search

BailiangJ / spine-ct-mr-registration

Weakly-supervised Biomechanically-constrained CT/MRI Registration of the Spine
MIT License
23 stars 3 forks source link

Segmentations for training #3

Open bhatrana opened 1 year ago

bhatrana commented 1 year ago

Hi @BailiangJ, Thanks for code and repo. What if there is no segmentation for training ? Is this repo customized for this case ? Is it possible to put step wise training process with commands in readme ?

Thanks Raj

BailiangJ commented 1 year ago

Hi @bhatrana , the method requires the segmentation mask of the rigid objects (at least in moving images) for training. Otherwise, it's just a deformable registration network VoxelMorph.

BailiangJ commented 1 year ago

If you are testing this method on spine CT data and need labels, it is recommended to use the webtool Anduin to generate the segmentations.

zhuczhuc commented 6 months ago

Hi @bhatrana , the method requires the segmentation mask of the rigid objects (at least in moving images) for training. Otherwise, it's just a deformable registration network VoxelMorph.

Hi @BailiangJ, how to only apply the moving (CT） mask? I can see the MR mask is necessary for the code.

BailiangJ commented 6 months ago

Hi @zhuczhuc , the MR mask is only for dice score tracking during training, it's not used in training loss computation.

You can comment out "mr_mask" related keys in the load_data script and "target_oh" related codes, dice metric computation in the epoch_run script.