ENHANCE-PET / MOOSE

MOOSE (Multi-organ objective segmentation) a data-centric AI solution that generates multilabel organ segmentations to facilitate systemic TB whole-person research.The pipeline is based on nn-UNet and has the capability to segment 120 unique tissue classes from a whole-body 18F-FDG PET/CT image.
https://enhance.pet
GNU General Public License v3.0
196 stars 28 forks source link

Added the new body model #86

Closed mprires closed 12 months ago

mprires commented 12 months ago

Pull Request

Description:

The last body model had holes when inferring the body model. Changed it with a new model trained on 1.4k manually verified datasets.

Changes Made:

Related Issues:

-None

Testing Done:

Tested the model. Screenshot below

Screenshots or GIFs (if applicable):

Screenshot 2023-11-27 at 15 40 37

Reviewer Checklist:

Additional Notes:

Reviewer(s): @[mention reviewers here]

Assignee: @[mention assignee here]

Labels: [List any labels that apply to this PR, such as bug, feature, enhancement, etc.]

Milestone: [Specify the milestone if applicable.]

Linked Pull Requests: None

Joon-Jung commented 8 months ago

Hi @mprires !

I am interested in which datasets were used in the training. Could you share more details about them (e.g., publicly available, annotations)? Thanks in advance and looking forward to your reply.

LalithShiyam commented 8 months ago

Hi @Joon-Jung, Thanks for reaching out. We are using our in-house datasets for training and currently they are not open-source. We are negotiating with the concerned parties to make it open source. But I can't give you a concrete answer until June. Cheers, Lalith

Joon-Jung commented 8 months ago

Hi Lalith (@LalithShiyam),

Thank you for your prompt response. It's unfortunate that the dataset is not yet publicly available. I look forward to good news in the near future. Thank you very much!

Best regards, Joon