KeepLargestConnectedComponent

[Meddebma]

Hi,

I was using the KeepLargestConnectedComponent as a post transform for segmentation tasks and it always worked well, since two weeks it shows me a new error saying:
NotImplementedError: Only for images of dimension 1-3 are supported, got a 4D one

I am wondering why, I did not change the dimensions of my data.
here is my notebook:
https://github.com/Meddebma/pyradiomics/blob/master/3D_Segmentation_eval.ipynb

thanks for your help

[Nic-Ma]

Hi @Meddebma ,

Thanks for your interest and experiments with MONAI.
We introduced some minor breaking change in v0.6 recently, please check this doc for more details:
https://github.com/Project-MONAI/MONAI/wiki/v0.5-to-v0.6-migration-guide.

Thanks.

[Meddebma]

@rijobro
did you have the opportunity to look at the notebook what should I change to use the keeplargestcomponent?
https://github.com/Meddebma/pyradiomics/blob/master/3D_Segmentation_eval.ipynb
thank you

[Nic-Ma]

Hi @Meddebma ,

As you are using MONAI v0.6, please change the postprocessing logic to decollate batch first, then execute transforms for every channel-first item.

from monai.data import decollate_batch

val_outputs = [post_trans(i) for i in decollate_batch(val_outputs)]

Thanks.

[Meddebma]

Hi @rijobro
thanks for your help. The postprocessing worked well. I still have a problem with the evaluation,
I would like to display for every data the segmentation mask as well as segmentation metrics,

for i, val_data in enumerate(val_loader):
val_images, val_labels = val_data["image"].to(device), val_data["label"].to(device)

TypeError: new(): invalid data type 'str'

here is the notebook
https://github.com/Meddebma/pyradiomics/blob/master/3D_Segmentation_eval.ipynb

[Nic-Ma]

I feel maybe you didn't execute transforms for image and label, so val_data["image"] and val_data["label"] are still file path strings, maybe you can print out to debug?

Thanks.

[Meddebma]

when I write
"for val_data in val_loader "
it works properly,
I did a loop into another and I get this result:
https://github.com/Meddebma/pyradiomics/blob/master/3D_Segmentation_eval.ipynb