CLPEX Chat Board

I keep an ear to the ground for mammography radiography diagnostics (aka diagnosing breast cancer from Xrays) just because there's a lot of overlap to what we do in terms of recognition, diagnosticity , error rates, and Bayesian Reasoning plus it has way more research published on the topic. I ran across this article this morning titled Artificial intelligence achieves near-human performance in diagnosing breast cancer and was wondering if anyone knew of similar research going on in our field or if anyone thought there'd be a move towards this direction in our discipline.

Very interesting. Especially the part about combining the human with the AI to get even better work.

I think the struggle for AI in latent print work is that success is measured a little differently. Diagnostics generally (as far as I can tell) look to sensitivity and specificity. They want an ANSWER that's ACCURATE. Latent comparisons (again, at least to me) are more interested in PPV. We want to be able to say that WHEN we give and answer that's +, we're right.

I wonder if there's a field out there (with more research than ours) where a high PPV is the most important stat and sens. and spec. don't really matter as much.