Limitless Medical Imaging podcast, ep.4 - with Brian Casey from The Imaging Wire


- Hello Brian, and thank you for accepting our invitation to Limitless Medical Imaging Podcast. We're happy to have you here! I will start by asking you to please introduce yourself.

- Great. Well, it's good to be here, Andra. My name is Brian Casey, I am managing editor of The Imaging Wire and The Imaging Wire is a website and newsletter that's dedicated to writing about medical imaging, radiology, and all things related to that. I've been in radiology for about 30 years, to a variety of publications. So I was at SIIM, last week, and I thought it was a great show. I think there was a lot of great discussions. A lot of people were really excited about what's been going on in image management, digital image management, enterprise imaging, and especially on AI.
So lots of good talks, lots of good networking there.

- This is why I wanted to talk to you about because I knew you attended SIIM, and I wanted to find out from the person who is the most knowledgeable about what happened there. What were the main themes that got discussed this year? What trends do you foresee?

- This was my first SIIM in a few years. The last one I went to was in Portland in 2016. And I guess my first one was probably in 2001, back then it was called Scan and it was in Salt Lake City. The number one thing that struck me most was how little discussion there was of COVID, you because all these meetings, you know, COVID has been such a huge shadow over medical meetings for the last three years. And this was the first meeting where it just didn't seem to be a big topic. And there weren't very many people wearing masks and there wasn't really much concern about COVID at all. So that was actually really nice. And I went to a talk one day, I think it was on Thursday, it was by Dr. Jim Whitfield, and he talked about the importance of being together, and the reason for that was, Dr. Whitfield was actually the president or chair of SIIM in 2020 when it was supposed to be held that year. And they had to cancel it, obviously because of COVID and just like all the medical meetings that year, and so they substituted it with a virtual meeting for it. And like a lot of other societies there, there's been a lot of virtual meetings. And a lot of us have kind of gotten used to that. And we've sort of assumed that, well, you know, is it really that much different, you know, meeting virtually as opposed to meeting in person? And he went through and he talked about some of the negative aspects of virtual technology and human relationships and kind of went through some of the impact that we've seen on adolescents and worker productivity.

So, when it comes to content, the thing that really dominated at SIIM was artificial intelligence and specifically ChatGPT. As I wrote in one of my articles about SIIM for The Imaging Wire, Artificial Intelligence has really given SIIM kind of a new mission, because, SIIM goes back a long time, originally it was the Society for Computer Applications and Radiology, and the original mission was to teach people about PACS and digital image management, and when PACS came along early nineties, people didn't know,  we were still using film in radiology. And so there was really a need to train people into that. But over the years, everybody has acquired PACS. And so, at least in Western hospitals, all the imaging is digital now, and so there isn't that need to learn what AI has done, it's this new technology that could potentially have as big of an impact on radiology as PACS has had, but it's really a mystery.

And so SIIM has pivoted to focus quite a bit on AI and to talk about how AI can be used in radiology. And so that's, that's really been nice to see at SIIM and, and a lot of the discussions were about AI and we can dive into that in more detail if you want to.

- Yes, of course. From the discussions that you've had,  what have you seen as challenges in developing algorithms in radiology?

- Well, there's a number of challenges and the positive thing is that most radiologists don't seem to be opposed to AI, to using AI anymore, when the first AI algorithms came along, a lot of radiologists got really worried about it. And there was that famous quote by Jeffrey Hinton, where he said that we should stop training radiologists right now. And so the good news is that radiologists appear to have gotten over that reticence about A. I. The problem is that it is taking a lot longer for A. I. to penetrate radiology than people expected. And it's just the clinical adoption is taking a long time. I was at another meeting A. I. Med in San Diego two weeks ago. Another theme there was, what do we have to do to get radiologists to start using more AI? And I think that some of the issues are, it needs to be easier for radiologists to put into play. You can't have radiologists having to choose between algorithms as they're reading a scan, that's just not going to work. So a lot of vendors are approaching this from a platform approach, rather than selling algorithms to radiologists on a one-off basis. We need to have them on a platform where ideally, the algorithm can just launch and analyze a study, in real-time without having a lot of interaction from the radiologist. So the platform approach is really big. There's also a lot of concern about the quality of, about access to data for training AI algorithms.

And there was an early nice talk by Dr Zied Obermeier on one of the days where he discussed how we need to get better data for AI training. You train algorithms on small data sets, that's problematic because a lot of times those algorithms aren't generalizable. They don't perform very well outside of the data set that they were trained on. We can't have an algorithm have a sensitivity of 90% on the training data. And then when you get it out in the real world, that drops down to 65 or 70. So that was another big issue at SIIM. We're still like super early days for ChatGPT and I don't know of anybody that's actually using it clinically right now.

I think that we're mostly at this phase where we're playing around with it. And I was at one talk where, one of the moderators asked the audience,  how many of you have, not used it clinically, but how many of you just played around with ChatGPT and,  most of the people in the audience said, yeah, they had tried it out. Having said that, I think everybody would agree that it's really not at a point at all where it can be used clinically. And we're still finding out ways where it can be used. So that's what's really exciting. I think that it has a lot of potential in terms of taking kind of mundane tasks and taking some load off the radiologist, for example, writing a summary of a radiology report that's maybe written at a level that a patient could understand. I think that's a possibility. And, the radiologist could go back over and fix things. The one big concern about Chat GPT is this concept called the hallucination effect where ChatGPT just makes things up. And it sounds real, you read it and you're like, Oh, okay, we're great. And then you dive into it and it's like, well, these things aren't true. So I think we're going to have to deal with that. We're going to have to deal with the hallucination effect. We're going to have to test this quite a bit, but it does seem that it's got a lot of potential. And it seems like every new iteration of it,  sees a much bigger improvement in performance. I think that's only going to get better. 

- Thank you so much for this perspective. From your discussions, what have you seen about clinicians, how are they seeing the future, the modern imaging workspace?

- Well, the big issue there is, is just workflow. One of the themes that was really hammered home at SIIM was just workflow, workflow, workflow, and radiologists all over the world are just being buried by this, this incredible volume of imaging studies that are being produced. We've got these scanners now, CT scanners that are producing just hundreds or thousands of slices per exam. And it's really challenging for radiologists to have to read all these things. And the population is getting older, and imaging is being used more often. Clinicians need tools to help them work more efficiently. I don't necessarily think that they feel like they need an AI algorithm to back themselves up because their sensitivity is already pretty high, but they need technology that's going to stay out of their way and help them do their jobs and help them deal with all this volume we're seeing. So I think that's their big concern. They need productivity tools and they are really reticent about tools that might get in their way and slow them down. You know, nobody wants to have to go through a mammogram and have to follow up on 20 marks of suspicious lesions. I think that's where the winners and losers are going to sort out, the tools that can be integrated with radiologists' workflow and really help them work more efficiently are going to be the ones that are going to win out here.

- What were the main concerns around storage and security, for radiology informatics?

- Yeah, these are both big issues, and they were really a topic of a lot of discussion at SIIM. There've been a number of really high-profile security breaches both at radiology departments or centers, facilities, and also in healthcare more broadly. And so security is a huge issue. It seems to me like a lot of these incidents are forcing healthcare providers to turn to cloud providers for their data, data storage and management, because they're realizing that they just, running a top flight cyber security department just isn't in there. Their skill set and that there are other people out there who can do this better. And, you know, these are the platform people like Amazon and Microsoft and Google. And so we're seeing a lot more interest in cloud-based image management, and I think that historically healthcare had been really concerned about having their data stored off-premises. I think that's starting to go away a little bit. And I think that the security issues are driving that. And so I think that we're going to see continued migration of health care to cloud hosting in the future.

- Okay. Amazing. Thank you so much, Brian, for taking part in this episode and for sharing with us your experience at SIM. Everyone, please send us any thoughts or questions, or ideas you might have for Brian or for us, for Medicai. Feel free to message us and subscribe to this podcast. Until next time. 

About the author - Andra Bria

Andra Bria is a marketing manager at Medicai. She is interested in health equity, patient experience and value-driven care pathways. She believes in interoperability and collaboration for a more connected healthcare industry.