Feedback on AI Interoperabilty in Imaging

[Doyle, Sean W.]

General comment: a very fine document!


What would be wonderful if some organization (IHE?) could maintain a list of selected DICOM private elements and what they mean. The one very strong case here is for magnetic field strength and direction. GE and Siemens both encode this in private fields; I believe that the Siemens format is binary and is not easily decoded. Once a researcher has learned how to reverse engineer this (or if the vendor nicely provides this information) it would be very good to share it in the community. A second value of this list of important private tags is that they could be excluded by de-identification algorithms (many de-identification routines simply delete all private fields). For example - the Agfa Impax system saves overlay information in private fields in the DICOM header. If you have a few years of residents marking pulmonary nodules - why not make this data available for training and/or study selection? But if the private fields are undocumented it’s difficult to use them. I think crowdsourcing this would be best since historically I’ve found vendors very hesitant to document their private fields.


More detailed comments in spreadsheet:

Sean Doyle