Active Learning - Delete previous label submission on 3DSlicer

[niklu-igloo]

Hi all,

thank you for the creation of this great project.

I have a technical question about Active Learning.

Is it possible to delete a previous submitted label of a series and restart the labeling with the Scribbles tools? In the uploaded image, I have for example two series and I have labeled one already (that is why 50% is shown under Active Learning).
Typical use case: If somebody had labeled incorrectly and pressed "Submit Label", couldn't she/he restore the original labels and restart labeling the first series?

Please let me know if the question is clear.

Best,
Nicholas

[diazandr3s]

Hi @niklu-igloo,

Glad to hear you find this framework useful.

Currently, MONAI Label does not have a label management mechanism. This means, that once the label is submitted, it can't be changed from the viewer.

Submitted labels go to the labels/final folder. You could remove that label from the folder and re-annotate the image.

Hope that helps,

[niklu-igloo]

Hi @diazandr3s,

thank you for your reply and the clear piece of information.

I have already tried to remove the label from ~/.cache/monailabel/dicom/d428d22fe8e85c96b930ae3d4bf173eb/labels/final and nothing changes from the 3DSlicer side: it still shows that 50% of the labels have been submitted. The above-mentioned folder is completely empty at the moment.

Have I modified the wrong folder?

[diazandr3s]

Hi @niklu-igloo,

I can see you are using DICOM images. For that, you'll need to set up a DICOM Web server (https://github.com/Project-MONAI/MONAILabel/tree/main/plugins/ohif#installing-orthanc-dicomweb) so MONAI Label can connect to it.
If you still want to use a local folder, you'll need to convert the images to either NIfTI or NRRD format.

Hope this helps,

[niklu-igloo]

Thank you for the hint. I will read the documentation in detail.
I have actually already set up a DICOM Web server from which I get the slices and I have already performed inference on them, which worked perfectly. The only problem is the active learning part. I guess I have to find the labels/final folder within the DICOM Web server and modify it from there, instead of changing it in the local machine on which MONAILabel is running.
Would that be the right approach?

[diazandr3s]

If you are using DICOM Web, there is no such thing as labels/final. That only applies to the local folder using the NIfTI or NRRD file format.

DICOM Web is a set of APIs for storing, retrieving, searching and deleting DICOM files. MONAI Label uses this DICOM Web client to communicate with the DICOM Web server.

Here is the dicom datastore as well: https://github.com/Project-MONAI/MONAILabel/blob/main/monailabel/datastore/dicom.py#L45

To delete a segmentation, you can either use those APIs or, if using Orthanc, you can manually delete them using the Orthanc UI

[niklu-igloo]

Thank you for the clear explanation. I have managed to delete through APIs the DICOM files that MONAILabel has stored into the DICOM Web server and the "Active Learning" bar has successfully been updated to 0%.

Thank you very much for solving this issue.

Best,
Nicholas