Should we consider FHIR support, push notification endpoint with persistency support, in the IG?

[MMelQin]

Let start the discussion, and hopefully draft up a roadmap.

[MMelQin]

Some AI models, notably, e.g. the COVID-19 EXAM model co-developed by NVIDIA DLMed and many global institutions, are multi modal, which has both radiology image (XRay, CT, etc) as well as lab data (patient's vital) which can come from EHR, either through HL7 or FHIR.

To support such a model in inference, we already know that the IG can handle DICOM input, great. But getting EHR is challenging, in many aspects, some of which will be called later. Some experimental implementation has been done to open up a method in the inference request API to specify the associated FHIR resources for a specific DICOM Study, e.g. URL's, and then rely on the application to "GET" the resources.

Supporting push notification of FHIR in the IG is non-trivial, as by the very nature of the information and the protocol, pieces have to be received, parsed, processed, and the persisted, in preparation of later association with other pieces of information, e.g. specifically DICOM Studies for the same patient (this is typically called reconciliation in medical imaging workflow, and has to be correct, otherwise, patient safety is at risk). One can not simply assume so long as the patient ID matches, all is good, but NO, clinical use has more to consider. Can dive into that later.

Anyway, given the small number of DICOM studies, if any, get the benefit of AI, having the FHIR push notification endpoint in the IG, persisting huge amount of patient EHR data hoping for later use in the AI, may seem to be a waste of resource while raising the HIPAA compliance concerns. There are some opensource FHIR servers and services, for experimental use, which we can leverage. Thoughts.