Deep Learning
+ Healthcare
Thomas Paula
May 24, 2018 - HCPA
=
Thomas Paula
● Machine Learning Engineer and Researcher @HP
● Msc in Computer Science
● POA Machine Learning Meetup
● @tsp_thomas
● tsp.thomas@gmail.com
Who am I?
2Deep Learning and Healthcare
Why should you bother
about this Deep
Learning thing?
3
300+
4
papers regarding deep
learning and medical
image analysis
Source: Litjens, Geert, et al. "A survey on deep learning in medical image analysis." Medical image analysis 42
(2017): 60-88.
5
Hinton said: "(...) people should stop training
radiologists now, it's just completely obvious that
in five years deep learning is going to do better
than radiologists, it might be ten years".
Agenda
● Introduction
● What is Deep Learning?
● Why now?
● Deep Learning applied to healthcare
● Challenges
● Conclusion
6Deep Learning and Healthcare
Introduction
7Deep Learning and Healthcare
What is Deep Learning?
Multiple definitions, however, these definitions have in common:
● Multiple layers of processing units;
● Supervised or unsupervised learning of feature representations in each layer, with
the layers forming a hierarchy from low-level to high-level features.
8Deep Learning and Healthcare
Data is Compositional
Sources: Convolutional Deep Belief Networks. Honglak Lee, et. al. and Large
Scale Deep Learning. Jeff Dean, joint work with Google.
9Deep Learning and Healthcare
Traditional Approaches
Input Feature Extraction Classification
● Expert knowledge
● Time-consuming hand-tuning
● In industrial applications, it is 90% of the time
● Domain-specific
10
“Intuition” of what features are
11
Traditional vs Deep Learning
Traditional
DeepLearning
12
Neuroscience: inspiration
13Deep Learning and Healthcare
Source: https://www.youtube.com/watch?v=QzkMo45pcUo&feature=youtu.be&t=6m7s
Neuroscience: inspiration
● Hubel and Wiesel Nobel Prize in 1981
14Deep Learning and Healthcare
First CNN application: digits recognition
15
1998
Source: LeCun, Yann, et al. "Gradient-based learning applied to document recognition." Proceedings of the IEEE 86.11 (1998).
Deep Learning is not
new… so what happened?
16
17
Data Hardware Algorithms
Deep Learning examples
18
Images
19Deep Learning and Healthcare
Image Translation
Audio
20Deep Learning and Healthcare
Speech Synthesis (Apple) Music Recommendation (Spotify)
Text
21Deep Learning and Healthcare
Text Summarization
Games
22
Deep Learning and Healthcare
examples
23
Pneumonia Detection on Chest X-Rays with Deep Learning
24Deep Learning and Healthcare
2017
Source: Rajpurkar, Pranav, et al. "CheXNet: Radiologist-Level Pneumonia Detection on Chest X-Rays with Deep Learning." (2017).
Classification of Skin Cancer with Deep Neural Networks
25Deep Learning and Healthcare
Source: Esteva, Andre, et al. "Dermatologist-level classification of skin cancer with deep neural networks." Nature (2017)
2017
Arrhythmia Detection
26Deep Learning and Healthcare
2017
Source: Rajpurkar, Pranav, et al. "Cardiologist-level arrhythmia detection with convolutional neural networks." (2017).
An Augmented Reality Microscope for Cancer Detection
27Deep Learning and Healthcare
Source: An Augmented Reality Microscope, Google Research Blog, 2018.
An Augmented Reality Microscope for Cancer Detection
28Deep Learning and Healthcare
Source: An Augmented Reality Microscope, Google Research Blog, 2018.
An Augmented Reality Microscope for Cancer Detection
29Deep Learning and Healthcare
Source: An Augmented Reality Microscope, Google Research Blog, 2018.
What are the challenges
and opportunities?
30
Challenges/Opportunities
● ML to be used as a tool (e.g. like stats)
● Unbalanced classes (e.g. more non-cancer than cancer examples)
● Data creation and annotation
○ Cumbersome process
○ Experts don’t always agree
● Variability of the data (e.g. MRI)
● Uncertainty
● How to leverage data from different sources (e.g. images and general
patient information)?
31Deep Learning and Healthcare
Take home message #1
Machine learning / deep
learning is here to stay.
32
Take home message #1
Machine learning / deep
learning is here to stay.
33
However, be aware of the hype!
Take home message #2
Cross-area collaboration
is essential.
34
Take home message #3
Data creation and sharing is a
cornerstone for the success of AI
in healthcare.
35
Questions?
Thank you!
Thomas Paula
May 24, 2018 - HCPA

Deep learning and Healthcare

  • 1.
    Deep Learning + Healthcare ThomasPaula May 24, 2018 - HCPA =
  • 2.
    Thomas Paula ● MachineLearning Engineer and Researcher @HP ● Msc in Computer Science ● POA Machine Learning Meetup ● @tsp_thomas ● tsp.thomas@gmail.com Who am I? 2Deep Learning and Healthcare
  • 3.
    Why should youbother about this Deep Learning thing? 3
  • 4.
    300+ 4 papers regarding deep learningand medical image analysis Source: Litjens, Geert, et al. "A survey on deep learning in medical image analysis." Medical image analysis 42 (2017): 60-88.
  • 5.
    5 Hinton said: "(...)people should stop training radiologists now, it's just completely obvious that in five years deep learning is going to do better than radiologists, it might be ten years".
  • 6.
    Agenda ● Introduction ● Whatis Deep Learning? ● Why now? ● Deep Learning applied to healthcare ● Challenges ● Conclusion 6Deep Learning and Healthcare
  • 7.
  • 8.
    What is DeepLearning? Multiple definitions, however, these definitions have in common: ● Multiple layers of processing units; ● Supervised or unsupervised learning of feature representations in each layer, with the layers forming a hierarchy from low-level to high-level features. 8Deep Learning and Healthcare
  • 9.
    Data is Compositional Sources:Convolutional Deep Belief Networks. Honglak Lee, et. al. and Large Scale Deep Learning. Jeff Dean, joint work with Google. 9Deep Learning and Healthcare
  • 10.
    Traditional Approaches Input FeatureExtraction Classification ● Expert knowledge ● Time-consuming hand-tuning ● In industrial applications, it is 90% of the time ● Domain-specific 10
  • 11.
    “Intuition” of whatfeatures are 11
  • 12.
    Traditional vs DeepLearning Traditional DeepLearning 12
  • 13.
    Neuroscience: inspiration 13Deep Learningand Healthcare Source: https://www.youtube.com/watch?v=QzkMo45pcUo&feature=youtu.be&t=6m7s
  • 14.
    Neuroscience: inspiration ● Hubeland Wiesel Nobel Prize in 1981 14Deep Learning and Healthcare
  • 15.
    First CNN application:digits recognition 15 1998 Source: LeCun, Yann, et al. "Gradient-based learning applied to document recognition." Proceedings of the IEEE 86.11 (1998).
  • 16.
    Deep Learning isnot new… so what happened? 16
  • 17.
  • 18.
  • 19.
    Images 19Deep Learning andHealthcare Image Translation
  • 20.
    Audio 20Deep Learning andHealthcare Speech Synthesis (Apple) Music Recommendation (Spotify)
  • 21.
    Text 21Deep Learning andHealthcare Text Summarization
  • 22.
  • 23.
    Deep Learning andHealthcare examples 23
  • 24.
    Pneumonia Detection onChest X-Rays with Deep Learning 24Deep Learning and Healthcare 2017 Source: Rajpurkar, Pranav, et al. "CheXNet: Radiologist-Level Pneumonia Detection on Chest X-Rays with Deep Learning." (2017).
  • 25.
    Classification of SkinCancer with Deep Neural Networks 25Deep Learning and Healthcare Source: Esteva, Andre, et al. "Dermatologist-level classification of skin cancer with deep neural networks." Nature (2017) 2017
  • 26.
    Arrhythmia Detection 26Deep Learningand Healthcare 2017 Source: Rajpurkar, Pranav, et al. "Cardiologist-level arrhythmia detection with convolutional neural networks." (2017).
  • 27.
    An Augmented RealityMicroscope for Cancer Detection 27Deep Learning and Healthcare Source: An Augmented Reality Microscope, Google Research Blog, 2018.
  • 28.
    An Augmented RealityMicroscope for Cancer Detection 28Deep Learning and Healthcare Source: An Augmented Reality Microscope, Google Research Blog, 2018.
  • 29.
    An Augmented RealityMicroscope for Cancer Detection 29Deep Learning and Healthcare Source: An Augmented Reality Microscope, Google Research Blog, 2018.
  • 30.
    What are thechallenges and opportunities? 30
  • 31.
    Challenges/Opportunities ● ML tobe used as a tool (e.g. like stats) ● Unbalanced classes (e.g. more non-cancer than cancer examples) ● Data creation and annotation ○ Cumbersome process ○ Experts don’t always agree ● Variability of the data (e.g. MRI) ● Uncertainty ● How to leverage data from different sources (e.g. images and general patient information)? 31Deep Learning and Healthcare
  • 32.
    Take home message#1 Machine learning / deep learning is here to stay. 32
  • 33.
    Take home message#1 Machine learning / deep learning is here to stay. 33 However, be aware of the hype!
  • 34.
    Take home message#2 Cross-area collaboration is essential. 34
  • 35.
    Take home message#3 Data creation and sharing is a cornerstone for the success of AI in healthcare. 35
  • 36.