AI-assisted segmentation using free tools - 3D Slicer and Nvidia Clara

Поділитися
Вставка
  • Опубліковано 10 лют 2025
  • This video shows how to do AI-assisted segmentation of tumors and organs on CT and MRI images using Nvidia Clara in 3D Slicer. The software is free, open-source, available on all desktop platforms. Please post any questions to 3D Slicer forum: discourse.slic...

КОМЕНТАРІ • 28

  • @gregortidholm
    @gregortidholm 4 роки тому +1

    This is very useful! Thank you for sharing!

  • @Selmaelgrini
    @Selmaelgrini 10 місяців тому

    Please, is there any extension for automatic segmentation in 3D Slicer

  • @Muzahem96
    @Muzahem96 3 роки тому +1

    Annotation ct liver is not showing , is there a solution?

  • @CraigDaly
    @CraigDaly 4 роки тому

    That looks fantastic. Thanks.

  • @mikolajbuchwald
    @mikolajbuchwald 2 роки тому

    Very impressive!

  • @radyolojigrubu6256
    @radyolojigrubu6256 2 роки тому

    I want to use these segmentatitons to calculate some texture parameters. How can I do it? Is it only the 2D/3D presentations?

    • @AndrasLasso
      @AndrasLasso 2 роки тому

      You can segment some regions using Segment Editor module and then compute texture features in each region using SlicerRadiomics extension.

  • @tylerchan5251
    @tylerchan5251 3 роки тому +1

    6 sides boundary points are marked, press Start, Sent. --> Status: 500; Response: 500 Internal Server Error.

    • @PerkLabResearch
      @PerkLabResearch  3 роки тому +1

      The demo server needs an upgrade. If this model keeps failing then you need to set up your own server or wait for the demo server"s upgrade (bu twe don't know when it will happen).

    • @thalialeytonreto3498
      @thalialeytonreto3498 3 роки тому

      @@PerkLabResearch I just just got the same error! How can we set up our own server?

    • @PerkLabResearch
      @PerkLabResearch  3 роки тому +1

      @@thalialeytonreto3498 See instructions here: github.com/NVIDIA/ai-assisted-annotation-client/tree/master/slicer-plugin#setup

  • @kenny_tukin
    @kenny_tukin 4 роки тому

    Do the user has to send the images to the NVIDIA server for the segmentation or the AI-segmentation tool is running completely in the local PC?

    • @AndrasLasso
      @AndrasLasso 4 роки тому +1

      It is up to you. You can either set up your own AIAA server (on your computer or anywhere else locally or remotely) or use Slicer's default server, which is currently hosted at Queen's University in Canada.

    • @davidsibrina598
      @davidsibrina598 4 роки тому

      @@AndrasLasso Can you please give me more information on where to connect to the Slicer default server. I would love to try this out. Thank you

    • @AndrasLasso
      @AndrasLasso 4 роки тому +2

      @@davidsibrina598 As described in the documentation (github.com/NVIDIA/ai-assisted-annotation-client/tree/master/slicer-plugin#nvidia-ai-assisted-annotation-aiaa-for-3d-slicer) leave the server field empty to use the default server.

  • @user46346bdtgry
    @user46346bdtgry 4 роки тому

    where can i get more pre-trained ai models for ct (healthy) organ segmentation?

    • @PerkLabResearch
      @PerkLabResearch  3 роки тому +2

      You can download models from NVidia NGC and there are various other model zoos (check out monai.io).

    • @brad8122
      @brad8122 2 роки тому

      ​@@PerkLabResearch Can we do this segmentation for a CT scan without any injected contrast, ? yours look like it has injected contrast for those organs?

    • @PerkLabResearch
      @PerkLabResearch  2 роки тому +1

      @@brad8122 Certain structures and pathologies are only visible when contrast is used. Whatever is visible, you can train a model to recognize it. There have been a lot of progress in the field since the video was created. I would recommend to check out TotalSegmentator (100+ structures segmented in 2 minutes from any CT - see discourse.slicer.org/t/26710) and MONAILabel (for training your own model - see docs.monai.io/projects/label/en/latest/quickstart.html)

    • @brad8122
      @brad8122 2 роки тому

      @@PerkLabResearch Thank you very very much but It raised lots of questions for me.
      I am seeing lots of wavy structures on a kidney surface, also lots of spikes?
      drive.google.com/file/d/1lcXFsDPfkMfdMLbz4-fylDFITw2lalTG/view?usp=share_link
      i.imgur.com/Eg6kwzp.png
      Why can it not show urinary tract canals/ureters for example? Can I train (or maybe just with a simple setting change) it to show those little details too?

    • @AndrasLasso
      @AndrasLasso 2 роки тому

      @@brad8122 There is no such thing as a perfect segmentation - it has to be just good enough for the specific clinical purpose. If you find that an AI model is not accurate enough for your application then you can add more and/or more accurate training data and and train more. Since all TotalSegmentator data sets are openly available, you can improve them and contribute back your changes and the maintainers can incorporate those in the next release.

  • @arazmed
    @arazmed Рік тому

    İ get server error. How can i solve this

    • @PerkLabResearch
      @PerkLabResearch  Рік тому +1

      We switched to TotalSegmentator - github.com/wasserth/TotalSegmentator - for CT segmentation, which segments 100+ structures in 2 minutes, even without GPU; and MONAILabel - github.com/Project-MONAI/MONAILabel/tree/main/plugins/slicer#readme - that allows you to train your own models.

    • @arazmed
      @arazmed Рік тому

      @@PerkLabResearch thank you

    • @JOc17KER
      @JOc17KER Рік тому

      so the nvdia AIAA cannot be use anymore?@@PerkLabResearch

    • @wanderingamerica5719
      @wanderingamerica5719 Рік тому

      @@PerkLabResearch do you have any tutorial for it?

    • @aakib_parray1470
      @aakib_parray1470 6 місяців тому

      It shows server error how to resolve it