Andre Ng, formerly at Google and Baidu and Adjunct Professor at Stanford, after getting really excited by training his computer to read chest X-rays tweeted this:
It is fair to say that in November 2017 we were on the rising part of ‘AI will eat radiologists’ lunch’ curve. This rhetoric continued unabated for a while but years later reality on the ground was that AI was not eating anyone’s lunch. The main reason was that most AI was only good in demos on small publicly available sets and failed rapidly as we tried to generalise it in any direction (case mix, acquisition technology, patient demographic).
Fast forward to 2021 and the landscape looks very different. Clinical AI community has moved well beyond the ‘my magic wand is bigger than yours’ phase. The best companies have embraced the challenges of the field and are now focused on two main areas:
- Robust, generalisable, products that meet regulatory standards and offer cost or quality advantage to payer/providers in existing pathways
- Smart AI that does things that humans can’t, sees things human eyes can’t to build new pathways from scratch