Support for the AI podcast comes from the GPU Technology Conference.
Eager to learn about AI? Then don't miss the Premier AI Conference.
At GTC, choose from over 600 sessions on dozens of hot topics.
Get hands-on training. Connect with experts from Google, Stanford, Amazon, Apple, Facebook, Epic, Microsoft, MIT, Pixar, Citi, BMW and hundreds more.
It all starts March 17th at the GPU Technology Conference in Silicon Valley.
To learn more, go to gputechconf.com. That's gputechconf.com.
Then save 20% on registration with code GMPODCAST.
All caps, GMPODCAST. We will see you there.
Hello and welcome to the NVIDIA AI Podcast.
I'm your host, Noah Kravitz. You probably recognize the name Nuance, Nuance Communications. from their work speech recognition and virtual assistant technology.
They've been doing this for decades, and they're one of the big players and pioneers in voice recognition tech.
Of course, they've gotten into AI because they're smart, they know what they're doing, and the world is headed towards AI.
But what you might not know is that they've also gotten into using AI to chart the course of the healthcare industry and how physicians can leverage technology to make their work better and to make people healthier.
Today's guest is Karen Holzberger. Karen is the Vice President and General Manager of Diagnostic Solutions for Nuance Healthcare.
Let's just open broadly. What is Nuance doing with AI in the healthcare world?
And how, if you can speak to this kind of a huge question, so we'll get there, but how did that grow out of Nuance's work with voice and speech recognition?
So if you think about it, and that's a great question because we're really relying on our roots and being pioneers in AI for the past 20 years, starting with as you said, speech recognition, natural language processing, and now moving forward with AI from virtual assistants, ambient intelligence and what we're doing here in radiology with natural language processing and AI, as well as the image characterization algorithms. it's amazing that we've been able to watch this shift and where we've always been as nuanced communications, whether it's in the automotive side of things and,
Our enterprise with what we're doing with touch commerce, the connected car, those pieces and being able to leverage those. foundational technologies around AI in healthcare.
If you think about why that's important in healthcare, you think about just the amount of information, data, and the need to be able to put that data to work, whether it's within radiology where 450 million studies Imaging studies are done on patients in the US and Canada each and every year, leveraging that so these tools whether it's machine algorithm can actually help assist the radiologist to actually take better care of that patient through that care pathway. to really tell the primary care physician how to treat that patient better.
So having these assistants, these virtual assistants, help them make better decisions around that care.
So I have a good friend who's a physician, relatively new out of out of med school and working in a hospital nearby.
And he's kind of joked with me that his dream job is to move to Hawaii and review x-rays and other images, radiology images remotely, and that that's the thing.
So how does this work? What does a radiologist do and how does Nuances Tech help?
It's really interesting because a radiologist doesn't necessarily spend time with patients and that care team. the radiologist is really considered the doctor's doctor.
And if you take a step back, And there's certainly so much pressure because I talked about the volume that's happening each and every year how much they need to be able to interpret and make those right decisions.
Because it's not always about what the study or the reason for the exam in front of them that they're answering.
But there's all these other questions around incidental findings, other things to make sure that they're giving the right synopsis and direction back to that care team so they can take care of the patient.
So they're feeling the same pressures of burnout, lower reimbursements of those types of things.
So I think that's good context to have. But if you think about where we play and support the radiologists is through the creation of their report.
And so the workflow is that a patient like you or myself comes in, you know, I'm complaining of a headache.
And the first thing that your primary care physician is going to say, let's get a CT of the head.
Let's make sure there's nothing going on, that there isn't pressure.
There's not a potential brain aneurysm behind the left eye.
Did you have any trauma? And that radiologist is going to interpret that CT of your head or my head in this case, and then make some recommendations.
We are the document creator and that document, that report after the radiologist has done his or her impression, recommendations, follow-up, those things telling what your primary care physician should do, whether go to a specialist. monitor all those things is through that report.
And we're really privileged because we have captured and served 75% of the radiologists within the U.S. and Canada each and every day as they do their documentation.
Again, it goes back to they don't see the necessarily the care team around the patient or the patient, but they're making these very important recommendations of what to do next.
Right. And so your ML algorithms, Do they assist in the document creation and reporting?
Do they assist in the interpretation of the x-rays?
Where does the ML come in? So let's take a step back.
If you think about the different aspects of AI and where AI can really be impactful, the first place was in speech recognition.
So where nuance has always played for the past 20 years is moving from the, if you can imagine, the typing of a report or the analog recording and having somebody transcribe that back to the radiologist to correct.
That was the workflow up until about 10 years ago.
Nuance's role through PowerScribe was to actually take speech and be able to translate that voice to text to be able to start creating these structured reports.
You then layer on natural language processing to help with the clarifications the standardization to drive evidence based medicine to make sure that the report was as accurate as this could be.
So, you know, you've translated from voice to text.
Now you're taking our next layer of AI through natural language processing, what we call our CLU, to make sure that you haven't made mistakes around laterality mismatch, like you're A radiologist makes sure he's talking about the left hand versus the right, the left kidney, the left lung, all those types of things.
And now in this brave new world of machine learning for image characterization, it's really taking the pieces from what we'll call the X-ray images, the CT images, the ultrasound images, to make sure that those results from those algorithms are placed in a way into the diagnostic reports so the radiologists don't have to repeat themselves.
And, you know, we think about this time save, they can get more work done, more accurate, more productive.
It's really this dynamic interaction between the algorithms, making sure that their results are getting into the diagnostic report in a way that saves time and really pressure for the RAD.
Right, and that's, as you pointed out, you're talking about a huge volume of images and documents every year. that are being looked at and created and for the radiologist, for the physician, You know, it's a huge time save, which leads to all the other good things that ultimately leads to better patient care, which is the point.
On the other end, are you able to do anything or are the physicians and care centers, hospitals you're working with, are they able to do anything with the data to help in the advance of evidence-based medicine.
Absolutely, and so what we're seeing now and it's really, it's phenomenal work that Partners is doing here in Boston through the Center for Clinical Data Science, The work with the American College of Radiology, who really advocates for the 36,000 radiologists that are here in the US.
You think about that, you take a step back.
36,000 radiologists are producing 450 million reports on patient images each and every year.
You wanna talk about volume. That's a lot of work, yeah. a lot of work.
And we've been doing this since the inception of really digital imaging for the past 20 years.
And places like CCDS, some of the work that's happening out of Stanford, other centers that are coming out with University of California, San Francisco, where they're actually helping the data scientists to actually start building these algorithms with the amount of annotated imaging data that we have.
So there's this really nice, I'd say, cross where there's a real problem in healthcare.
We have a lot of data that has been annotated by 36,000 radiologists that we can actually put these algorithms to be created and to work.
And so I think this is what makes this time so special that we're getting the trust and the confidence and the leaning in of the governing bodies, the leading academics and the data scientists, and also folks like NVIDIA who are building out really the compute power to be able to do this successfully. which we haven't had before.
And this is what's so exciting about the partnership that we've been working together.
As you were speaking, I was thinking about an episode we did some months back with some palliative care experts. and they were talking about how they're using machine learning and algorithms in their work and getting into this notion of helping basically the network of different positions at the hospital flag patients who would be good candidates to get some extra care.
In this case, have the palliative care team take a look at their cases. see if it might be appropriate to step in.
And we got into this notion of trust and skepticism and fear about the robots are going to take over to make light of it.
But this notion of folks needing to understand that this isn't a replacement for human care, but it's an assist. it's a tool to help process these reams of data that humans couldn't possibly get to, or perhaps in this case, more about reducing workload and burnout.
Have you encountered any significant skepticism or pushback about, even if it's based on misunderstanding of what you're doing, Anything, any stories, anything to talk about around folks kind of coming to the other side and seeing how this is actually helpful?
So, yeah, absolutely. If you go back to 2015, there's been a lot of articles and a lot of Dr. Kim, who is one of our chief medical information officers. talks a lot about the hope, the hype, the fear of AI within radiology.
There is many skeptics out there that the machine is smarter than the human around some of this.
The one thing that I will continue to say, and this is why I am continuing to be so hopeful about this and where Nuance and the partnership of NVIDIA will actually drive wide scale adoption.
There's going to be, let me take a step back.
There's going to be thousands and thousands and thousands of algorithms but each algorithm is only gonna be able to answer one question at a time.
The radiologist, when he sits in front of your images, And if you think about patients are older, patients are sicker, they're more complicated diseases.
There's multiple diseases of obesity with If you think about high blood pressure, diabetes, all the cancer, they're all coming together.
The radiologist is... never just answering that one question that the one algorithm can answer.
He or she is going to answer this question about shortness of breath looking for emphysema, smoking history, And this is what's so important of the angle that we're taking on this together about driving adoption and workflow and driving access of the algorithms as they're being built by the data scientists and the AI developers is to put it in workflow and put it into where they can use the daily practice of utilizing these algorithms each and every day.
But until The algorithm can answer the dozens of questions that a radiologist has to answer and is bound ethically to answer. we're going to see this as this virtual assistant, this virtual resident for here, which I think is an important context of why I'm so hopeful about this, that it's augment the radiologist, not replace the radiologist.
And for us as Nuance with the ability to give access to the RADs of these algorithms through our AI marketplace, the network that we've built because we've been connected to 5,800 facilities, the partnerships that we have with the American College of Radiology, folks like yourselves at Nvidia, through partners and the Center for Clinical Data Sciences, and our wide footprint of PowerScribe, the solution that creates the report, we feel we can focus on daily application to get the radiologist to use this in a meaningful way. and then be able to give that feedback back to the algorithm vendor to make this improve to continue to augment.
Our guest today is Karen Holzberger. Karen is the Vice President and General Manager of Diagnostic Solutions for Nuance Healthcare.
It's a division of Nuance Communications, who you probably know from their decades-long work in voice recognition, natural language processing, and artificial intelligence.
Karen, let's switch gears for a minute and talk a little bit about your background.
You've been working in healthcare IT for a long time and working in radiology in particular, is not new to you.
And if my little research here is correct, you're actually a mechanical engineer by training?
I am, if you can believe it. I do, and it's great.
So, you know, in 30 seconds or less, so to speak, tell us your story.
How did you get into healthcare, into healthcare IT, and now doing what you're doing with radiology.
I'm actually one of five engineers in my family.
And so my dad was an engineer, so there wasn't any real choice except to go to engineering school.
Right. you know, kind of grew up that way.
And I started with GE soon after college on the Edison engineering program.
So really from a technical background. You mentioned before, and many of the guests who I've talked to bring this up, that the increases in compute power In the past, people like to talk about maybe the past two years, past five years, past 10 years. have really sparked this revolution in what's possible or kind of this big leap in the evolution of what's possible.
The compute power is obviously a thing, but is there anything looking back either in your time at Nuance or even further back to the past five, 10 years, Anything that's really surprised you or stuck out as kind of a watershed moment in your industry and the work you're doing?
That's a great question. I think there are two pieces, as you've talked about, certainly the compute power I think the other piece is just this collection of imaging data that we have available to us that we've been collecting for the past 20 years of this annotated data by the radiologist.
The other piece is the cloud, which is so critical.
If you think about cloud, the network, the consumer side of things of adoption of cloud, the adoption of cloud-based applications, which really AI is.
We all go back to what Apple has done through the Apple Store, what they've done with the iPhone, building their network, building the cloud. we see the same thing with the partnership with NVIDIA and ourselves at Nuance is the fact that we've got it's funny to call it this way, but this last mile workflow to the radiologists connected through our cloud network of 5,800 facilities with the compute power through our AI marketplace to be able to get those algorithms into use.
And on top of having this more wide scale adoption of an acceptance of these newer technologies and the availability of all this imaging data.
So to unpack that just a little bit, is the data crunching happening in your cloud or you mentioned kind of this last mile network?
What are you referring to? So for us, we're not doing necessarily the data crunching.
We're gonna rely on the really smart data scientists, the AI developers.
What we see ourselves as the role is the moving of that data to the last mile workflow.
That's the important thing about connecting the developers of AI to the subscribers and users of AI.
It's really important and that's gotta be thought through.
And it's through the cloud, it's through our AI marketplace through our network.
So what is the AI marketplace? So the AI marketplace, and we've been working with the folks at NVIDIA on this too, It's our cloud-based platform through what we call our PowerShare network. we actually have a storefront where AI vendors can say AI doc, some of the partners that you have yourselves can actually publish their algorithms and gain access to the 75% of the market that we have of the users of those algorithms, the radiologists.
And so those are those connection points that we've been able to build through the marketplace.
Now flipping ahead of ourselves and looking forward, obviously the compute power, as it increases, more opportunities will evolve and the cloud marketplace.
I mean, I think that's something that people can very easily, you know, with your example of Apple and other developer marketplaces can very easily see the place in the world and its continued place and the work you're doing.
What else are you working on? Things that you can speak about.
Where do you see this headed for radiologists, for... physician-oriented workflow and other things that you're doing at Nuance.
Where do you see this headed over the next two to five years, let's say?
So for us, it's about driving adoption within their workflow.
You think about the pressures, again, I always go back to the pressures that the radiologist feel each and every day.
More work to be done, less time, less reimbursement, all leading to this burnout process the significant burnout problem that we have in the US around our physician community or physician base, the more that we can do to have AI algorithms in action help prioritize the work Help them see things that they don't see within the report.
Help them document more quickly, whether it's the mundane things around measurements. whether it's helping them see diseases or aneurysms or lung nodules, that they haven't seen to help them create this report in a different way than it always has been from speech to text.
For us, we know the more that we can drive rapid adoption within the tools that they're using each and every day so they're not distracted or have to go to another corner of their of their department to do this, and that they also have this opportunity to give feedback where Maybe they have discordance with the AI algorithm.
Those types of things is what's going to drive confidence and adoption over the next two to five years.
I think for me, it's such an exciting time for us and it's such an exciting time for radiology and diagnostics.
The radiologists have always been these pioneers around IT and technology.
If you think about it, they were the first ones to adopt front-end speech and natural language processing to drive their report creation.
They were the first to adopt CT and MR and get away from what everybody called analog and film-based interpretation to digital. the adoption of PACs, picture archiving communication systems, all of this.
And they're the right ones to drive this across healthcare too in this new paradigm of AI and AI adoption.
Well, thank you, first of all, for the work that you're all doing, because again, in the end, it's about... keeping people healthy, healthier, improving quality of life.
And there's nothing more important than that. and practical application standpoint, it's just fascinating.
So Karen, thank you for taking the time.
If folks want to find out more about what you're doing with AI at Nuance from a physician standpoint, a developer standpoint, or just an interested party standpoint, where should they go?
Nuance.com slash healthcare. Easy enough.
Karen Holzberger, thank you again for joining the podcast.
Stay warm out there on the East Coast. And best of luck.
Look forward to catching up with you again in the future.
It was terrific talking to you. Thanks so much.
Thank you. Thank you.
