so inspiration is a powerful force it’s
a force that can propel a teenager to
push past her limitations and create an
artificial neural network that can make
breast cancer diagnostics quicker
cheaper less invasive and most
importantly more accurate see I was
always that kid who never outgrew the Y
phase but it wasn’t until I found
science that I found my answers and more
questions because as we all know with
science the more you know the more you
wonder then when I was in middle school
I was taking this fascinating course on
futuristic thinking and I came across
the concept of artificial intelligence
the fact that computers could transcend
human knowledge I was enthralled I went
home I bought a coding textbook a
frequent visitor to all those internet
forums and I decided that I was going to
learn to code then when I was in tenth
grade I found another source of
inspiration my cousin was diagnosed with
breast cancer and I saw firsthand the
impact this disease really has on a
woman and her family I found out my
cousin wasn’t alone in fact one in eight
women will be diagnosed with breast
cancer I was inspired to get involved
and make a difference so I started
researching everything I could find
about breast cancer and I came across
public domain data from the University
of Wisconsin that was based on fine
needle aspirates fine needle aspirates
are these great tests they’re very
non-invasive they’re cheap and they’re
quick however they’re wildly
inconclusive so a lot of doctors refuse
to use them my goal was to create a
program that could really revive the
fine needle aspirates and provide a tool
for doctors to use in the analysis of
these fine needle aspirates in science
having that question that goal is always
a huge first step however being a
teenager trying to create a neural
network out of your bedroom doesn’t make
creating something of such a scale an
easy task I want to make it clear I’m
not some was kid I wasn’t born with a
beaker in my hand for me my journey in
science has been all about sheer
willpower and determination I started by
breaking the project down into
digestible phases I spent an entire year
just researching breast psychology and
optimizing commercially available neural
networks so I didn’t have to
code anything I found that even the best
are artificial neural networks that were
already out there only had erratic
success so I decided that I wanted to
move forward with coding my own network
in Java so naturally I researched
everything I could I spent months upon
months researching and I decided to
create a back propagation neural network
I was really excited I dove in and my
first attempt was not as excited as I
was there were thousands of errors of
code more errors than code and I ended
up scrapping the entire thing so I went
back to the drawing board I did further
research to make sure I really
understood what was going on I walked
through some tutorials videos etc and I
decided to try again
the attempts compiled I was so excited
but it proved to be solidly worse than
flipping a coin at diagnosing breast
cancer in fact it got worse as a
quote-on-quote
learned but the great thing about
science is you learn just as much from
those five experiments as you do from
your successful ones so at that point I
kind of took a step back and I did a lot
of soul-searching to figure out what
went wrong this was certainly a time
where I was doing constant pondering and
wondering if I could even create this
neural network without going off to
college and taking more advanced
computer science courses but then one
day it dawned on me you see the second
implementation although it compiled was
very object oriented so each node was
kind of doing its own thing and the
program wasn’t working well as a
cohesive unit
so I decided for my third attempt I
would make it more of a team effort as
well as have three novel different
implementations one I would change the
inconclusive logic — I would have heavy
malignant weighting and three I was
going to make sure that the network was
able to diagnose cancer patients
correctly for sure the network had
success I was over the moon and actually
proved to be 99.1% sensitive to
malignancy and I think the reason it had
so much success as I was able to learn
from those original experiments however
even though I had seen it half
successful once there was this nagging
voice that wondered is this a fluke so I
ran it again and a third time and my
results were
firming the prior results I decided that
the network was in fact working so I
turned my attention to another problem
you see neural networks learn based on
their experiences and mistakes so in
theory as they get more experiences they
should learn better I decided to run a
series of seven point six million trials
setting my alarm every three to four
hours day in and day out so that I could
determine the impact that having more
data would have on the network and what
was really exciting is every day I could
see my persistence paying off because
the line showing the correlation between
training success and sample size was
getting clearer and the inconclusive
rate was going down so at this point I
determined that I needed more data and
the way I decided to get that more data
was by deploying my program to the cloud
because the cloud which we just heard
about is this incredible elastic entity
that can scale to support usage by every
hospital in the world in order for my
program to be ready for cloud delivery I
really had to focus on making a better
user interface and also making it
quicker to avoid timeout limits after my
program was deployed as cloud for
cancer.com I actually got to go to the
Google science fair and have had this
incredible platform that I’ve been able
to use to share my research with the
world I recently got data from Lankenau
Medical Center which is a hospital in
Philadelphia and so far the program has
diagnosed all of the samples correctly
proving it can work with multiple
institutions I also recently built a
rest service and so I’ll be getting four
hundred dubious samples from an
institute in Italy so that I can see how
my network does against those I’m really
excited about the way the neural
networks working and my primary goal is
to get it into real hospitals helping
real patients saving real lives but I’ve
also been curious about whether these
same sort of tactics can apply to other
cancer diagnostics specifically I have
an interest in Imola leukemia because
MLL leukemia is a very aggressive form
of leukemia with a very poor prognosis
and there’s really no good treatment
option so I decided to see if cloud for
cancer could apply to that it took 11
attempts to get it right I ended up
creating a hybrid neural network but
what’s been really exciting is it
been able to diagnose all the mo L
samples in the data set correctly in
addition I’m working with genetic
expression profiles so it’s actually
been able to go through and determine
certain proteins that may be candidates
for drug targeting in the future so that
those prognosis rates can go up for me
my journey in science has very much been
about finding what inspires me and then
having the courage to pursue it even
through failure I’m going to be a
college freshman soon as and as I go off
to dual major in computer science and
biology I plan on living by the phrase
persistence pays off which means never
give up because if I hadn’t had the
persistence to push through this
experiment there’s no way I’d be
standing here at CERN talking about my
invention that could someday be in
hospitals we as young curious minds are
facing a lot of challenges from finding
the cures to cancer to creating
alternative energy sources to unraveling
the mysteries behind particle physics
together with hard work and passion we
can truly revolutionize the world around
us
and to me that’s what’s exciting about
science thank you so much
Brittany Wenger, 18, high school senior, brilliant young scientist and Grand Prize Winner 2012 Google Science Fair, for her project «Global Neural Network Cloud Service for Breast Cancer» talks about how she came to science in Research and Inspiration.
How to make a neural network in your bedroom | Brittany Wenger |a TEDxCERN