def test_store_as_scaled_ubyte_nifti_fail(test_output_dirs: TestOutputDirectories, input_range: Any) -> None:
image = np.random.random_sample((dim_z, dim_y, dim_x))
header = ImageHeader(origin=(1, 1, 1), direction=(1, 0, 0, 0, 1, 0, 0, 0, 1), spacing=(1, 2, 4))
with pytest.raises(Exception):
io_util.store_as_scaled_ubyte_nifti(image, header,
test_output_dirs.create_file_or_folder_path(default_image_name),
input_range)
def test_store_image_as_short_nifti(test_output_dirs: TestOutputDirectories,
norm_method: PhotometricNormalizationMethod,
image_range: Any,
window_level: Any) -> None:
window, level = window_level if window_level else (400, 0)
image = np.random.random_sample((1, 2, 3))
image_shape = image.shape
args = SegmentationModelBase(norm_method=norm_method, window=window, level=level, should_validate=False)
# Get integer values that are in the image range
image1 = LinearTransform.transform(data=image, input_range=(0, 1), output_range=args.output_range)
image = image1.astype(np.short) # type: ignore
header = ImageHeader(origin=(1, 1, 1), direction=(1, 0, 0, 0, 1, 0, 0, 0, 1), spacing=(1, 1, 1))
nifti_name = test_output_dirs.create_file_or_folder_path(default_image_name)
io_util.store_image_as_short_nifti(image, header, nifti_name, args)
if norm_method == PhotometricNormalizationMethod.CtWindow:
output_range = get_range_for_window_level(args.level, args.window)
image = LinearTransform.transform(data=image, input_range=args.output_range, output_range=output_range)
image = image.astype(np.short)
else:
image = image * 1000
t = np.unique(image)
assert_nifti_content(nifti_name, image_shape, header, list(t), np.short)
def test_store_as_binary_nifti(test_output_dirs: TestOutputDirectories, image: Any) -> None:
image = np.array(image)
header = ImageHeader(origin=(1, 1, 1), direction=(1, 0, 0, 0, 1, 0, 0, 0, 1), spacing=(1, 2, 4))
io_util.store_binary_mask_as_nifti(image, header,
test_output_dirs.create_file_or_folder_path(default_image_name))
t = np.unique(image)
assert_nifti_content(test_output_dirs.create_file_or_folder_path(default_image_name), image.shape, header, list(t),
np.ubyte)
def test_store_posteriors_nifti_invalid_entries(test_output_dirs: TestOutputDirectories) -> None:
image = np.array([0, 1, 2.71, np.nan])
header = ImageHeader(origin=(1, 1, 1), direction=(1, 0, 0, 1, 0, 0, 1, 0, 0), spacing=(1, 1, 1))
with pytest.raises(ValueError) as ex:
io_util.store_posteriors_as_nifti(image, header,
test_output_dirs.create_file_or_folder_path(default_image_name))
assert "invalid values" in ex.value.args[0]
assert "2.71" in ex.value.args[0]
assert "nan" in ex.value.args[0]
def test_store_as_ubyte_nifti(test_output_dirs: TestOutputDirectories) -> None:
image = np.random.random_sample((dim_z, dim_y, dim_x))
# get values in [0, 255] range
image = np.array((image + 1) * 255).astype(int)
header = ImageHeader(origin=(1, 1, 1), direction=(1, 0, 0, 0, 1, 0, 0, 0, 1), spacing=(1, 2, 4))
io_util.store_as_ubyte_nifti(image, header, test_output_dirs.create_file_or_folder_path(default_image_name))
t = np.unique(image).astype(np.ubyte)
assert_nifti_content(test_output_dirs.create_file_or_folder_path(default_image_name),
image.shape, header, list(t), np.ubyte)
def test_store_inference_results(
test_output_dirs: TestOutputDirectories) -> None:
np.random.seed(0)
num_classes = 2
posterior = torch.nn.functional.softmax(torch.from_numpy(
np.random.random_sample((num_classes, dim_z, dim_y, dim_x))),
dim=0).numpy()
segmentation = np.argmax(posterior, axis=0)
assert segmentation.shape == (dim_z, dim_y, dim_x)
posterior0 = to_unique_bytes(posterior[0], (0, 1))
posterior1 = to_unique_bytes(posterior[1], (0, 1))
spacing = (2.0, 2.0, 2.0)
header = ImageHeader(origin=(0, 0, 0),
direction=(1, 0, 0, 0, 1, 0, 0, 0, 1),
spacing=spacing)
inference_result = InferencePipeline.Result(epoch=1,
patient_id=12,
posteriors=posterior,
segmentation=segmentation,
voxel_spacing_mm=(1, 1, 1))
test_config = _create_config_with_folders(test_output_dirs)
assert test_config.class_and_index_with_background() == {
"background": 0,
"region": 1
}
results_folder = test_output_dirs.root_dir
store_inference_results(inference_result, test_config,
Path(results_folder), header)
assert_nifti_content(
os.path.join(results_folder, "012", "posterior_background.nii.gz"),
segmentation.shape, header, list(posterior0), np.ubyte)
assert_nifti_content(
os.path.join(results_folder, "012", "posterior_region.nii.gz"),
segmentation.shape, header, list(posterior1), np.ubyte)
assert_nifti_content(
os.path.join(results_folder, "012", "background.nii.gz"),
segmentation.shape, header, list([0, 1]), np.ubyte)
assert_nifti_content(os.path.join(results_folder, "012", "region.nii.gz"),
segmentation.shape, header, list([0, 1]), np.ubyte)
assert_nifti_content(
os.path.join(results_folder, "012", DEFAULT_RESULT_IMAGE_NAME),
segmentation.shape, header, list(np.unique(segmentation)), np.ubyte)
assert_nifti_content(
os.path.join(results_folder, "012", "uncertainty.nii.gz"),
inference_result.uncertainty.shape, header, list([248, 249, 253,
254]), np.ubyte)
def test_store_posteriors_nifti_fail(test_output_dirs: TestOutputDirectories,
image: Any) -> None:
image = np.array(image)
header = ImageHeader(origin=(1, 1, 1),
direction=(1, 0, 0, 1, 0, 0, 1, 0, 0),
spacing=(1, 1, 1))
with pytest.raises(Exception):
io_util.store_posteriors_as_nifti(
image, header,
test_output_dirs.create_file_or_folder_path(default_image_name))
def test_store_as_scaled_ubyte_nifti(test_output_dirs: TestOutputDirectories, input_range: Any) -> None:
image = np.random.random_sample((dim_z, dim_y, dim_x))
header = ImageHeader(origin=(1, 1, 1), direction=(1, 0, 0, 0, 1, 0, 0, 0, 1), spacing=(1, 2, 4))
io_util.store_as_scaled_ubyte_nifti(image, header,
test_output_dirs.create_file_or_folder_path(default_image_name),
input_range)
image = LinearTransform.transform(data=image, input_range=input_range, output_range=(0, 255))
t = np.unique(image.astype(np.ubyte))
assert_nifti_content(test_output_dirs.create_file_or_folder_path(default_image_name), image.shape, header, list(t),
np.ubyte)
def test_store_as_binary_nifti_fail(test_output_dirs: OutputFolderForTests,
image: Any) -> None:
image = np.array(image)
header = ImageHeader(origin=(1, 1, 1),
direction=(1, 0, 0, 1, 0, 0, 1, 0, 0),
spacing=(1, 2, 4))
with pytest.raises(Exception):
io_util.store_binary_mask_as_nifti(
image, header,
test_output_dirs.create_file_or_folder_path(default_image_name))
def test_store_posteriors_nifti(test_output_dirs: TestOutputDirectories,
image: Any, expected: Any) -> None:
image = np.array(image)
header = ImageHeader(origin=(1, 1, 1),
direction=(1, 0, 0, 0, 1, 0, 0, 0, 1),
spacing=(1, 1, 1))
io_util.store_posteriors_as_nifti(
image, header,
test_output_dirs.create_file_or_folder_path(default_image_name))
assert_nifti_content(
test_output_dirs.create_file_or_folder_path(default_image_name),
image.shape, header, list(expected), np.ubyte)
def test_store_as_nifti_fail(test_output_dirs: OutputFolderForTests, image_type: Any, scale: Any, input_range: Any,
output_range: Any) \
-> None:
header = ImageHeader(origin=(1, 1, 1),
direction=(1, 0, 0, 1, 0, 0, 1, 0, 0),
spacing=(1, 2, 4))
image = np.random.random_sample((dim_z, dim_y, dim_x))
with pytest.raises(Exception):
io_util.store_as_nifti(
image, header,
test_output_dirs.create_file_or_folder_path(default_image_name),
image_type, scale, input_range, output_range)
def test_save_dataset_example(test_output_dirs: OutputFolderForTests) -> None:
"""
Test if the example dataset can be saved as expected.
"""
image_size = (10, 20, 30)
label_size = (2, ) + image_size
spacing = (1, 2, 3)
np.random.seed(0)
# Image should look similar to what a photonormalized image looks like: Centered around 0
image = np.random.rand(*image_size) * 2 - 1
# Labels are expected in one-hot encoding, predictions as class index
labels = np.zeros(label_size, dtype=int)
labels[0] = 1
labels[0, 5:6, 10:11, 15:16] = 0
labels[1, 5:6, 10:11, 15:16] = 1
prediction = np.zeros(image_size, dtype=int)
prediction[4:7, 9:12, 14:17] = 1
dataset_sample = DatasetExample(epoch=1,
patient_id=2,
header=ImageHeader(origin=(0, 1, 0),
direction=(1, 0, 0, 0,
1, 0, 0, 0,
1),
spacing=spacing),
image=image,
prediction=prediction,
labels=labels)
images_folder = test_output_dirs.root_dir
config = SegmentationModelBase(
should_validate=False,
norm_method=PhotometricNormalizationMethod.Unchanged)
config.set_output_to(images_folder)
store_and_upload_example(dataset_sample, config)
image_from_disk = io_util.load_nifti_image(
os.path.join(config.example_images_folder, "p2_e_1_image.nii.gz"))
labels_from_disk = io_util.load_nifti_image(
os.path.join(config.example_images_folder, "p2_e_1_label.nii.gz"))
prediction_from_disk = io_util.load_nifti_image(
os.path.join(config.example_images_folder, "p2_e_1_prediction.nii.gz"))
assert image_from_disk.header.spacing == spacing
# When no photometric normalization is provided when saving, image is multiplied by 1000.
# It is then rounded to int64, but converted back to float when read back in.
expected_from_disk = (image * 1000).astype(np.int16).astype(np.float64)
assert np.array_equal(image_from_disk.image, expected_from_disk)
assert labels_from_disk.header.spacing == spacing
assert np.array_equal(labels_from_disk.image, np.argmax(labels, axis=0))
assert prediction_from_disk.header.spacing == spacing
assert np.array_equal(prediction_from_disk.image, prediction)
def test_scale_and_unscale_image(
test_output_dirs: TestOutputDirectories) -> None:
"""
Test if an image in the CT value range can be recovered when we save dataset examples
(undoing the effects of CT Windowing)
"""
image_size = (5, 5, 5)
spacing = (1, 2, 3)
header = ImageHeader(origin=(0, 1, 0),
direction=(-1, 0, 0, 0, -1, 0, 0, 0, -1),
spacing=spacing)
np.random.seed(0)
# Random image values with mean -100, std 100. This will cover a range
# from -400 to +200 HU
image = np.random.normal(-100, 100, size=image_size)
window = 200
level = -100
# Lower and upper bounds of the interval of raw CT values that will be retained.
lower = level - window / 2
upper = level + window / 2
# Create a copy of the image with all values outside of the (Window, Level) range set to the boundaries.
# When saving and loading back in, we will not be able to recover any values that fell outside those boundaries.
image_restricted = image.copy()
image_restricted[image < lower] = lower
image_restricted[image > upper] = upper
# The image will be saved with voxel type short
image_restricted = image_restricted.astype(int)
# Apply window and level, mapping to the usual CNN input value range
cnn_input_range = (-1, +1)
image_windowed = LinearTransform.transform(data=image,
input_range=(lower, upper),
output_range=cnn_input_range)
args = SegmentationModelBase(
norm_method=PhotometricNormalizationMethod.CtWindow,
output_range=cnn_input_range,
window=window,
level=level,
should_validate=False)
file_name = test_output_dirs.create_file_or_folder_path(
"scale_and_unscale_image.nii.gz")
io_util.store_image_as_short_nifti(image_windowed, header, file_name, args)
image_from_disk = io_util.load_nifti_image(file_name)
# noinspection PyTypeChecker
assert_nifti_content(file_name, image_size, header,
np.unique(image_restricted).tolist(), np.short)
assert np.array_equal(image_from_disk.image, image_restricted)
def test_store_as_nifti(test_output_dirs: TestOutputDirectories, image_type: Any, scale: Any, input_range: Any,
output_range: Any) \
-> None:
image = np.random.random_sample((dim_z, dim_y, dim_x))
spacingzyx = (1, 2, 3)
path_image = test_output_dirs.create_file_or_folder_path(default_image_name)
header = ImageHeader(origin=(1, 1, 1), direction=(1, 0, 0, 0, 1, 0, 0, 0, 1), spacing=spacingzyx)
io_util.store_as_nifti(image, header, path_image,
image_type, scale, input_range, output_range)
if scale:
linear_transform = LinearTransform.transform(data=image, input_range=input_range, output_range=output_range)
image = linear_transform.astype(image_type) # type: ignore
assert_nifti_content(test_output_dirs.create_file_or_folder_path(default_image_name),
image.shape, header, list(np.unique(image.astype(image_type))), image_type)
loaded_image = io_util.load_nifti_image(path_image, image_type)
assert loaded_image.header.spacing == spacingzyx
def test_register_and_score_model(
is_ensemble: bool, dataset_expected_spacing_xyz: Any,
model_outside_package: bool,
test_output_dirs: TestOutputDirectories) -> None:
"""
End-to-end test which ensures the scoring pipeline is functioning as expected by performing the following:
1) Registering a pre-trained model to AML
2) Checking that a model zip from the registered model can be created successfully
3) Calling the scoring pipeline to check inference can be run from the published model successfully
"""
ws = get_default_workspace()
# Get an existing config as template
loader = get_model_loader(
"Tests.ML.configs" if model_outside_package else None)
config: SegmentationModelBase = loader.create_model_config_from_name(
model_name="BasicModel2EpochsOutsidePackage"
if model_outside_package else "BasicModel2Epochs")
config.dataset_expected_spacing_xyz = dataset_expected_spacing_xyz
config.set_output_to(test_output_dirs.root_dir)
# copy checkpoints into the outputs (simulating a run)
stored_checkpoints = full_ml_test_data_path(
os.path.join("train_and_test_data", "checkpoints"))
shutil.copytree(str(stored_checkpoints), config.checkpoint_folder)
paths = [Path(config.checkpoint_folder) / "1_checkpoint.pth.tar"]
checkpoints = paths * 2 if is_ensemble else paths
model = None
model_path = None
# Mocking to get the source from the current directory
# the score.py and python_wrapper.py cannot be moved inside the InnerEye package, which will be the
# only code running (if these tests are run on the package).
with mock.patch('InnerEye.Common.fixed_paths.repository_root_directory',
return_value=tests_root_directory().parent):
try:
tags = {"model_name": config.model_name}
azure_config = get_default_azure_config()
if model_outside_package:
azure_config.extra_code_directory = "Tests" # contains DummyModel
deployment_hook = lambda cfg, azure_cfg, mdl, is_ens: (Path(
cfg.model_name), azure_cfg.docker_shm_size)
ml_runner = MLRunner(config,
azure_config,
model_deployment_hook=deployment_hook)
model, deployment_path, deployment_details = ml_runner.register_segmentation_model(
workspace=ws,
tags=tags,
best_epoch=0,
best_epoch_dice=0,
checkpoint_paths=checkpoints,
model_proc=ModelProcessing.DEFAULT)
assert model is not None
model_path = Path(
model.get_model_path(model.name, model.version, ws))
assert (model_path /
fixed_paths.ENVIRONMENT_YAML_FILE_NAME).exists()
assert (model_path / Path("InnerEye/ML/runner.py")).exists()
assert deployment_path == Path(config.model_name)
assert deployment_details == azure_config.docker_shm_size
# move test data into the data folder to simulate an actual run
train_and_test_data_dir = full_ml_test_data_path(
"train_and_test_data")
img_channel_1_name = "id1_channel1.nii.gz"
img_channel_1_path = train_and_test_data_dir / img_channel_1_name
img_channel_2_name = "id1_channel2.nii.gz"
img_channel_2_path = train_and_test_data_dir / img_channel_2_name
# download the registered model and test that we can run the score pipeline on it
model_root = Path(model.download(str(test_output_dirs.root_dir)))
# create a dummy datastore to store model checkpoints and image data
# this simulates the code shapshot being executed in a real run
test_datastore = Path(test_output_dirs.root_dir) / "test_datastore"
shutil.move(str(model_root / "test_outputs"),
str(test_datastore / RELATIVE_TEST_OUTPUTS_PATH))
data_root = test_datastore / DEFAULT_DATA_FOLDER
os.makedirs(data_root)
shutil.copy(str(img_channel_1_path), data_root)
shutil.copy(str(img_channel_2_path), data_root)
# run score pipeline as a separate process using the python_wrapper.py code to simulate a real run
return_code = SubprocessConfig(
process="python",
args=[
str(model_root / "python_wrapper.py"),
"--spawnprocess=python",
str(model_root / "score.py"),
f"--data-folder={str(test_datastore)}",
f"--test_image_channels={img_channel_1_name},{img_channel_2_name}",
"--use_gpu=False"
]).spawn_and_monitor_subprocess()
# check that the process completed as expected
assert return_code == 0
expected_segmentation_path = Path(
model_root) / DEFAULT_RESULT_IMAGE_NAME
assert expected_segmentation_path.exists()
# sanity check the resulting segmentation
expected_shape = get_nifti_shape(img_channel_1_path)
image_header = ImageHeader(origin=(0, 0, 0),
direction=(1, 0, 0, 0, 1, 0, 0, 0, 1),
spacing=(1, 1, 1))
assert_nifti_content(str(expected_segmentation_path),
expected_shape, image_header, [0], np.ubyte)
finally:
# delete the registered model, and any downloaded artifacts
shutil.rmtree(test_output_dirs.root_dir)
if model and model_path:
model.delete()
shutil.rmtree(model_path)
