Press "Enter" to skip to content

Artificial Intelligence Can Change the future of Medical Diagnosis | Shinjini Kundu | TEDxPittsburgh


[Music]
in 1895 a man took a picture on hand
this hand picture was so mind-boggling
that he won a Nobel Prize and it changed
the world his name Wilhelm roentgen and
I present to you the first medical image
ever taken this is the world’s first
x-ray this may not look very impressive
to you today but at the time it was
genius because this was the first time
we could look inside of the human body
without having to slice someone open and
indeed medical imaging has changed the
face of medicine in the last hundred
years since roentgen imaging still
represents one of the fastest growing
sources of medical data let me ask you a
question how many of you have gotten a
mammogram a cat scan an MRI an x-ray can
I see by a show of hands please that’s
incredible because it seems that almost
everyone here has benefited from medical
imaging technology in some way or
another
you see we begin to realize that imaging
can sometimes reveal a disease even
before you feel it and for diseases like
cancer the earlier we diagnosed the more
likely you’ll survive this is why we
started screening people who exhibit no
signs or symptoms of disease pap smears
mammograms colonoscopies all examples of
screening tests that are recommended to
everyone at certain ages we know that
diagnosing a disease early saves lives
it reduces treatment costs and it
reduces soum
much patient suffering unfortunately
though most diseases are still diagnosed
at advanced stages you see the problem
with detecting these diseases early is
that as you turn back the clock visible
evidence of disease becomes smaller and
smaller until it vanishes before the
naked eye so realistically how early can
we detect these diseases there’s hope
because it turns out that there is an
invisible side to imaging small changes
that your eyes can’t see still exist and
in my research I’ve shown that these
hidden patterns can sometimes be decoded
with the help of a computer and
sometimes these patterns contain secrets
about the imminence of disease that we
never knew so let me give you an example
osteoarthritis is one of those diseases
that we can’t detect until the damage
has been done in this room more than one
in ten of us will develop osteoarthritis
in our knee this is the disease in which
the thin cushion of cartilage between
two bones falls apart causing bone to
grind painfully on bone oh it makes me
cringe just thinking about it but today
doctors can’t see osteoarthritis until
after bone damage and pain have
developed now about to show you an x-ray
that shows some very subtle signs of
bone damage so be sure to look this is
what your x-ray might look like the
first time you’re diagnosed with
osteoarthritis but there’s evidence that
the disease begins long before that
inside the cartilage so why can’t we
just look at the cartilage to diagnose
the disease so let’s take a trip back in
time and let’s actually take a look some
images of cartilage these are magnetic
resonance images or MRIs
of knee cartilage these are all healthy
people each scan belongs to a different
subject the different colors here
represent the different ingredients that
make up cartilage and we know that three
years down the line one of these groups
will receive a diagnosis of
osteoarthritis but the other group won’t
and the question is which group is which
you see the problem is that our eyes
can’t seem to detect
a difference between these groups of
patients even experts can’t tell what’s
wrong here so does that mean that three
years before bone damage develops
there’s nothing to see in the cartilage
or could we possibly find evidence of
early disease that eludes the naked eye
with the help of a computer ten years
ago I started studying signals signals
are the way that computers understand
and process the world and I wanted to
teach a computer to recognize patterns
and medical images the same way that
your personal assistant Syrian Alexa can
recognize your voice today by using
something called machine learning the
basic premise behind machine learning is
to program a computer that can learn on
its own with experience so the way that
it works is that we feed in the images
and whether that person goes on to
develop osteoarthritis three years down
the line and the computer sips through
the images to try to determine whether
there is a pattern that can
differentiate osteoarthritis progresses
from healthy people in my research I
developed a technique called transport
based morphometry which teaches the
computer to learn interrelationships
between pixels rather than trying to
focus on finding one particular pixel
within
the image and the true elegance of this
technique is that transport based
morphometry can come back and teach us
the patterns that the computer has
learned okay so at this point you’re all
probably dying to know whether the
computer can teleport these images that
we couldn’t teleport right so it turns
out that on an image of knee cartilage
that the computer has never seen before
it can predict whether that person goes
on to develop osteoarthritis three years
down the line with 86 percent accuracy
and because these machines learn with
experience they’ll only get better and
better as we feed in more scams now
remember that doctors today have no way
of diagnosing osteoarthritis three years
prior to symptoms imagine what three
years means in a person’s life if I knew
that I would develop osteoarthritis
three years from now I would take steps
today to try to stave off the disease
for as long as possible
okay now you’re probably wondering well
what is the computer seeing that we
can’t see right well these images are
computer-generated MRIs and the red here
indicates the presence of water which we
already know but what we didn’t know was
that what the computer can use to
diagnose osteoarthritis in the future
is the diffusion of water you see to the
far left we see an image of a person is
completely healthy and the water here is
evenly distributed throughout the
cartilage on the far right we see an
image of a person who goes on to develop
osteoarthritis three years from now and
the water pulls to the center of the
cartilage just as potholes in the road
accumulate water the pooling of water
suggests that there’s a weakening of
cartilage at that location and it’s
incredible that this is the first time
we can see early evidence of disease in
the cartilage even though it’s been
there the whole time on literally
million
of scans that physicians have seen we
can finally see these changes through
the eyes of a computer teaching us what
to look for now you’re probably
wondering why do we need to look through
the eyes of a computer why couldn’t we
see this on our own to which I would
answer that remember that everyone’s
knees a little bit different so when you
look at a picture of knee cartilage your
eyes become saturated with the normal
variability that can happen from person
to person but when the computer looks at
the image it can really zero in on
Stevan subtle differences in water
content so this is why the computer can
recognize early signs of disease while
your eyes can’t cracking this puzzle was
no easy feat science is built on
accumulated knowledge and I’d like to
thank my collaborators and colleagues I
could not have done this without them
early detection of osteoarthritis was a
great starting place for me because for
me it highlighted the power of machine
learning to be able to push our human
capabilities we could finally find
things that we never knew to look for
and we never even knew existed and to
think of the myriad diseases where we
can apply this technique today we don’t
understand how many diseases arise and
develop particularly in the brain
Alzheimer’s autism schizophrenia are
examples but in the future we may be
able to understand how these diseases
start and identify new targets for
treatment today we can’t diagnose
diseases until after symptoms develop
and by then it might be too late in the
future we might be able to diagnose
diseases before they develop and we
might have the opportunity to halt the
disease before it even begins I’m
working to get this technology in the
hands of your doctors within the next 10
years but for that what needs to happen
first because these machines learn with
experience
we need vast amounts of patient data to
ensure that these computers become
smarter and more accurate in diagnosing
fortunately this computer lives in our
hospitals today second we need to build
a community of people who can accelerate
this work and bring benefits to patients
sooner rather than later
just as in 1895 with the first rent
Gunnar Graham the face of Medicine is
about to change again because this is
the first time we can see not the
visible but the invisible changes in the
human body that allow us to diagnose a
disease even before it develops now all
we need is curiosity and an open mind
thank you
[Applause]
[Music]
[Music]
you
Please follow and like us: