search

neuropoly / totalspineseg

Segmentation of vertebrae, intervertebral discs, spinal cord and CSF from MRI images.
5 stars 0 forks source link

Document how to train the model #19

Closed jcohenadad closed 4 months ago

jcohenadad commented 5 months ago

In the README, add a section "Train model", with everything needed to reproduce the model. Including data selection.

yw7 commented 5 months ago

@jcohenadad Is it ok to merge https://github.com/neuropoly/totalsegmentator-mri/pull/16 then we'll orgenize all?

I suggest orgenize into 5 main folders:

SynthSeg - For the first failed try with only synthetic data. Reg2PAM50 - For the code used to register the data to pam50. nnUNetV1 - For the code used to train the model we've sent to the ISMRM24. nnUnetSacrum - For the code @NathanMolinier used to train the model of the Sacrum. nnUnetV2 - For the code we'll use to train the next version including the sacrum and SC labels in the SPIDER dataset.

In each of them 'scripts', 'resources' and README. Also in the root folder: README for general explenation and 'scripts' for general scripts used in all of the projects.

Sould we keep the scripts from the faild first try with only synthethic data (SynthSeg above)? Or leave it only in the repo history?

NathanMolinier commented 5 months ago

I personally think that we should change the structure to have a package like this repository: https://github.com/spinalcordtoolbox/disc-labeling-benchmark

yw7 commented 4 months ago

https://github.com/neuropoly/totalsegmentator-mri/pull/16 solved this issue