Change segmentation directory structure

[ghost]

Hi,
I'm trying to save my completed segmentations to a different directory structure than default: rather than folders of name 0,1,2.. I'd like to use the input case/image names: S01,S02,S03...

The code I'm using is:

saver = NiftiSaver(output_dir=outputdata_dir, dtype=(['numpy.float16']))
for i, val_data in enumerate(test_loader):
    [other stuff]
    saver.save_batch(largest)

I realise there is now SegmentationSaver, but the project is wrapping up so I'd ideally get it working with my current implementation, if possible? I've tried reading the docs and spent ages trying to adjust it myself but with no luck.

Alternatively it is is easier I'd be happy to just save all the segmentations to the main output_dir but to change the output filename to each case name eg S01_seg, S02_seg..., but again I can't figure out how to modify the default 0,1,2 etc

Any assistance would be much appreciated!

Edit: To be clear, I know in principal how to set things to the case/image ID (eg f"{case_name[i]}") but my issue is how/where I can add this arguement.

[ghost]

I'm returning to this problem and any help would be much appreciated please. Thanks in advance

[Nic-Ma]

Hi @jnc13 ,

Thanks for your interest here.
I think maybe you missed to pass the meta data to the saver, so it uses index as file names:
https://github.com/Project-MONAI/MONAI/blob/dev/monai/data/nifti_saver.py#L118
Let me try to update the doc-string to make it more clear.

Thanks.

[Nic-Ma]

Hi @jnc13 ,

Thanks for your feedback.
Let me try to enhance the saver to make it possible to not save them in separate folders.

Thanks.

[ghost]

That would be great, thanks @Nic-Ma.
I'm still a little stuck getting the saver output filenname to be the input filename. I think maybe rather than using my own list I should try to do as you suggested: "If you call the save_batch() function of saver, it expects the meta_data is collated by DataLoader, so the filename_or_obj should be a list for 1 batch:

meta_dict = {'filename_or_obj': ['TEST', ...]}
"
Could you please clarify how I can go about accessing the meta_data collated by the DataLoader to use as the list for meta_dict?

Thanks as always, I really do appreciate your help!

[Nic-Ma]

Hi @jnc13 ,

I think you can check this example deeply to see how it works: https://github.com/Project-MONAI/tutorials/blob/master/3d_segmentation/torch/unet_inference_dict.py
Especially how the SaveImaged transform works.

Thanks.

[Nic-Ma]

BTW, I created a ticket to track your feature request: #2562.

Thanks.

[Nic-Ma]

Hi @jnc13 ,

Submitted PR to add support for your feature request: #2576.

Thanks.