Automated Segmentation using Total Segmentator
Total Segmentator is a 3D Slicer extension for fully automatic whole body CT segmentation using the TotalSegmentator AI model. Computation time is less than one minute.
Required extensions
- TotalSegmentator
- PyTorch
Pytorch
PyTorch is a python library for machine learning. Slicer uses this library to run Total Segmentator. Sometimes, this library needs to be updated to be compatible with the latest version of Total Segmentator.
The latest version of Total Segmentator requires a version Pytorch greater than 2.0. Bring up the PyTorch Utils module to check what version of PyTorch you have installed.
If the Torch version is 2.2.2 or greater, you should be good to go.
If your Torch version is < 2.0
Uninstall PyTorch
Restart the application
- Bring up PyTorch Utils again.
- Set the Torch version requirement to:
>=1.12
- If you have a Mac or and AMD Windows machine, set the Computation backend to
CPU
. Otherwise, if you have a NVIDIA graphics card, leave asautomatic
Install PyTorch
Total Segmentator Module
For this example, we will use the CTChest volume from the Sample Data module. Load that volume into Slicer
Bring up the Total Segmentator module
- Set the Input volume to
CTChest
- Set the Segmentation task: to
total
- Check Fast on
- Create new segmentation on Apply
- Click
Apply
- You may get a notification that certain python items are being installed. Just wait.
- You should get a notification that module is "Creating segmentations with the TotalSegmentator AI". Just wait.
- Just wait.
Finally, you should get some segmentations.
Review segmentations
Show the Segmentations in 3D
- Switch to the Segmentation Editor
- Click on Show 3D
Hide some Segmentations
- Switch to the Data Module
- There should be a new Segmentation Node called "CTChest segmentation"
- There should be a lot of segmentations in that node.
- Select all of the Segmentations
- Right-Click and select "Hide" - the eye icons should close for all of the segmentations
- Selectively click open the eye icons of those segmentations that you want to visualize