-
Floating Point Pixel Sample Data
-
IHE AI Results
-
Any success mapping to vendors?
-
No vendors support all codes
-
Trying to map from different vendors to a single standard
-
Lots of different coding systems
- Architecture should support multiple code sets
- Insurance vs registry
- Diagnostic vs billing
- Will alawys have multiple codes and always have different local interpretations of codes
- Architecture should support multiple code sets
-
Agreement signed collaboration agreement between LOINC(RadLex) and SNOMED to map between the two
- LOINC doesnt have good support for findings
- Radiology is very granular in findings - e.g. charactersitcs of nodules/tumors
- SNOMED-CT has better support for findings
-
IHE does not constrain to specific code schemes
- Focuses on how to share them
-
What people are hearing in different countries about reimbursements for AI?
- USA - nothing?
- One vendor looking to sell their data to help offset costs for AI
- Charge whoever is benefiting from AI
- healthcare system
- Rad group/practice
- Article on business models for AI in medical imaging:
- Canada? Ontario
- No revision in payments in a decade
-
Data Market
- People are more willing to sell this over the past 5 years
- De-identifying DICOM data is hard
- https://dandelionhealth.ai/
- licenses federated access to de-id’d data from Sharp and Sanford Health.
- https://www.mayoclinicplatform.org/
- is working on federated access to a bunch of de-id’d data
- nVidia FLAIR - federated learning
- https://developer.nvidia.com/flare
- https://github.com/NVIDIA/NVFlare
- https://nvflare.readthedocs.io/en/2.2.1/
- https://www.nature.com/articles/s41591-021-01506-3
- https://aws.amazon.com/blogs/apn/privacy-preserving-federated-learning-on-aws-with-nvidia-flare/
- https://towardsdatascience.com/practical-federated-learning-with-azure-machine-learning-8807f9bd1a7e
-
Generateive AI