Press "Enter" to skip to content

My Ride to the Cloud | Ernest Garcia | TEDxEmory


thank you are you ready for some magic
some technology magic that’s what I’m
here for
we’re gonna talk about my story of how
I’ve used technology and a bit of magic
to realize the vision in my career this
is my right to the cloud so the right is
really more like surfing like writing
very large technology waves in my case
to developed medical applications
software you too can write these waves
taking you very fast but also they’re
very the quality variety of technology
that you can use even today to fulfill
your overcome any challenges
particularly technologies challenges in
your career so if you already get on the
board and see what the wave takes us so
my story starts back in 1969 when I was
fortunate to get a job with a power
company where they actually taught me
computers and programming back in 69
this were not subjects that were being
taught in even in the top universities
back then computers like this were
called mainframe there were over a
million dollars each and you be amazed
to understand that the smartphone’s that
you have with you today
have faster processing and more storage
than this mainframe computer so we’ve
come a long way in 30 years and as part
of the theme they’re only how fast
technology moves but what you can do
with it by 1971 many computers came on
the scene and the smaller now you can
fit them in a closet and less expensive
a hundred thousand rather than a million
dollars but
still incredibly limited again consider
that the entire memory of these devices
both in to put applications and data is
less or was less than five megabytes
that is about the attachment of an email
today so it was challenging to program
them but because I’ve had the training I
was able to do that and it actually is a
transition for me because they were
being introduced in the medical imaging
field so I went into that and
specifically in nuclear cardiology now
as I started working with physicians I
noticed how proud aware of their art of
image interpretation to reach a
diagnosis yet I recognize that I will
give the experts the same images to
different experts would often come with
very different conclusions and that one
observation became the vision for my
career and that is to use technology to
transform the art of image
interpretation to the science of image
interpretation and everything else I’ll
tell you will be related to that so also
consider that in 1983 no one was paying
for software so you had $100,000 system
for the hardware they would throw in
whatever software was just necessary but
because no one wanted to pay for the
software it was limiting our clinical
medical applications so you can see in
this tutorial that I wrote back in 83
and appealed to the community to start
paying for software but now you have the
medical experts that are still reluctant
to change from art to science and you
have the fact that they were already
paying $100,000 for devices now we’re
asking them to pay for something else so
you can imagine this was part of the
challenges the challenge is not only
technologic
but it’s also financial so the only way
or the best way to convince physicians
to use the technology is to prove to
them how it helps their patients so let
me go a little more specifically into my
field of nucleic re ology so that you
understand the rest of the presentation
in in what we do in our field we inject
radioactive chemicals that go
specifically to the muscle of the
myocardium the muscle of the heart which
pumps the blood to the rest of the body
and in this in the top panel what I’m
showing you there is cutting through
slices that look like donors for the
cavity that pumps the blood and you can
see that the colors are white or yellow
consistent with high blood flow so this
is a normal patient and when you stack
the slices from tip of the heart to the
base and we plotted them in a polar map
this became our representation of what
the heart looks like a quantitative
resent ation that could acts
artistically compare as I’ll show you in
a minute
now why such a simple thing for such a
difficult subject because in 1983
computers were very slow and the
software didn’t really exist for example
for three-dimensional rendering so
consider this other patient where again
we’re looking at the polar maps the top
left is during exercise and you can see
thou no longer do the spoiler maps look
uniformly yellowed but around 12 o’clock
it looks like it’s blue which is
reducing blood flow and at that same
patient address improves so first of all
the physician has an easy representation
of the of the heart but also we could
compare these statistically to databases
that we had in our lab and we can hide
we could highlight in black you see in
the bottom left panel
the regions that were significantly
reducing blood flow during exercise and
in white we will highlight the regions
that’s significantly improved during
rest now again it appears to be quite a
simple concept it doesn’t seem to be
much magic so far but if I can just skip
to today just for a second
realize that this program there’s over
35,000 licenses of it all over the world
and that over 4 million patients are
analyzed every year with this software
so of course it’s a little more
complicated that when I showed you but
in fact this ideas can help a lot of
patients in a in a short time now by the
year 2000 computers have gotten faster
there was a lot more software so what
you’re seeing here is where we fuse our
technology with other imaging modalities
and now we can render in three
dimensions and of course it looks a lot
more realistic than than the polar maps
that we show you before even though the
physicians believe it or not they still
prefer the polar maps because they were
trained that way
now during those years that’s when the
personal computers came into the scene
now
think about it we went from a million
dollar computer to a hundred thousand
dollar computer to a thousand dollar
computer and this comes out and our
computer is tremendously more powerful
than either of the other two with a lot
more software storage and and
performance this is why one of the
things that we did spin off a company so
that we could take advantage of this and
offer it to end-users for a lot less
cost now it was also about this time
that we started experimenting with
artificial intelligence as you know this
is the technology where the computer
tries to emulate how humans make
decisions so in this approach that you
see
here is is called neural networks where
we try to do here is cluster simulate
how we can cluster neurons in the brain
to make a decision and the way we train
the computer to learn how to make these
interpretations was to offer it as input
or imaging data in terms of the blood
flow to the heart and clinical data and
the output would be the known Danone
interpretation or the non diagnosis of
the patient and you will offer many many
thousands of studies for the computer to
learn how to do it now the technique we
ended up using is the one and that is
using certainty in decision support
think of certainty as a probability if
you have a lot of data just like before
but if you don’t have a lot of data it
can be just an expert opinion so these
systems are built up in terms of rules
from experts and as patient information
comes in for a specific patient it
either is used as positive or negative
evidence that that the patient if it’s
positive evidences that the patient is
abnormal if it’s negative evidence is
that the patient is normal and as all
the input and all the evidence comes in
wherever you end up at that certainty
becomes the final diagnosis appreciate
that this is all using the English
language and both the rules are written
in English and the justification the
tracking of how we reach that conclusion
is also written in English which brings
us to the next huge technology wave that
all of you are experiencing and that the
area of natural language processing or
computerized language understanding and
because not only am i communicating with
you in natural languages but us
communicate with computers in natural
language either from the different
search engines to search something that
we want or
the virtual assistant to get something
that that we need the air this area
imagined in the last 10 years the
progress has been phenomenal now how can
we use this how can we combine this with
what with previous did with the
artificial intelligence systems to
generate clinical reports how can we put
them together and this is where the
magic comes in and the magic is we call
Merlin and and why did we did Merlin
appreciate that once we send the systems
out the system stops learning so this is
just like a physician that graduated and
never heard another article never went
to a meeting so it doesn’t stops doesn’t
know on any of the new things
so in Merlin Berlin we program that so
that it would read clinical research
articles in our domain it would extract
the knowledge any new knowledge that if
thought I found it would then test that
knowledge against large databases in our
systems and if the knowledge is the new
knowledge increase the accuracy of the
diagnosis it incorporated our knowledge
into the into this expert systems so
it’s quite a bit of progress in terms of
using natural language processing in our
field now to implement so think about it
Merlin could do this every night right
by the time you wake up in the morning
you have new data you have better
diagnosis you have new rules in red all
the scientific articles for you to
implement programs like Merlin and to
continue the progress in our particular
clinical application another huge
technology wave has been the increasing
the speed of the Internet and appreciate
that the scale here is an exponential
scale so in the last 20 years the speed
of the Internet
it’s been improving at the rate of 50%
per year again because all of us are
using it and it makes the right business
model so why don’t s facilitate it
facilitates cloud computing and this is
my my right to the cloud this is one of
the clouds that I was talking about and
as you know you use the cloud and you
think of it as a storage place in that
you can don’t only you can store but
retrieve movies and photos and songs and
things like that we do the same thing
but we also store we upload into the
cloud all of the patient imaging data
but for us the cloud is more than just a
place for storage it is a place most
clouds usually have hundreds of
computers and servers attached to it
which means that this little laptop in
the bottom right can access the all
those computers and and can have
billions of operations per second so now
we’ve been able to also load this
program in the cloud and have it process
read the data that’s already in the
cloud the program in the cloud and when
you’re in the laptop it appears to you
as everything is happening locally but
it’s actually happening remotely now
this strategy also soluz to use a huge
amount of imaging data because the the
institutions that have this service are
loading up all their images to the cloud
so now we have hundreds of thousands of
studies that we can use to train our
official intelligence system and even to
look at trends in the field so this
brings us to the fourth version of our
program which we call the amore cardiac
toolbox and which is cloud compatible
and as I mentioned it has the the
functions of artificial intelligence for
reporting has cloud storage retrieval
and it can be processing and this can be
done anywhere in the world so in fact a
physician can’t read the studies
in a different country or so they were
sitting at their office now there’s a
part of this story that I’ve skipped and
it’s an important part because success
or failure comes at a cost and for me
that cost was that in 2008 I had a chest
pain event which was initially
considered to being significant but when
I was submitted to our modality and
process with our program
you can see in the bottom right in
yellow it shows that 26 percent of the
blood flow to my heart had been reduced
at stress and address had improved
significantly but only 7 percent her
return meaning that I lost about 7
percent of the muscle of my heart to a
heart attack this could have been a
one-way ticket to the cloud also but but
fortunately my my function was preserved
the pumping function of the ventricle
that 7 percent did not influence it
enough to bother it so this shows a good
prognosis and when the rft artificial
intelligence report comes out totally
generated by the computer I won’t be
reading everything that he says but the
final thing is the impression that
abnormal high risk study yet I was
reassured because of the fact that I
knew how the program work how we’d reach
an objective conclusion that I really
had the disease and I understood the
treatment that I needed this brings us
to the theory of everything that find a
formula that solves all your or your
questions so think about it when we put
together cloud storage and retrieving
and processing increase internet speed
Big Data computerized natural language
understanding and artificial
intelligence for decision support not
only using one of those but putting it
all together you can pretty much answer
every problem that we have today and
many that we’ll have in the future so
just to conclude that this also brings
us to the present trend that we’re
having which becomes a future trend and
that is that a very inexpensive device
can access everything any service that
it needs everything data that it needs
but that from the business point of view
this also implies a training changing
from ownership to paying for service
because now you’re only gonna be paying
for what you need when you need it but
you have the latest the latest
technology available to you so with that
I hope you got something from this the
stock and we have enjoyed it right
I hope you enjoy your ride using one or
all these kinds of technology thank you
[Applause]
Please follow and like us: