[FEA] Attach meta data to the image object which is output by the segmentation operator

[AndreasHeumann]

Is your feature request related to a problem? Please describe.

The image object which is output by the segmentation operator has no metadata. It's not possible to query the pixel spacing or the orientation of the volume without metadata.

Describe the solution you'd like

The image object which is output by the segmentation operator should have the metadata of the input image attached. Values like pixel spacing or orientation.

Describe alternatives you've considered

Additional context

[MMelQin]

Besides saving the segmentation image in NIfTI file format (with the mandatory metadata) in the application's output folder under a known sub folder, a domain Image object is generated, right now containing image array only, as the in-memory output from the inference operator, which is meant to be used by the downstream segmentation writer operator. The known downstream operator has as its input the original DICOM metadata containing all the attributes required to interpret the array as a 3D image.

The above approach has been sufficient to v0.1 with known consumers of the inference operator, but definitely with more consumers being created, e.g. the viz operator, it cannot be safely assumed or required that the consumer operator need to tag DICOM metadata as its input, so populating the same metadata in Image object is needed as requested. Planned for v0.2 if not earlier.

For the record, the following is the set of metadata from the input in-mem Image object (that is first instantiated by the DICOMSeriesToVolumeOperator operator) for the inference operator.

Note:

• The image ndarray is shaped to be that of ndarray from DICOMSeriesToVolumeOperator (DIFFERENT shape from what is used in the inference transform dataset)
• The image is spacing and orientation is reverted back to those of the input, so the Image metadata is reused
• A few pydicom types are present in the metadata dict, not ideal, and should be changed in the DICOMSeriesToVolumeOperator.

series_instance_uid: 1.2.826.0.1.3680043.2.1125.1.68102559796966796813942775094416763, type <class 'pydicom.uid.UID'>
modality: CT, type <class 'str'>
series_description: No series description, type <class 'str'>
row_pixel_spacing: 1, type <class 'pydicom.valuerep.DSfloat'>
col_pixel_spacing: 1, type <class 'pydicom.valuerep.DSfloat'>
depth_pixel_spacing: 1.0, type <class 'float'>
row_direction_cosine: [-1.0, 0.0, 0.0], type <class 'list'>
col_direction_cosine: [0.0, -1.0, 0.0], type <class 'list'>
depth_direction_cosine: [0.0, 0.0, 1.0], type <class 'list'>
dicom_affine_transform: [[-1.  0.  0.  0.]
 [ 0. -1.  0.  0.]
 [ 0.  0.  1.  0.]
 [ 0.  0.  0.  1.]], type <class 'numpy.ndarray'>
nifti_affine_transform: [[ 1. -0. -0. -0.]
 [-0.  1. -0. -0.]
 [ 0.  0.  1.  0.]
 [ 0.  0.  0.  1.]], type <class 'numpy.ndarray'>

[MMelQin]

The fix is in this PR