How the incredible pace of growth for AI at the world’s largest meeting of radiologists has added a new layer of excitement to RSNA 2018, a famously innovative event.
George Shih, M.D. is Associate Professor and Vice-Chair for Informatics in the Department of Radiology at Weill Cornell Medical College and Associate Attending Radiologist at NewYork-Presbyterian Hospital-Weill Cornell Campus.
He’s also a veteran of the Radiological Society of North America’s (RSNA) annual meeting, having attended the annual meeting for over two decades.
The RSNA is an international society of radiologists, medical physicists and other medical professionals with more than 54,000 members from 146 countries across the globe. They see the potential for ML to automate initial detection (imaging screening) of potential pneumonia cases in order to prioritize and expedite their review.
We asked him to tell readers of AIMed Magazine what RSNA 2018 has to offer for fans of medical artificial intelligence (AI):
AI MED: How has AI become more of a feature at RSNA during recent years?
I’ve been going to RSNA long as I can remember, 20 years. Since I became a resident I’ve gone every year and been participating in different committees.
I’m currently chair of the informatics sub-committee, one of the scientific committees that deals with the abstracts which come in, and all the computer-focused abstracts come through my committee. In informatics we’ve seen such a huge increase in the number of AI abstracts.
The category that we have is machine learning, and in 2016-2017 there was a huge surge in the around the order of two to three times the number of scientific abstracts for machine learning and from 2017 -2018 it tripled again.
RSNA is a pretty international organisation and is the largest medical meeting in the world, around the order of 60,000 people. A lot of the talks which happen at the RSNA get distributed internationally, and the abstracts are submitted from all over the world. So, it’s a nice snapshot of what’s happening worldwide.
AI has really grown at an incredible pace. There are lots of sub-categories, but by far the most popular was the machine learning for pixel data, that is for images. In some ways it wasn’t surprising that the rate of growth was so large, but it still took a lot of people by surprise.
A lot of them are extremely high-quality abstracts. I’ve been joking around saying it’s easier to get into Harvard than it is to one of my abstract sessions for machine learning because of the high quality and high number of submissions.
AI MED: What trends have you noticed among these AI abstracts submitted to RSNA
Some interesting things I’ve observed: Machine learning is so popular and a lot of the other sub-categories are suffering, less on other types of informatics and projects.
Also, there are lots of scientific abstracts being submitted by vendors or companies. We allow them if they’re not overtly biased, i.e. they’re just like any scientific abstract, and we’re seeing a lot of commercial vendors presenting scientific abstracts. It’s a little bit unusual compared to other scientific experiences I’ve had.
They’re international, from all over the world. I think it can only be explained by lots of other AI research groups that are working on radiology related projects. I think that’s great for radiology, that we attract people that don’t usually innovate in our area. Now, because of machine learning a lot of people are very interested in radiology.
We’ve seen these abstracts cover a wide variety of other sub-specialities, whether it’s chest or abdomen. So, people are starting to apply machine learning widely on lots of different types of diseases. I think that’s partly intuitive, but I also think that’s a nice distribution.
AI MED: How does this surge in AI abstracts manifest onto the conference room floor?
Most of what people see at the meetings are abstracts that are accepted, and due to the high volume of AI abstracts we’re accepting a very small percentage. The view through the RSNA abstracts essentially shows the trends on what people are doing in the wider world beyond RSNA, not just what people see at the meeting.
In terms of what people see at the meeting, there’s been a continual and very dramatic increase in interest around AI. Previously, the informatics sessions were for a lot of the computer people. In my experience over the last couple of years these sessions have been much more widely attended and there’s been a ton of excitement.
I think an analogy would be, like lots of other important technologies that we use today – computer smartphone, nowadays no-one can live without their smartphone – we’re still in an early stage with AI. Relatively few people are working on it, but the thing we’ve seen over the last few years is that the numbers have increased dramatically.
This creates excitement both on the academic front but also commercially, as the vendors now have to have some kind of an AI strategy to really attract people to go to their booth. If they don’t have that, then people are gonna ask and maybe not be as excited about their product.
AI MED: How can fans of medical AI get the most out of RSNA 2018?
There are multiple ways to learn the most about AI:
– Attend the scientific sessions on AI. A lot of great research is happening and the people interested in academics will learn a lot about what people are working about on that front. These sessions are research-focused and hands on.
– The machine learning showcases, which are in the technical exhibit hall, are showcasing AI companies that are available. RSNA have expanded these this year. These sessions will be unique to RSNA 2018 and would give the participants a really nice view in what’s happening in radiology with AI.
– RSNA 2018 is hosting a Kaggle competition and they’ve got a data set on pneumonia. There are over 1,200 teams entered, and RSNA will announce the winners who will present at RSNA 2018. The top 10 will open source their algorithms. People will be able to analyse these algorithms to see high level AI algorithms for pneumonia. I think that’s pretty exciting. The RSNA will be hosting the Pneumonia Detection Challenge Recognition event on Monday, Nov. 26, 2:00-3:30 PM in the Machine Learning Showcase Theater, located in in the North Exhibits building, Level 3, booth #7367.
Be sure to see our whitepaper on multi-organ imaging driven by AI in the latest issue of AIMed Magazine, which you can access here.