January 23, 2025 • 26 mins
What if the future of healthcare weren’t about replacing clinicians but empowering them to do more, faster, and better? From addressing clinician shortages to processing massive volumes of medical data, AI is rapidly moving beyond diagnostics and imaging into areas like workflow optimization, multimodal input, and real-time decision-making.
In this podcast, we dive into how AI scales at the device level, enhancing patient outcomes, and uncovering actionable insights from one of the largest data-generating sectors in the world. Discover how edge computing enables faster, more secure data analysis right where care is delivered, and learn how these advancements shape a more efficient and innovative future for healthcare.
Join us as we explore these ideas with:
Ian Fogg, Research Director of Network Innovation, CCS Insight
Alex Flores, Head of Global Health Solutions Vertical, Intel
Christina Cardoza, Editorial Director, insight.tech
Ian and Alex answer our questions about:
- The latest CCS Insight healthcare AI report findings
- Healthcare industry trends and observations
- Adopting technologies safely and securely
- How AI at the edge addresses healthcare data
- Strategies to bring AI into the healthcare industry
- The value of working with Intel
- Examples and case studies of healthcare transformations
- How AI usage will evolve in healthcare
Related Content
To learn more about healthcare at the edge, read about the latest innovations in patient care, monitoring, and wellness. For the latest innovations from CCS Insight, follow them on X/Twitter at @ccsinsight and LinkedIn. For the latest innovations from Intel, follow them on X/Twitter at @intel and LinkedIn.
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
(00:00):
(upbeat music)
(00:12):
- Hello and welcome to"insight.tech Talk,"
where we explore the latest IoT, AI, edge,
and network technologytrends and innovations.
As always, I'm yourhost, Christina Cardoza,
Editorial Director of insight.tech,
and today, we're going to bediscussing AI in healthcare
with our good friend IanFogg from CCS Insight
and Alex Flores from Intel.
Hey, guys, thanks for joining us.
(00:33):
- Thank you for having me.
- Before we get started,
we'd love to learn a littlebit more about yourselves
and the companies that you work at.
So Alex, I'll start with you.
What can you tell us about yourself
and what you do at Intel?
- Sure, so my name is Alex Flores.
I'm the Director of the HealthSolutions Vertical at Intel.
At Intel, we're not cliniciansor former clinicians.
(00:54):
We're engineers and we're technologists,
and we're driven by the intersection
of technology and healthcare.
And what we do is wework with the ecosystem
to look at how some ofthe biggest challenges
can be solved with technology.
- Great, looking forward todiving into that in just a bit.
But before we get there, Ian,
you've been a friend and guest of the show
(01:17):
numerous of times now,
but for those of our listenerswho have not listened
to those other greatconversations we've had
on the "insight.tech Talk,"
what can you tell us aboutyourself and CCS Insight?
- So, CCS Insight,
we're essentially atechnology research firm,
an industry analyst firm.
I'm a Research Director here.
Our role really working with insights.tech
(01:37):
is to interview key players in the market,
research some of the changes
like, for example, AIarriving in healthcare,
and then communicatethose and write those up
in a form that's easy tokind of pick up and use.
- Of course, and on that note,
CCS Insight has aresearch paper coming out
(01:57):
on this very topic,
which is why we have you joining us today.
So what can you tell us about that report
and if there was any particularfindings in this space
that stood out to you?
- So I think just how extensive
AI usage in healthcare already is.
It's something that I think is recently
(02:18):
in the last couple of years really arrived
on the kind of the popularmindset in the mainstream media.
But it's clear that inhealthcare, it is well-embedded,
it's widely used, and it'sgrowing across all categories.
I mean, it was one ofthe striking statistics
that I think we found inthe research, the report,
was that as of August 2024,
(02:40):
there were 950 AI-enabled medical devices
that have been approved bythe FDA across categories.
That's an enormous number.
And of course, it's growing all the time.
I think the other thingthat's really striking
about what's happening here
is how much it's moving fromdiagnostics, and imaging,
and research into other of thekind of healthcare ecosystem.
(03:01):
So organizational tasks, room management,
tying together disparate systems,
and also things like multimodal input.
So, transcribing conversations
that would otherwise never be recorded.
There's just this enormousburgeoning range of activities
really right away across the sector.
- It certainly is interestingand you bring up a good point.
(03:22):
It's not only the devices
and not only necessarily directlyrelated to the healthcare,
but there are things in hospitals,
or in organizations, and offices,
and buildings that we can addAI within the healthcare space
to really improve operationsand inefficiencies.
And like you said, a lotof this has been ongoing.
We're only hearing a bitabout it in the media.
(03:44):
I think that's some ofthe best implementation
of the technology is when it's happening.
But as a consumer or as a user,
you don't see it happening right up front.
Alex, I'm curious,
based on some of thefindings Ian just mentioned,
is that what you're seeing in the space
from an intel and engineering perspective?
- Absolutely.
(04:04):
We are seeing AI beingrapidly adopted in healthcare,
and it's really going across the board,
whether it's a patient is registering
or checking in, for example,
there's AI analyticsgoing on in the background
trying to...
Whether it's pre-fillingin your forms and so forth
(04:26):
or gathering data frompast visits and so forth.
Or in the actual clinical workflow,
whether a patient is receivingsome type of healthcare,
whether the doctor, for example,
is transcribing notes and so forth.
(04:47):
So it goes across multiple workflows.
And I think what's interesting
is a lot of it is reallybehind the scenes,
which is where we want it to be,
because ultimately whatit's doing behind the scenes
is impacting the clinician's workflow,
allowing them to do theirjob faster, better, easier,
so they can spend moretime with the patient.
(05:08):
- That's great.
And those benefits are a lotof things that I've seen.
Writing for insight.tech andfor different industries,
manufacturing, retail, tryingto get those benefits from AI.
I think healthcare isan interesting space.
It presents a lot of interestingchallenges and complexities
just because you're dealingwith a different environment,
(05:31):
regulations, patient-sensitive data.
So can you talk a little bitabout how the healthcare space
is able to adopt these technologies
in a safe, secure, efficient way?
- Yeah, before I jump into that,
I think a couple of datapoints I wanted to mention.
I think what's unique about healthcare
(05:51):
is a lot of people don't realize this,
that from a data perspective,
roughly about one thirdof the world's data
is being driven out of healthcare.
And then there's evidence out there
that maybe roughly about 5% of that data
is actually turned intoactionable insights.
So, there's this tremendousopportunity to use AI
(06:13):
to really kind of unlockthose insights from that data.
The second thing is you layerin some of the macro trends
that are happeningglobally across healthcare.
Those include an aging population.
They also include...
People are getting sicker,
there's more and more people
that are being diagnosed withmultiple chronic diseases.
(06:35):
And then you also layer the fact
that there's a globalshortage of clinicians,
both doctors and nurses, for example.
So with that, the need for AIbecomes even more important
and the rapid adoption ofthat becomes more important
because what it's doing isit's allowing clinicians
(06:57):
to increase their efficiencies,
whether it's the workflows,
whether it's being able totriage patients, and so forth.
So for me, that's whereAI's going to be crucial
in order to continue to helpalleviate some of those issues
and really allow our cliniciansthe benefits to do that.
(07:20):
And then, if it's implemented correctly,
you don't have to worryabout some of the regulatory.
Again, it's really thereto benefit the clinicians
so that they can allow themselves
to really focus on thepatient and patient outcomes.
- That's an interesting perspective on it.
I want to go back a little bit
to the data points thatyou were talking about,
(07:40):
all the data that's comingfrom the healthcare sector.
And then I imagine withdevices coming online
or more devices being AI-enabled,
that's just giving us even more data,
which I'm sure AI is helpful,
being able to sortthrough some of that data.
But Ian, I'm curious,
if you can touch on this growing amount
of healthcare data that we have
(08:02):
and how AI and AI at the edge
is really going to come into play,
if there's anything fromthat report you found
in this space.
- I think there's a few things here.
I think just touching on something
that Alex mentioned there,
I think what AI is doing in many areas
is not replacing clinicians,
it's making the clinicians more efficient;
it's taking load off them.
And you can see that in the waythat data is being analyzed.
(08:24):
You can see data volumesgoing up enormously.
One study I think
was talking about thatthe size of a CT scan
could be 250 megabytes for an abdomen,
a staggering one gigabyte for the heart,
digital pathology could be even greater
if you're looking at cells of 2.5.
Those are enormous,enormous amounts of data
for a single scan.
If you compare that witha smartphone camera,
(08:47):
that might be a five-megabyte image.
And one of the other thingsthat's striking, though,
is you can't use the same techniques
to compress medical imaging data
that you can use for a photograph
because the tools usedto compress a photograph
are lousy tools.
They lose data
and they lose it based on
what the human eye doesn't perceive.
They're perceptual compression algorithms.
(09:07):
You can't do that for medical.
You have to look at the full image
because you need to have all that detail
so you can spotirregularities in the scan.
And so that just makesthe challenge even harder.
Then you've got...
Well, you've got thisenormous amount of data.
AI has two slightlycompeting implications there.
One is that it means you cananalyze that data quicker.
(09:29):
So you can have anefficiency of speed benefit.
But one of the companies we talked about
framed it the other way around and said,
"Look, actually,
because you've got this AI toolthat can analyze more data,
what you can actually do
is analyze a greater part of a biopsy."
Which means that ifthere's just a few cells
that are irregular in a cancer scan,
(09:49):
you are more like to spot them
because you've scanned a bigger sample,
and that means your scan is more accurate,
which means you'll identify problems
and healthcare issues earlier,
and you'll save costs andload on the healthcare system
kind of down the line.
So there's some interesting dynamics there
that are striking.
The other piece is that when you want that
(10:11):
to be a very responsiveexperience for the clinician,
if you can do it at theedge rather than the cloud,
you can make that faster and you can...
It's easier to make it private
because the data can staycloser to the patient,
closer to where it's being captured.
And that's a trend we'veseen in many areas with AI,
(10:31):
where we've seen thingsstart in the cloud,
and then as edge devices get more capable,
you move those thingsonto the edge devices
for that performance benefit.
So there's all kind ofinteresting dynamics here
when you start looking at the data.
Do you make it a fast experience
or do you use AI to analyzea greater amount of the data
to improve the qualityof what you're doing?
(10:53):
- Ian, you really bring up
some really interestingand valid points, too.
I think one of the thingsthat I did want to emphasize
is what you were talking about,
speed, quality, and so forth,
and I think that's where alot of new compute technology
comes into play as well that'sworking in the background.
So for example, latency.
(11:13):
When a radiologist brings up an image,
if they're triaging something,
they want to be able to see that image
in real-time or near real-time
because every second counts, obviously,
for a lot of these healthcare providers.
Technologies like compressionand decompression,
again, these are allworking in the background,
(11:35):
but as we work
with a lot of the differentecosystem players,
a lot of the leading medicaldevice manufacturers,
that's what Intel's doingkind of in the background
is really looking at how wecan optimize their technology,
their workflows, theiralgorithms, and so forth,
so it gives the clinician that real-time
or near real-timeexperience that they need,
(11:57):
and if it's done correctly, it's seamless,
so they can go about theirjob as quickly as possible.
- And I'm sure when you'rethinking of cloud versus edge,
it depends on the device
or depends on the outcomethat you're trying to get.
Do you need the real-timemetrics and insights
to have it on the edge,
or is it quality and beingable to go through all the data
(12:18):
and that being on the cloud?
So I know Ian, you were talkingabout different approaches
to dealing with things,
especially in healthcare,
and so we have theedge, we have the cloud,
but are there any other strategies
that healthcare providers orpeople in the healthcare space,
or even patients,
can implement to bring AIinto the healthcare industry
(12:39):
and any best practices there?
- So, two things jumps out.
I mean, one is just this use sense,
that it isn't just about imaging and scans
and that kind of computer vision piece.
We've seen a lot of examplesnow of AI being used
to make the organizational aspects
from healthcare to thehospital more efficient.
(13:01):
Things like operating theatersare incredibly costly assets,
and if you can schedule the cleaning
and sanitization teams efficiently,
you can reduce downtimebetween operations.
And that came up in oneof the interviews we did
for the report.
The other thing that I thinkthat came up very strongly
was what's called federated learning.
This idea that when you havea machine learning model,
(13:24):
you want to maintain the privacy,
but you want to use adiverse and broad data set
to improve the quality of the AI model.
And a federated learning approach
means you can havepotentially multiple hospitals
or multiple healthcare facilitiescontributing to the model,
but where the data that'sused to improve the model
remains within the facility.
(13:46):
And that's somethingwhich enables the AI model
to become much more capable,much more sophisticated,
but still works within the kindof the environment you need
around privacy and management.
But we've seen that in some other areas,
but it's particularly relevantin the healthcare space.
- It's interesting as you'retalking about these approaches,
(14:07):
and strategies,
and the benefits that thehealthcare space gets with this,
I'm brought back to what Alexwas saying in the beginning,
how you aren't healthcareproviders, you're engineers,
and then we have thesehealthcare providers
implementing this technology.
So, Alex, from an engineering perspective,
how can hospitals, healthcareproviders deploy AI?
(14:31):
What challenges or opportunitiesdo they face in this space
and working with an engineerlike Intel to make it happen?
- Yeah, I think oftentimes,
when we're working withcustomers in the ecosystem,
really, it starts with giving them choice,
giving them options whenthey're deploying AI.
As Ian mentioned,
(14:52):
a lot of organizationsare deploying in the cloud
and that's great, it's a tried method.
Innovation is happening all the time.
The cloud is obviouslybeen in use for decades.
There's other organizations
that are kind of takinga hybrid approach, right?
They want the benefits of the cloud,
but they also want the benefits
of being able to access datareal-time or near real-time
(15:14):
at the edge.
And then there's other customers
that are looking at an edge-only approach,
where concerned, as Ian was saying,
maybe it's cost, maybeit's security reasons
and privacy, and so forth.
So for Intel, what we really want to do
is kind of walk them throughthe different options.
And specifically when we get to the edge.
(15:37):
What makes the edge so attractive
is that access to that data,
that access to patient data at a real-time
or near real-time.
So clinicians can take advantage of it,
especially if they'retrying to triage a situation
for a patient.
So having that access to that data,
(15:57):
being able to run the correct analytics
for that data at thatmoment becomes very crucial.
And then at that point,they can determine,
"Okay, do I need to save this data?
Does it need to be stored in the cloud?
Can I send it to maybea local data center?"
and so forth.
So for us, it's really showingthe customers the ecosystems,
the choices, some of the benefits,
(16:19):
and then seeing what's best
for their particular implementation.
- So, to paint a picturefor the audience here,
do you have any examples
or case studies you can share with us?
And you don't have to name names,
but anyone that came to you,you gave them these options,
what the options were, what they chose,
and what the result was of that?
(16:40):
- Yeah, we have many different options,
which makes my job really exciting
to be able to see someof these technologies
being implemented.
And I've actually had the benefit
of seeing some of them actually in play.
One that comes to mindis patient positioning.
So for example, when apatient is getting a scan,
(17:01):
and they lay on the table,
what happened before inthe past is oftentimes,
the patient wasn't positioned correctly.
So then that would cause the technician
to have to redo the scans,
and then obviously for variousdifferent reasons, right?
(17:22):
Now it's taking longer
because the patient has toget rescanned, for example.
The patient may be exposedto additional radiation
that they shouldn't havebeen exposed to and so forth.
So, having AI-based algorithms
that help the technologyposition the patient correctly
before the scan.
So that's one example.
(17:43):
Second one is around accuratecontouring of organs at risk.
So one of the major bottlenecksfor radiation therapy
is doing this contouring of these organs,
and often based on theimage quality and so forth,
there can be a lot of error in that.
So, having AI-based contouring
(18:04):
is another area that reallycan help the clinicians
speed up their process,
it really can help automate
some of these differenttasks and so forth.
A last example that I have ison ultrasound, for example,
and this is a real story.
So my wife, she had aprocedure a couple years ago,
(18:25):
and I remember we were driving back
and I asked her how the procedure went,
and she started describinga situation where she said,
"The anesthesiologist came in
and they used an ultrasoundmachine to identify the vein
where the anesthesiawould be administered."
And I got really excited because I said,
"Oh, I know exactlywhat algorithm that was
(18:47):
because we were working withthe ultrasound manufacturer
to optimize that."
And essentially, that algorithm was used
to help identify the vein to avoid,
sitting there having theclinician do multiple insertions
of a needle before finding the vein.
So, those are just threeexamples, the list goes on and on.
(19:11):
That's why, again, I getso excited about my job
is seeing that practicalityof the technology
being implemented with a solution.
- That's great seeingit out in the wild, too,
and having a personal experience
with some of this technologythat you're working on.
Definitely somethingthat I could have used
or that I can't wait tosee out in the real world.
(19:32):
I've had three children,
and my second one, theypoked and prodded my arm.
It was black and blue,
they couldn't get an IV linethat my third one, I was like,
"I don't even want one,don't even put it near me."
So, it's really...
I can't wait to see some of this stuff
be more widely adopted.
And we're talking about likediagnostics, and imaging,
and other areas,
(19:53):
but like you said, there's so many options
and so many different placesAI and healthcare could go.
You guys mentioned a couple times
dictating notes for doctorsand things like that.
So Ian, I'm curious, froma research perspective,
where is this space going?
Do you have any future-looking ideas
(20:14):
on how AI usage is going tocontinue to evolve in healthcare?
- I think this could evolve in many areas.
I mean, that ultrasound example,
I think it's a fascinating example
because ultrasound'sa very cost-effective,
very accessible type of scanning.
And what you are doing with that
is you are making a tool
that's been around fordecades more effective,
and that that is augmentingan existing tool.
(20:34):
It's a fascinating example.
I think some things, we'reclearly going to see.
We're going to seecloud-based AI continue,
but we're going to see increasinguse of AI on the edge too
for that responsiveness piece.
The other I think we'll see
is we've seen these very large AI models,
we'll see more smaller-focusedmodels come to market
for particular task or use,
(20:55):
and they become more portable,
they become even easierto put onto edge devices.
And we've seen that in other fields
outside of healthcare too.
I think we'll see this multimodal element.
So, multimodal means audio-based,
video-based, still-image-based,and text-based.
And that means both a way ofinteracting with the model,
(21:17):
but also what the modelis able to understand
and perceive about the world.
So it might be able to perceive,
use a camera to identifyif there are queues forming
or people forming in certainparts of the hospital.
The transcription piece is interesting.
That means you are capturing information
that may otherwise neverhave been captured,
maybe a patient-doctor conversation.
(21:38):
And then you can summarizethat conversation,
so you can add thingsinto the medical record
that maybe aren't beingcaptured at the moment,
but also make itaccessible, and surfaceable,
and findable later.
I think there are otherthings beyond that.
AI is very good at correlating trends
across different data sets.
This could be used in apublic health context more.
(22:01):
AI models can't do causation.
So when you find those correlations,
you'll still need to go and push them
in front of a researcher, a clinician,
to validate that it's a real thing,
not just one of those random correlations,
but it will probably uncover
underlying causes for conditions,
new ways of approaching healthcare
(22:21):
that we haven't thought about before.
And then there's just so many uses.
You can look at this reallyright the way across,
everywhere that technology's being used
in a hospital facility, I think.
- Yeah, so many opportunities.
And in this conversation,
we stuck a little bit to the devices,
and the implementation,and the data aspect of it,
but I'm sure we can go offin many different directions.
(22:44):
We're talking about the AI models,
the size of the Amodels, what they can do,
but at risk of opening up a can of worms
because I'm sure we'veonly scratched the surface
and we can keep going on and on,
I'm going to end the conversation here.
But what I'd love to do isthrow it back to each of you
for any final thoughts
or key takeaways you want toleave our listeners with today
(23:07):
as they prepare for theAI evolution in healthcare
or what they can expect to see.
So, Ian, I'll start with you on this one.
- I mean, I think oneof the big things here
is we've seen a lot of hypehere around internet-based LLMs.
I would say don't be discouraged
by the quality of thosethings like ChatGPT,
or Gemini, or Claude.
I mean, when you start lookingat these medical AI models,
(23:30):
they're typically trainedon pre-validated data sets,
not the internet.
So the accuracy level is much greater.
I think additionally, we'veseen some things come through,
where you can use one AI model
to validate the outputof another AI model,
and that can raise thequality of the output, too.
So, that quality piece that you might see
when you are playing with stuff online
(23:51):
isn't applicable here.
This is a different kind of space.
And in some cases,
people are using in-house,open source-based models,
so they have greatercontrol ownership of it too.
So, don't be discouragedby what you might see
in other areas on your phone,
or on your computer, or on the internet.
This is a different space.
The quality here is much, much higher.
(24:14):
- Awesome, and Ian,
I'll make sure to providea link out to that report,
so that those listening,
they can learn and diginto some of the things
that we were talking about even further.
Alex, before we go, any final thoughts
or key takeaways you wantto leave our listeners with?
- Yeah, I think Ian mentioneda really good thing,
and that's the miniaturization of AI.
And essentially, we're goingto continue to see that pattern
(24:38):
and we're going to learn
what is the right AI at theright time at the right device.
And the other thing thatI wanted to also mention
is when you're doing AIat the edge on the device,
power becomes a really important feature,
because if you think about it,
it's kind of a snowball effect.
(24:58):
More power you need, the biggerfans you're going to need,
the bigger device, the newform factor, and so forth.
Oftentimes, you don't need that.
You can run the rightamount of AI at the edge
without needing to redesignor reconfigure your device.
There's new technology,
new compute that allows you to do that.
So, as we continue to evolve,
(25:20):
as more and more artificialintelligence goes to the edge,
it's going to be easier andeasier to run at the edge.
- And I think deploy AI at the edge also,
like you said, thesedevices and technology,
it's getting smaller.
It's amazing what you can
with the infrastructure you already have
and without a lot ofhardware, a lot of equipment.
(25:42):
So, I can't wait to seewhere else the space goes,
other innovations andtechnologies from Intel.
So, I just want to thank youboth again for joining us today
and for talking about this topic.
Thank you to our listeners alsofor coming in and listening.
Until next time, this hasbeen "insight.tech Talk."
(uplifting music)
(upbeat music)
(00:12):
- Hello and welcome to"insight.tech Talk,"
where we explore the latest IoT, AI, edge,
and network technologytrends and innovations.
As always, I'm yourhost, Christina Cardoza,
Editorial Director of insight.tech,
and today, we're going to bediscussing AI in healthcare
with our good friend IanFogg from CCS Insight
and Alex Flores from Intel.
Hey, guys, thanks for joining us.
(00:33):
- Thank you for having me.
- Before we get started,
we'd love to learn a littlebit more about yourselves
and the companies that you work at.
So Alex, I'll start with you.
What can you tell us about yourself
and what you do at Intel?
- Sure, so my name is Alex Flores.
I'm the Director of the HealthSolutions Vertical at Intel.
At Intel, we're not cliniciansor former clinicians.
(00:54):
We're engineers and we're technologists,
and we're driven by the intersection
of technology and healthcare.
And what we do is wework with the ecosystem
to look at how some ofthe biggest challenges
can be solved with technology.
- Great, looking forward todiving into that in just a bit.
But before we get there, Ian,
you've been a friend and guest of the show
(01:17):
numerous of times now,
but for those of our listenerswho have not listened
to those other greatconversations we've had
on the "insight.tech Talk,"
what can you tell us aboutyourself and CCS Insight?
- So, CCS Insight,
we're essentially atechnology research firm,
an industry analyst firm.
I'm a Research Director here.
Our role really working with insights.tech
(01:37):
is to interview key players in the market,
research some of the changes
like, for example, AIarriving in healthcare,
and then communicatethose and write those up
in a form that's easy tokind of pick up and use.
- Of course, and on that note,
CCS Insight has aresearch paper coming out
(01:57):
on this very topic,
which is why we have you joining us today.
So what can you tell us about that report
and if there was any particularfindings in this space
that stood out to you?
- So I think just how extensive
AI usage in healthcare already is.
It's something that I think is recently
(02:18):
in the last couple of years really arrived
on the kind of the popularmindset in the mainstream media.
But it's clear that inhealthcare, it is well-embedded,
it's widely used, and it'sgrowing across all categories.
I mean, it was one ofthe striking statistics
that I think we found inthe research, the report,
was that as of August 2024,
(02:40):
there were 950 AI-enabled medical devices
that have been approved bythe FDA across categories.
That's an enormous number.
And of course, it's growing all the time.
I think the other thingthat's really striking
about what's happening here
is how much it's moving fromdiagnostics, and imaging,
and research into other of thekind of healthcare ecosystem.
(03:01):
So organizational tasks, room management,
tying together disparate systems,
and also things like multimodal input.
So, transcribing conversations
that would otherwise never be recorded.
There's just this enormousburgeoning range of activities
really right away across the sector.
- It certainly is interestingand you bring up a good point.
(03:22):
It's not only the devices
and not only necessarily directlyrelated to the healthcare,
but there are things in hospitals,
or in organizations, and offices,
and buildings that we can addAI within the healthcare space
to really improve operationsand inefficiencies.
And like you said, a lotof this has been ongoing.
We're only hearing a bitabout it in the media.
(03:44):
I think that's some ofthe best implementation
of the technology is when it's happening.
But as a consumer or as a user,
you don't see it happening right up front.
Alex, I'm curious,
based on some of thefindings Ian just mentioned,
is that what you're seeing in the space
from an intel and engineering perspective?
- Absolutely.
(04:04):
We are seeing AI beingrapidly adopted in healthcare,
and it's really going across the board,
whether it's a patient is registering
or checking in, for example,
there's AI analyticsgoing on in the background
trying to...
Whether it's pre-fillingin your forms and so forth
(04:26):
or gathering data frompast visits and so forth.
Or in the actual clinical workflow,
whether a patient is receivingsome type of healthcare,
whether the doctor, for example,
is transcribing notes and so forth.
(04:47):
So it goes across multiple workflows.
And I think what's interesting
is a lot of it is reallybehind the scenes,
which is where we want it to be,
because ultimately whatit's doing behind the scenes
is impacting the clinician's workflow,
allowing them to do theirjob faster, better, easier,
so they can spend moretime with the patient.
(05:08):
- That's great.
And those benefits are a lotof things that I've seen.
Writing for insight.tech andfor different industries,
manufacturing, retail, tryingto get those benefits from AI.
I think healthcare isan interesting space.
It presents a lot of interestingchallenges and complexities
just because you're dealingwith a different environment,
(05:31):
regulations, patient-sensitive data.
So can you talk a little bitabout how the healthcare space
is able to adopt these technologies
in a safe, secure, efficient way?
- Yeah, before I jump into that,
I think a couple of datapoints I wanted to mention.
I think what's unique about healthcare
(05:51):
is a lot of people don't realize this,
that from a data perspective,
roughly about one thirdof the world's data
is being driven out of healthcare.
And then there's evidence out there
that maybe roughly about 5% of that data
is actually turned intoactionable insights.
So, there's this tremendousopportunity to use AI
(06:13):
to really kind of unlockthose insights from that data.
The second thing is you layerin some of the macro trends
that are happeningglobally across healthcare.
Those include an aging population.
They also include...
People are getting sicker,
there's more and more people
that are being diagnosed withmultiple chronic diseases.
(06:35):
And then you also layer the fact
that there's a globalshortage of clinicians,
both doctors and nurses, for example.
So with that, the need for AIbecomes even more important
and the rapid adoption ofthat becomes more important
because what it's doing isit's allowing clinicians
(06:57):
to increase their efficiencies,
whether it's the workflows,
whether it's being able totriage patients, and so forth.
So for me, that's whereAI's going to be crucial
in order to continue to helpalleviate some of those issues
and really allow our cliniciansthe benefits to do that.
(07:20):
And then, if it's implemented correctly,
you don't have to worryabout some of the regulatory.
Again, it's really thereto benefit the clinicians
so that they can allow themselves
to really focus on thepatient and patient outcomes.
- That's an interesting perspective on it.
I want to go back a little bit
to the data points thatyou were talking about,
(07:40):
all the data that's comingfrom the healthcare sector.
And then I imagine withdevices coming online
or more devices being AI-enabled,
that's just giving us even more data,
which I'm sure AI is helpful,
being able to sortthrough some of that data.
But Ian, I'm curious,
if you can touch on this growing amount
of healthcare data that we have
(08:02):
and how AI and AI at the edge
is really going to come into play,
if there's anything fromthat report you found
in this space.
- I think there's a few things here.
I think just touching on something
that Alex mentioned there,
I think what AI is doing in many areas
is not replacing clinicians,
it's making the clinicians more efficient;
it's taking load off them.
And you can see that in the waythat data is being analyzed.
(08:24):
You can see data volumesgoing up enormously.
One study I think
was talking about thatthe size of a CT scan
could be 250 megabytes for an abdomen,
a staggering one gigabyte for the heart,
digital pathology could be even greater
if you're looking at cells of 2.5.
Those are enormous,enormous amounts of data
for a single scan.
If you compare that witha smartphone camera,
(08:47):
that might be a five-megabyte image.
And one of the other thingsthat's striking, though,
is you can't use the same techniques
to compress medical imaging data
that you can use for a photograph
because the tools usedto compress a photograph
are lousy tools.
They lose data
and they lose it based on
what the human eye doesn't perceive.
They're perceptual compression algorithms.
(09:07):
You can't do that for medical.
You have to look at the full image
because you need to have all that detail
so you can spotirregularities in the scan.
And so that just makesthe challenge even harder.
Then you've got...
Well, you've got thisenormous amount of data.
AI has two slightlycompeting implications there.
One is that it means you cananalyze that data quicker.
(09:29):
So you can have anefficiency of speed benefit.
But one of the companies we talked about
framed it the other way around and said,
"Look, actually,
because you've got this AI toolthat can analyze more data,
what you can actually do
is analyze a greater part of a biopsy."
Which means that ifthere's just a few cells
that are irregular in a cancer scan,
(09:49):
you are more like to spot them
because you've scanned a bigger sample,
and that means your scan is more accurate,
which means you'll identify problems
and healthcare issues earlier,
and you'll save costs andload on the healthcare system
kind of down the line.
So there's some interesting dynamics there
that are striking.
The other piece is that when you want that
(10:11):
to be a very responsiveexperience for the clinician,
if you can do it at theedge rather than the cloud,
you can make that faster and you can...
It's easier to make it private
because the data can staycloser to the patient,
closer to where it's being captured.
And that's a trend we'veseen in many areas with AI,
(10:31):
where we've seen thingsstart in the cloud,
and then as edge devices get more capable,
you move those thingsonto the edge devices
for that performance benefit.
So there's all kind ofinteresting dynamics here
when you start looking at the data.
Do you make it a fast experience
or do you use AI to analyzea greater amount of the data
to improve the qualityof what you're doing?
(10:53):
- Ian, you really bring up
some really interestingand valid points, too.
I think one of the thingsthat I did want to emphasize
is what you were talking about,
speed, quality, and so forth,
and I think that's where alot of new compute technology
comes into play as well that'sworking in the background.
So for example, latency.
(11:13):
When a radiologist brings up an image,
if they're triaging something,
they want to be able to see that image
in real-time or near real-time
because every second counts, obviously,
for a lot of these healthcare providers.
Technologies like compressionand decompression,
again, these are allworking in the background,
(11:35):
but as we work
with a lot of the differentecosystem players,
a lot of the leading medicaldevice manufacturers,
that's what Intel's doingkind of in the background
is really looking at how wecan optimize their technology,
their workflows, theiralgorithms, and so forth,
so it gives the clinician that real-time
or near real-timeexperience that they need,
(11:57):
and if it's done correctly, it's seamless,
so they can go about theirjob as quickly as possible.
- And I'm sure when you'rethinking of cloud versus edge,
it depends on the device
or depends on the outcomethat you're trying to get.
Do you need the real-timemetrics and insights
to have it on the edge,
or is it quality and beingable to go through all the data
(12:18):
and that being on the cloud?
So I know Ian, you were talkingabout different approaches
to dealing with things,
especially in healthcare,
and so we have theedge, we have the cloud,
but are there any other strategies
that healthcare providers orpeople in the healthcare space,
or even patients,
can implement to bring AIinto the healthcare industry
(12:39):
and any best practices there?
- So, two things jumps out.
I mean, one is just this use sense,
that it isn't just about imaging and scans
and that kind of computer vision piece.
We've seen a lot of examplesnow of AI being used
to make the organizational aspects
from healthcare to thehospital more efficient.
(13:01):
Things like operating theatersare incredibly costly assets,
and if you can schedule the cleaning
and sanitization teams efficiently,
you can reduce downtimebetween operations.
And that came up in oneof the interviews we did
for the report.
The other thing that I thinkthat came up very strongly
was what's called federated learning.
This idea that when you havea machine learning model,
(13:24):
you want to maintain the privacy,
but you want to use adiverse and broad data set
to improve the quality of the AI model.
And a federated learning approach
means you can havepotentially multiple hospitals
or multiple healthcare facilitiescontributing to the model,
but where the data that'sused to improve the model
remains within the facility.
(13:46):
And that's somethingwhich enables the AI model
to become much more capable,much more sophisticated,
but still works within the kindof the environment you need
around privacy and management.
But we've seen that in some other areas,
but it's particularly relevantin the healthcare space.
- It's interesting as you'retalking about these approaches,
(14:07):
and strategies,
and the benefits that thehealthcare space gets with this,
I'm brought back to what Alexwas saying in the beginning,
how you aren't healthcareproviders, you're engineers,
and then we have thesehealthcare providers
implementing this technology.
So, Alex, from an engineering perspective,
how can hospitals, healthcareproviders deploy AI?
(14:31):
What challenges or opportunitiesdo they face in this space
and working with an engineerlike Intel to make it happen?
- Yeah, I think oftentimes,
when we're working withcustomers in the ecosystem,
really, it starts with giving them choice,
giving them options whenthey're deploying AI.
As Ian mentioned,
(14:52):
a lot of organizationsare deploying in the cloud
and that's great, it's a tried method.
Innovation is happening all the time.
The cloud is obviouslybeen in use for decades.
There's other organizations
that are kind of takinga hybrid approach, right?
They want the benefits of the cloud,
but they also want the benefits
of being able to access datareal-time or near real-time
(15:14):
at the edge.
And then there's other customers
that are looking at an edge-only approach,
where concerned, as Ian was saying,
maybe it's cost, maybeit's security reasons
and privacy, and so forth.
So for Intel, what we really want to do
is kind of walk them throughthe different options.
And specifically when we get to the edge.
(15:37):
What makes the edge so attractive
is that access to that data,
that access to patient data at a real-time
or near real-time.
So clinicians can take advantage of it,
especially if they'retrying to triage a situation
for a patient.
So having that access to that data,
(15:57):
being able to run the correct analytics
for that data at thatmoment becomes very crucial.
And then at that point,they can determine,
"Okay, do I need to save this data?
Does it need to be stored in the cloud?
Can I send it to maybea local data center?"
and so forth.
So for us, it's really showingthe customers the ecosystems,
the choices, some of the benefits,
(16:19):
and then seeing what's best
for their particular implementation.
- So, to paint a picturefor the audience here,
do you have any examples
or case studies you can share with us?
And you don't have to name names,
but anyone that came to you,you gave them these options,
what the options were, what they chose,
and what the result was of that?
(16:40):
- Yeah, we have many different options,
which makes my job really exciting
to be able to see someof these technologies
being implemented.
And I've actually had the benefit
of seeing some of them actually in play.
One that comes to mindis patient positioning.
So for example, when apatient is getting a scan,
(17:01):
and they lay on the table,
what happened before inthe past is oftentimes,
the patient wasn't positioned correctly.
So then that would cause the technician
to have to redo the scans,
and then obviously for variousdifferent reasons, right?
(17:22):
Now it's taking longer
because the patient has toget rescanned, for example.
The patient may be exposedto additional radiation
that they shouldn't havebeen exposed to and so forth.
So, having AI-based algorithms
that help the technologyposition the patient correctly
before the scan.
So that's one example.
(17:43):
Second one is around accuratecontouring of organs at risk.
So one of the major bottlenecksfor radiation therapy
is doing this contouring of these organs,
and often based on theimage quality and so forth,
there can be a lot of error in that.
So, having AI-based contouring
(18:04):
is another area that reallycan help the clinicians
speed up their process,
it really can help automate
some of these differenttasks and so forth.
A last example that I have ison ultrasound, for example,
and this is a real story.
So my wife, she had aprocedure a couple years ago,
(18:25):
and I remember we were driving back
and I asked her how the procedure went,
and she started describinga situation where she said,
"The anesthesiologist came in
and they used an ultrasoundmachine to identify the vein
where the anesthesiawould be administered."
And I got really excited because I said,
"Oh, I know exactlywhat algorithm that was
(18:47):
because we were working withthe ultrasound manufacturer
to optimize that."
And essentially, that algorithm was used
to help identify the vein to avoid,
sitting there having theclinician do multiple insertions
of a needle before finding the vein.
So, those are just threeexamples, the list goes on and on.
(19:11):
That's why, again, I getso excited about my job
is seeing that practicalityof the technology
being implemented with a solution.
- That's great seeingit out in the wild, too,
and having a personal experience
with some of this technologythat you're working on.
Definitely somethingthat I could have used
or that I can't wait tosee out in the real world.
(19:32):
I've had three children,
and my second one, theypoked and prodded my arm.
It was black and blue,
they couldn't get an IV linethat my third one, I was like,
"I don't even want one,don't even put it near me."
So, it's really...
I can't wait to see some of this stuff
be more widely adopted.
And we're talking about likediagnostics, and imaging,
and other areas,
(19:53):
but like you said, there's so many options
and so many different placesAI and healthcare could go.
You guys mentioned a couple times
dictating notes for doctorsand things like that.
So Ian, I'm curious, froma research perspective,
where is this space going?
Do you have any future-looking ideas
(20:14):
on how AI usage is going tocontinue to evolve in healthcare?
- I think this could evolve in many areas.
I mean, that ultrasound example,
I think it's a fascinating example
because ultrasound'sa very cost-effective,
very accessible type of scanning.
And what you are doing with that
is you are making a tool
that's been around fordecades more effective,
and that that is augmentingan existing tool.
(20:34):
It's a fascinating example.
I think some things, we'reclearly going to see.
We're going to seecloud-based AI continue,
but we're going to see increasinguse of AI on the edge too
for that responsiveness piece.
The other I think we'll see
is we've seen these very large AI models,
we'll see more smaller-focusedmodels come to market
for particular task or use,
(20:55):
and they become more portable,
they become even easierto put onto edge devices.
And we've seen that in other fields
outside of healthcare too.
I think we'll see this multimodal element.
So, multimodal means audio-based,
video-based, still-image-based,and text-based.
And that means both a way ofinteracting with the model,
(21:17):
but also what the modelis able to understand
and perceive about the world.
So it might be able to perceive,
use a camera to identifyif there are queues forming
or people forming in certainparts of the hospital.
The transcription piece is interesting.
That means you are capturing information
that may otherwise neverhave been captured,
maybe a patient-doctor conversation.
(21:38):
And then you can summarizethat conversation,
so you can add thingsinto the medical record
that maybe aren't beingcaptured at the moment,
but also make itaccessible, and surfaceable,
and findable later.
I think there are otherthings beyond that.
AI is very good at correlating trends
across different data sets.
This could be used in apublic health context more.
(22:01):
AI models can't do causation.
So when you find those correlations,
you'll still need to go and push them
in front of a researcher, a clinician,
to validate that it's a real thing,
not just one of those random correlations,
but it will probably uncover
underlying causes for conditions,
new ways of approaching healthcare
(22:21):
that we haven't thought about before.
And then there's just so many uses.
You can look at this reallyright the way across,
everywhere that technology's being used
in a hospital facility, I think.
- Yeah, so many opportunities.
And in this conversation,
we stuck a little bit to the devices,
and the implementation,and the data aspect of it,
but I'm sure we can go offin many different directions.
(22:44):
We're talking about the AI models,
the size of the Amodels, what they can do,
but at risk of opening up a can of worms
because I'm sure we'veonly scratched the surface
and we can keep going on and on,
I'm going to end the conversation here.
But what I'd love to do isthrow it back to each of you
for any final thoughts
or key takeaways you want toleave our listeners with today
(23:07):
as they prepare for theAI evolution in healthcare
or what they can expect to see.
So, Ian, I'll start with you on this one.
- I mean, I think oneof the big things here
is we've seen a lot of hypehere around internet-based LLMs.
I would say don't be discouraged
by the quality of thosethings like ChatGPT,
or Gemini, or Claude.
I mean, when you start lookingat these medical AI models,
(23:30):
they're typically trainedon pre-validated data sets,
not the internet.
So the accuracy level is much greater.
I think additionally, we'veseen some things come through,
where you can use one AI model
to validate the outputof another AI model,
and that can raise thequality of the output, too.
So, that quality piece that you might see
when you are playing with stuff online
(23:51):
isn't applicable here.
This is a different kind of space.
And in some cases,
people are using in-house,open source-based models,
so they have greatercontrol ownership of it too.
So, don't be discouragedby what you might see
in other areas on your phone,
or on your computer, or on the internet.
This is a different space.
The quality here is much, much higher.
(24:14):
- Awesome, and Ian,
I'll make sure to providea link out to that report,
so that those listening,
they can learn and diginto some of the things
that we were talking about even further.
Alex, before we go, any final thoughts
or key takeaways you wantto leave our listeners with?
- Yeah, I think Ian mentioneda really good thing,
and that's the miniaturization of AI.
And essentially, we're goingto continue to see that pattern
(24:38):
and we're going to learn
what is the right AI at theright time at the right device.
And the other thing thatI wanted to also mention
is when you're doing AIat the edge on the device,
power becomes a really important feature,
because if you think about it,
it's kind of a snowball effect.
(24:58):
More power you need, the biggerfans you're going to need,
the bigger device, the newform factor, and so forth.
Oftentimes, you don't need that.
You can run the rightamount of AI at the edge
without needing to redesignor reconfigure your device.
There's new technology,
new compute that allows you to do that.
So, as we continue to evolve,
(25:20):
as more and more artificialintelligence goes to the edge,
it's going to be easier andeasier to run at the edge.
- And I think deploy AI at the edge also,
like you said, thesedevices and technology,
it's getting smaller.
It's amazing what you can
with the infrastructure you already have
and without a lot ofhardware, a lot of equipment.
(25:42):
So, I can't wait to seewhere else the space goes,
other innovations andtechnologies from Intel.
So, I just want to thank youboth again for joining us today
and for talking about this topic.
Thank you to our listeners alsofor coming in and listening.
Until next time, this hasbeen "insight.tech Talk."
(uplifting music)
Popular Podcasts
United States of Kennedy
United States of Kennedy is a podcast about our cultural fascination with the Kennedy dynasty. Every week, hosts Lyra Smith and George Civeris go into one aspect of the Kennedy story.
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Follow now to get the latest episodes of Dateline NBC completely free, or subscribe to Dateline Premium for ad-free listening and exclusive bonus content: DatelinePremium.com