Search
  • Guy Tsafnat

AI Doc Episode 1: Why? What? and How?

Above probably all other professions, automating doctors has captured our imagination.


With the recent re-emergence of a particular class of machine learning algorithms called "Neural Networks", or in it's new incarnation "Deep Learning" (or sometimes "Deep AI") came a number of medical applications for this technology. As Deep Learning is particularly suitable for image processing, most of these applications are indeed in radiology. Deep Learning algorithms have been found to identify fractures in shoulders and cancerous nodules in lungs, more accurately than human radiologists. The New Yorker published the call of at least one technologist to stop training radiologists. The argument being that the technology will continue to improve until it will do better than radiologists for all applications before new radiologists can be trained. If the technology is better and cheaper that argument makes sense. If that makes sense, then the following statement would not: AI increases the need for radiologists.


All machine learning needs feedback to improve. Typically this feedback is provided by subject matter experts, like radiologists in this case. They also need data to be very clean and accurately annotated by radiologists. This creates a lot more work for radiologists than actual classification. While the promise of ML is that the training period is limited and return on investment will eventually come, this assumes that imaging technology itself will not change, because when it does training will need to be resumed.


So at best, Deep Learning can change what radiologists do, or how they do what they already do, it cannot replace them. It can help them make less errors, miss less cancers or shoulder fractures etc. It can help them improve practice gradually over time. It is therefore welcome technology in medicine but not for the obvious reason. The same goes for other disciplines and other tasks that doctors perform. This is not even limited to AI technologies: when keyhole surgery was first introduced it fundamentally changed how surgery was performed. It didn't replace surgeons. So why would AI replace physicians?


#ai #deeplearning #medicine #automation #radiology #evidence #ebm #surgery #imageprocessing #jobs #future #machinelearning


Deep Medicine: How Artificial Intelligence Can Make Healthcare Human Again by Eric Topol

Address

Level 1 E10 Building

UNSW Sydney

NSW 2052, Australia

Contact

linkedin.png
twitter.png

©2019 Evidentli