Totalsegmentator is an amazing AI tool and now in a great 3D Slicer extension. Very good tutorial. You can slow it down a bit if you are new to Slicer.
But is it real though? It raised lots of questions for me. I am seeing lots of wavy structures on a kidney surface, also lots of spikes? Why can it not show urinary tract canals for example? Can I train it to show those little details too?
@@brad8122 If you disable fast mode then the result is smooth. You can of course also improve the segmentation and retrain, but that may be significant effort.
@@PerkLabResearch sorry I have another question. Can we use MRI data and 3D segment it like this? Any Totalsegmentator alternatives but for MRI ? In MRI bones have very little signal as opposed to CT
When do you think something similar for full Brain MRI will be available? I know there's something similar in FreeSurfer, but it takes too long to segment everything...
It seems that nnunet and monai auto3dseg model can learn any 3d segmentation task - all you need is to provide training data. So, if you can segment the appendix in a few hundred CTs (and then apply that model to segment 2000 CTs and fix any errors that you find in those) then you can get a model that will segment the appendix fully automatically in practically any CT image.
TotalSegmentator is for whole-body CT scans. If the head is visible then it just aims for segmenting the face area so that you can remove the face from scans. For head CTs you need specialized scans and there is a wide variety of segmentation goals that you need to address with specialized models (e.g., segmenting teeth or detect bleeding in the brain are completely different tasks). Maybe specialized models will be added to TotalSegmentator, but probably they will not be freely available. You can use MONAILabel module in Slicer to train your own model from your data.
Loaded my ct scans on here to have a look. I get an error about geometry and spacing. Nothing but flat lines displayed yet when I scroll I can see there is data changing in the displayed line. Must be an issue with the data from the scanner at the hospital. Another software wouldn't load it either. Used the latest version not 3dslicer 5.2.1
Skewed images (where the image axes are not orthogonal) are rejected by many algorithms. You can harden any transform that is applied to the volume (e.g., due to tilted gantry or non-uniform slice spacing) and/or use Crop volume module to resample the image to a regular grid. Use latest Slicer Preview Release or latest Slicer Stable Release.
@@AndrasLasso i was requesting the extension of totalsegmentator for bone. I got the link to download the package of bone extension. But i don't how to install it to be used in 3D Slicer.
I just ran a simulation now from this extension and the output i got is so rough and low poly, what can i do to make or improve the output to something better and smoother. Please I need your help
@@AndrasLasso I didn't simulate with fast mode, that's why I'm surprised at the result I got. After almost 1hr of simulation it still came out low poly
@@chinonsopromise4962 TotalSegmentator does not generate polys. It creates a labelmap volume. Segmentation is converted to polygonal mesh (millions of triangles, I would not call low poly) and finally OpenAnatomy Export can decimate and export the meshes. You can set "Reduction factor" to a smaller value to preserve more triangles.
@@AndrasLasso I understand the other user frustration, I'm also not using the Fast option, and the voxel are visible larger than expect, like it need to use Smooth to look more like my normal results using Thresholds.
@@SilasC Thresholding uses the resolution of the segmented image. TotalSegmentator uses 1.25mm voxel size. If you need finer details for specific structures then you can use TotalSegmentator to extract the bounding box of the structure and then you can train a new model to extract only that structure in that small region.
Although I get excellent results with 3D Slicer sample, I see that with my patients CT it never do all their Scapula, what is weird for me as it is so easy to get bones with the Threshold.
Maybe something is special about your CT images, something looks different than in all the images that TotalSegmentator was trained on. You can improve the model by adding your images (with your segmentations) to the training data set and train a new model.
You may need to get the latest development version of TotalSegmentator. It is available for the latest Slicer Stable Release and latest Slicer Preview Release.
Awesome video...Quick with a lot of useful info. Thanks for sharing. I'll try segmenting my colon as soon as I get off work today.
Totalsegmentator is an amazing AI tool and now in a great 3D Slicer extension. Very good tutorial. You can slow it down a bit if you are new to Slicer.
But is it real though? It raised lots of questions for me.
I am seeing lots of wavy structures on a kidney surface, also lots of spikes? Why can it not show urinary tract canals for example? Can I train it to show those little details too?
@@brad8122 If you disable fast mode then the result is smooth. You can of course also improve the segmentation and retrain, but that may be significant effort.
@@PerkLabResearch sorry I have another question. Can we use MRI data and 3D segment it like this? Any Totalsegmentator alternatives but for MRI ? In MRI bones have very little signal as opposed to CT
When do you think something similar for full Brain MRI will be available? I know there's something similar in FreeSurfer, but it takes too long to segment everything...
It is possible to use with Jupyter notebook? where can I find the documentation? Thank you!
I tried exporting with OpenAnatomy export, it just outputs a file that has nothing inside. What could be the issue please
what is the last software you used to hide the meshes ??
Hi, when will you be able to introduce the function of appendix image segmentation
It seems that nnunet and monai auto3dseg model can learn any 3d segmentation task - all you need is to provide training data. So, if you can segment the appendix in a few hundred CTs (and then apply that model to segment 2000 CTs and fix any errors that you find in those) then you can get a model that will segment the appendix fully automatically in practically any CT image.
It works well for CT Abdomen, but I'm struggling to get good results with a CT brain. Any suggestions for that?
TotalSegmentator is for whole-body CT scans. If the head is visible then it just aims for segmenting the face area so that you can remove the face from scans. For head CTs you need specialized scans and there is a wide variety of segmentation goals that you need to address with specialized models (e.g., segmenting teeth or detect bleeding in the brain are completely different tasks). Maybe specialized models will be added to TotalSegmentator, but probably they will not be freely available. You can use MONAILabel module in Slicer to train your own model from your data.
Loaded my ct scans on here to have a look. I get an error about geometry and spacing. Nothing but flat lines displayed yet when I scroll I can see there is data changing in the displayed line. Must be an issue with the data from the scanner at the hospital. Another software wouldn't load it either. Used the latest version not 3dslicer 5.2.1
my ct scans are easily displayed using autorun on the dvd, but can't see them using this .
uninstalled recent version 5.7 or whatever and it worked fine with version 5.6.0
Skewed images (where the image axes are not orthogonal) are rejected by many algorithms. You can harden any transform that is applied to the volume (e.g., due to tilted gantry or non-uniform slice spacing) and/or use Crop volume module to resample the image to a regular grid. Use latest Slicer Preview Release or latest Slicer Stable Release.
I cannot find it in the Extensions Manager, is there any other way to install it?
You need to use latest Slicer Stable Release or latest Slicer Preview Release.
can we just select the tibial bone using totalsegmentator?
Does it show(render) urinary tract?
please someone enlighten me how to manually adding "weigth" to TotalSegmentator.
What do you mean by "weigth"?
@@AndrasLasso i was requesting the extension of totalsegmentator for bone. I got the link to download the package of bone extension. But i don't how to install it to be used in 3D Slicer.
I just ran a simulation now from this extension and the output i got is so rough and low poly, what can i do to make or improve the output to something better and smoother. Please I need your help
Make sure that you use the full-resolution mode (disable "fast" mode). You then get 1.25mm isotropic resolution.
@@AndrasLasso I didn't simulate with fast mode, that's why I'm surprised at the result I got. After almost 1hr of simulation it still came out low poly
@@chinonsopromise4962 TotalSegmentator does not generate polys. It creates a labelmap volume. Segmentation is converted to polygonal mesh (millions of triangles, I would not call low poly) and finally OpenAnatomy Export can decimate and export the meshes. You can set "Reduction factor" to a smaller value to preserve more triangles.
@@AndrasLasso I understand the other user frustration, I'm also not using the Fast option, and the voxel are visible larger than expect, like it need to use Smooth to look more like my normal results using Thresholds.
@@SilasC Thresholding uses the resolution of the segmented image. TotalSegmentator uses 1.25mm voxel size. If you need finer details for specific structures then you can use TotalSegmentator to extract the bounding box of the structure and then you can train a new model to extract only that structure in that small region.
Oh waw that's 🤩
Hello. very nice !
What's the system requierments for this extension ?
Any computer that you bought in the last 5 years can run it. No GPU is needed.
Although I get excellent results with 3D Slicer sample, I see that with my patients CT it never do all their Scapula, what is weird for me as it is so easy to get bones with the Threshold.
Maybe something is special about your CT images, something looks different than in all the images that TotalSegmentator was trained on. You can improve the model by adding your images (with your segmentations) to the training data set and train a new model.
can you install this in windows?
Yes, 3D Slicer works on Windows, Linux, and macOS. The application and this segmentation model are all free, open-source, restriction-free software.
Amazing!
Can someone explain to me how to install totalsegmentator?
See its documentation: github.com/lassoan/SlicerTotalSegmentator#setup Post any questions on the Slicer forum: discourse.slicer.org
Why can't we extract the lower arms and legs?
You may need to get the latest development version of TotalSegmentator. It is available for the latest Slicer Stable Release and latest Slicer Preview Release.
@@AndrasLasso But I see that the version of TotalSegementator in Slicer is Published on Wed Aug 02 2023, which is the latest!
@@KaMI_NIE Aug 2 is two weeks old. You need the latest.
I have erorr in download totalsgmentor
Please post your question with more details (e.g., the full error message) to the Slicer forum - discourse.slicer.org
It could not work well on ubuntu system.