New extension: Fully automatic whole-body CT segmentation in 2 minutes using TotalSegmentator

[Xavier_Tre]

Hello, I have problems to install totalsegmentator, already I have read the document several times where, and I have made several attempts to install it without results. Can someone explain to me how to install it? a tutorial video?

[lassoan (Andras Lasso)]

Please describe what you did, what you expected to happen, and what happened instead. Include screenshots and copy here all messages displayed in the log window of the TotalSegmentator module.

[zhang_ming (SimZhang)]

Will this plugin use our local weights to compute(refined or trained by my local dataset), and the weights uploaded to googledrive for users to download

[Julius_W (Watrinet)]

Congratulation for implementing TotalSegmentator to Slicer. I use Slicer to segment bones of the lower extremities and wondered, if you are planning on implementing the v2.0 of TotalSegmentator as well? I would greatly appreciate this work as the segmentation of the lower extremity is enabled in this version.

Thank you very much
Julius

[rbumm (Rudolf Bumm)]

Yes, it is planned to implement TotalSegmentator v2 as soon as possible.

[lassoan (Andras Lasso)]

We have a functional prototype in this pull request of TotalSegmentator v2 in Slicer. We need to tune a few small things before we release it, probably in a week or so. If you don’t want to wait for that then you can get the updated TotalSegmentator.py and .ui files from github and overwrite the files in your Slicer installation folder.

[Julius_W (Watrinet)]

That’s great thank you!

[Luca1 (Luca)]

Is there any chance to use “total segmentator” with cone beam CT acquired with angiography systems in the body (liver)? Thanks

[lassoan (Andras Lasso)]

Yes, it works on CBCT images as well. However, if the image quality is very bad (which is not uncommon, especially for low-dose acquisitions) then the accuracy will suffer.

[Tobias (Tobias Karlsson)]

Hi,
I am trying to make it run and when I look in pytorch utils it said torch version 1.8.1
However, when i run the totalsegmentator it says that it needs to install a higher version of torch (1.12.1) and restart. And then it restarts but nothing happens.
Does anyone have any idea what I am doing wrong?

[lassoan (Andras Lasso)]

You have some old pytorch 1.x version in your pip cache that gets installed by default.

You can go to Pytorch Utils module, uninstall this old version, restart Slicer, go to Pytorch Utils module again, and before you click to install Pytorch, write >=2 in the version requirement textbox.

These uninstall and install steps should be performed automatically if you use the latest Slicer Stable Release or latest Slicer Preview Release.

[Sandra_Garcia]

Hi @lassoan I have a quick question.

I am interested in conducting an experimental campaign on the functionality of TotalSegmentator. For this purpose, I want to modify some parameters and compare the results obtained. I have tried to make modifications from within the program, but regarding segmentation, it is practically impossible to change anything. Therefore, I wanted to ask if there are parameters in the program’s code that can be modified to obtain different results from the same data set. For example, changing the resolution, edge detection, or something like that. If you have any ideas on how I can achieve this, I would greatly appreciate it. It doesn’t have to be something complex, just to be able to compare various results. Thank you in advance for your response.

Using CT images.

Thank you in advance for your response

[lassoan (Andras Lasso)]

The result is anyways resampled in the end to the input image resolution, so if you want to change that then you can resample the input image before running TotalSegmentator.

This will not change things fundamentally, as the internal segmentation model had fixed resolution. You would need to retrain the entire model on higher-resolution data to get finer details.