BUSINESS CASE

An AI to Detect Breast Cancer

Challenge

Early screening for breast cancer brings 5-year survival rates to 99%, from an 85% average. Tissue analysis by a pathologist is the only way to identify breast cancer with certainty. Fewer and fewer pathologists are trained each year. Image recognition algorithms can assist pathologists.

Health, IA, Image Recognition, Cancer, Detection
health, logo, image recognition

Breast cancer is the most frequent cancer encountered by women. The earlier the screening, the better the survival odds are. Enhancement of screening techniques is one of the main courses of actions pursued by researchers to improve patient care.

87%

5-year survival rate

12 k

Deaths in 2017

60 k

New Cases in 2017

Our impact

300,000
Labelled images to train our algorithm
88%
Algorithm accuracy after 2 weeks
Background
Background
Quotes

It's our goal that these systems allow doctors to personalise screening and detection programs, so as to definitely get rid of late diagnoses.

Regina Barzilay, Professor, MIT, Image Recognition, Cancer

Pr. Regina Barzilay

What We Did

Sicara built an image recognition solution to assist doctors in their breast cancer diagnoses

neural network, image recognition, ai

What We Did

Sicara built an image recognition solution to assist doctors in their breast cancer diagnoses

Breast cancer cases where affected tissue is detected early have a 5-year survival rate of 99%, as opposed to an 85% survival rate on average. With the aid of a neural network we developed, after two weeks of training, it could detect sick tissue strips with an 88% accuracy, helping alleviate the workload of pathologists. In cancerology, image recognition techniques can analyze cancer tissue following the processes we applied to breast cancer. The same methods can help diagnose throat cancer (using radios as DeepMind did).

How We Did It

We developed a neural network

Health, AI, image recognition

How We Did It

We developed a neural network

We trained the neural network we developed with 300 strips cut into 300,000 labeled 50x50 px images.

keras, logo, manomano, sicara
logo, python, manomano, sicara
logo, tensorflow

Our team set up

We worked full speed thanks to agile methodology

Sicara, startup, team, teamwork

Our team set up

We worked full speed thanks to agile methodology

Our team was composed of 4 data scientists and 1 Agile Coach and brought added value to the client on a daily basis.

Adil

Centrale Paris

Picture of Raphaël

Raphaël

ENSTA, Polytechnique


Related articles written by Sicara data scientists

Edge detection, tutorial, knowledge

Edge Detection in Opencv 4.0, A 15 Minutes Tutorial

This tutorial will teach you, with examples, two OpenCV techniques in python to deal with edge detection.

TensorFlow, AI, Docker, GPU

Set up TensorFlow with Docker + GPU in Minutes

Why Docker is the best platform to use Tensorflow with a GPU.

blurry street

GAN with Keras: Application to Image Deblurring

A Generative Adversarial Networks tutorial applied to Image Deblurring with the Keras library.