App SDK updated to be compatible with monai 1.0.0 and its MetaTensor

[MMelQin]

The pull request is to make App SDK compatible with monai 1.0.0, specifically the use of MetaTensor

• Removed the use of ToTensorD in the inference pre-processing step
• Updated the InMemImageReader to make use the newly available MetaKeys class etc
• Tested the example apps successfully with pip installed monai 1.0.0
• Update the bundle inference operator to pass keyed input MetaTensors, e.g. image, to post_processing, since the metadata dict is now encapsulated in the MetaTensors
• Updated the Jupyter Notebooks. NOTE: testing Clara-Viz notebook needs to pip install ipywidgets==7.7.2 before the clara-viz package is updated to be compatible with newer ipywidgets
• Updated the PIN integration code where monai transforms are used
• Updated the Seg Writer to add Series Date, Time, and UTC time zone offset, as well as adding support for setting custom tags, e.g. SeriesDescription as it is missing from highdicom lib.

Signed-off-by: M Q mingmelvinq@nvidia.com

[vikashg]

@MMelQin Do we have the sample data for liver tumor somewhere that we can test.

[vikashg]

For the notebook tutorials I am getting the following error.

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
Cell In [6], line 1
----> 1 app = AISpleenSegApp()
      3 app.run(input="dcm", output="output", model="model.ts")

Cell In [5], line 4, in AISpleenSegApp.__init__(self, *args, **kwargs)
      3 def __init__(self, *args, **kwargs):
----> 4     super().__init__(*args, **kwargs)

File ~/Documents/MONAI-DEPLOY/monai-deploy-app-sdk/monai/deploy/core/application.py:126, in Application.__init__(self, runtime_env, do_run, path)
    123 self._builder()
    125 # Compose operators
--> 126 self.compose()
    128 if do_run:
    129     self.run(log_level=args.log_level)

Cell In [5], line 12, in AISpleenSegApp.compose(self)
     10 series_to_vol_op = DICOMSeriesToVolumeOperator()
     11 # Creates DICOM Seg writer with segment label name in a string list
---> 12 dicom_seg_writer = DICOMSegmentationWriterOperator(seg_labels=["Spleen"])
     14 # Creates the model specific segmentation operator
     15 spleen_seg_op = SpleenSegOperator()

TypeError: __init__() missing 1 required positional argument: 'segment_descriptions'

[vikashg]

Hi @MMelQin, @gigony do we have the liver image somewhere on google drive that I can use to run some tests. Notebooks are self contained so I can run and test. I got an error, I think its a small one.

[MMelQin]

Hi @MMelQin, @gigony do we have the liver image somewhere on google drive that I can use to run some tests. Notebooks are self contained so I can run and test. I got an error, I think its a small one.

Thanks @vikashg I'll update the Notebook. It was in fact a different issue, originated from the change of DICOM Seg Writer API. I was planning to get it address in a separate PR, but will include it in this one.

For Liver Tumor test dataset, the test data was from Medical Decathlon in NIfTI format originally, and then converted to DICOM using 3D Slicer. The original image was oriented using Neurology viewing convention, RAS, instead of LPS for Rad, so the DICOM image is shown flipped in MicroDicom.

[Nic-Ma]

Do we need deepcopy here?

Thanks.

[MMelQin]

The InvertD has changed to look for image dict name instead of relying on image_metadata_dict. It was not clear to me if it matters the [image] value is before or post inference or even preprocessing. So, the intend here is really just to keep a reference to the value part of dict for adding back the key, image, because of the pop of key, image. I tried not to use pop but somehow directly using inputs[first_input_name].to(self._device) did not work.

[sonarqubecloud]

Kudos, SonarCloud Quality Gate passed!   

0 Bugs
0 Vulnerabilities
0 Security Hotspots
2 Code Smells

No Coverage information
0.0% Duplication

[wyli]

Thanks, it looks good to me.