Update dicom-rt converter (#430)

* Update dicom-rt converter

* Fix base classes to be able to pass roi interpreted types

* Update score.py

* Add Changelog

* Update Changelog

* Fix config

* Fix config name

* Fix variable name

* Fix variable name

* Fix variable name

* Fix tests

* Fix inference

* Fix inference score.py tests

CHANGELOG.md

@@ -10,7 +10,9 @@ created.
 
 ## Upcoming
 
-### Added 
+### Added
+- ([#430](https://github.com/microsoft/InnerEye-DeepLearning/pull/430)) Update conversion to 1.0.1 InnerEye-DICOM-RT to add: manufacturer, 
+ SoftwareVersions, Interpreter and ROIInterpretedTypes.
 - ([#385](https://github.com/microsoft/InnerEye-DeepLearning/pull/385)) Add the ability to train a model on multiple
 nodes in AzureML. Example: Add `--num_nodes=2` to the commandline arguments to train on 2 nodes.
 - ([#366](https://github.com/microsoft/InnerEye-DeepLearning/pull/366)) and

InnerEye/ML/config.py

@@ -433,6 +433,15 @@ class SegmentationModelBase(ModelConfigBase):
  "output of the model for that class includes postprocessing to fill holes, "
  "in the same order as in ground_truth_ids_display_names")
 
+ roi_interpreted_types: List[str] = param.List(None, class_=str, bounds=(1, None), instantiate=False,
+ allow_None=True,
+ doc="List of str with the ROI interpreted Types. Possible values "
+ "(None, CTV, ORGAN, EXTERNAL)")
+
+ interpreter: str = param.String("Default_Interpreter", doc="The interpreter that created the DICOM-RT file")
+
+ manufacturer: str = param.String("Default_Manufacturer", doc="The manufacturer that created the DICOM-RT file")
+
  _inference_stride_size: Optional[TupleInt3] = IntTuple(None, length=3, allow_None=True,
  doc="The stride size in the inference pipeline. "
  "At most, this should be the output_size to "

InnerEye/ML/configs/segmentation/BasicModel2Epochs.py

@@ -27,6 +27,7 @@ def __init__(self, **kwargs: Any) -> None:
  ground_truth_ids_display_names=fg_classes,
  colours=[(255, 255, 255)] * len(fg_classes),
  fill_holes=[False] * len(fg_classes),
+ roi_interpreted_types=["ORGAN"] * len(fg_classes),
  mask_id="heart",
  norm_method=PhotometricNormalizationMethod.CtWindow,
  level=50,

InnerEye/ML/configs/segmentation/GbmBase.py

@@ -33,6 +33,7 @@ def __init__(self, **kwargs: Any) -> None:
  ground_truth_ids_display_names=fg_classes,
  colours=[(255, 255, 255)] * len(fg_classes),
  fill_holes=[False] * len(fg_classes),
+ roi_interpreted_types=["ORGAN"] * len(fg_classes),
  num_dataload_workers=8,
  mask_id=None,
  norm_method=PhotometricNormalizationMethod.MriWindow,

InnerEye/ML/configs/segmentation/HeadAndNeckBase.py

@@ -39,6 +39,7 @@ def __init__(self,
  ground_truth_ids_display_names: Optional[List[str]] = None,
  colours: Optional[List[TupleInt3]] = None,
  fill_holes: Optional[List[bool]] = None,
+ roi_interpreted_types: Optional[List[str]] = None,
  class_weights: Optional[List[float]] = None,
  slice_exclusion_rules: Optional[List[SliceExclusionRule]] = None,
  summed_probability_rules: Optional[List[SummedProbabilityRule]] = None,
@@ -53,6 +54,8 @@ def __init__(self,
  present then must be of the same length as ground_truth_ids.
  :param fill_holes: Optional list of fill hole flags. If
  present then must be of the same length as ground_truth_ids.
+ :param roi_interpreted_types: Optional list of roi_interpreted_types. If
+ present then must be of the same length as ground_truth_ids.
  :param class_weights: Optional list of class weights. If
  present then must be of the same length as ground_truth_ids + 1.
  :param slice_exclusion_rules: Optional list of SliceExclusionRules.
@@ -65,6 +68,7 @@ def __init__(self,
  num_structures = len(ground_truth_ids)
  colours = colours or generate_random_colours_list(RANDOM_COLOUR_GENERATOR, num_structures)
  fill_holes = fill_holes or [True] * num_structures
+ roi_interpreted_types = roi_interpreted_types or ["ORGAN"] * num_structures
  ground_truth_ids_display_names = ground_truth_ids_display_names or [f"zz_{x}" for x in ground_truth_ids]
  # The amount of GPU memory required increases with both the number of structures and the
  # number of feature channels. The following is a sensible default to avoid out-of-memory,
@@ -115,6 +119,7 @@ def __init__(self,
  largest_connected_component_foreground_classes=ground_truth_ids,
  colours=colours,
  fill_holes=fill_holes,
+ roi_interpreted_types=roi_interpreted_types,
  class_weights=class_weights,
  slice_exclusion_rules=slice_exclusion_rules,
  summed_probability_rules=summed_probability_rules,

InnerEye/ML/configs/segmentation/HelloWorld.py

@@ -72,6 +72,7 @@ def __init__(self, **kwargs: Any) -> None:
  # Post-processing - in this section we define our post processing configurations, in this case
  # we are filling holes in the generated segmentation masks for all of the foreground classes.
  fill_holes=[True] * len(fg_classes),
+ roi_interpreted_types=["ORGAN"] * len(fg_classes),
 
  # Output - in this section we define settings that determine how our output looks like in this case
  # we define the structure names and colours to use.

InnerEye/ML/configs/segmentation/Lung.py

@@ -37,6 +37,7 @@ def __init__(self, **kwargs: Any) -> None:
  ground_truth_ids_display_names=fg_display_names,
  colours=[(255, 255, 255)] * len(fg_classes),
  fill_holes=[False] * len(fg_classes),
+ roi_interpreted_types=["ORGAN"] * len(fg_classes),
  largest_connected_component_foreground_classes=["lung_r", "lung_l", "heart"],
  num_dataload_workers=8,
  norm_method=PhotometricNormalizationMethod.CtWindow,

InnerEye/ML/configs/segmentation/ProstateBase.py

@@ -22,6 +22,7 @@ def __init__(self,
  ground_truth_ids_display_names: Optional[List[str]] = None,
  colours: Optional[List[TupleInt3]] = None,
  fill_holes: Optional[List[bool]] = None,
+ roi_interpreted_types: Optional[List[str]] = None,
  class_weights: Optional[List[float]] = None,
  largest_connected_component_foreground_classes: Optional[List[str]] = None,
  **kwargs: Any) -> None:
@@ -34,13 +35,16 @@ def __init__(self,
  present then must be of the same length as ground_truth_ids.
  :param fill_holes: Optional list of fill hole flags. If
  present then must be of the same length as ground_truth_ids.
+ :param interpreted_types: Optional list of interpreted_types. If
+ present then must be of the same length as ground_truth_ids.
  :param class_weights: Optional list of class weights. If
  present then must be of the same length as ground_truth_ids + 1.
  :param kwargs: Additional arguments that will be passed through to the SegmentationModelBase constructor.
  """
  ground_truth_ids_display_names = ground_truth_ids_display_names or [f"zz_{name}" for name in ground_truth_ids]
  colours = colours or [(255, 0, 0)] * len(ground_truth_ids)
  fill_holes = fill_holes or [True] * len(ground_truth_ids)
+ roi_interpreted_types = roi_interpreted_types or ["ORGAN"] * len(ground_truth_ids)
  class_weights = class_weights or equally_weighted_classes(ground_truth_ids, background_weight=0.02)
  largest_connected_component_foreground_classes = largest_connected_component_foreground_classes or \
  ground_truth_ids
@@ -55,6 +59,7 @@ def __init__(self,
  ground_truth_ids_display_names=ground_truth_ids_display_names,
  colours=colours,
  fill_holes=fill_holes,
+ roi_interpreted_types=roi_interpreted_types,
  image_channels=["ct"],
  inference_batch_size=1,
  inference_stride_size=(64, 256, 256),

InnerEye/ML/configs/unit_testing/passthrough_model.py

@@ -44,6 +44,7 @@ def __init__(self, **kwargs: Any) -> None:
  ground_truth_ids_display_names=fg_display_names,
  colours=generate_random_colours_list(RANDOM_COLOUR_GENERATOR, len(fg_classes)),
  fill_holes=[False] * len(fg_classes),
+ roi_interpreted_types=["ORGAN"] * len(fg_classes),
  inference_batch_size=1,
  class_weights=equally_weighted_classes(fg_classes, background_weight=0.02),
  feature_channels=[1],

InnerEye/Scripts/submit_for_inference.py

@@ -174,7 +174,8 @@ def submit_for_inference(args: SubmitForInferenceConfig, azure_config: AzureConf
  # is then just the file relative to the data_folder
  "--data_folder", image.parent.name,
  "--image_files", image.name,
- "--use_dicom", str(args.use_dicom)],
+ "--use_dicom", str(args.use_dicom),
+ "--model_id", model_id],
  conda_dependencies_files=conda_files,
  )
  run_config = create_run_config(azure_config, source_config, environment_name=python_environment_name)

Tests/ML/configs/DummyModel.py

@@ -36,6 +36,7 @@ def __init__(self, **kwargs: Any) -> None:
  ground_truth_ids_display_names=self.fg_ids,
  colours=[(255, 255, 255)] * len(self.fg_ids),
  fill_holes=[False] * len(self.fg_ids),
+ roi_interpreted_types=["Organ"] * len(self.fg_ids),
  mask_id="mask",
  dataset_expected_spacing_xyz=(1.0, 1.0, 1.0),
  norm_method=PhotometricNormalizationMethod.CtWindow,

TestsOutsidePackage/test_score.py

@@ -27,7 +27,6 @@
  extract_zipped_files_and_flatten, convert_zipped_dicom_to_nifti, \
  convert_nifti_to_zipped_dicom_rt
 
-
 test_image = full_ml_test_data_path("train_and_test_data") / "id1_channel1.nii.gz"
 img_nii_path = full_ml_test_data_path("test_img.nii.gz")
 # Expected zipped DICOM-RT file contents, just DEFAULT_RESULT_ZIP_DICOM_NAME without the final suffix.
@@ -119,7 +118,8 @@ def test_unpack_one_set_zip(zip_file_contents: List[Path], test_output_dirs: Out
 @pytest.mark.parametrize("zip_file_contents", TEST_ZIP_FILE_PATHS_ALL2)
 def test_unpack_two_distinct_sets_zip(zip_file_contents: List[Path], test_output_dirs: OutputFolderForTests) -> None:
  """
- Test that a zip file containing two distinct set of files in two folders, but possibly in a series of nesting folders,
+ Test that a zip file containing two distinct set of files in two folders, but possibly in a series of nesting
+ folders,
  can be extracted into a folder containing only the files.
 
  :param zip_file_contents: List of relative file paths to create and test.
@@ -233,7 +233,7 @@ def test_convert_nifti_to_zipped_dicom_rt(test_output_dirs: OutputFolderForTests
  convert_zipped_dicom_to_nifti(zipped_dicom_series_path, reference_series_folder, nifti_filename)
  model_config = PassThroughModel()
  result_dst = convert_nifti_to_zipped_dicom_rt(HNSEGMENTATION_FILE, reference_series_folder, model_folder,
- model_config, DEFAULT_RESULT_ZIP_DICOM_NAME)
+ model_config, DEFAULT_RESULT_ZIP_DICOM_NAME, model_id="test_model:1")
  assert_zip_file_contents(result_dst, HN_DICOM_RT_ZIPPED, model_folder)
 
 
@@ -253,7 +253,8 @@ def test_score_image_dicom_two_inputs(test_output_dirs: OutputFolderForTests) ->
  image_files=[str(zipped_dicom_series_path), str(zipped_dicom_series_path)],
  result_image_name=HNSEGMENTATION_FILE.name,
  use_gpu=False,
- use_dicom=True)
+ use_dicom=True,
+ model_id="Dummy:1")
 
  with pytest.raises(ValueError) as e:
  score_image(score_pipeline_config)
@@ -278,7 +279,8 @@ def test_score_image_dicom_not_zip_input(test_output_dirs: OutputFolderForTests)
  image_files=[str(test_file)],
  result_image_name=HNSEGMENTATION_FILE.name,
  use_gpu=False,
- use_dicom=True)
+ use_dicom=True,
+ model_id="Dummy:1")
 
  with pytest.raises(zipfile.BadZipFile):
  score_image(score_pipeline_config)
@@ -309,7 +311,8 @@ def test_score_image_dicom_mock_all(test_output_dirs: OutputFolderForTests) -> N
  image_files=[str(zipped_dicom_series_path)],
  result_image_name=HNSEGMENTATION_FILE.name,
  use_gpu=False,
- use_dicom=True)
+ use_dicom=True,
+ model_id="Dummy:1")
 
  with mock.patch('score.init_from_model_inference_json',
  return_value=(mock_pipeline_base, model_config)) as mock_init_from_model_inference_json:
@@ -356,7 +359,8 @@ def test_score_image_dicom_mock_run_store(test_output_dirs: OutputFolderForTests
  image_files=[str(zipped_dicom_series_path)],
  result_image_name=HNSEGMENTATION_FILE.name,
  use_gpu=False,
- use_dicom=True)
+ use_dicom=True,
+ model_id="Dummy:1")
 
  with mock.patch('score.run_inference',
  return_value=mock_segmentation) as mock_run_inference:
@@ -398,7 +402,8 @@ def test_score_image_dicom_mock_run(test_output_dirs: OutputFolderForTests) -> N
  image_files=[str(zipped_dicom_series_path)],
  result_image_name=HNSEGMENTATION_FILE.name,
  use_gpu=False,
- use_dicom=True)
+ use_dicom=True,
+ model_id="Dummy:1")
 
  image_with_header = io_util.load_nifti_image(HNSEGMENTATION_FILE)
 
@@ -438,7 +443,8 @@ def test_score_image_dicom_mock_none(test_output_dirs: OutputFolderForTests) ->
  image_files=[str(zipped_dicom_series_path)],
  result_image_name=HNSEGMENTATION_FILE.name,
  use_gpu=False,
- use_dicom=True)
+ use_dicom=True,
+ model_id="Dummy:1")
 
  segmentation = score_image(score_pipeline_config)
  assert_zip_file_contents(segmentation, HN_DICOM_RT_ZIPPED, model_folder)

environment.yml

@@ -20,7 +20,7 @@ dependencies:
  - gitpython==3.1.7
  - gputil==1.4.0
  - h5py==2.10.0
- - InnerEye-DICOM-RT==1.0.0
+ - InnerEye-DICOM-RT==1.0.1
  - joblib==0.16.0
  - jupyter==1.0.0
  - jupyter-client==6.1.5

score.py

@@ -49,6 +49,9 @@ class ScorePipelineConfig(GenericConfig):
  "containing a set of DICOM files.")
  result_zip_dicom_name: str = param.String(DEFAULT_RESULT_ZIP_DICOM_NAME,
  doc="The name of the zipped DICOM-RT file if use_dicom set.")
+ model_id: str = param.String(allow_None=False,
+ doc="The AzureML model ID. This is added to the SoftwareVersions DICOM tag in the "
+ "DICOM-RT output")
 
 
 def init_from_model_inference_json(model_folder: Path, use_gpu: bool = True) -> Tuple[FullImageInferencePipelineBase,
@@ -196,7 +199,7 @@ def convert_rgb_colour_to_hex(colour: TupleInt3) -> str:
 
 
 def convert_nifti_to_zipped_dicom_rt(nifti_file: Path, reference_series: Path, scratch_folder: Path,
- config: SegmentationModelBase, dicom_rt_zip_file_name: str) -> Path:
+ config: SegmentationModelBase, dicom_rt_zip_file_name: str, model_id: str) -> Path:
  """
  Given a Nifti file and a reference DICOM series, create zip file containing a DICOM-RT file.
 
@@ -209,6 +212,7 @@ def convert_nifti_to_zipped_dicom_rt(nifti_file: Path, reference_series: Path, s
  :param scratch_folder: Scratch folder to extract files into.
  :param config: Model config.
  :param dicom_rt_zip_file_name: Target DICOM-RT zip file name, ending in .dcm.zip.
+ :param model_id: The AzureML model ID <model_name>:<ID>
  :return: Path to DICOM-RT file.
  """
  dicom_rt_file_path = scratch_folder / Path(dicom_rt_zip_file_name).with_suffix("")
@@ -218,7 +222,12 @@ def convert_nifti_to_zipped_dicom_rt(nifti_file: Path, reference_series: Path, s
  out_file=dicom_rt_file_path,
  struct_names=config.ground_truth_ids_display_names,
  struct_colors=[convert_rgb_colour_to_hex(rgb) for rgb in config.colours],
- fill_holes=config.fill_holes)
+ fill_holes=config.fill_holes,
+ roi_interpreted_types=config.roi_interpreted_types,
+ manufacturer=config.manufacturer,
+ interpreter=config.interpreter,
+ modelId=model_id
+ )
  # Log stdout, stderr from DICOM-RT conversion.
  logging.debug("stdout: %s", stdout)
  logging.debug("stderr: %s", stderr)
@@ -286,7 +295,7 @@ def score_image(args: ScorePipelineConfig) -> Path:
 
  if args.use_dicom:
  result_dst = convert_nifti_to_zipped_dicom_rt(result_dst, reference_series_folder, model_folder,
- config, args.result_zip_dicom_name)
+ config, args.result_zip_dicom_name, args.model_id)
 
  if not is_offline_run_context(run_context):
  upload_file_name = args.result_zip_dicom_name if args.use_dicom else args.result_image_name