- Joined
- Sep 12, 2008
- Messages
- 57
- Reaction score
- 35
My thought gets on the Harvard article:
Interventional radiology obviously will stay. Diagnostic radiology should go away, but medicine being a guild craft will probably manage to require human "overread" of images. VA and Kaiser, for example, have huge databases of images with associated readings. NLP software to parse the reading for an xray and ML to identify things should get around the claim of no repository of tagged images, etc. That approach will take longer to mature than using images with features tagged, though.
AI caption of image. Credit to F. Chollet's blog The limitations of deep learning