Add dicom_utils, test upload_annotation

mdai_utils/dicom_utils.py

@@ -0,0 +1,60 @@
+import json
+import os
+from pathlib import Path
+from shutil import copyfile
+from tempfile import TemporaryDirectory
+from typing import NamedTuple
+
+import SimpleITK as sitk
+
+
+def file_list_from_directory(directory, extension=".dcm"):
+    """Return a list of files with the given extension in the given directory."""
+    return [str(f) for f in Path(directory).glob(f"*{extension}")]
+
+
+def write_image_and_metadata(image, metadata, output_image_filename):
+    """Write the image to a file, along with the metadata."""
+    sitk.WriteImage(image, output_image_filename)
+    p = Path(output_image_filename)
+    metadata_filename = p.parent / (p.stem + "_SOPInstanceUIDs.json")
+    with open(metadata_filename, "w") as f:
+        json.dump(metadata, f, indent=2)
+
+
+class ImageAndMetadata(NamedTuple):
+    image: sitk.Image
+    metadata: dict
+
+
+def read_dicoms_into_volume(valid_dcm_file_list) -> ImageAndMetadata:
+    """Convert a list of DICOM files to a image volume. Also returns metadata
+    (SOPInstanceUID) for each slice in the volume.
+
+    The metadata generated by this function is a map from slice index to
+    SOPInstanceUID. And can be used to upload annotations from non-dicom formats
+    back to MDai, which requires the SOPInstanceUID to identify the slice.
+
+    PRECONDTIION: the input dicom files are in the same series, they are
+    valid DICOM files, with no SCOUTS or other special slices.
+    """
+    # copy to temp directory in case multiple series occupy the same directory
+    with TemporaryDirectory() as temp_dir:
+        src_file_lookup = {}
+        for src_file in valid_dcm_file_list:
+            dst_file = os.path.join(temp_dir, os.path.basename(src_file))
+            copyfile(src_file, dst_file)
+            src_file_lookup[dst_file] = src_file
+        reader = sitk.ImageSeriesReader()
+        # allow for reading of metadata
+        reader.SetMetaDataDictionaryArrayUpdate(True)
+        dicom_names = reader.GetGDCMSeriesFileNames(temp_dir)
+        reader.SetFileNames(dicom_names)
+        image = reader.Execute()
+        uids = [reader.GetMetaData(idx, "0008|0018") for idx in range(len(dicom_names))]
+        # keep info about slice, original file, and SOPInstanceUID
+        metadata = {
+            slice_idx: {"dicom_file": src_file_lookup[fn], "SOPInstanceUID": uid}
+            for slice_idx, (fn, uid) in enumerate(zip(dicom_names, uids))
+        }
+        return ImageAndMetadata(image=image, metadata=metadata)

mdai_utils/upload_annotations.py

@@ -69,8 +69,11 @@ def upload_image_annotation_slice(
         failed_annotations (list): List of failed annotations. If empty, all annotations were uploaded successfully.
     """
     data_np = read_data_image(segmentation_image_path)
+    if data_np.ndim == 3:
+        # The perpendicular dimension is at index 0 in the numpy array.
+        data_np = data_np.squeeze(0)
     return upload_data_annotation_slice(
-        data_np=data_np.T,
+        data_np=data_np,
         mdai_client=mdai_client,
         mdai_project_id=mdai_project_id,
         mdai_dataset_id=mdai_dataset_id,

requirements.in

@@ -3,3 +3,4 @@ pydicom>=2.4.3
 itk>=5.3.0
 bidict>=0.22.1
 opencv-python>=4.8.1.78
+simpleITK>=2.3.0

requirements.txt

@@ -120,6 +120,8 @@ scipy==1.11.3
     # via
     #   dicom2nifti
     #   scikit-image
+simpleitk==2.3.0
+    # via -r requirements.in
 six==1.16.0
     # via
     #   python-dateutil

tests/conftest.py

@@ -56,11 +56,11 @@ def mdai_setup(
     with open(parameters_file) as f:
         parameters = json.load(f)
 
-    fixture_dir = _current_dir / "fixtures"
+    fixtures_dir = _current_dir / "fixtures"
     mdai_domain = parameters.get("mdai_domain") or "md.ai"
     mdai_client = mdai.Client(domain=mdai_domain, access_token=token)
     return {
         "mdai_client": mdai_client,
         "parameters": parameters,
-        "fixture_dir": fixture_dir,
+        "fixtures_dir": fixtures_dir,
     }

tests/test_parameters.json

@@ -10,6 +10,6 @@
     "mdai_annotations_only": true,
     "mdai_no_fixing_metadata": false,
     "labels": [
-        "mylabel",
+        "mylabel"
     ]
 }

tests/upload_test.py

@@ -1,5 +1,6 @@
 import pytest
 
+from mdai_utils.upload_annotations import upload_image_annotation_slice
 from mdai_utils.upload_dataset import upload_dataset
 
 
@@ -13,11 +14,11 @@ def test_pytest_fixture(mdai_setup):
     reason="Only need to upload once. run pytest tests with --upload-only to run it."
 )
 def test_upload_dataset(mdai_setup):
-    mdai_parameters = mdai_setup["parameters"]
-    mdai_dataset_id = mdai_parameters.get("mdai_dataset_id")
-    fixture_dir = mdai_setup["fixture_dir"]
-    dicom_dir = fixture_dir / "humanct_0002_1000_1004"
-    assert fixture_dir.exists()
+    parameters = mdai_setup["parameters"]
+    mdai_dataset_id = parameters.get("mdai_dataset_id")
+    fixtures_dir = mdai_setup["fixtures_dir"]
+    dicom_dir = fixtures_dir / "humanct_0002_1000_1004"
+    assert dicom_dir.exists()
     completed_process = upload_dataset(mdai_dataset_id, dicom_dir)
     process_message = completed_process.stdout.strip()
     print(process_message)
@@ -26,4 +27,22 @@ def test_upload_dataset(mdai_setup):
 
 
 def test_upload_annotation(mdai_setup):
-    mdai_setup["parameters"]
+    parameters = mdai_setup["parameters"]
+    fixtures_dir = mdai_setup["fixtures_dir"]
+    mdai_client = mdai_setup["mdai_client"]
+    # sop_instance_uid can be acquired from mdai, or from the metadata generated
+    # by the function dicom_utils.read_dicoms_into_volume.
+    sop_instance_uid = "1.2.826.0.1.3680043.2.1125.1.75064541463040.2005072610414630768"
+    mdai_label_ids = parameters.get("mdai_label_ids")
+    labels_to_upload = parameters.get("labels")
+    label_id = mdai_label_ids.get(labels_to_upload[0])
+
+    failed_annotations = upload_image_annotation_slice(
+        segmentation_image_path=fixtures_dir / "humanct_0002_1000_seg.nii.gz",
+        sop_instance_uid=sop_instance_uid,
+        mdai_client=mdai_client,
+        mdai_project_id=parameters.get("mdai_project_id"),
+        mdai_dataset_id=parameters.get("mdai_dataset_id"),
+        mdai_label_id=label_id,
+    )
+    assert len(failed_annotations) == 0