How AI is making it easier to diagnose disease

Today’s AI algorithms require tens of thousands of expensive medical images to detect a patient’s disease. What if we could drastically reduce the amount of data needed to train an AI, making diagnoses low-cost and more effective? Medical technologist and TED Fellow Pratik Shah is working on a clever system to do just that. Using an unorthodox AI approach, Shah has developed a technology that requires as few as 50 images to develop a working algorithm -- and can even use photos taken on a doctor's cell phone to provide a diagnosis. Learn more about how this new way to analyze medical information could lead to earlier detection of life-threatening illnesses and bring AI-assisted diagnosis to many more people worldwide.

Computer algorithms today
are performing incredible tasks

with high accuracies, at a massive scale,
using human-like intelligence.

And this intelligence of computers
is often referred to as AI

or artificial intelligence.

AI is poised to make an incredible impact
on our lives in the future.

Today, however,
we still face massive challenges

in detecting and diagnosing
several life-threatening illnesses,

such as infectious diseases and cancer.

Thousands of patients every year

lose their lives
due to liver and oral cancer.

Our best way to help these patients

is to perform early detection
and diagnoses of these diseases.

So how do we detect these diseases today,
and can artificial intelligence help?

In patients who, unfortunately,
are suspected of these diseases,

an expert physician first orders

very expensive
medical imaging technologies

such as fluorescent imaging,
CTs, MRIs, to be performed.

Once those images are collected,

another expert physician then diagnoses
those images and talks to the patient.

As you can see, this is
a very resource-intensive process,

requiring both expert physicians,
expensive medical imaging technologies,

and is not considered practical
for the developing world.

And in fact, in many
industrialized nations, as well.

So, can we solve this problem
using artificial intelligence?

Today, if I were to use traditional
artificial intelligence architectures

to solve this problem,

I would require 10,000 --

I repeat, on an order of 10,000
of these very expensive medical images

first to be generated.

After that, I would then go
to an expert physician,

who would then analyze
those images for me.

And using those two pieces of information,

I can train a standard deep neural network
or a deep learning network

to provide patient's diagnosis.

Similar to the first approach,

traditional artificial
intelligence approaches

suffer from the same problem.

Large amounts of data, expert physicians
and expert medical imaging technologies.

So, can we invent more scalable, effective

and more valuable artificial
intelligence architectures

to solve these very important
problems facing us today?

And this is exactly
what my group at MIT Media Lab does.

We have invented a variety
of unorthodox AI architectures

to solve some of the most important
challenges facing us today

in medical imaging and clinical trials.

In the example I shared
with you today, we had two goals.

Our first goal was to reduce
the number of images

required to train
artificial intelligence algorithms.

Our second goal -- we were more ambitious,

we wanted to reduce the use
of expensive medical imaging technologies

to screen patients.

So how did we do it?

For our first goal,

instead of starting
with tens and thousands

of these very expensive medical images,
like traditional AI,

we started with a single medical image.

From this image, my team and I
figured out a very clever way

to extract billions
of information packets.

These information packets
included colors, pixels, geometry

and rendering of the disease
on the medical image.

In a sense, we converted one image
into billions of training data points,

massively reducing the amount of data
needed for training.

For our second goal,

to reduce the use of expensive medical
imaging technologies to screen patients,

we started with a standard,
white light photograph,

acquired either from a DSLR camera
or a mobile phone, for the patient.

Then remember those
billions of information packets?

We overlaid those from
the medical image onto this image,

creating something
that we call a composite image.

Much to our surprise,
we only required 50 --

I repeat, only 50 --

of these composite images to train
our algorithms to high efficiencies.

To summarize our approach,

instead of using 10,000
very expensive medical images,

we can now train the AI algorithms
in an unorthodox way,

using only 50 of these high-resolution,
but standard photographs,

acquired from DSLR cameras
and mobile phones,

and provide diagnosis.

More importantly,

our algorithms can accept,
in the future and even right now,

some very simple, white light
photographs from the patient,

instead of expensive
medical imaging technologies.

I believe that we are poised
to enter an era

where artificial intelligence

is going to make an incredible
impact on our future.

And I think that thinking
about traditional AI,

which is data-rich but application-poor,

we should also continue thinking

about unorthodox artificial
intelligence architectures,

which can accept small amounts of data

and solve some of the most important
problems facing us today,

especially in health care.

Thank you very much.

(Applause)