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
191 stars 28 forks source link

Added PUMA 4mm model #102

Closed mprires closed 8 months ago

mprires commented 8 months ago

Pull Request

Description:

Added the new model for PUMA 4mm isometric model

Changes Made:

None

Testing Done:

Tested on one CT

Screenshots or GIFs (if applicable):

Screenshot 2024-03-07 at 11 54 46

Reviewer Checklist:

LalithShiyam commented 8 months ago

@Keyn34, the segmentation looks quite jagged, should we smooth before registration?

Keyn34 commented 8 months ago

@LalithShiyam, I think we should do that as a postprocessing measure. Although I'm unsure if we should smooth all the segmentations or each label individually, @mprires, could you try that or let me know if I should do it?

LalithShiyam commented 8 months ago

@Keyn34 This postprocessing shouldn't be in moosez (so not @mprires's job anymore), we should move it to pumaz. I would just use c3d, since we already have it downloaded for the package.