Jump to content

Quality of models


Candace Moore

Recommended Posts

Over the past few weeks I've tried several models- mesh files- on the website. They were all free. The quality was below anything I would have produced by careful segmentation. Are the free models all of lower quality? Has anyone seen a higher quality mesh file that can be downloaded for free?

Link to comment
Share on other sites

  • Allen changed the title to Quality of models

What exactly are you looking for Candace? Bones? Organs? Vessels? Some structures are amenable to automated segmentation (like bones) but others are much harder and must be done manually (organs, brain, tumors). There are models that were both automatically and manually segmented in the library. In general, if the thumbnails have a blue background they are probably auto segmented. Hope this helps.

 

Mike

Link to comment
Share on other sites

The ease with which you segment depends on the algorithms and tools you use. You don't actually have to do much manually. I have some expertise in this. Have you tried Philips Intellispace Portal? The European version 10s have pretty good segmentation algorithms, and you can put in your own scripts. Anyways, thanks for actually taking the time to respond. I have now found some well segmented ones. By the way, some of the bone segmentations clearly had the HU "border" set in a non-optimal way.

Link to comment
Share on other sites

I use 3D slicer at home, but I was in a course where Philips Intellispace was taught. Both have advantages and disadvantages. I'm sure the Intellispace costs differently depending upon how it is negotiated. If you want I can send you my contact there who might know what kind of deal was negotiated for my course. These things are inherently local although the company a big multi-national. I currently live in Israel, and still had to route all my questions through that contact in Spain where the course came from, as Philips here is different. The versions of Intellispace are even slightly different depending on your region. You live in the USA, and they are sort of hyper-capitalists over there. If I were you I might try to negotiate with Europe, and claim your site is intended to be a global resource.

Link to comment
Share on other sites

It looks shiny, but in my honest opinion, Philips are not famous with their software. They have notorious "fame" among the bulgarian radiological community as the worst software with the most non-user-friendly interface. My colleagues are all into Siemens and Esaote here. Besides, Slicer 3D is free, open source program. From the commercial ones, I really like Osirix and Materialise, but the best one in my opinion is Autodesk Within Medical. The price of 17k euro per year is a bit stiff, though.

Link to comment
Share on other sites

At my department we have Intellispace Portal 7 and I am quite pleased with it, depending what you want to do with it. The major thing is that there is possibility to export .stl file from volume rendered recons (short VRT) from version 6 or 7 and above, which is not possible from Siemens Syngo. Also we have Philips Azurion C-arm and when you perform rotational intrarterial angiography it is possible to make VRT 3D model and also export it into Intellispace Portal (I did that only once when Philips aplicator was at my department but will have to do it more when I am back after my final exam). Also, I tried version Intellispace 9 or 10  (not sure which one) at RSNA meeting and it looks pretty nice, too.

The thing is that all software in workstations have the same algorithms for automated segmentation and generation VRT models that are based on different threshold  values for different tissue density and as said before, the best visualized tissues are those that have significant contrast to other tissue (like bone or contrast blood to surrounding soft tissue). Organs like liver, kidneys etc. are composed of different density tissues that have different density voxels on CT scan that can range from higher HU values in one voxel (blood vessel in post contrast scan of liver) to lower HU values (let say small area of lipids accumulated in hepatocytes), and if that area is near liver capsule where adjacent tissue is fat, you will have artefacts in your automated segmentation of liver.

That is why I love manual or semi-automated segmentation for now until AI makes automated segementation more accurate (there was post about application of AI and segmentation in 3D slicer in some other topic).

Link to comment
Share on other sites

  • Allen featured this topic

 Not all the algorithms are based on different threshold values. I don't mean to get super-technical, but I am a somewhat skilled programmer for a very limited number of things. It's a super-simple algorithm to simply put a threshold on HU. There are a bunch of other approaches algorithmically. You can also threshold based on texture analyses even in 3D slicer. I could go on here for ten paragraphs, but trust me there are a lot. AI based segmentation that is pretty accurate is available for some things...but the ones I have seen were custom built by companies for a particular entity looking to segment something specific e.g. just segment out the lungs into bronchopulmonary segments.

Link to comment
Share on other sites

  • Dr. Mike unfeatured this topic
  • 8 months later...

CT images are segmented by a difference of threshold values. Therefore, tissue that present less contrast to each other are more difficult to generate good results with automatic segmentation. 

As mentioned above, for bone biomodels, many of the software have good results, but even with them it is necessary to perform a computational modeling afterwards, to clean noise and unwanted parts that appear in the segmented biomodel. I have used the InVesalius software (https://invesalius.github.io) for many of the biomodels that I have created. I believe that to achieve High quality in the final result, dedication to the creation of the biomodel is still necessary. 

I hope I contributed to the discussion.

 

Regards, 

Vitor Urel.

Link to comment
Share on other sites

20 minutes ago, VitorUrel said:

CT images are segmented by a difference of threshold values. Therefore, tissue that present less contrast to each other are more difficult to generate good results with automatic segmentation. 

In order to go around this issue, I'm using the following technique:
- First I'm making a segmentation of the main structures of interest.
- Then, I'm exporting the segmentation and removing the unwanted parts with a surface modelling software (Meshmixer).
- Finally, I'm reimporting the model as a segmentation and I'm using it as a mask with threshold techniques.
Sometimes, I have to export/import a segmentation multiple times, but this helps me to model even the smallest and most impossible structures. Here is my latest work - a liver with a multiple necrotic lesions and vena cava.
About the software - Slicer 3D is better than inVesalius because it have more powerful options for denoising, normalization of the dataset and analysing tools. inVesalius is shinier, prettier and easier for work though.

Link to comment
Share on other sites

18 hours ago, valchanov said:

In order to go around this issue, I'm using the following technique:
- First I'm making a segmentation of the main structures of interest.
- Then, I'm exporting the segmentation and removing the unwanted parts with a surface modelling software (Meshmixer).
- Finally, I'm reimporting the model as a segmentation and I'm using it as a mask with threshold techniques.
Sometimes, I have to export/import a segmentation multiple times, but this helps me to model even the smallest and most impossible structures. Here is my latest work - a liver with a multiple necrotic lesions and vena cava.
About the software - Slicer 3D is better than inVesalius because it have more powerful options for denoising, normalization of the dataset and analysing tools. inVesalius is shinier, prettier and easier for work though.

Thanks for the answer. Certainly, when performing these subsequent segmentation steps, the biomodel will be better defined. The next biomodel that I do I will adopt this methodology and see the result. The work done with the liver and vena cava is impressive. Do you recommend me any material or video to go deeper into the practical use of segmentation techniques? 

Thank you very much.

 

Regards, 

Vitor Urel.

Link to comment
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
×
×
  • Create New...