Présentation collégiens (Projet CooLearning et Segmentation des poumons)

CooLearning + segmentation poumons

Emmanuel Roux (équipe MYRIAD)

Plan de présentation

Histoire des réseaux de neurones

Un réseau de neurone en pratique : CooLearning

Application à la segmentation des poumons

Machine Learning

Deep Learning

Artificial Intelligence

Inspired (and simplified) from the deeplearningbook.org
(I. Goodfellow and Y. Bengio, A. Courville, 2016)

Machine Learning

Deep Learning

Artificial Intelligence

Inspired from Sebastian Raschka's deep-learning course

Artificial Intelligence

Inspired from (Cardon D., Cointet J.-P., Mazieres A., 2018)

Artificial Intelligence

Inspired from (Cardon D., Cointet J.-P., Mazieres A., 2018)

DEDUCTIVE

rule-based

no need of examples

INDUCTIVE

example based

adaptation

Symbolic AI

connectionism

Cybernetics (40’s to 60’s)

Inspired from (Cardon D., Cointet J.-P., Mazieres A., 2018)

connexionism

Perceptron (Rosenblatt)

ADALINE (Widrow & Hoff)

Homeostat, 1948

(W. Ross Ashby)

source wikipedia

source reddit

https://isl.stanford.edu/~widrow/papers/t1960anadaptive.pdf

Symbolic Artificial Intelligence (60’s to 80’s)

Inspired from (Cardon D., Cointet J.-P., Mazieres A., 2018)

Symbolic AI

MYCIN (Shortliffe): medical diagnoses (bacteria identification)

GUIDON (Clancey): teaching medical diagnostic strategy

CADUCEUS (Pople): internal medicine expert system

Transcript of an INTERNIST-I Consultation (Myers)

machine learning (00's to 10's)

deep learning (10's - now)

Image by Dake, Mysid

connexionism (80’s to 00’s)

Plan de présentation

Histoire des réseaux de neurones

Un réseau de neurone en pratique : CooLearning

Application à la segmentation des poumons

2.0

0.5

x_1 = 2

x_2 = 3

y =

un neurone

2.0

0.5

x_1 = 2

x_2 = 3

y = 0.5 \times

un neurone

2.0

0.5

x_1 = 2

x_2 = 3

y = 0.5 \times 2 +

un neurone

2.0

0.5

x_1 = 2

x_2 = 3

y = 0.5 \times 2 +

un neurone

2.0

0.5

x_1 = 2

x_2 = 3

y = 0.5 \times 2 + 2.0 \times

un neurone

2.0

0.5

x_1 = 2

x_2 = 3

y = 0.5 \times 2 + 2.0 \times 3

un neurone

2.0

0.5

un neurone

x_1 = 2

x_2 = 3

y = 0.5 \times 2 + 2.0 \times 3 = 7

deux neurones ?

2.0

0.5

2.0

1.0

x_1 = 2

x_2 = 3

y = ?

1.0

deux neurones ?

2.0

0.5

2.0

1.0

x_1 = 2

x_2 = 3

y = ?

2 \times 0.5 = 1

1.0

deux neurones ?

2.0

1.0

0.5

2.0

1.0

x_1 = 2

x_2 = 3

y = ?

2 \times 1 = 2

deux neurones ?

2.0

1.0

0.5

2.0

1.0

x_1 = 2

x_2 = 3

y = ?

2 \times 1 = 2

3\times2 = 6

deux neurones ?

2.0

1.0

0.5

2.0

1.0

x_1 = 2

x_2 = 3

2 + 6 =8

y = ?

deux neurones ?

2.0

0.5

2.0

1.0

x_1 = 2

x_2 = 3

y = ?

1.0

deux neurones ?

2.0

0.5

2.0

1.0

x_1 = 2

x_2 = 3

1 \times 2 = 2

8 \times 1.0 = 8

y = ?

1.0

deux neurones ?

2.0

0.5

2.0

1.0

x_1 = 2

x_2 = 3

y = 2+8=10

1.0

0.5

un neurone + fonction d'activation

x_1 = 2

x_2 = 3

https://en.wikipedia.org/wiki/Sigmoid_function

\sigma

\sigma(x)=\frac{1}{1+e^{-x}}

1.0

0.5

un neurone + fonction d'activation

x_1 = 2

x_2 = 3

https://en.wikipedia.org/wiki/Sigmoid_function

\sigma

\sim 0.982

\sigma(x)=\frac{1}{1+e^{-x}}

1.0

0.5

un neurone (avec biais)
+ fonction d'activation

x_1 = 2

x_2 = 3

https://en.wikipedia.org/wiki/Sigmoid_function

\sigma

\sigma(x)=\frac{1}{1+e^{-x}}

-2.0

1.0

0.5

un neurone (avec biais)
+ fonction d'activation

x_1 = 2

x_2 = 3

https://en.wikipedia.org/wiki/Sigmoid_function

\sigma

\sim 0.88

\sigma(x)=\frac{1}{1+e^{-x}}

-2.0

CooLearning : expérimenter sur

un réseau de neurones ... interactif !

> SELECTEUR <

> CHANGE LES POIDS <

> SELECTEUR <

> SELECTEUR <

> SELECTEUR <

> CHANGE LES POIDS <

1.0

> CHANGE LES POIDS <

-0.9

> CHANGE LES POIDS <

0.2

> CHANGE LES POIDS <

1.0

EXPERIMENTEZ !

> SELECTEUR <

> CHANGE LES POIDS <

Plan de présentation

Histoire des réseaux de neurones

Un réseau de neurone en pratique : CooLearning

Application à la segmentation des poumons

Segmentation des poumons

work by Ludmilla Penarrubia (PhD)

Respiratory Intensive Care

ARDS : Acute Respiratory Distress Syndrom

ARDS patient

Mechanical ventilator

ARDS Lungs

Healthy Lungs

Images : shutterstock.com, CT4ARDS dataset, MIDP dataset

work by Ludmilla Penarrubia (PhD)

Density and aeration changes between

end-expiration and end-inspiration

Images : CT4ARDS dataset

Prerequisite :

Segmented lungs

CT image

Quantified CT

work by Ludmilla Penarrubia (PhD)

Images : CT4ARDS dataset, Doctor with Patient Cartoon.svg from Wikimedia Commons by Videoplasty.com, CC-BY-SA 4.0

currently goal

vs.

~1h30 interactive

segmentation/pair of volumes

< 1h automatic segmentation

work by Ludmilla Penarrubia (PhD)

Images : CT4ARDS dataset

ACHIEVEMENT

work by Ludmilla Penarrubia (PhD, post-doc)

< 1 min automatic segmentation

PERSPECTIVES

interactive mesh correction

Images : CT4ARDS dataset

ON GOING WORK

work by María Marquez-Sosa (PhD student)

MERCI POUR VOTRE ECOUTE !

Des questions ?

Emmanuel Roux (équipe MYRIAD)