DICOM-RT to mask python package used by https://github.com/microsoft/InnerEye-deeplearning
Перейти к файлу
Anton Schwaighofer 51721527f8
Update README.md (#19)
2024-03-21 09:55:06 +00:00
.github/workflows 🐛 Reactivate testpypi jobs when tagging (#11) 2022-05-17 13:00:17 +01:00
Echo Init 2021-03-25 16:08:33 +00:00
RTConvert Fix DICOM Eclipse Issue (#16) 2022-10-12 16:19:50 +01:00
build_ado Init 2021-03-25 16:08:33 +00:00
src/InnerEye_DICOM_RT Init 2021-03-25 16:08:33 +00:00
tests Fix DICOM Eclipse Issue (#16) 2022-10-12 16:19:50 +01:00
.flake8 Init 2021-03-25 16:08:33 +00:00
.gitattributes Init 2021-03-25 16:08:33 +00:00
.gitignore Init 2021-03-25 16:08:33 +00:00
CODE_OF_CONDUCT.md Init 2021-03-25 16:08:33 +00:00
LICENSE Init 2021-03-25 16:08:33 +00:00
MANIFEST.in Init 2021-03-25 16:08:33 +00:00
Makefile Init 2021-03-25 16:08:33 +00:00
README.md Update README.md (#19) 2024-03-21 09:55:06 +00:00
SECURITY.md Init 2021-03-25 16:08:33 +00:00
SUPPORT.md Init 2021-03-25 16:08:33 +00:00
build.sh Init 2021-03-25 16:08:33 +00:00
build_requirements.txt Change Segmentation timestamp from UTC to local time (#15) 2022-10-12 15:58:58 +01:00
mypy.ini Init 2021-03-25 16:08:33 +00:00
pytest.ini Init 2021-03-25 16:08:33 +00:00
setup.py 📌 Pin dotnetcore2 version (#8) 2022-05-16 11:10:53 +01:00
test_requirements.txt Init 2021-03-25 16:08:33 +00:00

README.md

This project is now archived

This project is no longer under active maintenance. It is read-only, but you can still clone or fork the repo. Check here for further info. Please contact innereye_info@service.microsoft.com if you run into trouble with the "Archived" state of the repo.

Introduction

InnerEye-DICOM-RT contains tools to convert medical datasets in NIFTI format to DICOM-RT. Datasets converted using this tool can be consumed directly by InnerEye-DeepLearning.

Most of the work is done by a .NET Core 2.1 project in RTConvert, written in C#. There is a very lightweight wrapper around this so that it can be consumed from Python. The wrapper relies on the PyPI package https://pypi.org/project/dotnetcore2/ which wraps up .NET Core 2.1.

Installing

Git for Windows

Get the installer from Git for Windows

The installer will prompt you to "Select Components". Make sure that you tick

  • Git LFS (Large File Support)
  • Git Credential Manager for Windows

After the installation, open a command prompt or the Git Bash:

  • Run git lfs install to set up the hooks in git
  • Run git config --global core.autocrlf true to ensure that line endings are working as expected

Clone the InnerEye-DICOM-RT repository on your machine: Run git lfs clone --recursive https://github.com/microsoft/InnerEye-DICOM-RT

Visual Studio / .NET Core

The C# components can be built with the .NET Core SDK. We use version 2.1 for compatibility with the PyPI package dotnetcore2. Installation instructions are here: https://docs.microsoft.com/en-us/dotnet/core/install/. Visual Studio is not required to build, but if you wish to use it then for .Net Core 2.1 you need at least: Visual Studio 2017 version 15.7.

RTConvert

RTConvert can be built from a .NET Core command line:

dotnet build RTConvert

There are unit tests:

dotnet test RTConvert

Note that the unit tests have a dependency on System.Drawing and that on Linux System.Drawing requires a native package:

apt-get -s install libgdiplus

Finally, for consumption by the Python wrapper, this solution must be published:

dotnet publish RTConvert --configuration Release -p:Platform=x64

This should create a folder with all the requirements for RTConvert at: RTConvert/Microsoft.RTConvert.Console/bin/x64/Release/netcoreapp2.1/publish/*

Echo

Echo is a very simple application that takes 1 or 2 arguments. The first is echoed to stdout, and if a second argument is supplied then it is echoed to stderr. This is only required for units tests to establish that a .NET Core application can be called.

Echo can be built from a .NET Core command line:

dotnet build Echo

There are no unit tests.

Finally, for consumption by the Python wrapper, this solution must be published:

dotnet publish Echo --configuration Release -p:Platform=x64

This should create a folder with all the requirements for Echo at: Echo/Echo/bin/x64/Release/netcoreapp2.1/publish/*

Python

The Python wrapper is in src/InnerEye_DICOM_RT/nifti_to_dicom_rt_converter.py. It simply uses subprocess.Popen to invoke the .NET Core application passing in the relevant dll and command line arguments.

It does require that the RTConvert and Echo published packages are copied to the folder: src/InnerEye_DICOM_RT/bin/netcoreapp2.1.

Note that the github build action does this automatically, but if testing then this needs to be done manually.

The Python package is created with:

python setup.py sdist bdist_wheel

which builds a source distribution and wheel to the dist folder.

To run the Python tests:

pip install pytest dotnetcore2
pytest tests

Usage

To consume this package:

pip install InnerEye-DICOM-RT

To call RTConvert:

    from InnerEye_DICOM_RT.nifti_to_dicom_rt_converter import rtconvert

    (stdout, stderr) = rtconvert(
        in_file=NiftiSegmentationLocation,
        reference_series=DicomVolumeLocation,
        out_file=OutputFile,
        struct_names=StructureNames,
        struct_colors=StructureColors,
        fill_holes=FillHoles,
        roi_interpreted_types=ROIInterpretedTypes,
        manufacturer=Manufacturer,
        interpreter=Interpreter,
        modelId=ModelId
    )

where:

  • in_file is the path to the input Nifti file. This file is a 3D volume in Nifti format.
  • reference_series is the path to the input folder containing the reference DICOM series;
  • out_file is the path to the output DICOM-RT file;
  • struct_names is a list of structure names like: ["External", "parotid_l", "parotid_r", "smg_l"]. Each structure name corresponds to a non-zero voxel value in the input volume. In the example External corresponds to voxel value 1, parotid_l to 2, etc. Voxels with value 0 are dropped. If there are voxels without a corresponding structure name, they will also be dropped. The structure name will become its ROI Name in the Structure Set ROI Sequence in the Structure Set in the DICOM-RT file.
  • struct_colors is a list of structure colors in hexadecimal notation like: ["000000", "FF0080", "00FF00", "0000FF"]. Each color in this list corresponds to a structure in struct_names and will become its ROI Display Color in the ROI Contour Sequence in the ROI Contour in the DICOM-RT file. If there are less colors than struct_names, or if an entry is empty, the default is red (FF0000);
  • fill_holes is a list of bools like: [True, False, True]. If there are less bools than struct_names, or if an entry is empty, the default is false. If True then any contours found per slice will have their holes filled, otherwise contours will be returned as found.
  • modelId Model name and version from AzureML. E.g. Prostate:123
  • manufacturer Manufacturer for the DICOM-RT (check DICOM-RT documentation)
  • interpreter Interpreter for the DICOM-RT (check DICOM-RT documentation)
  • roi_interpreted_types is a list of ROIInterpretedType. Possible values (None, CTV, ORGAN, EXTERNAL).

Segmentation Timestamps

In the process of converting from NIFTI masks to DICOM, InnerEye-DICOM-RT adds a timestamp tag to the series. This timestamp will be set to the local time of the compute resource that rtconvert() is run on. This is as some DICOM readers do not handle timezones / UTC offsets from these tags and will result in erroneous times being displayed. Therefore, if your compute resource is running in a different timezone to the timezone you are viewing the DICOM files in, the timestamps may not be what you are expecting.

License

MIT License

You are responsible for the performance, the necessary testing, and if needed any regulatory clearance for any of the models produced by this toolbox.

Contributing

This project welcomes contributions and suggestions. Most contributions require you to agree to a Contributor License Agreement (CLA) declaring that you have the right to, and actually do, grant us the rights to use your contribution. For details, visit the Micrsoft CLA page.

When you submit a pull request, a CLA bot will automatically determine whether you need to provide a CLA and decorate the PR appropriately (e.g., status check, comment). Simply follow the instructions provided by the bot. You will only need to do this once across all repos using our CLA.

This project has adopted the Microsoft Open Source Code of Conduct. For more information see the Code of Conduct FAQ or contact opencode@microsoft.com with any additional questions or comments.

Trademarks

This project may contain trademarks or logos for projects, products, or services. Authorized use of Microsoft trademarks or logos is subject to and must follow Microsoft's Trademark & Brand Guidelines. Use of Microsoft trademarks or logos in modified versions of this project must not cause confusion or imply Microsoft sponsorship. Any use of third-party trademarks or logos are subject to those third-party's policies.