Create a specification for how to implement broad support for annotations and segmentations within OHIF, Orthanc, and Sonador client
DICOMseg and OHIF annotations are essential to having a set of open source tools capable of working with ML/AI from DICOM images. They appear to be supported within OHIF and Orthanc, but the implementation is incomplete. It is not possible, for example, to use tools such as the "angle annotation" utility in OHIF; or to apply rectangular/circular regions of interest.
More information is needed about the following points:
- How are annotations applied in OHIF written into Orthanc? What needs to be added to the Orthanc configuration (or to the OHIF codebase) to allow for all DICOM utilities to work as expected?
- How are DICOMseg annotations created? How can they be written to Orthanc so that they can be read by OHIF?
- Based on the way in which the data is structured, how can we write Python tools to help read annotations and segmentations from Orthanc for use in ML pipelines?
- What tools will be needed to allow for the data to be used from within OpenCV, PyTorch, and TensorFlow?