saving segmentation with original spacing

[Meddebma]

Hi,
I much appreciated the newly added original image spacing to the spleen tutorial:

https://github.com/Project-MONAI/tutorials/blob/master/3d_segmentation/spleen_segmentation_3d.ipynb

I had the problem until now that the saved the segmentation masks did not fit to original images in the Nifti Reader.
Can you please tell me (or add it in the tutorial) how to save the segmentations with the original image spacing?

thanks

[Nic-Ma]

Hi @Meddebma ,

Thanks for your interest here.
Please refer to this example tutorial for saving after inverting:
https://github.com/Project-MONAI/tutorials/blob/master/3d_segmentation/torch/unet_inference_dict.py#L88

Thanks.

[Meddebma]

Hi @Nic-Ma,

thanks for your answer it worked perfectly, I get though two channels (Segmentation and background), is it possible to save only the segmentation result?
I saw there is a Parameter squeeze_end_dims in SaveImaged

[Nic-Ma]

Hi @Meddebma ,

You can add a SplitChanneld transform to split the data.

Thanks.

[Meddebma]

Hi @Nic-Ma,
so I tried this:
post_transforms = Compose([
EnsureTyped(keys="pred"),
Invertd(
keys="pred",
transform=val_org_transforms,
orig_keys="image",
meta_keys="pred_meta_dict",
orig_meta_keys="image_meta_dict",
meta_key_postfix="meta_dict",
nearest_interp=False,
to_tensor=True,
),
KeepLargestConnectedComponentd(keys="pred",applied_labels=[1]),
AsDiscreted(keys="pred", argmax=True, to_onehot=True, n_classes=2),
AsDiscreted(keys="label", to_onehot=True, n_classes=2),
SplitChanneld(keys="pred", channel_dim=0),
SaveImaged(keys="pred", meta_keys="pred_meta_dict", output_dir="/content/drive/My Drive/Spleen_AI/CH Data/mask", output_postfix="seg", resample=False)
])
It did not change anything, then I changed the channel_dim to 1, and I've got also two channels saved out

[Nic-Ma]

Hi @Meddebma ,

When splitting channels, the transform will save the results into different keys, so please set the key in SaveImaged accordingly instead of the original pred.
Refer to:
https://github.com/Project-MONAI/MONAI/blob/dev/monai/transforms/utility/dictionary.py#L345

Thanks.

[Nic-Ma]

Hi @rijobro ,

I am not very familiar with the 3D view tools, do you have any idea why the split prediction is all black in viewer?
@Meddebma , is there a way to print out the saved raw data to see what's in the array?

Thanks.

[Meddebma]

Hi,
sure here are the saved data diplayed in the notebook
https://github.com/Meddebma/pyradiomics/blob/master/Untitled1.ipynb

[Nic-Ma]

Could you please help try again with below postprocessing transforms?

post_transforms = Compose([
    EnsureTyped(keys="pred"),
    Invertd(
        keys="pred",
        transform=val_org_transforms,
        orig_keys="image",
        meta_keys="pred_meta_dict",
        orig_meta_keys="image_meta_dict",
        meta_key_postfix="meta_dict",
        nearest_interp=False,
        to_tensor=True,
    ),
    AsDiscreted(keys="pred", argmax=True),
    SaveImaged(keys="pred", meta_keys="pred_meta_dict", output_dir="/content/drive/My Drive/Spleen_AI/CH Data/mask1", output_postfix="seg", resample=False)
])

I checked your code and I think you don't need to change the prediction to One-Hot after Argmax because you don't compute metrics.

Thanks.

[Meddebma]

Dear @Nic-Ma
thank you so much it worked perfectly and the saved output correspond exactly to the original image in the Nifti Readers. Appreciate your continuous help and tremendous work with MONAI!

[Nic-Ma]

NP, glad to see your good progress!
Thanks.