MONAI - Local Project "Tree" and Configuration - Results

Support
, , ,
1

Dear colleagues,
I have a question about the results of MONAI Label Service which I’m using on my tower-server with Asus Nvidia GeForce RTX 3080 10GiB based on Windows Server 2019 OS.

  1. We are testing two project structures:

  1. Results of v1 (not bad results)
    res_v1_1.jpg - Google Drive
    res_v1_2.jpg - Google Drive

  2. Results of v2 (bad results)
    res_v2.png - Google Drive

  3. The questions are:
    4.1. What do we need to put in FINAL if we are searching to find the solution for multiple-segmentational project structure?
    4.2. Do we need to put a nifti file with SCALAR RANGE 0…15 (equal to labels quantity) in FINAL?
    4.3. Where are all of this files needs to be located? Is there any rules or manuals about to know this part of Monai workflow? Smth like: “Working with labels on the local Monai server. Guide to action”?
    4.4. As I’ve understood correctly, the correct abbreviation for these nifti files is labels? Is it right?
    4.5. Where and how to I need to correct configuration or code to tell the Monai my research structure? Because this is brand new type.
    4.6. What algorithm I need to use in future for reconstruction the segmentation of missing tooth?
    It means that I need to generate the segmentation of missed tooth by NN. And it will be locate on the proper place.
    4.7. How to learn NN for possibilities to create segmentations in blank places?

Kind regards,
Dalv Silvermann.

1 Like
2

Hi @dalv.silvermann,

It is great to hear you’re using MONAI Label.

Tooth segmentation is a nice use case - a bit challenging though.

Of the tree images you’ve shared, MONAI Label supports version 1. Segmentations per volume should be in one file as a label map.

If you already have annotations for some of the volumes, they should be located in the folder dataset/labels/final with the same name as the volumes.

Each segment (tooth) should be consistently annotated. This means, it should have the same index/scalar in all images. If there are teeth missing, indexes/scalars representing the available teeth should still be the same in all images.

Here you can find a video tutorial for MONAI Label: MONAI Label - Installation with PyPi, Docker, and GitHub - YouTube

Additionally, here you can find more information regarding the tree: MONAI Label Workshop - Project Week - YouTube

BTW, I’d suggest you post these MONAI Label-related questions directly in the MONAI Label repository under the discussion section: Discussions · Project-MONAI/MONAILabel · GitHub
You’ll get faster replies from the other MONAI Label developers as well.

Hope this helps,

Andres

1 Like