DICOM
erich: Josh and I attended FHIR imaging group. Read through their docs. They're working on a mapping to FHIR.
… Done quite a bit. Important, but not the same as castingDICOm to RDF. There's overlap between the models. Need to map them.
… Still need an official casting of DICOM to RDF.
… But I was able to convert DICOM to JSON. DCM dump to JSON is based on the spec. Everything is multivalued.
… Taking that, going into JSON-LD playground, and adding an @context, it turns into RDF as quads.
… But there's a problem. In the doc, when you put nulls into an array in json-ld they get filtered out, but DICOM requires them to be kept.
… Null resets values instead of not asserting anything.
… Want to hold a position in the list.
… The default vocab covers everything. I got an @context for the spec. But custom tags would not be passed through. But @vocab can give a default.
… Also considered a differeent approach, 20B triples had perf issues. Need date strings to be dates, etc.
… But once you have it in there, there isn't much need for the value rep, w extra blank node etc.
… Another possibility would be to make custom datatypes for DICOM, and tag that onto the triple.
… Jena and Virtuoso both can map custom datatypes to native types.
… File comes out looking just like the original DICOM.
ericp: Checked w Gregg Kellogg?
… In this context null has meaning.
erich: Also, stay w these OIDs vs convert to URN?
eric: Pref the idea of deep curation. FHIR had datatypes that are union types, eg date datetime, and we separatted them to different types.
… Suspect a similar rationalization would make it more RDF friendly.
erich: SPARQL UPDATE query could transform the versoin w custom datatypes to the form that has properties.
… Bnefit of matching DICOM form, is that tools could convert it back ti DICOM binary.
… DICOM was adopted because of ease of using it. DICOM to JSON shows how to map them.
… If we can transform back it would be good.
detlef: Checked how they transform DICOM sequences?
erich: They have nested array.
… In DICOM JSON, everything is multivalued.
… Turns into an RDF list. Played w RDF sequence w explicit index.
… But JSON-LD only maps to an RDF list.
… Also for everything w a cardinality of one, everything in DICOM JSON is an array.
… I've only ever seen this in Drupal.
erip: FHIR itself requires preserving list order.
erich: They say "if a property exists in the source but no value, they want it preserved in the JSON, but no value". That works fine for RDF, except if there a vield w only one value.
… How to represent that in RDF if there's no value?
detlef: In RDF there are mandatory properties, even if there is no value.
dbooth: Could have a distinquished value meaning null.
erich: Could also reach out to DICOM group about how to move it forward. But could do community group first.
erich: JSON-LD framing is pretty powerful. Mibht convert optimized RDF to JSON.
ericp: One potential choice is to try for something close to what you want, and show exciting use cases.
… Nice if we can build an ecosystem where all the things in healthcare are RDFy. Tempted to try it and back off if needed.
ericp: Our preprocessing step started to get too complex, not worth it.
… Re framing, anyplace w nested structures won't work well. They'll come out arbitrarily.
erich: Since JSON-LD exists, if they matched it would be a neat way to go. Would like to try as RDFy as possible. Shold be fully transformable back into the original.
… Play w Framing to see if it can do it.
erich: Start a W3C DICOM community group?
ericp: You need 3-4 people to say they're interested.
detlef: I'm interested. At the moment our code is proprietary, need to get permission.
erich: I'm happy to publish my stuff.
… Working to go directly from DICOM to RDF. That's harvesting our covid data.
… 350TB.
ACTION: Erich to initiate DICOM community group
erich: Makes sense to coordinate w FHIR imaging group.
… I'll keep sitting in on mtgs.
ADJOURNED