Many of the x-ray machines I work with, particularly the newer ones, have some capability of recording the amount of radiation used during a procedure, which can be used to estimate the radiation dose to patients.
A dose report is usually generated and stored with the patient images, and on some systems it can be sent to the PACS. This is good.
In almost all cases, the information is sent as just a screen capture wrapped up in DICOM headers. This is not so good. Good for archiving, lousy for searching.
It occurred to me that I should be able to set up a DICOM receiver that the machines could send the dose reports to, do some OCR on the images to extract the numbers that I need and then stash it all in a database. This is essentially the way ACR's Dose Index Registry (DIR) works but it relies on getting the dose reports as a DICOM Structured Report. A lot easier to get at dose data that way and no need to OCR and parse images. Data in the DIR is anonymized, but for my purposes the dose data would only be partially anonymized.
The OCR part would require some kind of machine specific mapping to identify what parts of the image contain what information. Sadly, some machines are completely incapable of sending the dose reports anywhere (I'm looking at you Siemens). Some way of dealing with those machines will be needed. Studies with lots of series will have multi-page dose reports which is another case that will need to be dealt with.
Nation wide some of this info is being collected in the National Radiology Dose Registry, but all of that data has to come from somewhere, and has to be extracted somehow. This would help us see what goes on locally, and would be a useful source of information. If a dose estimate needed to be done, we could look up the numbers, do a little bit of crunching and come up with something reasonably accurate. It would also be a good data source for studying radiation exposures.
I need to work on this.