Comment

DICOM meta-objects need to be more tightly defined

Currently machine learning output in radiology is not stored in a standard format - often it is encapsulated as a 'secondary capture' image or in a proprietary format. This means that ML output is not in a format that could be reused and that it is not portable. Instead - this section should specify that algorithm output below the level of an image specified using the FHIR ImagingStudy object should use either a DICOM SR using Template TID 1500 (for graphical annotations) or DICOM Segmentation objects for segmentations.