KENNETH SHINOZUKA: Good
afternoon everyone. I'm really happy to be
here today to talk to you about my project,
Wearable Sensors, a novel health care solution
for the aging society. So growing up in a family
with three generations, I've always been very
close to my grandfather. When I was four years
old, my grandfather and I were walking in a park in Japan
when he suddenly got lost. It was one of the
scariest moments I've ever experienced in my life. And it was also
the first incident that inform
ed us that
my grandfather had Alzheimer's disease. Over the years, his condition
got worse and worse. And his wandering,
in particular, caused my family
significant stress. About two years ago, my
grandfather's wandering out of bed at night had
became much more frequent. And my aunt, his
primary caregiver, really struggled to stay awake
at night to keep an eye on him. And even then, often failed
to catch him leaving the bed. I became really concerned
about my aunt's well being, as well as my
gra
ndfather's safety. And I searched extensively
for, but really couldn't find a solution to
my family's problems. Later, I found out that the
struggles faced by my family were just a snapshot of a
much larger societal burden. There are currently 5.2
million Alzheimer's patients the United States alone and
over 65% of them wander. Alzheimer's disease is the
fastest growing health threat to the United States. Every 67 seconds, someone
in the US gets the disease. Caring for these patients
cost the US
$220 billion in 2013 alone. And that number is expected to
multiply five times to about $1.2 trillion by the year 2050. So this overwhelming
societal challenge, coupled with my concern
for my family's struggles, really motivated me
to find a solution. But how? One night, I was looking
after my grandfather and I saw him stepping
out of the bed. The moment his foot landed on
the floor, a light bulb flashed in my head. I thought, why don't I
put a pressure sensor on the heel of his foot? As soon a
s he steps
onto the floor, the pressure sensor
would detect an increase in pressure caused
by body weight, and then wirelessly
send an audible alert to my aunt's smartphone,
waking her up. That way, she could sleep
much better at night. My desire to create
a sensor-based system perhaps stemmed from my
lifelong passion for sensors. When I was six years old,
an elderly family friend fell down in the bathroom
and suffered severe injuries. I became concerned about
my own grandparents. And I decided
to invent
a smart bathroom, a motion-sensor
system that detects the falls of elderly
patients and alerts caregivers wirelessly
at their wrist watches. This and other
research experience prepared me to create my system. However, my
seemingly simple idea later proved to be very
challenging for me to realize. When I laid out my
plan, I realized that I faced three main
challenges-- creating a sensor, designing a circuit, and
coding a smartphone app. So first, I had to create
a wearable sensor that w
as thin and flexible enough
not to affect the comfort of the patient's walking. After extensive
research, and testing multiple different
materials, I decided to print a film sensor with
pressure sensitive electrically conducted particles. Once pressure's applied,
the connectivity between the particles increases. So therefore, I could
design a circuit that would measure
force by measuring electrical resistance. Next, I had to design a
wearable wireless circuit. But wireless signal transmission
co
nsumes lots of power and requires heavy
bulky batteries. Thankfully, I was
able to find out about the cutting edge Bluetooth
low-energy technology, which consumes very little
power and can be driven by a coin-sized battery. Third, I had to code
a smartphone app that would transform the
caregiver's smartphone into a remote monitor. For this, I had to
expand upon my existing knowledge of Java and Xcode. And I also had to
learn a lot about how to code for Bluetooth
low-energy devices. Integrating t
hese
components, I was able to create two
prototypes, a sensor sock and a reattachable
sensor assembly. I've tested both prototypes
on my grandfather for nearly 10 months now. And so far, they've
had a 100% success rate in detecting the known
cases of his wandering. Encouraged by these
results, I decided to form my own startup, SensoRx,
to commercialize my device, which I named Safe Wander. Over the summer, I
beta tested my system at several residential
care facilities, and I'm incorporating
the
feedback to improve the technology
into a marketable product. Through my beta test,
I realized that I need to increase the
single transmission distance of my sensor. When you place a tiny
antenna on the human body, the signal transmission
distance becomes limited. And this is actually
a common challenge that's faced by a lot
of wearable systems. So I've decided to incorporate
Wi-Fi in a cloud server to not only enable longer
transmission distance, but also to enable
data analytics for better he
alth care. For example, we can
examine correlations between the frequency of a
patient's nightly wandering and his daily
activities and diet. So that's all for me today. And thank you very much. [APPLAUSE] DANIELA LEE: Hi. I'm Daniela Lee. And I'm 17 years old. SADHIKA MALLADI:
Hi, Sadhika Malladi. I'm 16 years old. And today we're here to talk
about our experiences working on the project that entered
the Google Science Fair Global Finals. DANIELA LEE: So as a duo working
on a computer science p
roject, we came across many
hurdles in collaboration. However, this project,
and working as a partner, was really exciting, in
that it brought two people with very different
interests in science together to tackle a small part
of a massive problem, cancer. SADHIKA MALLADI:
And so our project addresses the deficiencies
in current standard practice for treatment decisions for
triple negative breast cancer patients. DANIELA LEE: We used machine
learning algorithms along with image analysis to
see h
ow the patient will respond to certain therapies,
especially chemotherapy. And this knowledge is very
important to optimize treatment and to make sure the patient
is getting the right treatment so we save valuable time. SADHIKA MALLADI:
And so what we do is we use a non-invasive
scan in combination with image analysis
and machine learning to create this prediction of
whether a patient will respond to chemotherapy. DANIELA LEE: We decided to work
with Triple Negative Breast Cancer, or TNBC, becau
se
the patients foster an aggressive
tumor and they also lack certain hormone receptors
that most chemotherapies target. So as you can see here, 15%
to 25% of all breast cancers right now have triple
negative breast cancer. And this is alarming because
the treatment options are sort of uncertain. And they're not as efficient
as they should be right now. SADHIKA MALLADI: And so
current standard practice right now is to take surgical
samples of the tumor. But even with these
biopsies, doctors tend
to use a one-size-fits-all regimen,
which is alarming because each cancer is different. DANIELA LEE: So we were able
to look at the breast cancers through these images, which
were very non-invasive. And they were DCE-MRI
images, or Dynamic Enhanced. Now, before we go into what we
did, we also want to point out, when you even have
a simple cold, you want to make sure you
get the right medicine and you want to make sure that
it's the best medicine you get. For cancer patients,
this urgency is amp
lified because they
don't have much time. And they can also have a
potentially fatal outcome. SADHIKA MALLADI:
And so our project was, essentially, a
lot of trial and error because we had to
optimize both the machine learning algorithms and the
image analysis algorithms. So essentially, you can
see what we did here through a basic workflow. And so we used image
analysis, like I said, with machine learning models. And we translated that
to a biological impact on the patient. Some of the stumbling
blocks
we faced included collaborating on our work because I would
sometimes have versions of code that Daniela wouldn't. And so when we were
testing our models, it would get a
little bit confusing. DANIELA LEE: So our
models that used scans or image of the patients
from just before chemotherapy were able to predict
whether or not the patient was going to
respond before you even had any cycles of chemotherapy. Our models that used scans from
before and during the treatment were able to assess h
ow well
the patient was responding and whether or not this was
the right path for them. So we were able to
achieve a 74% accuracy in predicting whether or
not the patient will respond to treatment before going
through any chemotherapy, and seeing if this is the
right option for them. We were also able to
achieve an 83% accuracy in seeing how the patient
will respond while we're doing the treatment. And to see, like
mentioned before, whether they will
achieve a full response and whether they're g
oing to
need surgery in the future. Now, our best model so
far has an 86% accuracy. And this model can tell you,
using mid-treatment assessment, how the patient will finish the
chemotherapy, whether they're going to have residual
tumors or not. SADHIKA MALLADI: And so when
we were working as a team, like I mentioned before, it
was really difficult for us to sync our versions of code
because there's no Google Docs, per se, for coding. There's nothing
with active syntax highlighting that we can bo
th
look at at the same time. While there is GitHub, we
both couldn't edit code at the same time. And so sometimes,
like I said, I would have optimized the
algorithms but Daniela would be running a previous
version of the algorithms. DANIELA LEE: And so we also
ran a lot of different models. So sharing the results with
each other was also a bit messy. However, despite
these struggles, we were able to save time by
working together as a team. Because we can run the different
models at different tim
es and all at once. SADHIKA MALLADI: And so because
we were different people, we had varying interpretations
on the results that we got. I was on the computer
science side of the project. And so I was really interested
in how machine learning works, how we can teach a computer
to recognize patterns that's lost on the human eye. DANIELA LEE: I was
more of-- well, I like to think that I
was like the biologist or the medical doctor
of the project, and maybe even
science popularizer. Because I was r
eally intrigued
by how our models would impact the patient life starting
with the non-invasiveness of the imaging to the
patient's changing morphology throughout treatment,
their emotions, and how this application
can be clinically viable. Our aha! moment was a little
bit different from Kenneth's. Actually, like I said,
I wasn't as well versed in computer science
as Sadhika is. So one time I accidentally
deleted half of the code. So because I was still
a little bit confused about the image analy
sis
and about machine learning, I, one day, was not able to
compile one of the models. And I had no idea
what was going on. So I went to our mentor. And he was a little
bit perplexed as well. But we were somehow
able to force the model to work by fixing
some parts of it. And throughout
this whole process, I had no idea that
I'd just deleted half of the texture measurements
or half of the image analysis algorithms. But I ran it anyway. And after a few hours,
this model was completed. The accuracy
rate of this
model, 86%, our best model yet. SADHIKA MALLADI: And so you
should have seen my face when I got the email from her. I was like, what? I didn't get anything like this. We were, like, around
maybe 74% before this. And so it just goes to show that
even the strangest mistakes-- I mean, I don't know if that
mistake's really replicable. But I mean, even the
strangest mistakes lead to the greatest insights. And it's with these insights
that we went to the Google Science Fair Global Finals
. And so you can see some pictures
from our experience here. I'm on the Google bike too. I mean, we just
had a lot of fun. We got to present our project to
judges, meet amazing finalists, receive feedback. And most importantly, our
work got on this global stage. DANIELA LEE: Because our
project involved machine learning algorithms, one of
the main limitations we had was the lack of patient data. And this is how we
think the cloud can factor into our project. Institutions around the world
have pe
ople gathering data, but they aren't able to
share this data effectively. We think that with more data our
models can become more refined. SADHIKA MALLADI: And so one of
the emerging trends that we all know about right now in
computer science is big data and harnessing the power of
multiple minds to do what one mind cannot. We recently heard about a
very interesting project in Switzerland, where they
put all of their information online. And essentially, coders
from around the world are able to c
ontribute
to the solution. DANIELA LEE: This data is
hosted on secure servers. So it's all compliant with
the HIPAA regulations. And the programmers
are invited to code and to harness the data to find
solutions to certain problems. If you think about this in
the context of our work, these people can come. They can contribute their ideas
to how we can find responses to chemotherapy. And this can completely find us
a new direction in our project and help us advance our models. SADHIKA MALLADI: And
so as you can see, the influence of the
cloud and big data can really take our
project to the next level. Ideally, in five years, we'd
be able to apply our model to other cancers besides
triple negative breast cancer, and make it clinically viable. Thank you. [APPLAUSE] MALE SPEAKER: How
about over one more? SADHIKA MALLADI: Oh, yeah. MALE SPEAKER: Let's get
everybody down on one side. Thanks. Here, Daniela, I'll
take that from you. Thank you. All right, so not bad? Do you guys think? So great
job, guys. How does it feel to
destroy the old saying, you know, that parents
have, where-- well, I can disrupt them from
what they're doing. It's not like they're
curing cancer or anything. You kind of threw that one
out the window, didn't you? So I understand, Kenneth, your
grandfather is suffering from Alzheimer's. But why did you guys tackle such
difficult challenges as, oh, cancer and Alzheimer's? KENNETH SHINOZUKA: So
is your question why? MALE SPEAKER: Yeah, why
not-- I understand there w
as a personal connection. But despite that, why would you
tackle something so daunting? Why something so difficult as
you know, detecting cancer? KENNETH SHINOZUKA: Sure. MALE SPEAKER: There are
labs spending billions of dollars on this stuff. And you guys just sort of cooked
this up in your kitchen, right? KENNETH SHINOZUKA:
So yeah, I think there was definitely a
personal connection there. And seeing the struggles
of my aunt, I think, was what really inspired
me to create my solution. My aunt
had to stay
awake all night to keep an eye on
my grandfather. She was feeling very
fatigued during the daytime, since she got no sleep. She had to sacrifice time
that she would usually spend on her work
and with her family. And seeing that struggle
was what really inspired me. It sort of broke
my heart, I think. And so that emotional
connection was there. And I think that was
what really inspired me to tackle the
problem because I really wanted to help my aunt. MALE SPEAKER: Daniela,
what about
you? DANIELA LEE: We also had a
lot of personal relations to our project. Both of us had
had family members who have passed away due
to hit-and-miss treatments with cancer. And since triple
negative breast cancer is one of the most
aggressive ones, we decided to tackle one of
the most aggressive ones. MALE SPEAKER: So
damn the obstacles, why not tackle the stuff, right? DANIELA LEE: Yeah. MALE SPEAKER: Right? Fresh eyes and fearlessness. So this actually
goes to the point we all face, which is w
hy
is it as we get older do we just become
so risk averse? You know, why not
tilt at the windmill? So each of you demonstrated
that you tackled the problem with fresh eyes, created stuff. If the solutions didn't
exist, you created them. You had the flexible sensors. You guys both came out with an
analysis, the machine learning, to help you break down the
patterns that were otherwise eluding people. And one of the things
I noticed, Sadhika, you and Daniela talked about,
was embracing the lessons
from your mistakes. So is that something
that came naturally? Or did you guys have
to just reconcile? Did you have to teach
yourselves to learn from it? Or were you just
frustrated that you kept having all these errors and
not what you wanted to achieve? SADHIKA MALLADI: Right. So I would say that
when I was learning these complicated topics,
like machine learning is not something you learn in
high school, generally. And so when I was learning
these topics, it was very hit and miss. I mean, I wa
s just
running things. I was trying things. I mean, I think that was kind
of the stage where I realized it's no longer about me. It's no longer about
my intelligence. It's more about tackling
another problem. It's not about advancing myself,
but really advancing the field. MALE SPEAKER: That's what
I hear from most teenagers [LAUGHTER] Yeah, you've got to hang out
with my daughters, I think. OK. Speaking of which,
have you ever, in the process of
your creating, did you ever think, hey, this
is g
oing to have potentially a commercial impact to me? Or did you approach this purely
from a scientific perspective? What was your
motivation down deep? I understand, Kenneth,
yours was helping your aunt, your grandfather, et cetera. But did that thought of
commercialization factor in? Was that a motivator ever? DANIELA LEE: Not really. In fact, when we got our
results we didn't even know it was like, a
really good result. We just heard from our mentor,
oh, wow, this is a good result. You could pu
blish it. So we were aiming more at
publication in a journal. We weren't even really thinking
about Google Science Fair. We just tried that, oh,
let's just see what happens. And now we're getting
comments about how we should try commercializing
it and making it actually clinically viable,
which we never imagined we would ever even try at all. MALE SPEAKER: You
were just looking to pad the resume for your
college applications weren't you? DANIELA LEE: We were looking
at having fun researching, no
t really what could happen. MALE SPEAKER: Fun? Passion. But this is fantastic. This is what I have found
in common with people who really make progress, right? They're passionate
about what they do. So despite the rejection
and the failures that you guys have
seen, you were always encouraged to continue
to the next level, right? SADHIKA MALLADI: Yeah. I'd say, definitely. I mean, there was a lot of
encouragement from our families and from our community
and from our mentors. And I mean, we hit a
lot
of roadblocks, definitely. I mean, Daniela
deleting half the code. MALE SPEAKER:
You're still holding that against her, aren't you? SADHIKA MALLADI: Yeah. MALE SPEAKER: Yeah, you
know, don't be bitter. It worked out. SADHIKA MALLADI: Yeah. it did work out. KENNETH SHINOZUKA: When I when I
launched my website-- which you can all visit,
SafeWander.com-- I was really surprised by the number of
people who are suffering from this problem of caring
for wandering patients and really struggling
to k
eep an eye on them. And I received a
lot of inquiries asking when my device
would be on the market and when they could
purchase the device. And that really
further motivated me to commercialize the device. So really, I felt that there
was a really enthusiastic need for the device, beyond what
I'd originally expected. MALE SPEAKER: Oh I can
think of a fantastic-- I need two strips,
one in a size eight for my 16-year-old daughter. Right, I need one that
says girl on the run. So all right, any ques
tions
for the participants? Anybody else wondering about
do they make house calls? Yes, Andy. AUDIENCE: So I've
got teen-aged sons. And it's obviously
very humbling. MALE SPEAKER: It
worked, didn't it? AUDIENCE: Absolutely. It just blows you away. What was the
self-learning experience you've taken from
all this that I could share with my sons and
their friends, at their age? KENNETH SHINOZUKA: Well, I
think that one of the things that I've always encouraged
other people to do when they ask me wh
at can we
learn from your experience? Is that I think the first step
to creating any solution that helps other people
is to first observe the problems around you. And then once you have a firm
understanding of the problem, you can begin to think about a
solution, a creative solution that approaches the problem. And then once you once you think
of a solution, work hard at it. And then, with
some perseverance, you can definitely create a
solution that helps people. One thing that I've definitely
l
earned from my experience is that you can be a
totally ordinary kid like me and still be able to
create something that helps the people around you. So you don't need
to be a genius. You don't need to
be anybody special. You just need to be someone
who recognizes a problem and then works hard at it. MALE SPEAKER:
There's still hope. [APPLAUSE] Fantastic. Any bits of wisdom, Sadhika? SADHIKA MALLADI: I would
say, I kind of took the opposite approach
from Kenneth. And so I kind of thought
about wha
t I was interested in and what I really liked doing. And then I thought about,
well, OK, I like doing it. But how can I make sure that
other people benefit from this? And so that's kind of
the opposite direction. But you can approach
it either way. MALE SPEAKER: Sort of a
pay-it-forward investment. SADHIKA MALLADI: Yeah. MALE SPEAKER: Excellent. See? There's hope for all of us. Now, by the way,
Kenneth, I don't think either of the three of
you are ordinary, by the way. So congratulations
on your
achievements. You guys have done
wonderful work. You're inspiring. And we wish you the best. SADHIKA MALLADI: Thank you. KENNETH SHINOZUKA:
Thank you so much. [APPLAUSE]
Comments