmentation?
Any guidance or suggestions would be greatly appreciated!
The general pathway would be to provide improved manual segmentations (i.e. use the Segment Editor) and then retrain a model like TotalSegmentator either using nnU-Net, MONAI Label, or MONAI Auto3DSeg. There’s a lot of documentation about these methods, but they do require some effort to learn and use. It would be great if you could create and share improved segmentations.
Note there is also this dataset of already segmented vertebrae that could also be used to train models that may be more robust / higher resolution than TotalSegmentator and it would be great if people could work on that.
I haven’t tried it myself, but this tool looks very promising for the task of correcting model outputs: