Subscribe
Share
Share
Embed
Eye on AI is an Applied Radiology podcast series exploring how artificial intelligence is advancing medical imaging and addressing the practical needs of today’s imaging professionals. Through conversations with clinical experts and industry leaders, the series highlights real-world applications of AI and the technologies shaping the future of imaging.
Kieran Anderson:
Welcome to the Applied Radiology Podcast and another installment of Eye
on AI, where we explore how artificial intelligence and emerging
technologies are changing medical imaging and the practice of radiology.
I’m Kieran Anderson, group publisher here at Applied Radiology. And
today we’re going to look beyond the individual MRI examination to
really consider a much bigger question. What if MRI could become a
longitudinal tool for understanding how an individual patient’s path and
health is changing over time? So as healthcare continues to move toward
earlier detection, prevention, and more personalized care, the ability
to acquire consistent comparable imaging across different scanners,
sites, and points in time could become really an interesting thing to
consider. And AI may be one of the technologies that could really help
us make that possible. So joining me today to talk about this is
Dr. Sean Raj, chief medical officer and information officer at SimonMed
Imaging, and Gina Park, chief strategy officer at AIRS Medical.
I’m going to start with you, Sean, because SimonMed has really been
investing significantly in preventative and longevity focused imaging
from my understanding. So from your perspective, what is driving that
shift and how do you see MRI fitting into a more proactive approach to
healthcare?
Sean Raj, MD:
Yeah, I would say that healthcare, as you mentioned, has historically
been extraordinarily good at diagnosing disease. It’s really been a very
episodic care effectively for as long as we know imaging. But what we’re
seeing now more and more, the opportunity has now really moved upstream
where we as imaging providers have a tremendous opportunity to provide
quite a bit of additional value to patients. And I think you asked
what’s causing this. I’d say there’s, in my mind, three major drivers,
three. I’d say first of all, patients, ever since COVID especially,
patients are becoming increasingly proactive. They want to understand
their health before something becomes clinically apparent. And that may
be a combination of patients hungry for insights through Google, but
really these ChatGPT and Claude and all these tools at the fingertips of
patients, they’re seeking information and becoming much more educated.
Point two would be the technology has improved dramatically.
You mentioned MRI, so I’ll go in on that, but MRI is faster, higher
quality, it’s radiation free. Patients know this, and MRI is becoming
increasingly capable of evaluating multiple organs in one shot. And then
finally, we have the advent of AI. I mean, AI has been part of imaging
for, you could say a decade if you really open up your definition of AI,
and particularly started in women’s health with mammography. But in the
last several years, particularly last three, we have seen such an
acceleration in quality of AI and the capabilities transforming from a
point solution. At the previous version to now we have what’s called
foundation models that are AIs that are thinking like humans available
for all different parts of the body. And so now we can go beyond just
simply describing anatomy to where it’s quantification, risk assessment,
and personalized insights. To me, that is what’s game-changing to the
field.
Kieran Anderson:
Sean, traditionally we’ve tended to think about MRIs answering that
clinical question at a particular moment, but my question for you here
is how does that change when you begin thinking about imaging more
longitudinally and looking at how that same patient over months or even
years and those imaging results? What are your thoughts there?
Sean Raj, MD:
Yeah. Well, to take one step back from that, this is literally the main
driver, one of the main drivers of why we created SimonMed Longevity.
It’s because patients have changed their mindset. They have this now
consumer mindset. Think about this. Everyone has some sort of wearable
or some sort of guide. They have their cell phone. I mean, even if
you’re not buying an Oura Ring or Apple Watch, you have a cell phone and
you go into your health app on your cell phone, you can track your steps
over the years you’ve had that cell phone. Consumers, patients are
tracking sleep, glucose, heart rate, fitness continuously. So it isn’t
surprising that we’re being asked for the richest source of information,
which is the body. And the way to do that is via MRI. And so this right
here is one of the most important evolutions in imaging, in my opinion.
We are able now, just like the cell phone can track the steps over
years, we can use MRI to create longitudinal imaging and change the
question from, is this abnormal to how is this patient changing? And
that opens up all the insights a patient needs to change their
trajectories to live a happier, healthier life. We can now say, look, is
the lesion that we saw previously actually growing is a very important
question that patients ask about their visceral fat, like my body
composition, is my visceral fat increasing or decreasing? Is my liver,
which kind of shows your metabolic health, is the liver fat changing? Is
muscle mass changing? Are vascular structural abnormalities getting
bigger or smaller? These are all these insights that previously we were
kind of giving qualitative insights on, but now have the ability to be
more quantitative and track longitudinal. So we think that trajectory is
what will guide actual change in patients.
Kieran Anderson:
In order to do this, I’m guessing you’ve got to be able to capture these
exams in sort of a consistent manner, and maybe you will speak more
about how important that is. And so when we think about that,
particularly when patients may be imaged, I guess on different scanners
at different sites and using different protocols over a period of time
and even over a period of several years, if you’re going to use this to
track it, it’s got to be consistent. So that longitudinal approach would
seem to depend heavily on that consistency. So Gina, bringing you into
the conversation, from your perspective, why is standardization so
important if we want to really and reliably compare a patient’s imaging
over time? What do you think about that?
Jina Park, PharmD, MS:
Yeah, I mean, I think Dr. Raj said it really well, but just to iterate,
when we are in medical school or pharmacy school, they always teach us
that the trend of a patient, whether it’s a lab value or imaging, trend
is more important that that’s one static number because that day the
patient may have had something or maybe that day they’re just abnormal
numbers. And so that’s why in medical schools, they always teach you to
see the trend, not the number itself. And at AIRS Medical, we have
always tried to figure out how to improve the proactive health portion
of it. And so that’s always been our philosophy and when we build our AI
softwares. And essentially what we try to do as a third party, because
that’s the unique part of being a third party is that we understand
Siemens and GE and Phillips, how they image and how the images come out
and they have its own characteristics.
And we try to see if, I don’t want to say hack it, maybe part of it is a
little bit of a reverse engineering of how they do it. We really try to
learn the characteristics and try to reverse that and see, can we try to
normalize that as much as possible? And the current data do show that we
can reduce on the variation. So part of the testing that we’ve been
doing in the past one year since we have developed the update for Swift
Mar is that we’ve been testing to see can we reduce on the variation?
And it does show that pretty significantly.
Sean Raj, MD:
Yeah, this is an enormously important point you’re making, especially at
scale. At SimonMed, we have close to 200 centers. We have a humongous
installed base, no consistency between scanners. I mean, we’ve been in
business for over 20 years. We have different vendors, we have different
generations of scanners, we have different field strengths, we have
different coils, different protocols that we try to harmonize, obviously
different technologists, different patient populations. And so this
really leads to the major point. If the scanner changes more than the
patient does, this longitudinal data set that we’re trying to unlock and
provide tremendous value to the patient is pretty worthless. If I’m
trying to say this lesion that we saw changed 10% over two years, when
I’m making that statement, you have to know whether that 10% change is
related to the actual underlying abnormality or is it the natural
variation between different scanners?
And so this is all about consistency of measurements, which will drive
the change that we can tell patients, and therefore the patient and the
providers can make the best clinical decisions based off real
information.
Kieran Anderson:
Well, let’s talk a little bit more about that then, because I think this
is really where AI becomes particularly interesting, at least to me,
because much of the discussion around AI, particularly in MRI, is always
focused on image quality and acquisition time, workflow efficiency.This
may fit into that a little bit, but we’re really talking about can AI
create better consistency across examinations and its potential role as
we look at this longitudinal imaging as part of this discussion and what
that really means to patient care over time. So Gina, coming back to
you, in your opinion, where does AI begin to change this equation and
how can technologies like those that AIRS is developing and bringing to
market help move MRI from simply producing higher quality images towards
creating more consistent comparable information over time?
Jina Park, PharmD, MS:
I don’t think that people have thought about how do we create this
longitudinal information so that they can see the trend and then make a
better prediction or make a better decision about their care. Right now,
it’s a static number, you don’t know what the trend is, you’re not
confident about the trend, and you say, “What am I supposed to do about
it?” And I think that’s the reaction that we’re getting from a lot of
the providers and the patients. And what we’re saying is that with this
reliable trend that we see, you should be taking actions differently.
And I think that’s, I think, where the AI can go as a next step. And I
do believe that we don’t plan on necessarily making every software to
track these patient progression and then be able to predict that future.
But I do believe that more AI companies will have to think about
consistency just because that’s what truly matters in being able to
track and then be able to make some kind of a suggestion to the patient
or the providers.
And also as a further step of being able to predict, and that way that
actually changes someone’s behavior and making recommendations to change
so that they can change that future that they may have with their
condition. And so that’s how I see the field shifting.
Sean Raj, MD:
Yeah. And I would just go on to say that the first era of imaging AI
really got very, very good at answering the question, can AI find the
abnormality? We are now entering this chapter here where we can have AI
help us understand the trajectory of the abnormality. And I think that,
I mean, we’re still early days, but that is truly game-changing. Look, I
think the vision that we’re painting is very realistic, though I do
double down on the statement that I made earlier, which is this is
really, really early days. We should very much be hellbent on trying to
distinguish what is possible from what is clinically validated. So we
already know what is something we already know. We already know that
longitudinal information is clinically powerful. The challenge here is
creating these standardized data sets which validate what change yields
what outcome, and then demonstrating that acting on these predictive
signals actually improves patient’s health.
So I mean, this is a whole massive opportunity that we have in imaging
to provide tremendous value to patients. We just got to focus on these
reproducible acquisitions, collecting these large longitudinal data
sets, and then clinically validating the findings. And then we are in a
tremendous position to help our patients.
Kieran Anderson:
We’ve talked a lot about where we are today. I’d like to finish with
what’s ahead and what does the future look like? If longitudinal imaging
and AI and preventative medicine continue to converge and this role of
MRI itself could look quite differently from the way we think about it
today. So my question really to the both of you is if we look five or 10
years ahead, what does success look like and how different could the
role of MRI be if we can consistently follow patients over time and
combine these imaging histories with the AI and other health
information?
Sean Raj, MD:
I think the biggest change will be that imaging becomes less episodic.
Today, again, we take a snapshot, a symptom happens, we get a study, we
give an answer. In the future state, at least hopefully for some
patients, imaging is just going to be part of its longitudinal health
record alongside labs, genomics, family history, wearable data. Let me
just paint you this picture. Can you imagine a patient could have a
baseline MRI exam when they’re completely healthy and now several years
later when that symptom occurs, they then come back for another MRI. And
now we have the AI technology along with our superstar subspecialized
radiologists evaluate that scan when they’re having symptoms compared to
their baseline when they were completely healthy. Now, importantly,
we’re comparing this today’s scan not to a reference population, but
we’re comparing that patient to most importantly, what they were when
they were healthy.
That is the ultimate promise of longitudinal imaging. It’s
personalization of medicine in the most literal sense, because instead
of asking whether you’re looking normal compared to the population, we
are now going to be asking whether you look meaningfully different
compared to yourself.
Jina Park, PharmD, MS:
And like Dr. Haraj is saying exactly what I would’ve said in terms of
personalized healthcare, I do believe that this is the pathway, but I do
think that at least from AI company perspective, what I do think will
change is the way that we develop AI moving forward. We also don’t plan
on developing AI ourselves all by ourselves, but what we actually plan
to do is essentially starting to work with other AI companies that want
to build on SwiftMR process data and then be able to build these
solutions together and then validate it together so that we can truly
build validated softwares really quickly in the market. The goal for us
is to continue to partner with AI companies to build these solutions
quickly so that we can get this out there. And then in addition to that,
I do think that partnering with larger institution, I do think will be a
really critical piece.
And so I do think that in the next three to five years, how AI companies
approach development and the R&D in general, I do think will shift a
little bit just because the need for the longitudinal data and at the
patient level, personalized medicine cannot happen without having a
really robust data to do that. And so I do believe that that will
change.
Kieran Anderson:
As I said a moment ago, this has really been a fascinating conversation
as we talk about longitudinal analysis here. Rather than just viewing
MRI simply as a snapshot sort of used to diagnose one specific problem,
these advances in AI, the standardization we’ve talked about, the
longitudinal analysis that we’ve talked about really is beginning to
allow imaging to provide a much more continuous picture of a patient’s
health, potentially helping clinicians, as Dr. Raj said, to recognize
more meaningful changes earlier and make more informed decisions about
prevention and care, which is really the goal. So Sean, Gina, I want to
thank you both for taking the time to sit down and chat with us here at
Applied Radiology about the future of longitudinal MRI and preventative
care. Thank you again, and as always, I want to thank our listeners for
joining us here for another edition of Eye on AI podcast from Applied
Radiology.
I’m Kieran Anderson, your host here at Applied Radiology. Thank you for
joining us.