If you can’t beat ‘em.

Top Line: So are we doomed?
The Study: This in-depth review of machine-learning advancements in radiology over the past decade thinks not. In fact, it highlights many of the challenges faced by AI as it enters the highly-regulated healthcare arena. The argument basically goes like this. Ok, sure, machine learning can pick up on pattern recognition to better and better degrees (eh hm, Gleason score, eh hm), but real patients love to break the mold. And what’s going to break AI? Ambiguity. The first tenet to clinical medicine is the differential diagnosis, often honed through patient discussions and occasionally even gestalt—that real intelligence working in the background piecing together patient cues the don’t even register in the conscience brain, which is otherwise occupied writing notes or contouring normal structures. Speaking of, rather than poo-pooing AI altogether, utilizing it to lessen the burden of our most menial mind-numbing tasks is a great way to incorporate available machine learning algorithms in 2020. In any case, this may all become a lot less high stakes if Elizabeth Warren erases our medical student loan debt.
TBL: Learn to embrace, not fight, the emergence of AI in oncology. | Chan, Br J Radiol 2019


Popular Posts