> self interested in sabotaging AI is tinfoil hat stuff
Agree this in nonsense. Not a radiologist but have worked with many.
The big barriers to AI impact in radiology are a) translation is a lot harder than people think, b) access to enough high quality data with good cohort characteristics c) good labeling (most of the interesting problems aren't really amenable to unsupervised) a d) generalization, as always.
It doesn't help that for the most part medical device companies aren't good at algorithms and algorithms companies aren't good at devices, lots of rookie mistakes made on both sides.
> Also PACS isn't designed to implement algorithms.
That doesn't really matter too much from the implementing-ML point of view, you can just use it as a file store. DICOM files themselves are annoying too (especially if they bury stuff in private tags), as are HL7 (and EMR integrations) but .. that's mostly just work.
Agree the viewers lack flexibility but that's a lot more solvable than say the morass of EMR. If you are just looking at image interpretation visualizing things isn't so bad, if you had the models to visualize.
Agree this in nonsense. Not a radiologist but have worked with many.
The big barriers to AI impact in radiology are a) translation is a lot harder than people think, b) access to enough high quality data with good cohort characteristics c) good labeling (most of the interesting problems aren't really amenable to unsupervised) a d) generalization, as always.
It doesn't help that for the most part medical device companies aren't good at algorithms and algorithms companies aren't good at devices, lots of rookie mistakes made on both sides.