SlideShare una empresa de Scribd logo
1 de 41
Open Science for the Challenges
of Medical Imaging AI
Stephen R. Aylward, Ph.D.
Chair of MONAI External Advisory Board
Senior Directory of Strategic Initiatives, Kitware
The professor…
COVID face
mask
…and his students
Bob (the “perfect” student)
Bob
graduates
$
$
$
I cannot find
Bob’s code.
Bob’s code
won’t compile.
I cannot
replicate Bob’s
results.
I emailed Bob a
question, but he
hasn’t responded.
Open Science is the Answer!
9
Why is deep learning succeeding?
● Performance
● Open-ness
10
-- Forbes.com
Deep Learning Success: Performance
< Left as an exercise for the audience >
11
Deep Learning Success: Open-ness
12
● Open science is pervasive in deep learning
○ Open access publications: arXiv
○ Open access data: ImageNet
○ Open access algorithms: Open source: PyTorch, MONAI
Why is Open Science Important?
13
“DOUBT EVERYTHING and only believe
in those things that are evidently true
(reproducible).”
-- Descartes 1637
Discourse on the (Scientific) Method
Failure of Open Science
Nature (March 2012)
-- Glenn Begley: Head of cancer research at pharma giant Amgen
-- Lee M. Ellis: Cancer researcher at the University of Texas
● Identified 53 'landmark' publications.
● Sought to double-check the findings before building on them for
drug development.
● Result: 47 of the 53 could not be replicated.
Three Challenges to Open Science
1) Sense of competition
2) Inadequacy of journal articles
>>
3) Priorities: I have more important things to do…
Benefits of Open Science Outweigh the Costs
18
● Funding (Tenure)
○ VTK
■ NIH R01 $3.4M
● Citations
○ An Object-Oriented Approach To 3D Graphics (Schroeder) Object-oriented modeling and
design (Lorensen)
■ 13,000+ Citations
● Impact
○ CMake, VTK, ParaView, 3D Slicer, ITK
■ 150,000 downloads per month
Open Science in Deep Learning
19
Medical Open Network for A. I. (MONAI):
Goal: Accelerate the pace of research and development
by providing a common software foundation and a vibrant
community for medical imaging deep learning.
■ Began as a collaboration between Nvidia and KCL
■ Freely available and community-supported
■ PyTorch-based
■ Optimized for medical imaging
■ Reference implementation of best practices
Accelerate Pace of Research and Innovation
With a Common Foundation
Data
Augmentation
Neural
Network
Loss
FunctionData
Samples
MONAI
● Integrate rather than compete
● Build a community through value
Current Conditions
● Many options
● Incompatible interfaces and formats
● Extended learning curves
Validation
Data Evaluation
NiftyNet
(KCL)
DeepNeuro
(Harvard)
DLTK
(ICL)
Clara Train
(NVIDIA)
End2End workflow
facilitated by MONAI
. . . . . .
Primary focus of
MONAI
Linkage with MONAI
Why is MONAI Needed?
• Biomedical applications have specific requirements
• Image modalities (MR, CT, US, etc.) require specific processing methods
• Data formats (DICOM, NIfTI, etc.) and meta-data (e.g., voxel spacing)
require special support
• Certain network architectures are designed for, or are highly suitable for,
biomedical applications
• Data augmenting, sample size limitations, annotation uncertainties, etc. are
domain specific
Why is MONAI Needed?
• Reproducibility is vital to clinical decision support
• Reduce re-implementation
• Provide baseline implementations
• Demonstrate best practices
• Stand on the shoulders of giants
How Does MONAI Address These Needs?
• MONAI provides flexible, Pytorch-compatible methods:
•Data loading and handling library for biomedical file types
•Data transforms to process, regularize, and augment image data
•Library of general-purpose network and loss function definitions
•Numerous metrics for evaluating results during and after training
•Jupyter Notebooks and Ignite Workflows to simplify training
•Efficient and fast data loaders for fast batch generation
•Support for multi-GPU and multi-node multi-GPU training
Open Science: Reproducibility
• MONAI aims to help researchers write reproducible experiments more easily
• Reusable and accessible code in a modular design (pick-and-choose)
• Accessible data
• Ready-made training paradigms as well as an extensible platform
• Permissive licensing
• Reference implementations of published methods and experiments with
proven replicable performance
Ease-of-use Example
net = monai.networks.nets.UNet(
dimensions=2, # 2 or 3 for a 2D or 3D network
in_channels=1, # number of input channels
out_channels=1, # number of output channels
channels=[8, 16, 32], # channel counts for layers
strides=[2, 2] # strides for mid layers
)
2D UNet network
• 2 hidden layers: outputs has 8 channels, and the bottom (bottleneck) layer has outputs with
32 channels
• Stride values state the stride for the initial convolution, ie. downsampling in down path and
upsampling in up path
Liaison with the community:
Recommend policies and priorities to development team
Working Groups of MONAI
1. IMAGING I/O – Stephen Aylward
2. DATA DIVERSITY – Brad Genereaux
3. REPRODUCIBILITY – Lena Maier-Hein
4. TRANSFORMATIONS – Jorge Cordoso
5. FEDERATED LEARNING – Jayashree Kalapathy
6. ADVANCED RESEARCH – Paul Jaeger
7. COMMUNITY ADOPTION – Prerna Dogra
Goal: Define how data is read into and written out from memory in MONAI.
I/O Working Group
Tasks
#1: Support reading and writing common research image file formats: NRRD, Nifti, DICOM (objects), JPG,
PNG, TIFF, MetaIO, ...
#2: Simplify interfacing MONAI with clinical systems such as PACS and health records systems: DICOM, HL7,
...
Members
➢ Stephen Aylward (Kitware)
➢ Marco Nolden (DKFZ)
➢ Jayashree Kalpathy-Cramer (MGH)
➢ Brad Genereaux (Nvidia)
➢ Ben Murray (KCL)
➢ Wenqi Li (Nvidia)
➢ Jorge Cardoso (KCL)
➢ Prerna Dogra (Nvidia)
Proposal #1: Create a single reader that handles a diversity of file formats
in a consistent manner and that user-defined “experiments” can
override to use a specific reader when I/O reproducibility is a concern.
Solution #1: Default image loader uses ITK Python. Scripts can specify
alternative loaders. Available starting in MONAI v0.3.
Suggestions for next steps and alternatives:
Stephen.Aylward@Kitware.com
MONAI DATA WORKING GROUP
Data: Images, Biomarkers, Demographics, Outputs and Outcomes
Working Group Homepage:
https://github.com/Project-
MONAI/MONAI/wiki/Data-Working-Group
Reach Out if you have any feedback,
questions, or thoughts on data
structures to be processed in MONAI.
We want to hear from you for your
preferred data types, data you need to
capture, and where the challenges are!
MONAI WORKING GROUP:
EVALUATION, REPRODUCIBILITY, BENCHMARKS
Working Group Homepage: https://github.com/Project-MONAI/MONAI/wiki/Evaluation,-Reproducibility,-Benchmarks-Working-Group-Meeting-Notes
Lena Maier-Hein
Working Group Lead
German Cancer Research Center
(DKFZ)
Annika Reinke
Lead of task force
Benchmarking
German Cancer Research Center
(DKFZ)
Jens Petersen
German Cancer Research Center
(DKFZ)
M. Jorge Cardoso
King’s College London (KCL)
S. Kevin Zhou
Working Group Lead
Chinese Academy of Science (CAS)
David Zimmerer
German Cancer Research Center
(DKFZ)
Carole Sudre
Lead of task force
Metrics
University College London (UCL)
Nicola Rieke
NVIDIA GmbH
Technical University of Munich (TUM)
Michela Antonelli
Lead of task force
Quick data access
King’s College London (KCL)
Dan Tudosiu
King’s College London (KCL)
Jayashree Kalpathy-Cramer
Harvard Medical School (MGH)
The core goal of the Working Group is to provide the software infrastructure and tools for
quality-controlled validation and benchmarking of medical image analysis methods.
We will address open technical challenges and research questions related to a variety of topics,
ranging from best practices to performance aspects and implementation efficiency.
Mission
MONAI WORKING GROUP:
EVALUATION, REPRODUCIBILITY, BENCHMARKS
You have further feedback or questions? Use the slack channel or contact us via monai-evaluation-benchmarking-wg@dkfz.de
Your feedback is important!
What do you want from our MONAI Working Group?
• What are current problems with respect to metrics,
benchmarking and challenges?
• What should we prioritize?
• Best practice recommendations for metrics?
• Easy access to existing data sets?
• …
Let your voice be heard:
https://www.surveylegend.com/s/2mes
Word cloud created from the Working Group’s meeting notes.
MONAI WORKING GROUP:
TRANSFORMATION & AUGMENTATION
Medical Specific Transformations
- LoadNifti | Spacing | Orientation
- RandGaussianNoise |Normalize Intensirt
- Rand2DElastic | Rand3DElastic
Fused Spatial Transforms & GPU Optimization
- Affine Transform
- Random sampling: Class balanced fixed Ratio
- Deterministic training controlled by setting random seed
Generic | Vanilla |Dictionary-based Transforms
LoadNifti AsChannelFirst Scale Intensity ConcatItems
AsChannelFirst
AsChannelFirst
Scale Intensity
Scale Intensity Network
Nifty Images
Brain
Window
Subdural Window
Bone
Window
Load Image from File Make 2 Copies Different ranges &Scale Concat Together
MONAI WORKING GROUP:
TRANSFORMATION & AUGMENTATION
3rd Party OSS Packages & MONAI adapter Tools
- Interoperability with other open source packages
- Accommodate different data for 3rd party Transforms
- Utility Transforms: ToTensor, ToNumpy, SqueeseDim
- BatchGenerator
- TorchIO
- Rising
Post-Processing & Integrate Third Party Transforms
Compose all Transforms
Goal: Support federated learning and other collaborative learning approaches.
Federated Learning Working Group
Tasks
#1: Develop reference implementations of MONAI FL with Clara-FL
#2: Develop example implementations of MONAI train with Substra/PySyft/??
#3: Perform a real-life federated learning experiment with interested members of community
Members
➢ Daniel Rubin (Stanford)
➢ Jayashree Kalpathy-Cramer (MGH)
➢ Marco Nolden (DKFZ)
➢ Stephen Aylward (Kitware)
➢ Holger Roth (Nvidia)
➢ Wenqi Li (Nvidia)
➢ Jonas Scherer (DKFZ)
➢ Andriy Myronenko (Nvidia)
➢ Jorge Cardoso (KCL)
➢ Prerna Dogra (Nvidia)
➢ Nicola Rieke (Nvidia)
➢ Shadi Albarqouni (TUM)
Proposal #1: Create interfaces to allow external FL packages to
connect to MONAI trainers
jkalpathy-cramer@mgh.Harvard.edu
rubin@stanford.edu
● Goal: Guide the integration of new methodologies and tools into MONAI by focusing on
community-driven progress and good scientific practice
Research Working Group
Previous Work:
Future Plans:
Before filling MONAI with Research Content,
We focused on establishing a systematic process
behind integration of methodologies in MONAI including:
• Clear definition and categorization of methodologies:
Components / Tutorials / Reserach Contributions
• Code location and entry points via the monai.io webpage
• Internal structure / taxnomoy ensuring scalability
• Guidelines for contributions by the community
• Prioritize and guide implementation of new methods such
as Object Detection, Registration, etc.
• Contribute towards an online benchmarking platform in
MONAI including standardized datasets, Baselines/Model-
Zoo and evaluations.
Members:
• Paul Jaeger (DKFZ)
• Dan Tudosiu (KCL)
• Dong Yang (Nvidia)
• Michael Baumgartner (DKFZ)
• Jorge Cardoso (KCL)
• Holger Roth (Nvidia)
• Ziyue Xu (Nvidia)
• Wentao Zhu (Nvidia)
• Chayanin Tangwiriyasakul (KCL)
• Ben Murray (KCL)
For feedback/questions
please contact p.jaeger@dkfz.de
See also our site on the MONAI wiki
36
Goal: Establish MONAI as a common software foundation that the Medical
Imaging research and development community can build upon.
Community Outreach Working Group
Tasks:
• Developing technical training and onboarding content
• Organize workshops and boot camps to educate and engage
• Define the infrastructure and processes needed to support and
grow a transparent, honest, and open community
• Establishing best practices for giving appropriate recognition to
those who are promoting and making significant contributions
back to MONAI
Members:
• Michael Zephyr (Nvidia)
• Prerna Dogra (Nvidia)
• Stephen Aylward (Kitware)
• Julia Levites (Nvidia)
• Paul Jaeger (DKFZ)
• Wenqi Li (Nvidia)
• Nicola Rieke (Nvidia)
How to Engage:
• See our Community Onboarding guide:
https://github.com/Project-MONAI/MONAI#community
• For feedback or questions, contact mzephyr@nvidia.com
Understand the different options you have to engage with the MONAI
community, developers, and working groups
How to Engage with MONAI
Where to go to get started with MONAI
• Getting Started (Installation, Examples, Demos, etc.) https://monai.io/start.html
• GitHub
• Community Guide: https://github.com/Project-MONAI/MONAI#community
• Contributing Guide: https://github.com/Project-MONAI/MONAI#contributing
• Use GitHub Issues for Small Concrete Tasks and use the community tag.
• Looking to get started contributing? Look at our Good First Issue tag! https://github.com/Project-
MONAI/MONAI/labels/good%20first%20issue
• PyTorch Forums. Tag @monai or see the MONAI user page. https://discuss.pytorch.org/u/MONAI/
• Stack Overflow. See existing tagged questions or create your own: https://stackoverflow.com/questions/tagged/monai
• Follow us on Twitter for the latest updates on releases, research, demos and conference engagements. https://twitter.com/ProjectMONAI
• Join our Slack Channel. Fill out the Google Form here: https://forms.gle/QTxJq3hFictp31UM9
• Transitioning away from Google Groups. Instead use one of the above options.
• For feedback or questions, contact mzephyr@nvidia.com
MONAI IS A GROWING COMMUNITY
Contributors: 1.4 K+ GitHub stars & growing engagement
BOOTCAMP – IN NUMBERS
A LOT OF INTEREST IN THE COMMUNITY!
• Number of applicants: 563 attendance applications
• Accepted participants with cluster access (60)
• Additionally other participants “observers”
(140)
• From 40 different countries:
Australia, Austria, Belgium, China, Cyprus,
Czechia, Egypt, Ethiopia, France, Ghana,
Germany, Greece, Guatemala, Hong Kong,
India, Israel, Iran, Malta, Mexico, Nepal,
Netherlands, Norway, Oman, Peru, Poland,
Portugal, Saudi Arabia, Slovenia, South Korea,
Spain, Sweden, Switzerland, Turkey, United
Arab emirates, United Kingdom, United States
of America
A truly global event!
MONAI Demonstration Applications
G. Wang, X. Liu, C. Li, Z. Xu, J. Ruan, H. Zhu, T. Meng, K. Li, N. Huang,
S. Zhang. (2020) "A Noise-robust Framework for Automatic
Segmentation of COVID-19 Pneumonia Lesions from CT Images."
IEEE Transactions on Medical Imaging. 2020.
DOI: 10.1109/TMI.2020.3000314
MONAI Jupyter Notebook Tutorial: Spleen 3D segmentation
Conclusion
Stephen R. Aylward, Ph.D.
Chair of MONAI External Advisory Board
Senior Directory of Strategic Initiatives, Kitware

Más contenido relacionado

La actualidad más candente

Game Engine Overview
Game Engine OverviewGame Engine Overview
Game Engine Overview
Sharad Mitra
 

La actualidad más candente (20)

Emotion detection using cnn.pptx
Emotion detection using cnn.pptxEmotion detection using cnn.pptx
Emotion detection using cnn.pptx
 
flutter intro.pptx
flutter intro.pptxflutter intro.pptx
flutter intro.pptx
 
ECS (Part 1/3) - Introduction to Data-Oriented Design
ECS (Part 1/3) - Introduction to Data-Oriented DesignECS (Part 1/3) - Introduction to Data-Oriented Design
ECS (Part 1/3) - Introduction to Data-Oriented Design
 
MONAI: Medical imaging AI for data scientists and developers @ 3D Slicer Proj...
MONAI: Medical imaging AI for data scientists and developers @ 3D Slicer Proj...MONAI: Medical imaging AI for data scientists and developers @ 3D Slicer Proj...
MONAI: Medical imaging AI for data scientists and developers @ 3D Slicer Proj...
 
Big Data Analytics Lab File
Big Data Analytics Lab FileBig Data Analytics Lab File
Big Data Analytics Lab File
 
Fast stone capture key
Fast stone capture keyFast stone capture key
Fast stone capture key
 
Scene Graphs & Component Based Game Engines
Scene Graphs & Component Based Game EnginesScene Graphs & Component Based Game Engines
Scene Graphs & Component Based Game Engines
 
Intrinsics: Low-level engine development with Burst - Unite Copenhagen 2019
Intrinsics: Low-level engine development with Burst - Unite Copenhagen 2019 Intrinsics: Low-level engine development with Burst - Unite Copenhagen 2019
Intrinsics: Low-level engine development with Burst - Unite Copenhagen 2019
 
Introduction to DirectX 11
Introduction to DirectX 11Introduction to DirectX 11
Introduction to DirectX 11
 
Next generation open world gaming (Latest Games)
Next generation open world gaming (Latest Games)Next generation open world gaming (Latest Games)
Next generation open world gaming (Latest Games)
 
Mongo db workshop # 01
Mongo db workshop # 01Mongo db workshop # 01
Mongo db workshop # 01
 
Big data-cheat-sheet
Big data-cheat-sheetBig data-cheat-sheet
Big data-cheat-sheet
 
Android: the Single Activity, Multiple Fragments pattern | One Activity to ru...
Android: the Single Activity, Multiple Fragments pattern | One Activity to ru...Android: the Single Activity, Multiple Fragments pattern | One Activity to ru...
Android: the Single Activity, Multiple Fragments pattern | One Activity to ru...
 
Game Engine Overview
Game Engine OverviewGame Engine Overview
Game Engine Overview
 
Distributed computing with Spark 2.x
Distributed computing with Spark 2.xDistributed computing with Spark 2.x
Distributed computing with Spark 2.x
 
Dx11 performancereloaded
Dx11 performancereloadedDx11 performancereloaded
Dx11 performancereloaded
 
Forts and Fights Scaling Performance on Unreal Engine*
Forts and Fights Scaling Performance on Unreal Engine*Forts and Fights Scaling Performance on Unreal Engine*
Forts and Fights Scaling Performance on Unreal Engine*
 
DX12 & Vulkan: Dawn of a New Generation of Graphics APIs
DX12 & Vulkan: Dawn of a New Generation of Graphics APIsDX12 & Vulkan: Dawn of a New Generation of Graphics APIs
DX12 & Vulkan: Dawn of a New Generation of Graphics APIs
 
Map Reduce
Map ReduceMap Reduce
Map Reduce
 
An Introduction to Apache Hadoop Yarn
An Introduction to Apache Hadoop YarnAn Introduction to Apache Hadoop Yarn
An Introduction to Apache Hadoop Yarn
 

Similar a MONAI and Open Science for Medical Imaging Deep Learning: SIPAIM 2020

Deep Learning for Recommender Systems
Deep Learning for Recommender SystemsDeep Learning for Recommender Systems
Deep Learning for Recommender Systems
Marcel Kurovski
 
Deep Learning for Recommender Systems
Deep Learning for Recommender SystemsDeep Learning for Recommender Systems
Deep Learning for Recommender Systems
inovex GmbH
 

Similar a MONAI and Open Science for Medical Imaging Deep Learning: SIPAIM 2020 (20)

Deep Learning for Recommender Systems
Deep Learning for Recommender SystemsDeep Learning for Recommender Systems
Deep Learning for Recommender Systems
 
Deep Learning for Recommender Systems
Deep Learning for Recommender SystemsDeep Learning for Recommender Systems
Deep Learning for Recommender Systems
 
Novi sad ai event 1-2018
Novi sad ai event 1-2018Novi sad ai event 1-2018
Novi sad ai event 1-2018
 
Docker in Open Science Data Analysis Challenges by Bruce Hoff
Docker in Open Science Data Analysis Challenges by Bruce HoffDocker in Open Science Data Analysis Challenges by Bruce Hoff
Docker in Open Science Data Analysis Challenges by Bruce Hoff
 
Ontology-based data access: why it is so cool!
Ontology-based data access: why it is so cool!Ontology-based data access: why it is so cool!
Ontology-based data access: why it is so cool!
 
AI and Deep Learning
AI and Deep Learning AI and Deep Learning
AI and Deep Learning
 
SC6 Workshop 1: Big Data Europe platform requirements and draft architecture:...
SC6 Workshop 1: Big Data Europe platform requirements and draft architecture:...SC6 Workshop 1: Big Data Europe platform requirements and draft architecture:...
SC6 Workshop 1: Big Data Europe platform requirements and draft architecture:...
 
"Deep Learning Beyond Cats and Cars: Developing a Real-life DNN-based Embedde...
"Deep Learning Beyond Cats and Cars: Developing a Real-life DNN-based Embedde..."Deep Learning Beyond Cats and Cars: Developing a Real-life DNN-based Embedde...
"Deep Learning Beyond Cats and Cars: Developing a Real-life DNN-based Embedde...
 
Using Deep Learning to do Real-Time Scoring in Practical Applications - 2015-...
Using Deep Learning to do Real-Time Scoring in Practical Applications - 2015-...Using Deep Learning to do Real-Time Scoring in Practical Applications - 2015-...
Using Deep Learning to do Real-Time Scoring in Practical Applications - 2015-...
 
Synergy of Human and Artificial Intelligence in Software Engineering
Synergy of Human and Artificial Intelligence in Software EngineeringSynergy of Human and Artificial Intelligence in Software Engineering
Synergy of Human and Artificial Intelligence in Software Engineering
 
Using Crowdsourced Images to Create Image Recognition Models with Analytics Z...
Using Crowdsourced Images to Create Image Recognition Models with Analytics Z...Using Crowdsourced Images to Create Image Recognition Models with Analytics Z...
Using Crowdsourced Images to Create Image Recognition Models with Analytics Z...
 
Applications for Social Networking Strategies in an Agency Context: Exploitin...
Applications for Social Networking Strategies in an Agency Context: Exploitin...Applications for Social Networking Strategies in an Agency Context: Exploitin...
Applications for Social Networking Strategies in an Agency Context: Exploitin...
 
Deep Learning Jeff-Shomaker_1-20-17_Final_
Deep Learning Jeff-Shomaker_1-20-17_Final_Deep Learning Jeff-Shomaker_1-20-17_Final_
Deep Learning Jeff-Shomaker_1-20-17_Final_
 
Software Analytics: Data Analytics for Software Engineering
Software Analytics: Data Analytics for Software EngineeringSoftware Analytics: Data Analytics for Software Engineering
Software Analytics: Data Analytics for Software Engineering
 
MediaEval 2016 - COSMIR and the OpenMIC Challenge: A Plan for Sustainable Mus...
MediaEval 2016 - COSMIR and the OpenMIC Challenge: A Plan for Sustainable Mus...MediaEval 2016 - COSMIR and the OpenMIC Challenge: A Plan for Sustainable Mus...
MediaEval 2016 - COSMIR and the OpenMIC Challenge: A Plan for Sustainable Mus...
 
How a Social Knowledge Graph Improves Remote Working by Capturing Context fro...
How a Social Knowledge Graph Improves Remote Working by Capturing Context fro...How a Social Knowledge Graph Improves Remote Working by Capturing Context fro...
How a Social Knowledge Graph Improves Remote Working by Capturing Context fro...
 
Software Analytics: Towards Software Mining that Matters (2014)
Software Analytics:Towards Software Mining that Matters (2014)Software Analytics:Towards Software Mining that Matters (2014)
Software Analytics: Towards Software Mining that Matters (2014)
 
Applications for Social Networking Strategies in an Agency Context
Applications for Social Networking Strategies in an Agency ContextApplications for Social Networking Strategies in an Agency Context
Applications for Social Networking Strategies in an Agency Context
 
Platform for Big Data Analytics and Visual Analytics: CSIRO use cases. Februa...
Platform for Big Data Analytics and Visual Analytics: CSIRO use cases. Februa...Platform for Big Data Analytics and Visual Analytics: CSIRO use cases. Februa...
Platform for Big Data Analytics and Visual Analytics: CSIRO use cases. Februa...
 
Semantic Web research anno 2006:main streams, popular falacies, current statu...
Semantic Web research anno 2006:main streams, popular falacies, current statu...Semantic Web research anno 2006:main streams, popular falacies, current statu...
Semantic Web research anno 2006:main streams, popular falacies, current statu...
 

Último

🌹Attapur⬅️ Vip Call Girls Hyderabad 📱9352852248 Book Well Trand Call Girls In...
🌹Attapur⬅️ Vip Call Girls Hyderabad 📱9352852248 Book Well Trand Call Girls In...🌹Attapur⬅️ Vip Call Girls Hyderabad 📱9352852248 Book Well Trand Call Girls In...
🌹Attapur⬅️ Vip Call Girls Hyderabad 📱9352852248 Book Well Trand Call Girls In...
Call Girls In Delhi Whatsup 9873940964 Enjoy Unlimited Pleasure
 
Call Girl In Pune 👉 Just CALL ME: 9352988975 💋 Call Out Call Both With High p...
Call Girl In Pune 👉 Just CALL ME: 9352988975 💋 Call Out Call Both With High p...Call Girl In Pune 👉 Just CALL ME: 9352988975 💋 Call Out Call Both With High p...
Call Girl In Pune 👉 Just CALL ME: 9352988975 💋 Call Out Call Both With High p...
chetankumar9855
 
Russian Call Girls Lucknow Just Call 👉👉7877925207 Top Class Call Girl Service...
Russian Call Girls Lucknow Just Call 👉👉7877925207 Top Class Call Girl Service...Russian Call Girls Lucknow Just Call 👉👉7877925207 Top Class Call Girl Service...
Russian Call Girls Lucknow Just Call 👉👉7877925207 Top Class Call Girl Service...
adilkhan87451
 

Último (20)

Models Call Girls In Hyderabad 9630942363 Hyderabad Call Girl & Hyderabad Esc...
Models Call Girls In Hyderabad 9630942363 Hyderabad Call Girl & Hyderabad Esc...Models Call Girls In Hyderabad 9630942363 Hyderabad Call Girl & Hyderabad Esc...
Models Call Girls In Hyderabad 9630942363 Hyderabad Call Girl & Hyderabad Esc...
 
🌹Attapur⬅️ Vip Call Girls Hyderabad 📱9352852248 Book Well Trand Call Girls In...
🌹Attapur⬅️ Vip Call Girls Hyderabad 📱9352852248 Book Well Trand Call Girls In...🌹Attapur⬅️ Vip Call Girls Hyderabad 📱9352852248 Book Well Trand Call Girls In...
🌹Attapur⬅️ Vip Call Girls Hyderabad 📱9352852248 Book Well Trand Call Girls In...
 
Call Girls Service Jaipur {8445551418} ❤️VVIP BHAWNA Call Girl in Jaipur Raja...
Call Girls Service Jaipur {8445551418} ❤️VVIP BHAWNA Call Girl in Jaipur Raja...Call Girls Service Jaipur {8445551418} ❤️VVIP BHAWNA Call Girl in Jaipur Raja...
Call Girls Service Jaipur {8445551418} ❤️VVIP BHAWNA Call Girl in Jaipur Raja...
 
Low Rate Call Girls Bangalore {7304373326} ❤️VVIP NISHA Call Girls in Bangalo...
Low Rate Call Girls Bangalore {7304373326} ❤️VVIP NISHA Call Girls in Bangalo...Low Rate Call Girls Bangalore {7304373326} ❤️VVIP NISHA Call Girls in Bangalo...
Low Rate Call Girls Bangalore {7304373326} ❤️VVIP NISHA Call Girls in Bangalo...
 
Call Girls Rishikesh Just Call 9667172968 Top Class Call Girl Service Available
Call Girls Rishikesh Just Call 9667172968 Top Class Call Girl Service AvailableCall Girls Rishikesh Just Call 9667172968 Top Class Call Girl Service Available
Call Girls Rishikesh Just Call 9667172968 Top Class Call Girl Service Available
 
Call Girl In Pune 👉 Just CALL ME: 9352988975 💋 Call Out Call Both With High p...
Call Girl In Pune 👉 Just CALL ME: 9352988975 💋 Call Out Call Both With High p...Call Girl In Pune 👉 Just CALL ME: 9352988975 💋 Call Out Call Both With High p...
Call Girl In Pune 👉 Just CALL ME: 9352988975 💋 Call Out Call Both With High p...
 
Best Rate (Guwahati ) Call Girls Guwahati ⟟ 8617370543 ⟟ High Class Call Girl...
Best Rate (Guwahati ) Call Girls Guwahati ⟟ 8617370543 ⟟ High Class Call Girl...Best Rate (Guwahati ) Call Girls Guwahati ⟟ 8617370543 ⟟ High Class Call Girl...
Best Rate (Guwahati ) Call Girls Guwahati ⟟ 8617370543 ⟟ High Class Call Girl...
 
Mumbai ] (Call Girls) in Mumbai 10k @ I'm VIP Independent Escorts Girls 98333...
Mumbai ] (Call Girls) in Mumbai 10k @ I'm VIP Independent Escorts Girls 98333...Mumbai ] (Call Girls) in Mumbai 10k @ I'm VIP Independent Escorts Girls 98333...
Mumbai ] (Call Girls) in Mumbai 10k @ I'm VIP Independent Escorts Girls 98333...
 
Coimbatore Call Girls in Coimbatore 7427069034 genuine Escort Service Girl 10...
Coimbatore Call Girls in Coimbatore 7427069034 genuine Escort Service Girl 10...Coimbatore Call Girls in Coimbatore 7427069034 genuine Escort Service Girl 10...
Coimbatore Call Girls in Coimbatore 7427069034 genuine Escort Service Girl 10...
 
Trichy Call Girls Book Now 9630942363 Top Class Trichy Escort Service Available
Trichy Call Girls Book Now 9630942363 Top Class Trichy Escort Service AvailableTrichy Call Girls Book Now 9630942363 Top Class Trichy Escort Service Available
Trichy Call Girls Book Now 9630942363 Top Class Trichy Escort Service Available
 
Jogeshwari ! Call Girls Service Mumbai - 450+ Call Girl Cash Payment 90042684...
Jogeshwari ! Call Girls Service Mumbai - 450+ Call Girl Cash Payment 90042684...Jogeshwari ! Call Girls Service Mumbai - 450+ Call Girl Cash Payment 90042684...
Jogeshwari ! Call Girls Service Mumbai - 450+ Call Girl Cash Payment 90042684...
 
9630942363 Genuine Call Girls In Ahmedabad Gujarat Call Girls Service
9630942363 Genuine Call Girls In Ahmedabad Gujarat Call Girls Service9630942363 Genuine Call Girls In Ahmedabad Gujarat Call Girls Service
9630942363 Genuine Call Girls In Ahmedabad Gujarat Call Girls Service
 
8980367676 Call Girls In Ahmedabad Escort Service Available 24×7 In Ahmedabad
8980367676 Call Girls In Ahmedabad Escort Service Available 24×7 In Ahmedabad8980367676 Call Girls In Ahmedabad Escort Service Available 24×7 In Ahmedabad
8980367676 Call Girls In Ahmedabad Escort Service Available 24×7 In Ahmedabad
 
Top Quality Call Girl Service Kalyanpur 6378878445 Available Call Girls Any Time
Top Quality Call Girl Service Kalyanpur 6378878445 Available Call Girls Any TimeTop Quality Call Girl Service Kalyanpur 6378878445 Available Call Girls Any Time
Top Quality Call Girl Service Kalyanpur 6378878445 Available Call Girls Any Time
 
All Time Service Available Call Girls Marine Drive 📳 9820252231 For 18+ VIP C...
All Time Service Available Call Girls Marine Drive 📳 9820252231 For 18+ VIP C...All Time Service Available Call Girls Marine Drive 📳 9820252231 For 18+ VIP C...
All Time Service Available Call Girls Marine Drive 📳 9820252231 For 18+ VIP C...
 
Russian Call Girls Lucknow Just Call 👉👉7877925207 Top Class Call Girl Service...
Russian Call Girls Lucknow Just Call 👉👉7877925207 Top Class Call Girl Service...Russian Call Girls Lucknow Just Call 👉👉7877925207 Top Class Call Girl Service...
Russian Call Girls Lucknow Just Call 👉👉7877925207 Top Class Call Girl Service...
 
(Low Rate RASHMI ) Rate Of Call Girls Jaipur ❣ 8445551418 ❣ Elite Models & Ce...
(Low Rate RASHMI ) Rate Of Call Girls Jaipur ❣ 8445551418 ❣ Elite Models & Ce...(Low Rate RASHMI ) Rate Of Call Girls Jaipur ❣ 8445551418 ❣ Elite Models & Ce...
(Low Rate RASHMI ) Rate Of Call Girls Jaipur ❣ 8445551418 ❣ Elite Models & Ce...
 
VIP Hyderabad Call Girls Bahadurpally 7877925207 ₹5000 To 25K With AC Room 💚😋
VIP Hyderabad Call Girls Bahadurpally 7877925207 ₹5000 To 25K With AC Room 💚😋VIP Hyderabad Call Girls Bahadurpally 7877925207 ₹5000 To 25K With AC Room 💚😋
VIP Hyderabad Call Girls Bahadurpally 7877925207 ₹5000 To 25K With AC Room 💚😋
 
Saket * Call Girls in Delhi - Phone 9711199012 Escorts Service at 6k to 50k a...
Saket * Call Girls in Delhi - Phone 9711199012 Escorts Service at 6k to 50k a...Saket * Call Girls in Delhi - Phone 9711199012 Escorts Service at 6k to 50k a...
Saket * Call Girls in Delhi - Phone 9711199012 Escorts Service at 6k to 50k a...
 
Call Girls Madurai Just Call 9630942363 Top Class Call Girl Service Available
Call Girls Madurai Just Call 9630942363 Top Class Call Girl Service AvailableCall Girls Madurai Just Call 9630942363 Top Class Call Girl Service Available
Call Girls Madurai Just Call 9630942363 Top Class Call Girl Service Available
 

MONAI and Open Science for Medical Imaging Deep Learning: SIPAIM 2020

  • 1. Open Science for the Challenges of Medical Imaging AI Stephen R. Aylward, Ph.D. Chair of MONAI External Advisory Board Senior Directory of Strategic Initiatives, Kitware
  • 3. …and his students Bob (the “perfect” student)
  • 8. I emailed Bob a question, but he hasn’t responded.
  • 9. Open Science is the Answer! 9
  • 10. Why is deep learning succeeding? ● Performance ● Open-ness 10 -- Forbes.com
  • 11. Deep Learning Success: Performance < Left as an exercise for the audience > 11
  • 12. Deep Learning Success: Open-ness 12 ● Open science is pervasive in deep learning ○ Open access publications: arXiv ○ Open access data: ImageNet ○ Open access algorithms: Open source: PyTorch, MONAI
  • 13. Why is Open Science Important? 13 “DOUBT EVERYTHING and only believe in those things that are evidently true (reproducible).” -- Descartes 1637 Discourse on the (Scientific) Method
  • 14. Failure of Open Science Nature (March 2012) -- Glenn Begley: Head of cancer research at pharma giant Amgen -- Lee M. Ellis: Cancer researcher at the University of Texas ● Identified 53 'landmark' publications. ● Sought to double-check the findings before building on them for drug development. ● Result: 47 of the 53 could not be replicated.
  • 15. Three Challenges to Open Science 1) Sense of competition
  • 16. 2) Inadequacy of journal articles >>
  • 17. 3) Priorities: I have more important things to do…
  • 18. Benefits of Open Science Outweigh the Costs 18 ● Funding (Tenure) ○ VTK ■ NIH R01 $3.4M ● Citations ○ An Object-Oriented Approach To 3D Graphics (Schroeder) Object-oriented modeling and design (Lorensen) ■ 13,000+ Citations ● Impact ○ CMake, VTK, ParaView, 3D Slicer, ITK ■ 150,000 downloads per month
  • 19. Open Science in Deep Learning 19 Medical Open Network for A. I. (MONAI): Goal: Accelerate the pace of research and development by providing a common software foundation and a vibrant community for medical imaging deep learning. ■ Began as a collaboration between Nvidia and KCL ■ Freely available and community-supported ■ PyTorch-based ■ Optimized for medical imaging ■ Reference implementation of best practices
  • 20. Accelerate Pace of Research and Innovation With a Common Foundation Data Augmentation Neural Network Loss FunctionData Samples MONAI ● Integrate rather than compete ● Build a community through value Current Conditions ● Many options ● Incompatible interfaces and formats ● Extended learning curves Validation Data Evaluation NiftyNet (KCL) DeepNeuro (Harvard) DLTK (ICL) Clara Train (NVIDIA) End2End workflow facilitated by MONAI . . . . . . Primary focus of MONAI Linkage with MONAI
  • 21. Why is MONAI Needed? • Biomedical applications have specific requirements • Image modalities (MR, CT, US, etc.) require specific processing methods • Data formats (DICOM, NIfTI, etc.) and meta-data (e.g., voxel spacing) require special support • Certain network architectures are designed for, or are highly suitable for, biomedical applications • Data augmenting, sample size limitations, annotation uncertainties, etc. are domain specific
  • 22. Why is MONAI Needed? • Reproducibility is vital to clinical decision support • Reduce re-implementation • Provide baseline implementations • Demonstrate best practices • Stand on the shoulders of giants
  • 23. How Does MONAI Address These Needs? • MONAI provides flexible, Pytorch-compatible methods: •Data loading and handling library for biomedical file types •Data transforms to process, regularize, and augment image data •Library of general-purpose network and loss function definitions •Numerous metrics for evaluating results during and after training •Jupyter Notebooks and Ignite Workflows to simplify training •Efficient and fast data loaders for fast batch generation •Support for multi-GPU and multi-node multi-GPU training
  • 24. Open Science: Reproducibility • MONAI aims to help researchers write reproducible experiments more easily • Reusable and accessible code in a modular design (pick-and-choose) • Accessible data • Ready-made training paradigms as well as an extensible platform • Permissive licensing • Reference implementations of published methods and experiments with proven replicable performance
  • 25. Ease-of-use Example net = monai.networks.nets.UNet( dimensions=2, # 2 or 3 for a 2D or 3D network in_channels=1, # number of input channels out_channels=1, # number of output channels channels=[8, 16, 32], # channel counts for layers strides=[2, 2] # strides for mid layers ) 2D UNet network • 2 hidden layers: outputs has 8 channels, and the bottom (bottleneck) layer has outputs with 32 channels • Stride values state the stride for the initial convolution, ie. downsampling in down path and upsampling in up path
  • 26.
  • 27. Liaison with the community: Recommend policies and priorities to development team Working Groups of MONAI 1. IMAGING I/O – Stephen Aylward 2. DATA DIVERSITY – Brad Genereaux 3. REPRODUCIBILITY – Lena Maier-Hein 4. TRANSFORMATIONS – Jorge Cordoso 5. FEDERATED LEARNING – Jayashree Kalapathy 6. ADVANCED RESEARCH – Paul Jaeger 7. COMMUNITY ADOPTION – Prerna Dogra
  • 28. Goal: Define how data is read into and written out from memory in MONAI. I/O Working Group Tasks #1: Support reading and writing common research image file formats: NRRD, Nifti, DICOM (objects), JPG, PNG, TIFF, MetaIO, ... #2: Simplify interfacing MONAI with clinical systems such as PACS and health records systems: DICOM, HL7, ... Members ➢ Stephen Aylward (Kitware) ➢ Marco Nolden (DKFZ) ➢ Jayashree Kalpathy-Cramer (MGH) ➢ Brad Genereaux (Nvidia) ➢ Ben Murray (KCL) ➢ Wenqi Li (Nvidia) ➢ Jorge Cardoso (KCL) ➢ Prerna Dogra (Nvidia) Proposal #1: Create a single reader that handles a diversity of file formats in a consistent manner and that user-defined “experiments” can override to use a specific reader when I/O reproducibility is a concern. Solution #1: Default image loader uses ITK Python. Scripts can specify alternative loaders. Available starting in MONAI v0.3. Suggestions for next steps and alternatives: Stephen.Aylward@Kitware.com
  • 29. MONAI DATA WORKING GROUP Data: Images, Biomarkers, Demographics, Outputs and Outcomes Working Group Homepage: https://github.com/Project- MONAI/MONAI/wiki/Data-Working-Group Reach Out if you have any feedback, questions, or thoughts on data structures to be processed in MONAI. We want to hear from you for your preferred data types, data you need to capture, and where the challenges are!
  • 30. MONAI WORKING GROUP: EVALUATION, REPRODUCIBILITY, BENCHMARKS Working Group Homepage: https://github.com/Project-MONAI/MONAI/wiki/Evaluation,-Reproducibility,-Benchmarks-Working-Group-Meeting-Notes Lena Maier-Hein Working Group Lead German Cancer Research Center (DKFZ) Annika Reinke Lead of task force Benchmarking German Cancer Research Center (DKFZ) Jens Petersen German Cancer Research Center (DKFZ) M. Jorge Cardoso King’s College London (KCL) S. Kevin Zhou Working Group Lead Chinese Academy of Science (CAS) David Zimmerer German Cancer Research Center (DKFZ) Carole Sudre Lead of task force Metrics University College London (UCL) Nicola Rieke NVIDIA GmbH Technical University of Munich (TUM) Michela Antonelli Lead of task force Quick data access King’s College London (KCL) Dan Tudosiu King’s College London (KCL) Jayashree Kalpathy-Cramer Harvard Medical School (MGH) The core goal of the Working Group is to provide the software infrastructure and tools for quality-controlled validation and benchmarking of medical image analysis methods. We will address open technical challenges and research questions related to a variety of topics, ranging from best practices to performance aspects and implementation efficiency. Mission
  • 31. MONAI WORKING GROUP: EVALUATION, REPRODUCIBILITY, BENCHMARKS You have further feedback or questions? Use the slack channel or contact us via monai-evaluation-benchmarking-wg@dkfz.de Your feedback is important! What do you want from our MONAI Working Group? • What are current problems with respect to metrics, benchmarking and challenges? • What should we prioritize? • Best practice recommendations for metrics? • Easy access to existing data sets? • … Let your voice be heard: https://www.surveylegend.com/s/2mes Word cloud created from the Working Group’s meeting notes.
  • 32. MONAI WORKING GROUP: TRANSFORMATION & AUGMENTATION Medical Specific Transformations - LoadNifti | Spacing | Orientation - RandGaussianNoise |Normalize Intensirt - Rand2DElastic | Rand3DElastic Fused Spatial Transforms & GPU Optimization - Affine Transform - Random sampling: Class balanced fixed Ratio - Deterministic training controlled by setting random seed Generic | Vanilla |Dictionary-based Transforms LoadNifti AsChannelFirst Scale Intensity ConcatItems AsChannelFirst AsChannelFirst Scale Intensity Scale Intensity Network Nifty Images Brain Window Subdural Window Bone Window Load Image from File Make 2 Copies Different ranges &Scale Concat Together
  • 33. MONAI WORKING GROUP: TRANSFORMATION & AUGMENTATION 3rd Party OSS Packages & MONAI adapter Tools - Interoperability with other open source packages - Accommodate different data for 3rd party Transforms - Utility Transforms: ToTensor, ToNumpy, SqueeseDim - BatchGenerator - TorchIO - Rising Post-Processing & Integrate Third Party Transforms Compose all Transforms
  • 34. Goal: Support federated learning and other collaborative learning approaches. Federated Learning Working Group Tasks #1: Develop reference implementations of MONAI FL with Clara-FL #2: Develop example implementations of MONAI train with Substra/PySyft/?? #3: Perform a real-life federated learning experiment with interested members of community Members ➢ Daniel Rubin (Stanford) ➢ Jayashree Kalpathy-Cramer (MGH) ➢ Marco Nolden (DKFZ) ➢ Stephen Aylward (Kitware) ➢ Holger Roth (Nvidia) ➢ Wenqi Li (Nvidia) ➢ Jonas Scherer (DKFZ) ➢ Andriy Myronenko (Nvidia) ➢ Jorge Cardoso (KCL) ➢ Prerna Dogra (Nvidia) ➢ Nicola Rieke (Nvidia) ➢ Shadi Albarqouni (TUM) Proposal #1: Create interfaces to allow external FL packages to connect to MONAI trainers jkalpathy-cramer@mgh.Harvard.edu rubin@stanford.edu
  • 35. ● Goal: Guide the integration of new methodologies and tools into MONAI by focusing on community-driven progress and good scientific practice Research Working Group Previous Work: Future Plans: Before filling MONAI with Research Content, We focused on establishing a systematic process behind integration of methodologies in MONAI including: • Clear definition and categorization of methodologies: Components / Tutorials / Reserach Contributions • Code location and entry points via the monai.io webpage • Internal structure / taxnomoy ensuring scalability • Guidelines for contributions by the community • Prioritize and guide implementation of new methods such as Object Detection, Registration, etc. • Contribute towards an online benchmarking platform in MONAI including standardized datasets, Baselines/Model- Zoo and evaluations. Members: • Paul Jaeger (DKFZ) • Dan Tudosiu (KCL) • Dong Yang (Nvidia) • Michael Baumgartner (DKFZ) • Jorge Cardoso (KCL) • Holger Roth (Nvidia) • Ziyue Xu (Nvidia) • Wentao Zhu (Nvidia) • Chayanin Tangwiriyasakul (KCL) • Ben Murray (KCL) For feedback/questions please contact p.jaeger@dkfz.de See also our site on the MONAI wiki
  • 36. 36 Goal: Establish MONAI as a common software foundation that the Medical Imaging research and development community can build upon. Community Outreach Working Group Tasks: • Developing technical training and onboarding content • Organize workshops and boot camps to educate and engage • Define the infrastructure and processes needed to support and grow a transparent, honest, and open community • Establishing best practices for giving appropriate recognition to those who are promoting and making significant contributions back to MONAI Members: • Michael Zephyr (Nvidia) • Prerna Dogra (Nvidia) • Stephen Aylward (Kitware) • Julia Levites (Nvidia) • Paul Jaeger (DKFZ) • Wenqi Li (Nvidia) • Nicola Rieke (Nvidia) How to Engage: • See our Community Onboarding guide: https://github.com/Project-MONAI/MONAI#community • For feedback or questions, contact mzephyr@nvidia.com
  • 37. Understand the different options you have to engage with the MONAI community, developers, and working groups How to Engage with MONAI Where to go to get started with MONAI • Getting Started (Installation, Examples, Demos, etc.) https://monai.io/start.html • GitHub • Community Guide: https://github.com/Project-MONAI/MONAI#community • Contributing Guide: https://github.com/Project-MONAI/MONAI#contributing • Use GitHub Issues for Small Concrete Tasks and use the community tag. • Looking to get started contributing? Look at our Good First Issue tag! https://github.com/Project- MONAI/MONAI/labels/good%20first%20issue • PyTorch Forums. Tag @monai or see the MONAI user page. https://discuss.pytorch.org/u/MONAI/ • Stack Overflow. See existing tagged questions or create your own: https://stackoverflow.com/questions/tagged/monai • Follow us on Twitter for the latest updates on releases, research, demos and conference engagements. https://twitter.com/ProjectMONAI • Join our Slack Channel. Fill out the Google Form here: https://forms.gle/QTxJq3hFictp31UM9 • Transitioning away from Google Groups. Instead use one of the above options. • For feedback or questions, contact mzephyr@nvidia.com
  • 38. MONAI IS A GROWING COMMUNITY Contributors: 1.4 K+ GitHub stars & growing engagement
  • 39. BOOTCAMP – IN NUMBERS A LOT OF INTEREST IN THE COMMUNITY! • Number of applicants: 563 attendance applications • Accepted participants with cluster access (60) • Additionally other participants “observers” (140) • From 40 different countries: Australia, Austria, Belgium, China, Cyprus, Czechia, Egypt, Ethiopia, France, Ghana, Germany, Greece, Guatemala, Hong Kong, India, Israel, Iran, Malta, Mexico, Nepal, Netherlands, Norway, Oman, Peru, Poland, Portugal, Saudi Arabia, Slovenia, South Korea, Spain, Sweden, Switzerland, Turkey, United Arab emirates, United Kingdom, United States of America A truly global event!
  • 40. MONAI Demonstration Applications G. Wang, X. Liu, C. Li, Z. Xu, J. Ruan, H. Zhu, T. Meng, K. Li, N. Huang, S. Zhang. (2020) "A Noise-robust Framework for Automatic Segmentation of COVID-19 Pneumonia Lesions from CT Images." IEEE Transactions on Medical Imaging. 2020. DOI: 10.1109/TMI.2020.3000314 MONAI Jupyter Notebook Tutorial: Spleen 3D segmentation
  • 41. Conclusion Stephen R. Aylward, Ph.D. Chair of MONAI External Advisory Board Senior Directory of Strategic Initiatives, Kitware

Notas del editor

  1. Hello Today speaking about the MONAI toolkit, an platform for the application of deep learning to medical image analysis However, the goal of this talk is much broader. My goal is to spark your interest in open science, and in particular, show you the value of open source software for deep learning. For my talk I will first, provide a brief overview of why deep learning is succeeding in our field. Second, I will then provide a brief history of open science Third, I will present How you and MONAI can work together to both benefit from and contribute to open science. That is, show you the value of using MONAI.
  2. So, why is deep learning succeeding Two reasons First, performance. This is the most obvious answer Second, less obvious but, I will argue, just as important as the first, is that dl is succeeding because of open ness
  3. So, what are some examples… I leave that up to you, the audience, to fill in the blank Numerous examples of outperforming traditional image analysis methods More applications every day However, much research remains for explaining and improving generalization, small sample size, etc.
  4. Second reason why is deep learning succeeding..openness. In particular, open science. It is pervasive in deep learning. Open science practices include…
  5. The history of open science is quite long and rich. It arguably began with the work by Descartes in 1637. He wrote… This was the foundation – the formal specification of open science. As true today as it was then
  6. Open science, nevertheless, faces many challenges or obstacles. Here I will cover three. First, competition. It is inherent in us. But – you really don’t want to be these guys. Collaboration, sharing, and learning together are key. However, it could be argued that much of our lives and education are based on competition. It is a hard trait to overcome.
  7. Second challenge for open science…publications. They are woefully inadequate at describing most scientific experiments. The gaps impact reproducibility. Hinder progress. Result in wasted effort. We need richer descriptions – luckily we are computer scientists, so we have our code to share and overcome this challenge.
  8. Third, open science is rarely a priority…we have many goals that pull at us…
  9. However, if you take a step back you’ll realize that the benefits outweigh the costs…in fact, open science will get you to most of your goals faster than closed approached… If you are after funding… If you are after citations… If you want to have impact, educate, generate a following…
  10. The benefits of open science led to the creation of MONAI.
  11. MONAI is focused on Deep Learning Data Aug, Neural Network, Loss Function and Validation It arouse from collaboration between multiple other toolkits: (Niftynet,… Currently – many options… Goal of MONAI is to integrate rather than compete Build a community through value