Can we design a healthcare system for all? | Microsoft Azure and NVIDIA | Catalyst E5

0:03 It's fine for us to fly in when you've fallen off

0:05 the cliff and pick you up and take you into town.

0:08 But what's better is to put a fence up

0:10 the top so you don't fall in the first place.

0:13 That’s good primary health care and preventative health care.

0:19 Healthcare challenges are all about getting the right care to the right

0:23 individual at the right time Early detection in most cases,

0:27 is the difference between life and death.

0:30 Everyone on the planet, irrespective of where they live,

0:32 can participate in healthcare equally, affordably, and regularly.

0:36 And both Microsoft and Nvidia provide critical technical capabilities

0:39 to realize that at the sort of scale we're talking about.

0:45 We have a moment in time today,

0:47 and we have a technology that, if deployed responsibly and executed right,

0:54 has the potential to reset the trajectory of human health.

1:23 The tyranny of distance.

1:25 Well, that something we live with every day in our service.

1:31 The distances between healthcare services and patients

1:33 in this country can be up to 2000km.

1:37 So how do you bridge that gap?

1:40 Here at Flying Doctors we caring for people in rural and remote communities

1:44 where there aren't necessarily healthcare services

1:46 or the healthcare services are quite restricted.

1:50 An example of what we're doing at the moment is

1:52 we're putting free access to telehealth services in small communities.

1:56 What we haven't really seen is

1:58 widespread coordinated utilizations of these things

2:01 in the medical sense to not only allow for screening,

2:06 but to incorporate into a healthcare service

2:10 delivery that allows monitoring and management in home.

2:16 Australia's healthcare challenges, are driven in part by its geography.

2:21 If you think about that geography and spread and if you just

2:24 think about the fact that healthcare is delivered as a physical infrastructure,

2:27 somebody needs to go somewhere to do something, to see a doctor, to get a check.

2:31 It's impossible for the current healthcare system to cover all of us

2:35 in a way that provides the sort of access that we need to healthcare.

2:53 The further you get from regional towns, the less and less the basic screening

2:58 type services and maintenance type services are available.

3:02 So the rates of unplanned hospitalizations

3:05 and preventable hospitalizations in rural and remote parts

3:09 of this country are about three times higher than the rates in the city.

3:14 We need to change our idea of what healthcare is,

3:16 to provide access to the broadest possible population.

3:20 Is there another way we can do healthcare?

3:22 Is there a way that we can make health far more convenient,

3:25 certainly cheaper, and more accessible, you know, for everybody?

3:29 Could we commute all the health to your phone?

3:37 Helfie is a preventative health system for early detection and prediction.

3:44 We pair everyone with a personal health AI that is their constant

3:50 companion sitting on their mobile

3:52 devices and continuously available and learning.

3:56 Today, Helfie checks for over 20

3:59 plus health conditions through multimodal health checks.

4:02 So Helfie essentially is your human health platform that is in your pocket.

4:08 The ability just to look at your phone and take a selfie,

4:11 which is where the name Helfie originates from, and check

4:14 all your vital indicators in a couple seconds.

4:16 Recording your voice or coughing into your phone

4:19 to check on respiratory conditions like COPD or croup,

4:22 or bronchitis, or Covid, or tuberculosis.

4:25 For your personal context to be considered in your health,

4:27 is the capability that, we would love for every human to be able to access,

4:31 because we know that when people can get access to services

4:35 like that, it makes a profound difference to their health.

4:42 For me, Helfie has a unique proposition.

4:46 We’re not trying to treat your problem.

4:48 We're trying to find your problem before it's a problem.

4:52 We're trying to use all of that information you are gathering every day,

4:56 with every device we have on us,

4:58 and turning that into useful predictive information.

5:04 I love mountain biking.

5:05 I do it on regular and competitive level.

5:09 But 18 months ago, I had a widowmaker heart attack.

5:15 It was totally unexpected.

5:19 That was a real eye opener for me because I was fit.

5:23 I had a low resting heart rate.

5:24 I had low blood pressure.

5:26 I know from my performance stats on the six months

5:30 leading up to my heart attack that the signs were there,

5:33 but they just weren’t read.

5:36 Data’s not active in healthcare in the way it could and should be.

5:40 Data’s siloed, data is hard to get, it doesn't sort of flow.

5:44 It's one of the reasons I think that we have a systemic decline

5:48 in life expectancy as well as that we can't put data that we have available,

5:51 to active use in our healthcare.

5:54 And just to give you a sense of the magnitude of that data,

5:56 30 or 40% of all the data generated on the internet today is human health data.

6:01 Yeah.

6:01 So not selfies, not entertainment content,

6:04 not all the sort of stuff that the internet

6:06 is active with, but it's human health data,

6:08 and that’s growing exponentially with sensors and wearables.

6:11 So we've got enough data to solve enormous problems in healthcare.

6:29 When we think of the data that we capture from a user,

6:32 that comes from different modalities.

6:34 It comes from the data the user gives Helfie.

6:38 It comes from the data that are

6:40 coming from image analysis or your document analyzes.

6:44 It also comes from many different angles like the tone of your voice,

6:48 the things that you have been journaling in Helfie.

6:52 We have the full profile of the user that we manage,

6:55 which has your medical history and where you're coming from, your culture,

6:59 your ethnic background.

7:01 And it's used to give like an insightful output for the user.

7:05 Now imagine this for the entire world.

7:08 That's a lot of computation power that we need.

7:15 Leveraging Microsoft and Nvidia technology helps us build smarter models,

7:19 more scalable models, better inference pathways.

7:26 Helfie is currently using Azure Machine

7:28 Learning as its primary production inference mechanism.

7:32 The same infrastructure supports high throughput,

7:34 low latency, serving of large multimodal models.

7:38 Nvidia H100’s on Azure’s higher compute density

7:42 and memory bandwidth significantly reduce our training

7:45 time cutting what was taking multiples of hours down to approximately an hour.

7:50 Which means we could train full models in under 24 hours rather than 48 to 72.

7:56 That really, really helps us with our ability to iterate on ideas.

8:04 One of the advantages for us being on the Azure

8:06 stack is that we have a global presence.

8:09 We can put the compute where we need it.

8:11 We’re not talking to an LLM with general knowledge.

8:14 We're talking to, and creating inference around, your specific data.

8:18 and what matters to you.

8:20 And the next thing you need to do, and that needs to be fluid.

8:24 And the only way that that's going to be fluid is to be able to run it globally,

8:28 get information as close to the user as possible, and act on that information.

8:38 We're living in a world where we hear about cyber threats every day.

8:42 And healthcare is one of the biggest targets

8:45 for acquiring and then extorting companies around patients' personal data.

8:52 Humanity right now is rightly concerned about who has access

8:57 to their data and how it can be used against them.

9:01 Examples with healthcare would be say, well,

9:04 my insurance premium might go up if you know that you know,

9:08 my whole family had heart attacks when they were 40.

9:10 Or, I might not get insurance because you know that.

9:14 And that's a very, very slippery and dangerous slope for us to go down.

9:18 And trust in healthcare is everything.

9:21 If you don't have that, people won't engage.

9:24 And if people don't engage, then you've failed.

9:27 You can have the best model of care and the best tech,

9:29 if people don't trust it, they won't use it.

9:33 We take every possible opportunity to secure health data.

9:38 And we do that by using enterprise grade infrastructure,

9:42 on the Microsoft Azure stack.

9:45 This is infrastructure that is utilized

9:47 by governments and regulators and healthcare systems globally.

9:52 We have an entire AI OPs team,

9:55 that puts AI guardrails around how the AI utilizes data

9:58 and what data the AI is allowed to get access to.

10:02 And all of this is based on our relationship

10:06 with this incredible infrastructure that we are on, being Azure,

10:10 whereby we are able to deploy data privacy by design on that stack.

10:20 We only capture what we need to capture in terms of data use.

10:23 So we will only save the data that is

10:26 absolutely critical to provide the hyper personalization that we promise.

10:32 So this data belongs to the end user.

10:35 It is never marketed, It's never harvested.

10:38 Its always working for the end user's health benefits.

10:56 I've always had a passion for trying to understand

10:59 how infections in a population can be controlled.

11:02 So I began as an infectious disease physician.

11:04 I did a PhD in epidemiology.

11:07 I worked in vaccination and healthcare, and for the last 25 years,

11:11 I've been the director of Melbourne Sexual Health.

11:14 So we had wondered for a while whether or not you could use key questions

11:19 from individuals and show them images and ask

11:23 them whether their condition look like this or not,

11:25 and provide an accurate answer, then without any input from a doctor.

11:30 Having done the academic work,

11:31 we went looking for someone who could commercialize this and had

11:35 a vision for what this sort of AI healthcare might look like.

11:40 And Helfie was the organization that found us,

11:44 and we found them at the same sort of time.

11:49 So the way the Helfie app works, is an individual goes on to the app

11:54 concerned they have a sexually transmitted infection, they've got symptoms.

11:57 And they will then answer the same questions that a sexual

12:01 health physician like myself would ask them in the clinic.

12:05 They then will look at different images.

12:07 and they'll tell us whether it's exactly like that or not like that.

12:10 And then they might upload an image that they've got,

12:13 and the AI program will analyze that as well.

12:17 As part of their experience on the app,

12:19 they essentially get a clinical consultation at the end

12:23 of it without having to take their clothes off,

12:26 without having to tell a doctor intimate details about their life.

12:33 If healthcare is more accessible, more people get treated.

12:35 The reason it's more accessible is firstly, you can do it online.

12:41 The second thing it does, is it makes it less expensive because the use

12:47 of AI in healthcare will shorten the duration of a consultation.

12:53 Healthcare is a significant cost to any country Most countries in the Western

12:58 world spend between 8 and up to 17% of their GDP on health care.

13:03 And that is clearly, not enough.

13:07 A lot of people, do without.

13:08 There's a lot of costs in providing,

13:11 infrastructure, personnel, R&D and then therapies.

13:18 The ability to fund that for everyone, everywhere,

13:20 all the time is limited by any country.

13:25 If I was to give you an example on skin cancer rates here in Australia,

13:30 Two out of three Australians,

13:32 will have a skin cancer diagnosis in their lifetime.

13:36 And unfortunately, we still see around 2000 deaths a year.

13:41 But we know that with early prevention and early detection,

13:44 the survival rate is over 95%.

13:50 Imagine this.

13:51 What if you could screen every adult Australian,

13:55 20 million plus people, from their sofas,

13:58 in their homes, for under 25 cents, at a cost of 5 million dollars.

14:05 And through that program, you're able to get instant feedback on some

14:10 of the earliest signs of this horrendous disease.

14:15 Now, with the Australian government spending

14:18 almost $2 billion on this condition alone,

14:21 you'd think a $5 million screening program that you

14:26 might even run every quarter if you wanted,

14:29 would be a really great use of money.

14:33 Not only would it save lives,

14:35 but also it would have an incredible economic impact.

14:39 Think about broader conditions cardiovascular disease, hypertension, stroke,

14:44 other types of cancers, and the equation just simply multiplies.

14:53 So the benefits of treating people with AI are huge,

14:57 beyond the fact that you've just stopped a doctor having to go to them,

15:00 or them having to go to a doctor.

15:04 I think that's where we are heading.

15:07 Healthcare at the time and place,

15:10 where the patient is, at their need and discretion,

15:13 not built around the old structures of 9

15:17 to 5 outpatient services and hospitals in certain locations.

15:22 And that's the challenge.

15:23 And that's a really exciting challenge.

15:28 Too often, access to healthcare is governed by where you live.

15:33 People deserve the same level of healthcare wherever they are.

15:40 Our mission at Helfie is to reset the trajectory

15:43 of human health for all 8 billion humans on this planet,

15:47 by building the world's first human health network.

15:52 By building a system that can be truly accessed, by all humans,

15:58 The hope is that, we make it possible

15:59 for all of us not just to participate in healthcare,

16:02 but have health outcomes, and an impact on our health that’s early and positive,

16:06 and makes a difference to the way we all live.

16:09 This is the first time we've been able to do something like this.

16:12 This is the first time that humans and humanity's been

16:15 able to to design a system for all of us.

16:23 In our world, almost everyone is a Helfie user.

16:28 We have a saying, most of the things you're doing in day to day life,

16:33 you're halfway to Helfie.