Quotes from Jeremy Howard at AIMed North America 2017
A question I was asked at AIMed 2017 was: “When will imaging machine learning (ML) reach high enough accuracy to be clinically useful?”
For me, that’s not the interesting question facing this space. We’re now at the point where the idea that we can do AI for medical imaging should not be at all weird or controversial.
In the last couple of years academic research has shown that deep learning (DL) is better at recognising what objects are in an image than humans are, given that enough data is available to train an appropriate model.
You should assume that for most kinds of an image that a human looks at, a computer could recognize objects in that image more quickly and more accurately than you can.
A study released in the end of 2017 showed a model was diagnosing breast cancer from pathology slides with an ROC of 0.994, so basically nearly perfect, and it was built from less than 200 slides.
Perhaps the most surprising thing about building these DL models is how incredibly easy it is. DL is so straightforward nowadays and that’s not where the technology challenge is.
Really the interesting question to me is: What can we or should we do with it?
Personally, my belief is the best thing to do with it is to help radiologists, pathologists, all the other folks who use medical imaging to be both more efficient and more effective.
We can use it to diagnose strokes as soon as somebody comes out of the CT with a notification that alerts the radiologist straight away. That would be technically extremely simple thing to do and it would save a whole lot of time.
Another question is what’s it gonna take for the regulatory pathway to become more accessible? Currently it’s really poorly aligned with the needs of healthcare innovators and very few have actually gone through the process of getting FDA approved.
And perhaps the most pivotal question is how do we more quickly and effectively get this technology into the clinic? That’s where we should focus.
You can see Jeremy Howard’s full talk in the recording of Session II at AIMed North America 2017, which is free to access in our resources page here.
Jeremy Howard – Founding Researcher at fast.ai
Jeremy Howard is an entrepreneur, business strategist, developer, and educator. Jeremy is a founding researcher at fast.ai, a research institute dedicated to making deep learning more accessible. He is also a Distinguished Research Scientist at the University of San Francisco, a faculty member at Singularity University, and a Young Global Leader with the World Economic Forum.
Jeremy’s most recent startup, Enlitic, was the first company to apply deep learning to medicine, and has been selected one of the world’s top 50 smartest companies by MIT Tech Review two years running. He was previously the President and Chief Scientist of the data science platform Kaggle, where he was the top ranked participant in international machine learning competitions 2 years running. He was the founding CEO of two successful Australian startups (FastMail, and Optimal Decisions Group–purchased by Lexis-Nexis). Before that, he spent 8 years in management consulting, at McKinsey & Co, and AT Kearney. Jeremy has invested in, mentored, and advised many startups, and contributed to many open source projects.
He has many television and other video appearances, including as a regular guest on Australia’s highest-rated breakfast news program, a popular talk on TED.com, and data science and web development tutorials and discussions.