Episode Transcript
Transcripts are displayed as originally observed. Some content, including advertisements may have changed.
Use Ctrl + F to search
0:36
Welcome to Who
0:36
would have thought my name is
0:38
Sacha Heppell, Chief Marketing
0:38
Officer of SmartTab. And I'm
0:42
hosting this podcast with Robert
0:42
Niichel our Founder and CEO.
0:46
Robert's experience and
0:46
leadership and management of
0:49
pharmaceutical research and
0:49
development led to the founding
0:52
of SmartTab in 2016 to combine
0:52
wireless technology with
0:56
pharmaceutical drug delivery.
0:56
Today we explore artificial
1:00
intelligence, otherwise known as
1:00
AI, and how it can empower
1:04
doctors. We will speak with an
1:04
extraordinary machine learning
1:07
specialist and AI mastermind
1:07
about the current and future
1:11
applications of artificial
1:11
intelligence for COVID-19 and
1:15
beyond. I'll pass it over to
1:15
you, Robert to introduce our
1:19
guests today, behind the
1:19
artificial intelligence
1:22
revolution, Artur Adib and Dr.
1:22
George Hauser.
1:26
Thank you, Sacha, I'd like to introduce Artur Adib, PhD Founder and CEO
1:28
of Biocogniv Artur has built the
1:32
flight controller software for
1:32
the world's largest electric
1:35
flying car and was a senior
1:35
engineer at Twitter, a former
1:40
faculty member at the National
1:40
Institutes of Health and leads
1:43
the vision and AI for Biocogniv.
1:43
We'd also like to welcome Dr.
1:48
George Hauser, Chief Medical
1:48
Scientist at Biocogniv. Dr.
1:52
Hauser is the author of over 20
1:52
peer reviewed publications, and
1:56
specializes in research
1:56
involving artificial
1:59
intelligence and electronic
1:59
health records. At Biocogniv,
2:03
their model for COVID-19
2:03
screening is the largest
2:06
application of artificial
2:06
intelligence to COVID yet, and
2:09
they're leading the push to
2:09
empower doctors with AI. Okay,
2:14
good morning. Hello, Artur and
2:14
George, welcome to the show.
2:17
Thank you for joining us today.
2:17
Artur, can you tell us your
2:20
story, your background in AI?
2:20
And what inspired you to pursue
2:25
Biocogniv as the Founder/CEO?
2:28
Yeah, absolutely. It's a great pleasure to be here, Robert, and
2:29
Sacha, thank you for the
2:32
opportunity. My background is, I
2:32
came to this country for
2:36
graduate school in computational
2:36
and statistical physics about 20
2:41
years ago, that's back when AI
2:41
was sort of a fringe topic was
2:45
mostly talked about in academia,
2:45
to my knowledge, at the time,
2:49
there was very little being done
2:49
outside of academia. Biocogniv
2:53
started out of my catching up
2:53
with the literature of the last
2:56
few years of the medical
2:56
literature, applying AI into
3:00
diagnostics in particular. I
3:00
have even framed here in our
3:05
office, some of the early papers
3:05
that that I read that really
3:08
left an impression. Those were
3:08
papers that applied a state of
3:12
the art machine learning to
3:12
image classification in the
3:15
specific case of pathology. And
3:15
I found that it was one of the
3:20
most powerful things that one
3:20
could, simply by presenting
3:24
examples of disease state
3:24
images, and healthy in that
3:29
specific case for skin
3:29
conditions, skin lesions, that
3:34
you could train a model simply
3:34
by showing examples and it
3:37
outperformed, in some cases,
3:37
dermatologists. And I felt like
3:42
that was such a milestone for AI
3:42
20 years later, that I wanted to
3:48
be part of it. And as we now
3:48
know, there's quite the
3:51
conversation about AI in this
3:51
space. And so that's sort of the
3:55
the gist of how that all started.
3:59
And then moving
3:59
over to George, could you share
4:01
with us about your background
4:01
and your story in machine
4:04
learning and and how the
4:04
platform at about Biocogniv, how
4:08
that works?
4:09
Yeah, I echo
4:09
what Arthur said, I'm happy to
4:12
be here. Thanks for putting this
4:12
together. I think it's great to
4:14
have people start discussions
4:14
around this and see where it
4:17
leads us. So my background is in
4:17
biomedical engineering. And when
4:23
I started undergrad, wasn't that
4:23
long ago, and there was only a
4:27
few biomedical engineering
4:27
programs in the country. And I
4:30
went to one at the University of
4:30
Michigan. And when I graduated,
4:34
I went into medicine. And I was
4:34
using a lot of just like
4:39
computer technology in general.
4:39
And it felt like there was a
4:43
real need to be able to
4:43
synthesize all this information
4:47
that was being and I still think
4:47
it is, poorly organized and just
4:51
just saw an opportunity there.
4:51
So graduated medical school, and
4:55
then I went to train as a
4:55
clinical pathologist, which is
4:59
subdomain in medicine that
4:59
focuses on laboratory testing,
5:03
and in that space, I've
5:03
continued my interest in AI. And
5:07
then so when when Artur came up
5:07
with his proposal, it just
5:10
seemed like a natural fit. So
5:10
you know, we're working together
5:13
in this space now.
5:17
And then two
5:17
questions. The first one is,
5:19
what exactly does your platform
5:19
around the AI? What exactly does
5:24
it provide for the physicians?
5:26
Yeah, so the
5:26
company started last year. And
5:30
one of the things I guess I
5:30
didn't mention is, after
5:34
graduate school, I spent several
5:34
years at the National Institutes
5:36
of Health as a faculty there,
5:36
and then spent some time in
5:39
Silicon Valley, I decided that
5:39
academia wasn't quite right for
5:43
me, and then spent some time
5:43
Silicon Valley at a company,
5:46
companies like Twitter, Magic
5:46
Leap, and more recently, an
5:50
aerospace company. But the way I
5:50
think about Biocogniv, is this
5:54
marriage of software
5:54
engineering, and academic
5:57
research, and when I started
5:57
working at Biocogniv last year,
6:01
obviously, there was no
6:01
Coronavirus, and the goal of the
6:05
company was to identify unmet
6:05
needs, in hospital systems in
6:09
particular, tertiary care, and
6:09
try to help address those unmet
6:14
needs with AI. So we had some
6:14
targets we started working on
6:19
before Coronavirus hit, but then
6:19
it happened, COVID happened. And
6:23
we saw how, early, how much of
6:23
an impact it was going to have
6:27
in hospital systems here and
6:27
whatnot. So I think this is a
6:31
testament to the power of AI,
6:31
combined with the availability
6:36
of electronic medical records.
6:36
So we pivoted our efforts
6:40
towards COVID. And we were able
6:40
to come up in probably less than
6:45
five months, we were able to go
6:45
from zero to a model, product,
6:52
that could rule out COVID with
6:52
extremely high sensitivity based
6:57
only on blood test data like
6:57
complete blood count and
7:00
comprehensive metabolic panel,
7:00
which are routine labs that are
7:03
performed at hospital systems.
7:03
Again, I think this is a
7:06
testament to the power of AI.
7:06
Typically when when one wants to
7:10
come up with a new biomarker,
7:10
traditionally, this is a multi
7:14
year process tens, maybe
7:14
hundreds of millions of dollars
7:17
to understand the
7:17
pathophysiology of the disease
7:21
and then identify the molecules
7:21
that take part in the process
7:25
and then isolate the molecule
7:25
and then come up with an assay
7:28
for it and so on. I think this
7:28
is a great example of how AI can
7:31
be really helpful, like George
7:31
was alluding to, there's
7:35
hospitals in the country, in
7:35
general, has a ton of data
7:38
already. And it's generally
7:38
poorly organized and poorly
7:42
leveraged. So we think that AI
7:42
can really unlock the power of
7:45
that data and in what we call
7:45
AI-COVID. That is our first
7:49
product and is a great example
7:49
of how that can happen. And
7:54
answering your question around
7:54
the utility and how it can be
7:59
used in clinical practice. The
7:59
main reason we did this was in
8:03
response to the ongoing shortage
8:03
of kits, of testing kits, across
8:08
the country. So FDA has approved
8:08
over 200 test kits. Despite that
8:14
there is an ongoing issue with
8:14
supply chains across the
8:18
country, it varies from week to
8:18
week, one week, a hospital
8:22
system will have plenty of
8:22
swabs, plenty of reagents, and
8:25
so on. But another week, they
8:25
will run out of one of those
8:27
components. And that will set
8:27
them back. Obviously, AI doesn't
8:32
have that issue in particular,
8:32
especially because AI in
8:37
particular lives on top of very,
8:37
very routine laboratory data
8:40
that you generally don't run out
8:40
of reagents for so that's
8:44
essentially it. That's the use
8:44
case is tertiary care hospital
8:47
systems, is to help them triage
8:47
really COVID patients when
8:51
they're struggling with their
8:51
PCR kit supplies.
8:56
Talking some
8:56
more about the COVID-19. Like
8:59
you said, you went from zero to
8:59
you have on your web page listed
9:03
over 60 US based hospitals, over
9:03
200,000 patients. That's a
9:07
significant, that's pretty heavy
9:07
lift. And so, maybe you could
9:11
talk about some, number one,
9:11
challenges through that. And
9:15
then, it looks like COVID-19 is
9:15
going to be around for a while
9:18
so how do you see your
9:18
technology being continued to be
9:21
scaled up? What are your plans
9:21
there to, you know, say move it
9:25
from 200,000 patients to a
9:25
million to two million etc?
9:28
Yeah, great
9:28
question. I mean, obviously,
9:31
there's a lot of papers, a lot
9:31
of AI work being done. To your
9:35
point, like we think that one of
9:35
the key things that sets us
9:38
apart from other studies is the
9:38
sheer size. The multicenter
9:43
nature of it, 66 hospitals. Some
9:43
of it was heavy lifting in terms
9:49
of getting IRB approvals with
9:49
hospital systems and whatnot,
9:53
but the vast majority was
9:53
through a vendor or a partner
9:56
and we believe that it's the
9:56
largest, perhaps not only for
10:01
COVID, but probably for
10:01
infectious diseases application
10:05
of AI to infectious diseases and
10:05
whatnot. In terms of
10:07
multicenter, we can spend some
10:07
time talking about the, you
10:12
know, as you probably know,
10:12
there's a lot of concern around
10:15
the blackbox nature of AI,
10:15
especially physicians and
10:19
clinicians having the ability to
10:19
understand what is going on,
10:23
before they accept a certain
10:23
result. A great, probably the
10:28
largest part of earning that
10:28
trust with the community, the
10:33
medical community is making sure
10:33
that the product generalizes
10:37
well to different populations in
10:37
different health systems, often,
10:40
you will see models and AI
10:40
models that have been trained on
10:43
one or two hospital systems that
10:43
are not for a variety of
10:47
reasons, not representative of
10:47
the wild, right. So the the
10:51
field in general, where it would
10:51
be applied. And so we believe
10:55
that we can mitigate some of
10:55
these concerns by making sure
10:58
that A we look at as many
10:58
hospital systems as we can but
11:03
also B do a deeper analysis of
11:03
how does the algorithm perform
11:09
in different population groups
11:09
by demographics, for example, is
11:13
one one way that we look at it?
11:17
And then let's
11:17
talk a little bit about, how you
11:20
expand. So you started in a
11:20
certain way, and then COVID-19
11:24
came along, and you're focusing
11:24
on that? And then do you see
11:27
this technology as you expand it
11:27
being more demographic focused
11:31
or on particular, right, so
11:31
COVID-19 is a particular disease
11:36
or virus, you focus on that? And
11:36
then is there like something
11:39
else, like and does this get
11:39
broken out per disease? Or how
11:44
do you see it expanding per
11:44
category?
11:48
Like you mentioned, we think that COVID is gonna be here for a while,
11:50
unfortunately. So that's keeping
11:53
us plenty busy. But that said,
11:53
we started the company with a
11:55
with a bigger mission. And we
11:55
continue to believe in that
11:58
mission. And that is that there
11:58
is a variety of conditions for
12:02
which AI can deliver value. I'll
12:02
give you one example that we're
12:07
particularly interested in, that
12:07
will be pulmonary embolism. If
12:10
you take something like
12:10
Troponin, that is the biomarker
12:15
of choice, obviously, for
12:15
myocardial infarction, virtually
12:18
every patient that presents with
12:18
chest pain in the hospital
12:20
system will get your Troponin
12:20
ordered. We try to be the
12:22
Troponin for pulmonary embolism.
12:22
Yes, there's a marker for ruling
12:27
out PE, D-dimer. But it doesn't
12:27
have the same characteristic
12:32
performance as Troponin. So it's
12:32
not as widely used, we believe
12:35
that we can move the needle on
12:35
that. And as a result of that,
12:39
we can avoid the over
12:39
utilization of CT scans, for
12:43
example, which is considered by
12:43
probably the majority of
12:47
emergency physicians, one of the
12:47
top issues that they're facing
12:51
is again, the over utilization
12:51
of imaging, which has been
12:55
actually reflected in how payers
12:55
see this. The largest payers in
12:59
the country are not paying for
12:59
outpatient imaging, CT imaging
13:03
and it's not getting reimbursed
13:03
because they feel like it's been
13:05
over utilized. So we think we
13:05
can help with that.
13:09
And then as you
13:09
expand the pulmonary market,
13:12
that's a huge market right
13:12
there, just that one. And then
13:15
do you go back to these same
13:15
60-66 hospitals you're in? Or do
13:19
you focus more on ones that are
13:19
more advanced with cardio or
13:22
emergency rooms? Or how did how
13:22
does that roll out?
13:25
That's a great
13:25
question, I think you're getting
13:27
to sort of the core of a
13:27
successful AI product. And that
13:32
is, it has to be representative
13:32
of the data that it will
13:36
actually see in clinical
13:36
practice. And one way to do
13:40
that, obviously, is to just try
13:40
to get as many hospital systems
13:44
as you can, not only in
13:44
different geographic regions,
13:48
but also in different resource
13:48
sort of circumstances, like
13:52
small community, rural
13:52
hospitals, versus large academic
13:56
centers, and so on, because
13:56
they, they'll have different
13:58
instruments, they will have
13:58
different practices and whatnot.
14:00
So you want the model to be
14:00
robust against that. We are
14:03
working with some collaborators
14:03
at different institutions, this
14:06
will probably provide them the
14:06
most depth in terms of the
14:10
quality of the data. But it's
14:10
impractical for a startup like
14:14
ourselves to go after, let's
14:14
say, 60, hospital systems one at
14:18
a time, getting one IRB at a
14:18
time, and so on. And so we are
14:22
lineal partners, and so on for
14:22
this.
14:26
The larger
14:26
strategic partners are always
14:29
helpful, especially if we get
14:29
the correct one. So I have one
14:32
last question here, and then I'll turn it over to Sacha. So really, where do you see your
14:34
technology as it rolls out, how
14:37
it gets built out? So number
14:37
one, your technology,
14:40
specifically in the next five
14:40
years, and then in general, more
14:43
just the entire AI market? What
14:43
does that look like in five
14:47
years from now?
14:48
Yeah, that's
14:48
that's a timely question. How do
14:50
you roll this out? So I was just
14:50
reading the news yesterday and
14:54
today. The country has been hit
14:54
with a massive cyber attack in
14:58
hospital systems in the country.
14:58
Over 400 are being hit with
15:01
cyberattack, including one here
15:01
in our backyard and the
15:04
University of Vermont. And we
15:04
are very, very aware of this, we
15:09
are going to be a software based
15:09
deployment in hospital systems.
15:13
And we have hired two of some of
15:13
the the top engineering talent
15:18
out of Silicon Valley, including
15:18
one who's an expert in security.
15:21
So I would say that one of the
15:21
key issues, it's not talked
15:25
about too much, in addition to
15:25
all the blackbox nature of AI
15:29
and all of this. One of the key
15:29
issues will be if you're
15:31
connecting to a hospital system,
15:31
how will you make sure that this
15:34
is a secure connection that not
15:34
only will protect patient data,
15:38
but it'll prevent hackers from
15:38
entering the hospital system
15:41
through your system. And so
15:41
we're very much on top of this,
15:44
it's one of the top concerns
15:44
that FDA has. So we've been
15:47
having almost bi weekly meetings
15:47
with FDA to get our product
15:51
through the finish line. And
15:51
this is top of mind for them.
15:53
And right now, like literally in
15:53
the last 24 hours this this
15:57
massive breach has happened, I
15:57
think it's gonna only going to
16:00
become more important that we
16:00
address this. So I would say
16:03
that answers a part of your
16:03
question like our deployment
16:06
strategy is safety and security
16:06
first, the second part is
16:10
projecting out into the future.
16:10
I'm extremely optimistic about
16:14
what's what's happening with AI
16:14
across the board. I know that
16:18
there's there's some well
16:18
received skepticism. And I think
16:21
that's the that's necessary,
16:21
it's healthy. It forces us to
16:25
think hard, long and hard about
16:25
how we can prove safety and
16:29
effectiveness, which is sort of
16:29
FDA per view for medical
16:32
devices. But 5 to 10 years from
16:32
now, I think it's easy to
16:36
imagine a world where radiology
16:36
for example, right now, you
16:40
don't even have to project that
16:40
out that far. Like we're already
16:42
starting to see unprecedented
16:42
reimbursement pathways, a
16:46
company like Viz AI, for
16:46
example, in the last month or
16:49
two has received reimbursement
16:49
as sort of a reimbursement
16:54
pathway from CMS that for their
16:54
pure AI software, which I think
16:59
is unprecedented. I think we're
16:59
going to start seeing a lot of
17:02
folks are saying that this has
17:02
opened the floodgates for AI
17:05
reimbursement, I think we're going to start seeing more and more of this, that's certainly a
17:07
path that we want to pursue, we
17:10
want to make sure, ultimately
17:10
what we feel is that AI has the
17:15
potential to become as safe as
17:15
effective as any other medical
17:21
device, we just have to show
17:21
that that's the case. And I
17:24
think that there's mounting
17:24
evidence now that that's
17:27
becoming the case. And so in 5
17:27
years from now, 10 years from
17:30
now, I think pathology is going
17:30
to be, and I'll let George speak
17:35
somewhat to this, as a
17:35
pathologist, he knows that
17:38
there's a lot of activity in
17:38
this space, radiology is another
17:41
one. And we think Laboratory
17:41
Medicine, this is sort of our
17:45
niche, folks talk about 70% of
17:45
medical decisions being made
17:50
through laboratory results. And
17:50
imagine how much you can, how
17:54
much farther you can go by
17:54
combining multiple laboratory
17:59
results, and analyzing them and
17:59
identifying patterns that
18:03
humanly were nearly impossible
18:03
to identify these patterns. And
18:07
so that's essentially what our
18:07
product for COVID has been able
18:10
to do, is to identify patterns
18:10
in routine laboratory tests that
18:14
doctors are used to seeing every
18:14
day. But by presenting large
18:18
numbers of these results, and
18:18
ground truths, as we call them
18:23
for who's sick, who's not, then
18:23
the models have been able to
18:27
pick up these patterns. So I think we're going to start seeing more and more of this,
18:29
and how that helps hospital
18:32
systems in a variety of ways. We
18:32
already talked about avoidance
18:36
of CT scans and whatnot, we
18:36
think there's also an
18:39
opportunity to to expedite
18:39
results. Cultures, for example,
18:43
are notoriously slow, they take
18:43
on the order of days, bacterial
18:48
cultures and whatnot to come
18:48
back. Potentially, we could
18:51
shorten that time to minutes or
18:51
hours and really change patient
18:56
outcomes there. Because with
18:56
that information, doctors can
19:00
intervene accordingly.
19:03
Well, that's,
19:03
that's great. Now, that's it for
19:05
my questions. And then Sacha, go
19:05
ahead.
19:09
Thanks, Bob. So
19:09
George, I'd like to ask you
19:12
about what do we have yet to
19:12
discover? You know, as you're a
19:16
machine learning specialist, I
19:16
just like to know, kind of like
19:20
what our Artur was just saying
19:20
about identifying those patterns
19:24
and all of that data coming
19:24
together. What do we have yet to
19:28
discover about the benefits of
19:28
that, of this technology?
19:32
Yeah, I think
19:32
you can take the technology in a
19:35
lot of different directions. I
19:35
think one of the avenues that
19:38
we're taking it in is sort of
19:38
this ability to find patterns
19:44
and which otherwise may be
19:44
interpreted as noise, in the
19:47
diagnostic realm. But I think in
19:47
general, you know, depending on
19:51
how you want to box in AI, it's
19:51
going to be used in a lot of
19:54
different ways, throughout
19:54
healthcare, in the diagnostic
19:58
realm, which we've been talking
19:58
about now, but in a lot of other
20:02
ways, including more mundane
20:02
things like scheduling,
20:07
workflow, and I think you're
20:07
going to start seeing it more
20:10
integrated outside of this sort
20:10
of Crystal Palace and more
20:15
democratized to things like
20:15
visits online, no sort of
20:20
virtual doctors. And then that's
20:20
a huge shift for medicine, which
20:24
has traditionally been a very
20:24
sort of brick and mortar type of
20:30
experience that that people have
20:30
had. And I think there's a lot
20:35
of change that's going to
20:35
happen, you know, it's sort of
20:37
been rippling through a lot of
20:37
different fields, you know, from
20:40
finance for quite a while and
20:40
moving into medicine more
20:45
recently. But I think I think
20:45
we're just standing on the tip
20:48
of the iceberg here, it's going
20:48
to affect essentially every
20:51
process within a healthcare
20:51
system, at some point.
20:55
Yeah. And
20:55
focusing on one of those areas
20:59
that actually we work on here at
20:59
SmartTab, we're developing drug
21:02
delivery systems that would work
21:02
inside of some of these AI
21:06
systems. How do you see that
21:06
kind of playing out? How would
21:09
this potentially integrate with
21:09
digital medicine?
21:12
I think the
21:12
system you're describing has a
21:15
lot of different sources of
21:15
information from like image
21:18
analysis to patient records,
21:18
diagnoses. And anytime you start
21:23
to have those sort of inputs,
21:23
you get a certain amount of
21:26
complexity. And that's what AI
21:26
is good for, is sorting through
21:31
that level of complexity. So you
21:31
know, one of the additional
21:35
streams that you could add to
21:35
that would be some of the
21:38
diagnostic products that we're
21:38
developing and integrating that
21:41
with, with what you're doing to
21:41
optimize the patient outcome.
21:46
That's, I think that's what
21:46
we're trying to do here is about
21:48
that, that data integration step
21:48
and how to do that efficiently.
21:53
When it becomes overwhelming. I
21:53
mean, you guys are probably
21:57
taking quite a few images, as
21:57
your device travels through the
22:02
body and being able to process
22:02
those, analyze them and optimize
22:06
it is a difficult task for a
22:06
person to do by themselves.
22:11
And so then looking at, you know, virtual care as it's become a true
22:13
delivery model. Now, how do you
22:17
see AI really supporting that in
22:17
the short term, but also down
22:22
the road? And like, how this
22:22
will evolve?
22:25
Yeah, sure, I
22:25
think it's probably going to
22:27
speed up the whole transition to
22:27
AI, you know, if you want to
22:31
think of AI is like a train, you
22:31
know, it has a certain amount of
22:34
momentum. And I think early on
22:34
the gatekeepers of that
22:40
momentum, we're not familiar
22:40
with the technology, it was sort
22:44
of foreign to them. And so it
22:44
kept the pace of the train
22:48
moving pretty slow. And as the
22:48
the new leadership comes up, and
22:53
they've grown up with these
22:53
devices and and use them, you
22:57
know, basically their whole
22:57
lives, I think the speed of the
23:01
train is starting to pick up.
23:01
And as we transition more to
23:05
virtual care, what that's doing
23:05
is taking away the elements that
23:10
may have been traditionally
23:10
difficult to capture in sort of
23:13
structured data, things like the
23:13
physical exam, for instance. And
23:18
then that dialogue that you have
23:18
with the patient. So what the
23:22
virtual care is likely doing, if
23:22
you're looking at the big
23:26
picture of AI, it's creating
23:26
more data. And as you create
23:31
more data, and especially in a
23:31
in a way that can be analyzed,
23:35
then that's going to be able to
23:35
feed back and train the
23:40
algorithms and allow them to be
23:40
improved. And just keep the
23:45
train moving. And I think it's
23:45
starting to pick up the momentum
23:48
from that. And it'll likely keep
23:48
doing so for for a number of
23:52
factors, from just keeping costs
23:52
in line to outperforming current
23:58
standards of care. You know,
23:58
healthcare is a very human
24:02
intensive field. We have a lot
24:02
of support staff, nurses,
24:07
technologists, technicians who
24:07
have specialty training and in
24:11
each of those fields, and you
24:11
know, AI certainly has a lot of
24:15
different niches that it can
24:15
fill in any number of these
24:19
applications.
24:21
What is one that
24:21
you see is like the most
24:23
important that would really
24:23
that's really going to empower
24:27
the doctors and in the work that
24:27
they're doing. Which area would
24:31
you say is like the most
24:31
important?
24:34
I've always
24:34
tried to try to stress the
24:37
integration of computers with
24:37
with healthcare providers rather
24:41
than use words that suggest a
24:41
replacement. Because I feel like
24:48
AI does dumb things well, and
24:48
people do some things well, so a
24:52
computer might not be able to
24:52
solicit information from a
24:55
person who is a little reticent
24:55
to give that information. But
25:01
like I said, computers much
25:01
better integrating the large
25:04
amount of information. So I
25:04
think they have to work
25:07
together. And I think you've
25:07
seen some instances where, like
25:12
with the Watson, I think one of
25:12
the ways that they sort of
25:15
approached it the wrong way was
25:15
they try to pitch it as sort of
25:19
like a replacement for the for
25:19
the doctor. And I think it
25:22
didn't, it didn't work out so
25:22
well in that sort of pitch. And
25:26
remember watching the Watson
25:26
performing the Jeopardy against
25:30
Ken Jennings, I think you can
25:30
still find that on YouTube. And
25:34
the clip that stands out to me
25:34
was this point in time where,
25:38
the question was asked, and then
25:38
somebody rang in and they gave
25:41
an answer. And then it was
25:41
wrong. And Watson rang in right
25:47
after, and gave the wrong
25:47
answer, the same wrong answer
25:50
that the person before it had
25:50
done. And so it seemed like they
25:53
hadn't trained Watson to listen
25:53
to what the other person was
25:57
saying it was more sort of like
25:57
a no give its answer
26:00
independently. So you know, the
26:00
AI can fail spectacularly. And I
26:04
think it's important to keep in
26:04
mind that we're all working
26:08
together, and everyone has sort
26:08
of their advantages, and just
26:13
sort of go about it with that
26:13
general mindset.
26:16
Yeah. And as you
26:16
develop your system, are you how
26:21
do you keep that mindset alive?
26:21
While you're innovating your
26:25
system? Like, what what are some
26:25
ways that you make sure that
26:28
you're really having it be
26:28
physician-centered, or
26:32
patient-centered, that it's
26:32
going to actually really make a
26:35
difference, and not just be a
26:35
great piece of technology?
26:38
yeah, that's the goal, that's the holy grail there is to integrate it in with
26:40
what you with, with the
26:44
healthcare environment. You
26:44
don't want it to be this sort of
26:47
like warp that people don't want
26:47
to touch. It needs to be
26:51
integrated well, and you do that
26:51
with getting feedback from the
26:54
people who are going to use it,
26:54
and making sure that they
26:58
believe in it. And when they
26:58
give you honest feedback, that
27:01
it doesn't work, you try to
27:01
improve it. I think one of the
27:05
things that we really strive to
27:05
do is to actually change
27:09
outcomes. And it's very hard to
27:09
change clinical outcomes,
27:15
whether it be extending life, or
27:15
whether it be improving quality
27:19
of life, or whether it be, you
27:19
know, showing a clear monetary
27:24
savings. Those are all very hard
27:24
outcomes to achieve. And I think
27:30
working towards, you know,
27:30
proving that an AI system can do
27:34
that, in a real world setting.
27:34
That's what needs to be
27:39
demonstrated and demonstrated
27:39
repeatedly and in many different
27:43
areas, to keep that train moving
27:43
as fast as it can.
27:47
Yeah, I would
27:47
echo that, I think that there's
27:49
a lot of, there has been an
27:49
attempt to bring AI and I think
27:54
the Watson might be an example
27:54
of that, where it's, and maybe
27:58
Google, you know, it's a catch
27:58
all type of thing, like you're
28:01
bringing AI, and that's going to
28:01
be this massively disruptive
28:05
thing that is gonna, you know,
28:05
do XYZ all at once. And so one,
28:10
we feel like, it's, we're gonna
28:10
have to win this one battle at a
28:13
time, like, like George was
28:13
saying, like, boy, it's so hard
28:16
to prove that he actually moved
28:16
the needle on outcomes. So I
28:21
like to think of us as more of a
28:21
diagnostic company first, that
28:25
happens to be using AI. And so
28:25
that keeps us more grounded in
28:29
terms of the deliverable to our
28:29
customer. It's not just a fancy
28:35
gadget, it's something that will
28:35
actually improve the quality of
28:39
your diagnosis. But you can't do
28:39
that. It's incredibly hard to do
28:43
this. If you try to do
28:43
everything at once you got to
28:47
you got to carve out every
28:47
little unmet need at a time. And
28:53
boy, does it take effort to
28:53
prove that you can actually do
28:56
something in that little space.
28:56
Right. But I think that
28:59
ultimately, that's that's going
28:59
to be how the the puzzle is
29:03
going to be solved. It's like
29:03
one piece at a time and not like
29:07
everything at once, if that makes sense.
29:09
Yeah, totally,
29:09
totally makes sense. If you put
29:12
yourself in the shoes of a
29:12
doctor, maybe there's an example
29:14
that you could paint a picture?
29:14
What would that look like, as
29:18
they're doing their work?
29:19
I can speak to for example, the the COVID product that we have today. It's
29:19
going to be seamlessly integrated with the Electronic
29:21
Health Record system and we
29:23
think that all of our products are going to be like that. So it's not going to be the case
29:24
where a physician is going to have to interrupt their workflow
29:25
and to go learn like an app or
29:37
whatnot. And they're going to be
29:37
able to place an order just like
29:39
they place an order for
29:39
laboratory tests today. And
29:43
rather than that taking however
29:43
long that would normally take,
29:47
you know, we're seeing some some
29:47
hospital systems, taking days to
29:51
get a result back because they
29:51
have to do a send out and so on.
29:54
They're going to be able to just
29:54
do a routine blood test and get
29:56
the result in under one hour. So
29:56
from their from their
29:59
perspective. If it will work
29:59
just like any other lab, except
30:03
that it will be a faster
30:03
turnaround, and potentially even
30:07
more accurate than, than the
30:07
existing available tests.
30:11
And so what's
30:11
next for Biocogniv? What do you
30:14
sees the next steps for you and
30:14
what your focus is, like you
30:20
say, focusing one step at a
30:20
time, what's the next steps for
30:23
you?
30:24
Yeah, we're
30:24
laser focused on getting this
30:28
product through the FDA. We
30:28
want to really help with the
30:34
current pandemic, like we talked
30:34
about, there's, this is not
30:37
going to go away anytime soon.
30:37
And so we want to make sure that
30:40
we can help and move the needle
30:40
on testing capabilities across
30:44
the country. And the first step
30:44
towards that is really to just
30:48
prove the safety and
30:48
effectiveness of the product
30:51
with FDA. Then obviously,
30:51
pursuing reimbursement
30:54
strategies and so on. But then
30:54
we do have a pipeline, and we
30:59
like to think of us as more of
30:59
like a machinery that produces
31:02
these models, reproducible
31:02
machinery. And so we're building
31:07
a system internally that will
31:07
enable us to more efficiently
31:12
produce these models, whether
31:12
they're, you know, for
31:15
infectious diseases, or for
31:15
cardiovascular conditions, like
31:19
we talked about shouldn't
31:19
matter, but we want to build a
31:21
platform that can actually
31:21
produce them. And so we have
31:26
sort of the alpha version of
31:26
that, that we've used to, to
31:30
build AI-COVID. But ultimately,
31:30
we're building this bigger
31:34
platform for for other
31:34
conditions. But like I said, I
31:37
think it's really important to
31:37
stay focused, one piece of the
31:40
puzzle, at a time. I think COVID
31:40
is gonna keep like 90% of the
31:45
company busy for the next year
31:45
or so. And then the remaining
31:48
10% is going to be like one step
31:48
forward thinking about this
31:52
platform and in the new
31:52
indications.
31:56
Thank you for
31:56
taking that on and your
31:59
commitment to really focusing on
31:59
COVID-19. And, and making a
32:04
difference for transforming
32:04
healthcare and your commitment
32:08
to advancing this technology for
32:08
that. And we really, we admire
32:11
that and and your team for the
32:11
work you do in innovation and
32:16
sticking to it and figuring it
32:16
out. You know, this is so we
32:20
really appreciate that. Is there
32:20
anything else that you'd like to
32:26
share with us?
32:27
No, I would
32:27
just echo what you're saying, we
32:30
just went through Y Combinator
32:30
over the summer, which is
32:33
probably the number one startup
32:33
incubator in the country, and I
32:36
got to meet phenomenal
32:36
entrepreneurs there. I would
32:39
just say that, it goes right
32:39
back to you guys. Innovation is
32:43
so incredibly hard. And I have
32:43
great admiration for people who
32:48
go out of their way to pursue
32:48
innovation under very uncertain
32:53
circumstances and so on. And so
32:53
it's a rough road, but we think
32:58
it's a it's a worthy road. And
32:58
kudos to you guys. I mean, the
33:02
the product you guys are
33:02
envisioning sounds out of like a
33:05
science fiction movie. So very
33:05
much appreciate your work as
33:09
well. So thanks for having us.
33:11
Yeah, thank you for coming on today.
33:12
And how can
33:12
potential partners contact you?
33:15
Yeah,
33:15
absolutely. So probably the best
33:17
way is to just email me directly
33:17
for now, at my email address:
33:22
33:32
Okay, awesome.
33:32
Thanks so much for spending the
33:34
time with us. We look forward to
33:34
staying up to date on your
33:36
progress. I appreciate your
33:36
time.
Podchaser is the ultimate destination for podcast data, search, and discovery. Learn More