A script loads 3D images, but 4D are expected for preprocessing

[fepegar]

I am trying to run inference locally, using the following command:

$ PYTHONPATH=/tmp/Hippocampus_ python \
    ~/git/RadiomicsNN/innereye-deeplearning/score.py \
    --data_folder /tmp/inputs \
    --model_folder /tmp/Hippocampus_ \
    --no-use_gpu

I got the following error:

Traceback (most recent call last):
  File "/home/fernando/git/RadiomicsNN/innereye-deeplearning/score.py", line 311, in <module>
    main()
  File "/home/fernando/git/RadiomicsNN/innereye-deeplearning/score.py", line 307, in main
    score_image(ScorePipelineConfig.parse_args())
  File "/home/fernando/git/RadiomicsNN/innereye-deeplearning/score.py", line 289, in score_image
    segmentation = run_inference(images, inference_pipeline, config)
  File "/home/fernando/git/RadiomicsNN/innereye-deeplearning/score.py", line 132, in run_inference
    photo_norm_images = [photo_norm.transform(image_with_header.image) for image_with_header in images_with_header]
  File "/home/fernando/git/RadiomicsNN/innereye-deeplearning/score.py", line 132, in <listcomp>
    photo_norm_images = [photo_norm.transform(image_with_header.image) for image_with_header in images_with_header]
  File "/home/fernando/git/RadiomicsNN/innereye-deeplearning/InnerEye/ML/photometric_normalization.py", line 99, in transform
    self.output_range, self.sharpen, self.tail, self.debug_mode
  File "/home/fernando/git/RadiomicsNN/innereye-deeplearning/InnerEye/ML/photometric_normalization.py", line 300, in mri_window
    min(max_foreground, level + tail[ichannel] * std_i * sharpen))
IndexError: list index out of range

In photometric_normalization.py:

InnerEye-DeepLearning/InnerEye/ML/photometric_normalization.py

Line 276 in 95d8b72

nchannel = image_in.shape[0]

Hence, this function expects a 4D image with the channels dimension being the first. However, images are loaded in

InnerEye-DeepLearning/InnerEye/ML/utils/io_util.py

Lines 186 to 201 in 95d8b72

	def read_image_as_array_with_header(file_path: Path) -> Tuple[np.ndarray, ImageHeader]:
	"""
	Read image with simpleITK as a ndarray.
	:param file_path:
	:return: Tuple of ndarray with image in Z Y X and Spacing in Z X Y
	"""
	image: sitk.Image = sitk.ReadImage(str(file_path))
	img = sitk.GetArrayFromImage(image) # This call changes the shape to ZYX
	spacing = reverse_tuple_float3(image.GetSpacing())
	# We keep origin and direction on the original shape since it is not used in this library
	# only for saving images correctly
	origin = image.GetOrigin()
	direction = image.GetDirection()
	return img, ImageHeader(origin=origin, direction=direction, spacing=spacing)

which returns a 3D image.

I'm not sure what an appropriate solution would be. Maybe just adding a channels dimension at the beginning of PhotometricNormalization.transform and removing it before returning?

AB#5331