[BIRDS CHIRPING]

[BREATHING]

[RAIN]

LEWIS WANJOHI: Sound analysis is used in so many fields--

[HORNS]

[PLAYING TROMBONES]

--such as identification of music--

[MUSIC PLAYING]

--and the identification of the type of animal--

[MEOW]

--based on the sounds that they produce.

Sound is critical to what we are doing

because physiological sounds are very important in medicine.

But the traditional stethoscope hasn't

changed for close to 200 years.

The doctor is limited by the human ear.

And they cannot hear specific frequencies.

This method is very inaccurate and causes

a lot of misdiagnoses.

Our mission is to use machine learning and TensorFlow

to revolutionize the diagnosis and treatment

of respiratory diseases in low resource areas

such as sub-Saharan Africa.

The Tambua app is a powerful screening tool

that helps doctors make decisions quickly.

The core technology is trying to mimic the human auditory

system.

Once the patient walks in, the doctor

collects the lung sounds with symptoms, risk factors,

and vitals of the patient, and then the Tambua app

combines all that information and gives the doctor

a probability of the patient having a specific respiratory

disease.

SULEIMAN NDORO: [NON-ENGLISH] breathe sounds?

SPEAKER: Yeah.

LEWIS WANJOHI: Eric introduced TensorFlow to us

because he felt that we can use TensorFlow

to go through all the stages of development

to deployment of our model.

ERIC KIRIMA: Our model uses spectograms.

We take sound data from the digital stethoscope

and convert it into a visual problem

that the computer can best identify.

We have worked with a number of clinics and pathologists.

And we are able to collect data from 621 patients.

And then we use that data now to build our machine learning

model.

LEWIS WANJOHI: Once we had trained and evaluated

our machine learning model, we deployed it on our Tambua app.

TensorFlow Lite helps us to perform

an inference on our mobile device

without the need of a connection,

so doctors can use the Tambua app offline

without connecting to the cloud.

There are 216 health care facilities

that are using the Tambua app and the Tambua devices.

These clinics are spread out, and some

are very rural clinics.

SULEIMAN NDORO: It's certainly a nice project, I think.

I can use it [INAUDIBLE] plus devoid of sounds.

Then when I add the clinical history

and other physical examination, there's

no need for me to take this patient for an X-ray.

Misdiagnosis, which is one of the problems leading

to deaths in Kenya, is really creating a menace.

Tambua is helping me.

It shows me even the things that I would have missed in case

I had the traditional way of doing it.

LEWIS WANJOHI: Over 2.5 million people

die each year because of pneumonia, asthma, COPD,

and pulmonary tuberculosis.

I believe that we can use machine

learning in treatment and management

of these respiratory diseases.