This past Friday, I was fortunate enough to attend the 11th annual Medical Imaging Informatics and Teleradiology Conference (MIIT).
All of the sessions were very information and, based on what I saw this year, I would highly recommend you make an effort to attend next year’s 12th annual MIIT.
For the purpose of this Blog Post, I’ve decided to comment in particular on Brad Erickson’s presentation on Deep Learning. I’m really interested in how AI will fit into our future.
Deep Learning is something most of us access multiple times a day. Anyone using Siri, Google, Facebook, Twitter, Skype and the list goes on…, is witnessing the power of Deep Learning. Here is an article from WIRED on how AI is changing Google Searches. The article reinforces what Brad Erickson’s presentation stated on the power of systems that can ‘learn’. Here’s one quote from the article:
“Google’s search engine was always driven by algorithms that automatically generate a response to each query. But these algorithms amounted to a set of definite rules. Google engineers could readily ch ange and refine these rules. And unlike neural nets, these algorithms didn’t learn on their own. As Lau put it: “Rule-based scoring metrics, while still complex, provide a greater opportunity for engineers to directly tweak weights in specific situations.”
Something as common-place as searching for the best Thai restaurant in your neighborhood has been positively impacted by AI. What will the impact be to Imaging Informatics?
It’s been 5 years since Watson won Jeopardy and since then we have watched IBM improve Watson and prepare ‘it’ for healthcare. It seems likely that in 15-20 years (or quite a bit earlier) most diagnostic imaging will been ‘seen’ by a deep learning solution. What will it take for the culture to change for us to feel comfortable with a computer performing the primary read on our diagnostic imaging?
I think there’s an analogy between self-driving cars and AI in radiology. We have Tesla’s that can drive us to work, but it’s illegal for us to sleep at the wheel. Regardless of the fact that Google’s self-driving cars had driven over 1.3 million miles before causing its first accident, many people felt like this accident emphasized why we shouldn’t have self-driving cars.
Imagine the Public’s reaction on a system’s first misdiagnosed read that caused a negative impact to patient care. Even if AI makes the odds of accidents low, it feels more natural to have humans cause accidents rather than computers.
Moore’s Law may state that computing power will approximately double every 18 months – but can the general public’s comfort level with technology keep up with this pace?
This talk really whet my appetite for the SIIM 2016’s closing talk ‘Peering Into the Future through the Looking Glass of Artificial Intelligence’.
Hope to see you there in Portland.
Feel free to share your comments below on how you think AI will fit into radiology, and how the ‘court of public opinion’ will impact the technology moving forward.