Good morning,
I’m interested in performing segmentation of large nodule from CT dicom images.
Is Lung CT Segmenter Module useful to identify large nodule or other abnormalities?
Can you please suggest me the better segmentation process to do it?
Thanks
Lung CT Segmenter itself segments lungs, lobes, airways and a vesselmask. It creates an empty segment for “tumor” and we do not yet have nodule or tumor recognition.
For segmenting the tumor you could for example use the “Segment Editor” and it’s “Local Threshold” effect as shown here:
CTRL + left click into the tumor
Thank you very much for reply and for information.
Just another question in order to be sure to make sure I understood.
Is Lung CT Segmenter Module able to independently identify a region that is anomalous compared to the anatomical regions it segments?
Thanks
The Segmenter just segments the basic structures, but you can use the Lung CT Analyzer after the Segmenter to quantify infilrations or lung collapse of any cause (evaluated for COVID) as well as fibrosis and emhysema.
Dear @rbumm,
thank you very much for support and suggestion.
I performed segmentation using Lung CT segmenter Module. It created a volume “tumor” that was empty. At this point, I tried the Local Threshold segmentation, as you suggested, but after CTRL + left click into the tumor, the region was not identified. Extraeffects module was installed.
Can you please suggest me a solution?If you want, I can share with you private copy of the dicom.
Thanks
No problem, you can DM a link to your data, I can give it a try.
Best
r
So this is quite a noisy dataset - but the processing works.
- Run the dataset through the Lung CT Segmenter with airway segmentation and AI TotalSegmentator lung basic.
This will give you the lungs, lobes and the unelegant airway.
If you do not need the lungs or lobes skip this part. But at least admire the quality that AI can segment the lobes from these noisy data.
Then go Segment editor and follow these steps very carefully.
- Disable the upper lobe
- Select the “tumor” segent
- Make it visible (check the box)
- Select the local threshold effect
- Allow Overlap
- Set minimum diameter to 3 mm
- Now the most important part: Adjust the threshold in a way that only bones and the tumor are flickering. (-339 to 3071)
- Then CTRL + left click into the tumor and wait.
- This clip is unedited.
- You will have to do some postprocessing to remove the vascular “fingers” from the tumor.
Good luck.
It you fail, just DM me for a Zoom call.
Best
r
Thanks a lot.
I will follow all your suggestions and I will inform you on results.
Thanks
Dear @rbumm,
can you please share with me also the video of the first part (Lung Segmenter module).
I have a different output also from the first part.
Did you use Interactive segmentation module?I have no identification of upper and lower lobes.
Concerning the step “CTRL + left click into the tumor and wait”, it takes a long time.
I’m waiting ;-(
Can you share your system specifications ? What computer, operating system, graphic system, does your system have a GPU ? What brand and model ?
For the first step, did you use Interactive Lobe Segmentation module?
Thanks
No, I used the Lung CT Segmenter, part of the Lung CT Analyzer module. You will get this kind of automatic lobe segmentation by using the AI functions and selecting TotalSegmentator lung basic. However, unfortunately, your computer is too weak to run such sophisticated software. You will need a Windows or Linux system with an NVIDIA GPU with at least 8 GB of GPU memory. Alternatively, you could rent a GPU-enabled system in the cloud (Google, Amazon). But this would require a lengthy setup process and running costs of around 2-3 dollars per hour. I would recommend investing in a future-proof hardware system. The 3D Slicer guys should be able to provide recommendations for such a 3D Slicer and AI optimized configuration.
Thank you very much for all.
Your support and kindness are absolutely precious.
Thanks again
How many datasets do you have to analyze?
Dear @rbumm,
It depends on how well this pipeline performs.
Is there a way to improve the nodule segmentation result (except having better quality images)?
Does 3D Slicer include a module for the dimensional analysis of segmentations? In the Statistics module you can view volumes and surfaces. Any other parameters?
There are a few things that initially work as intended by the researcher. But 3D Slicer is one if not the most cited medical image application to date.
This is just a start for experimentation, even commercial software has no reliable tumor or nodule recognition.
Yes, you have segment statistics and radiomics.
Try it out 
there are many others.
I would like to compare the segmentation performed using the pipeline you suggested me (Lung Segmentation with AI and airways and Local Threshold for nodule) with the manual segmentation of large nodule from TC using the free database SIMBA (Patient SL0058).
Results from the first case showed that a volume of nodule segment is around 18164 mm^3 from manual segmentation compared to 11695.2 mm^3 from Slicer (in attach the two results).
The difference seems to be large.
Can you please suggest a way to improve this problem?
Do you think that with a cleaner database we can improve this difference? Do you have information on another free CT database in which nodules are evaluated?
Thanks
Is this is the same dataset ? The one you have sent is SL0058 ? Because the one you sent is is identified S02A01 …
Yes, is the same I shared with you @rbumm. The master folder is named SL0058.
It contains a subfolder S02A01.