SlideShare a Scribd company logo
How deep learning reshapes
medicine
Hongyoon Choi
Department of Nuclear Medicine, Seoul National University Hospital
Current
Finding
Hypermetabolic solid mass in the RLL (8.3), suggesting
lung cancer
Small LNs in mediastinal 4R, 7 without hypermetabolism.
Otherwise, no abnormal hypermetabolic lesion
suggesting metastasis.
Our Future?
Google’s plenary lecture at
AACR 2018
Our Future?
Vinyals, Oriol, et al. Proceedings of the IEEE Conference
on Computer Vision and Pattern Recognition. 2015.
Vision
Deep CNN
Language
Generating
RNN
“A group of people shopping at
an outdoor market.
There are many vegetables at
the fruit stands.”
Demo: http://vqa.cloudcv.org/
CONTENTS
INTRODUCTION
Era of Medical Big Data
DEEP LEARNING IN
MEDICINE
Brief overview of Deep learning
Real application to medicine
PERSPECTIVES
What to solve
New roles as a physician
Era of medical big data
INTRODUCTION
 Single target
 Pharmaceutical
companies
 Systemic, multimodal,
multidimensional
 Tech companies
Current
Future
INTRODUCTION
INTRODUCTION
Flood of data
$1,000 Genome sequencing
INTRODUCTION
Flood of data
Wearable polysomnography
Wearable EKG
Blood glucose monitoring system
INTRODUCTION
Previous Near Future
Use Herceptin! ?
What is the best choice for chemotherapy?
Integrating multi-omics
Genome
Phenome
Proteome
Imaging
Metabolome
Biosensor
CNN for imaging
RNN for health care records
Multidimensional data clustering
INTRODUCTION
+
Deep learning
Tsunami in medical fields
INTRODUCTION
2016
1 billion USD
2024
AI Industry in Healthcare
(from Global Market Insights)
First Target is Image…
INTRODUCTION
ROC curve
- better than dermatologists
Esteva, Andre, et al. Nature 2017
Computer tech papers invade to Nature/Cell/Science & NEJM/Lancet/JAMA
INTRODUCTION
Diabetic Retinopathy
 Better or equivalent
to ophthalmologists
Normal DM
Gulshan, Varun, et al. JAMA 2016
ChestXnet
 Equivalent/Superior to radiologists (?)
Pranav Rajpurkar, … Andrew Ng, 2017. Arxiv
INTRODUCTION
FDA approve a device for diagnosing
diabetic retinopathy (2018.4)
AI-aided system (CT angiography
for large vessel occlusion)
Year of AI invasion to clinic
FDA approve OsteoDetect(2018.5)
INTRODUCTION
• Really replace?
– Imaging specialists’ role
– Find clinical unmet needs
?
3 min Overview of Deep learning
3 min overview of deep learning
• Deep learning: y=f(x) where y: label. x: data
f cat
3 min overview of deep learning
• Conventional Perceptron
Lesion size
Circularity
Hounsfield unit
x1
x2
x3
Y
w1
w2
w3
b0
Activation function
Output = 1 if Y >0
Output = 0 if Y <0
Output
1: Malignancy
0: Benign
Find optimized W for minimized error
3 min overview of deep learning
• Limitation in previous perceptron
0 1 1 3 5 7 8 8
0 0 0 1 3 3 5 3
0 0 1 2 4 7 7 1
0 0 2 3 8 5 7 6
2 5 8 8 8 4 9 5
0 0 8 8 6 4 2 3
128x128
=16,384
Output
- Require good manual features
- Raw data  Too big.
- More layers?  Difficulty in learning
3 min overview of deep learning
• MLP to Deep learning
- Require good manual features
- Raw data  Too big.
- More layers?  Difficulty in learning
• Automatic feature extraction from raw data
CNN and RNN
• New methods for deep layer training
New activation function & Stochastic gradient descent
• Methods for reducing overfitting
3 min overview of deep learning
• Convolutional Neural Network
• Skim locally, instead of look all things
3 min overview of deep learning
identify line / some texture
identify head lights and wheels
identify Car!
Hierarchical Recogntion
3 min overview of deep learning
• Convolutional Neural Network
3 min overview of deep learning
• Convolutional Neural Network
ImageNet Challenge Results
28.2%
2010
25.8%
2011
16.4%
2012
Shallow model
AlexNet
11.7%
2013
6.7%
2014
3.57%
2015
GoogleNet
ResNet
8-layers
22-layers
152-layers
3 min overview of deep learning
• Recurrent Neural Network
3 min overview of deep learning
• More modules to train deep layers
Problem of Vanishing Gradient
Solved by nonlinearity function
Sigmoid  ReLU , tanh, ELU, Leaky ReLU
Sigmoid
ReLU
3 min overview of deep learning
• More modules to reduce overfitting
Problem of overfitting
Apple
3 min overview of deep learning
• More modules to reduce overfitting
Dropout
Overview of deep learning
• Current Concept of Deep learning
Deep layered
neural network
Output
+
Data type-specific
layers
Convolution
Recurrent
Modification for
training
+
ReLU activation
SGD training
Dropout
Batch normalization
Variable Loss
…
Deep learning in Medicine
Particularly for nuclear medicine
Deep learning in Medicine
DL for medical imaging:
Supervised learning using CNN
f cat
f: CNN
f Lung cancer
Simple application of CNN for diagnosis
Deep learning in Medicine
Diabetic Retinopathy
 Better or equivalent
to ophthalmologists
Normal DM
Gulshan, Varun, et al. JAMA 2016
ChestXnet
 Equivalent/Superior to radiologists (?)
Pranav Rajpurkar, … Andrew Ng, 2017. Arxiv
DL for medical imaging:
Supervised learning using CNN
Deep learning in Medicine
DL for medical imaging:
Supervised learning using CNN
AD & NC
MCI-converter & non-converter
FDG and amyloid PET to predict future cognitive decline
Choi H and Jin KH KSNM 2016;
Behav Brain Res 2018
Deep learning in Medicine
DL for medical imaging:
Supervised learning using CNN
FDG and amyloid PET to predict future cognitive decline
Accuracy
AD vs. NC
96.0% 85.4% 80.7%
Deep CNN FDG quantification AV45 quantification
Accuracy
MCI conversion
84.2% 75.4% 80.1%84.2%
Deep CNN-based
82%
Moraldi et al. 2016
(MRI+Clinical Variables)
78%
Zhang et al. 2013
(FDG, MRI+Clinical)
Feature Extraction + Machine learning
72%
Shaffer et al. 2013
(FDG, MRI, CSFl)
Deep learning in Medicine
Output score measured by
baseline PET
& 3-year cognitive score
f p(Alzheimer|X)
Direct biomarker
A single parameter
Choi H and Jin KH KSNM 2016;
Behav Brain Res 2018
Deep learning in Medicine
https://adfdgpet.appspot.com
Online Demo
Input file
Web application
Output: likelihood for AD
& predicted cognitive score
Output:
Cognitive dysfunction-related map
p(Alzheimer|X)
Deep learning in Medicine
SWEDD:
Clinically PD,
Image normal
Expert1 Expert2
Conventional
quantification
SWEDD
Deep learning-
abnormal
2-year follow-up
80%
Abnormal
Deep learning : Redefine SWEDD diagnosis
Choi, H., … ,Lee. D.S. NeuroImage: Clinical 2017
Dopamine transporter imaging for Parkinson’s disease diagnosis
3D CNN model
normal Parkinson
Deep learning in Medicine
Laborious Work Replaced by DL:
Segmentation
Choi, H., & Jin, K. H. J Neurosci Methods 2016 de Brebisson, et al. CVPR 2015.
Deep learning in Medicine
Laborious Work Replaced by DL:
Detection
Liu Y, et al. Arxiv 2017
Deep learning in Medicine
Enhance Image Acquisition & Quality
Dahl et al. Arxiv 2017
Standard dose
1/200 Dose 1/200 Dose + CNN
Deep learning in Medicine
Image Generation
Cat
Cat
Common deep learning model Generative model
z = f(x)
where x: data, z: discriminative features
f: classifier model
x = g(z)
where x: data, z: latent
g: generation function
Deep learning in Medicine
Generative Adversarial Network
z~N(0,1)
G:
generator
Karras T, et al. Arxiv 2017
G:
generator
Isola P, et al. arxiv, 2016.
Generative Adversarial Network
Structural MR generation from PET
Florbetapir PET
Generator:
U-net
Skip connection
Generated MR
PETandgeneratedMRPETandrealMR
Discriminator Real or Fake
Generative Adversarial Networks
for MR generation
z G(z)
z & G(z)
z & x
Choi H and Lee DS, J Nucl Med 2017.
Deep learning in Medicine
RealMRI
Generated
MRI
18F-Florbetapir
PET
Deep learning in Medicine
PET-based normalization
SUVR
measurement
MRI-based normalization
SUVR
measurement
MR generation
Generated
MRI-based normalization
transformation apply to PET
SUVR
measurement
Choi H and Lee DS, J Nucl Med 2017.
Generated-MRbased
quantification
PreviousPET-based
quantification
Gold standard
Gold standard
Deep learning in Medicine
Conditional Generation
Antipov G, Arxiv 2017
Encoder
Latent
features
Generator
+ Age
Latent
features
+ Age
VAE model for brain PET
generation
 Brain metabolism aging movie
Choi H,… Lee DS. Biorxiv 2017
Deep learning in Medicine
Conditional Generation
Choi H,… Lee DS. Biorxiv 2017
Estimating normal population distribution
Deep learning in Medicine
Deep learning in Medicine
• Individual vs population
– Aging of individual brain : Comparing with virtual population
Population distribution of brain
metabolism at each age by iterative
generating from the VAE model
Generator
Latent
features
N(0,1) Random sampling
from normal distribution
+ Age
Age
50
55
60
65
70
75
• Evaluating individual brain’s aging
compared with general population
• Define ‘pathologic aging’
Choi H,… Lee DS. Biorxiv 2017
Practical Issues & Perspectives
Practical Issues
Gap between Tech-initiated DL and Clinician’s need
We just need BMI to know obesity!
Practical Issues
– Various purposes
• More than simple diagnosis: Prognosis, Disease
status monitor, Response prediction.
– Ambiguous ground-truth
• Diagnosis is not a simple classification
Gap between Tech-initiated DL and Clinician’s need
Practical Issues
Clinical diagnosis:
Spectrum instead of
a clear-cut classification
Blood pressure = a type of biomarker
Cardiovascular
eventrisk
140/90
HypertensionNormal
Blood Pressure
(one-dimensional)
Quantitative biomarker
reflecting prognostic outcome
Direct
Most diseases…
Multiple domain and high-dimensional
Not rely on single measurement
Practical Issues
SUVmax
A value
maximize
information
of whole voxels
to represent
patient’s prognosis
What we expect: Deep learning as a ‘best parameter’ extractor
from high-dimensional data
Simple Diagnosis << Biomarker
Practical Issues
Data
Data
size
Unlabeled
data
Where to
get
Practical Issues
Where to get
Hospital
• Best source
• Practical model validation: Real world
problems
• Ethical issues and relatively small data size
Public database
• Easy to get
• Red ocean
• Limitation in real-world validation
Practical Issues
• Public databases
– NM images included
• The Cancer Imaging Archives
• PPMI (Parkinson), ABIDE (Autism), ADHD-200 (ADHD), OASIS
(Brain MR), HCP (Brain MR)
• ADNI (Alzheimer PET, MRI)
– Chest X-ray8 dataset
– DDSM (mammography)
– Kaggle diabetic retinopathy
– MICCAI brain tumor segmentation challenges
Practical Issues
Data Size
– Does deep learning require big data?
– Always more than 100,000 images?
Answer : No
Answer : Yes
Practical Issues
Data Size
– Voxel-based training:
Segmentation/Superresolution etc.
– Image generation (Pix2Pix)
:~100 3d volumes
– Image augmentation
: rotation/flipping
(but, cautious)
– Task and complexity dependent:
AD vs normal was trained by ~300 cases.
Practical Issues
Image labels
– Unlabeled data >> Labeled data
• e.g. all FDG PET/CT >> Baseline breast cancer FDG PET/CT
– Unbalanced label
• e.g. Lymphoma PET >> Adrenocortical carcinoma PET
 Importance of unsupervised & semi-
supervised learning
Practical Issues
Tumor metabolism
estimated by
FDG PET image
Gene networks
Gene expression data with
FDG PET image
 ~ 40 pairs
Gene expression data alone
 ~ 1000 cases
Image labels
Example of semisupervised learning in practical problem
Choi H and Na KJ. Theranostics 2018
Practical Issues
Expressiondata
Tumor metabolism
(maximum SUV)
Encoder
Decoder
HiddenFeatures
626 64 1
Supervised Training
(n = 20)
Unsupervised Training
(n = 226)
Dimensions
Example of semisupervised learning in practical problem
Semisupervised learning
- Supervised training
combined with autoencoder
Choi H and Na KJ. Theranostics 2018
Practical Issues
At report…
- Lung cancer cannot be excluded.
Rec> “Clinical correlation is recommended”
Issues of Uncertainty
Suggestive of cancer
Probably, cancer
Possibly, cancer
Cancer cannot be excluded
Recommend
further exam
Deep learning trained
by Natural Images
“Cat”
Practical Issues
Data distribution
Dataset for model training/validation
Alzheimer Normal
Real data
at Hospital
Normal
Alzheimer
FTD
DLB
PSP
Depression
Real data
at Community Normal
• Will DL model work at real clinic?
• How to deal with rare/unseen disease?
Aged person with
subjective cognitive decline
Perspectives as a Physician
 Medicine is too complex
 Problem of responsibility
 Too many empirical things
 Didn’t you see AlphaGo?
 Domination is already started
 At least, many doctors will lose
their job
Perspectives as a Physician
Tsunami in medical fields
Deep Learning
Big Data
Perspectives as a Physician
• Gaps between Tech. & Hosp.
• Become an expert of data
• Roles will be changed
– Not just replace MD’s role
– Laborious things replaced by AI
– New information produce new jobs
Perspectives as a Physician
• Roles will be changed
– Laborious things replaced by AI
– New information produce new jobs
• Performed by AI ~1 sec.
• New information for medical decision
(Cortical thickness, Tumor volume,
etc
as a clinical routine)
Perspectives as a Physician
Disruptive Innovation: Raw medical & healthcare data
Diet + Previous Glucose Level
Future Glucose Level &
Scheduling Insulin
Sugar.iq from Medtronic
(FDA approved 2016.9)
Perspectives as a Physician
Disruptive Innovation: Raw medical & healthcare data
Perspectives as a Physician
Disruptive Innovation: Raw medical & healthcare data
HTC DeepQ Tricoder
Predicting PVC from daily EKG
Diagnosis of otitis media
Perspectives as a Physician
• Accelerating changes to daily healthcare
from hospital-based health
Deep learning facilitates left-shifting
Perspectives as a Physician
Future medical decision
ConcatenatingFeatures
• Diagnosis
• Management Plan
AbnormalNormal
Integrated biomarker based on DL
RiskatDeath
Human
Cannot Do!
Perspectives as a Physician
Future medical decision
Previous classification
Single target
Based on some receptors
Breast cancer
classification
Current classification
Multiple targets
Based on ~50 transcripts
Perspectives as a Physician
Future medical decision
Breast cancer
classification
Genome
Phenome
Proteome
Imaging
Metabolome
Biosensor
Unsupervised learning
Integrative classification
Future classification
Multiomics targets
Based on unsupervised learning
Perspectives as a Physician
Empirical
-based
Evidence
-based
Data
-driven
“약을 써보니 낫더라
“RCT를 해보니
이 약은 효과가
유의하게 있다
“모든 이 사람의 활동 데이터,
Omics 데이터, 영상 정보 등을
통합하여 볼 때 이 약이 가장
적합할 것이다
Deep Learning

More Related Content

What's hot

Public Databases for Radiomics Research: Current Status and Future Directions
Public Databases for Radiomics Research: Current Status and Future DirectionsPublic Databases for Radiomics Research: Current Status and Future Directions
Public Databases for Radiomics Research: Current Status and Future Directions
CancerImagingInforma
 
MedicalResearch.com: Medical Research Exclusive Interviews December 14 2014
MedicalResearch.com:  Medical Research Exclusive Interviews December 14 2014MedicalResearch.com:  Medical Research Exclusive Interviews December 14 2014
MedicalResearch.com: Medical Research Exclusive Interviews December 14 2014
Marie Benz MD FAAD
 
Brain Tumor Detection using Neural Network
Brain Tumor Detection using Neural NetworkBrain Tumor Detection using Neural Network
Brain Tumor Detection using Neural Network
ijtsrd
 
Twenty Years of Whole Slide Imaging - the Coming Phase Change
Twenty Years of Whole Slide Imaging - the Coming Phase ChangeTwenty Years of Whole Slide Imaging - the Coming Phase Change
Twenty Years of Whole Slide Imaging - the Coming Phase Change
Joel Saltz
 
Study on Physicians Request for Computed Tomography Examinations for Patients...
Study on Physicians Request for Computed Tomography Examinations for Patients...Study on Physicians Request for Computed Tomography Examinations for Patients...
Study on Physicians Request for Computed Tomography Examinations for Patients...
IRJESJOURNAL
 
Medical Computer Vision: Current Limitations of Vision Datasets | CVPR 2021
Medical Computer Vision: Current Limitations of Vision Datasets | CVPR 2021 Medical Computer Vision: Current Limitations of Vision Datasets | CVPR 2021
Medical Computer Vision: Current Limitations of Vision Datasets | CVPR 2021
Asma Ben Abacha
 
Integrated Modelling Approach for Enhancing Brain MRI with Flexible Pre-Proce...
Integrated Modelling Approach for Enhancing Brain MRI with Flexible Pre-Proce...Integrated Modelling Approach for Enhancing Brain MRI with Flexible Pre-Proce...
Integrated Modelling Approach for Enhancing Brain MRI with Flexible Pre-Proce...
IJECEIAES
 
Framework for comprehensive enhancement of brain tumor images with single-win...
Framework for comprehensive enhancement of brain tumor images with single-win...Framework for comprehensive enhancement of brain tumor images with single-win...
Framework for comprehensive enhancement of brain tumor images with single-win...
IJECEIAES
 
The Patient Journey with AI
The Patient Journey with AIThe Patient Journey with AI
The Patient Journey with AI
Imagia Cybernetics
 
AI in translational medicine webinar
AI in translational medicine webinarAI in translational medicine webinar
AI in translational medicine webinar
Pistoia Alliance
 
Radiomics: Novel Paradigm of Deep Learning for Clinical Decision Support towa...
Radiomics: Novel Paradigm of Deep Learning for Clinical Decision Support towa...Radiomics: Novel Paradigm of Deep Learning for Clinical Decision Support towa...
Radiomics: Novel Paradigm of Deep Learning for Clinical Decision Support towa...
Wookjin Choi
 
M sc research_project_report_x18134599
M sc research_project_report_x18134599M sc research_project_report_x18134599
M sc research_project_report_x18134599
MansiChowkkar
 
Digital pathology in developing country
Digital pathology in developing countryDigital pathology in developing country
Digital pathology in developing country
Dr. Ashish lakhey
 
Ph.D. Defense Presentation Slides (Changhee Han) カリスの東大博論審査会(公聴会)発表スライド Patho...
Ph.D. Defense Presentation Slides (Changhee Han) カリスの東大博論審査会(公聴会)発表スライド Patho...Ph.D. Defense Presentation Slides (Changhee Han) カリスの東大博論審査会(公聴会)発表スライド Patho...
Ph.D. Defense Presentation Slides (Changhee Han) カリスの東大博論審査会(公聴会)発表スライド Patho...
カリス 東大AI博士
 
Prospects of Deep Learning in Medical Imaging
Prospects of Deep Learning in Medical ImagingProspects of Deep Learning in Medical Imaging
Prospects of Deep Learning in Medical Imaging
Godswll Egegwu
 
University of Toronto - Radiomics for Oncology - 2017
University of Toronto  - Radiomics for Oncology - 2017University of Toronto  - Radiomics for Oncology - 2017
University of Toronto - Radiomics for Oncology - 2017
Andre Dekker
 
The Complementary Roles of Computer-Aided Diagnosis and Quantitative Image A...
 The Complementary Roles of Computer-Aided Diagnosis and Quantitative Image A... The Complementary Roles of Computer-Aided Diagnosis and Quantitative Image A...
The Complementary Roles of Computer-Aided Diagnosis and Quantitative Image A...
Carestream
 
C0344023028
C0344023028C0344023028
C0344023028
inventionjournals
 
Digital webinar master deck final
Digital webinar master deck finalDigital webinar master deck final
Digital webinar master deck final
Pistoia Alliance
 

What's hot (20)

Public Databases for Radiomics Research: Current Status and Future Directions
Public Databases for Radiomics Research: Current Status and Future DirectionsPublic Databases for Radiomics Research: Current Status and Future Directions
Public Databases for Radiomics Research: Current Status and Future Directions
 
MedicalResearch.com: Medical Research Exclusive Interviews December 14 2014
MedicalResearch.com:  Medical Research Exclusive Interviews December 14 2014MedicalResearch.com:  Medical Research Exclusive Interviews December 14 2014
MedicalResearch.com: Medical Research Exclusive Interviews December 14 2014
 
Brain Tumor Detection using Neural Network
Brain Tumor Detection using Neural NetworkBrain Tumor Detection using Neural Network
Brain Tumor Detection using Neural Network
 
Twenty Years of Whole Slide Imaging - the Coming Phase Change
Twenty Years of Whole Slide Imaging - the Coming Phase ChangeTwenty Years of Whole Slide Imaging - the Coming Phase Change
Twenty Years of Whole Slide Imaging - the Coming Phase Change
 
Study on Physicians Request for Computed Tomography Examinations for Patients...
Study on Physicians Request for Computed Tomography Examinations for Patients...Study on Physicians Request for Computed Tomography Examinations for Patients...
Study on Physicians Request for Computed Tomography Examinations for Patients...
 
Medical Computer Vision: Current Limitations of Vision Datasets | CVPR 2021
Medical Computer Vision: Current Limitations of Vision Datasets | CVPR 2021 Medical Computer Vision: Current Limitations of Vision Datasets | CVPR 2021
Medical Computer Vision: Current Limitations of Vision Datasets | CVPR 2021
 
Integrated Modelling Approach for Enhancing Brain MRI with Flexible Pre-Proce...
Integrated Modelling Approach for Enhancing Brain MRI with Flexible Pre-Proce...Integrated Modelling Approach for Enhancing Brain MRI with Flexible Pre-Proce...
Integrated Modelling Approach for Enhancing Brain MRI with Flexible Pre-Proce...
 
Framework for comprehensive enhancement of brain tumor images with single-win...
Framework for comprehensive enhancement of brain tumor images with single-win...Framework for comprehensive enhancement of brain tumor images with single-win...
Framework for comprehensive enhancement of brain tumor images with single-win...
 
The Patient Journey with AI
The Patient Journey with AIThe Patient Journey with AI
The Patient Journey with AI
 
AI in translational medicine webinar
AI in translational medicine webinarAI in translational medicine webinar
AI in translational medicine webinar
 
Radiomics: Novel Paradigm of Deep Learning for Clinical Decision Support towa...
Radiomics: Novel Paradigm of Deep Learning for Clinical Decision Support towa...Radiomics: Novel Paradigm of Deep Learning for Clinical Decision Support towa...
Radiomics: Novel Paradigm of Deep Learning for Clinical Decision Support towa...
 
M sc research_project_report_x18134599
M sc research_project_report_x18134599M sc research_project_report_x18134599
M sc research_project_report_x18134599
 
Digital pathology in developing country
Digital pathology in developing countryDigital pathology in developing country
Digital pathology in developing country
 
Ph.D. Defense Presentation Slides (Changhee Han) カリスの東大博論審査会(公聴会)発表スライド Patho...
Ph.D. Defense Presentation Slides (Changhee Han) カリスの東大博論審査会(公聴会)発表スライド Patho...Ph.D. Defense Presentation Slides (Changhee Han) カリスの東大博論審査会(公聴会)発表スライド Patho...
Ph.D. Defense Presentation Slides (Changhee Han) カリスの東大博論審査会(公聴会)発表スライド Patho...
 
Prospects of Deep Learning in Medical Imaging
Prospects of Deep Learning in Medical ImagingProspects of Deep Learning in Medical Imaging
Prospects of Deep Learning in Medical Imaging
 
323462348
323462348323462348
323462348
 
University of Toronto - Radiomics for Oncology - 2017
University of Toronto  - Radiomics for Oncology - 2017University of Toronto  - Radiomics for Oncology - 2017
University of Toronto - Radiomics for Oncology - 2017
 
The Complementary Roles of Computer-Aided Diagnosis and Quantitative Image A...
 The Complementary Roles of Computer-Aided Diagnosis and Quantitative Image A... The Complementary Roles of Computer-Aided Diagnosis and Quantitative Image A...
The Complementary Roles of Computer-Aided Diagnosis and Quantitative Image A...
 
C0344023028
C0344023028C0344023028
C0344023028
 
Digital webinar master deck final
Digital webinar master deck finalDigital webinar master deck final
Digital webinar master deck final
 

Similar to How deep learning reshapes medicine

Deep learning application to medical imaging: Perspectives as a physician
Deep learning application to medical imaging: Perspectives as a physicianDeep learning application to medical imaging: Perspectives as a physician
Deep learning application to medical imaging: Perspectives as a physician
Hongyoon Choi
 
Big Data in Biomedicine: Where is the NIH Headed
Big Data in Biomedicine: Where is the NIH HeadedBig Data in Biomedicine: Where is the NIH Headed
Big Data in Biomedicine: Where is the NIH Headed
Philip Bourne
 
Big Data and the Promise and Pitfalls when Applied to Disease Prevention and ...
Big Data and the Promise and Pitfalls when Applied to Disease Prevention and ...Big Data and the Promise and Pitfalls when Applied to Disease Prevention and ...
Big Data and the Promise and Pitfalls when Applied to Disease Prevention and ...
Philip Bourne
 
Large Language Models, No-Code, and Responsible AI - Trends in Applied NLP in...
Large Language Models, No-Code, and Responsible AI - Trends in Applied NLP in...Large Language Models, No-Code, and Responsible AI - Trends in Applied NLP in...
Large Language Models, No-Code, and Responsible AI - Trends in Applied NLP in...
David Talby
 
Recent Advances in Deep Learning Techniques for Electronic Health Record
Recent Advances in Deep Learning Techniques for Electronic Health RecordRecent Advances in Deep Learning Techniques for Electronic Health Record
Recent Advances in Deep Learning Techniques for Electronic Health Record
kingstdio
 
tranSMART Community Meeting 5-7 Nov 13 - Session 2: Creating a Comprehensive ...
tranSMART Community Meeting 5-7 Nov 13 - Session 2: Creating a Comprehensive ...tranSMART Community Meeting 5-7 Nov 13 - Session 2: Creating a Comprehensive ...
tranSMART Community Meeting 5-7 Nov 13 - Session 2: Creating a Comprehensive ...
David Peyruc
 
Talk on reproducibility in EEG research
Talk on reproducibility in EEG researchTalk on reproducibility in EEG research
Talk on reproducibility in EEG research
Dorothy Bishop
 
DEEP FACIAL DIAGNOSIS: DEEP TRANSFER LEARNING FROM FACE RECOGNITION TO FACIAL...
DEEP FACIAL DIAGNOSIS: DEEP TRANSFER LEARNING FROM FACE RECOGNITION TO FACIAL...DEEP FACIAL DIAGNOSIS: DEEP TRANSFER LEARNING FROM FACE RECOGNITION TO FACIAL...
DEEP FACIAL DIAGNOSIS: DEEP TRANSFER LEARNING FROM FACE RECOGNITION TO FACIAL...
IRJET Journal
 
2011 11 16 - Vreeman - Corralling Creativity with Standards
2011 11 16 - Vreeman - Corralling Creativity with Standards2011 11 16 - Vreeman - Corralling Creativity with Standards
2011 11 16 - Vreeman - Corralling Creativity with Standardsdvreeman
 
Machine Learning in Biology and Why It Doesn't Make Sense - Theo Knijnenburg,...
Machine Learning in Biology and Why It Doesn't Make Sense - Theo Knijnenburg,...Machine Learning in Biology and Why It Doesn't Make Sense - Theo Knijnenburg,...
Machine Learning in Biology and Why It Doesn't Make Sense - Theo Knijnenburg,...
Seattle DAML meetup
 
IFMIA 2019 Plenary Talk : Deep Learning in Medicine; Engineers' Perspectives
IFMIA 2019 Plenary Talk : Deep Learning in Medicine; Engineers' PerspectivesIFMIA 2019 Plenary Talk : Deep Learning in Medicine; Engineers' Perspectives
IFMIA 2019 Plenary Talk : Deep Learning in Medicine; Engineers' Perspectives
Namkug Kim
 
Ccids 2019 cutting edges of ai technology in medicine
Ccids 2019 cutting edges of ai technology in medicineCcids 2019 cutting edges of ai technology in medicine
Ccids 2019 cutting edges of ai technology in medicine
Namkug Kim
 
14 00-20171207 rance-piv_c
14 00-20171207 rance-piv_c14 00-20171207 rance-piv_c
14 00-20171207 rance-piv_c
Bertrand Tavitian
 
BiTeM / SIBTex @ TREC CDS 2014
BiTeM / SIBTex @ TREC CDS 2014BiTeM / SIBTex @ TREC CDS 2014
BiTeM / SIBTex @ TREC CDS 2014
Julien Gobeill
 
Final_Presentation.pptx
Final_Presentation.pptxFinal_Presentation.pptx
Final_Presentation.pptx
SudeekshaKoricherla
 
인공지능 논문작성과 심사에관한요령
인공지능 논문작성과 심사에관한요령인공지능 논문작성과 심사에관한요령
인공지능 논문작성과 심사에관한요령
Namkug Kim
 
Machine learning in biology
Machine learning in biologyMachine learning in biology
Machine learning in biology
Pranavathiyani G
 
Basics of Data Analysis in Bioinformatics
Basics of Data Analysis in BioinformaticsBasics of Data Analysis in Bioinformatics
Basics of Data Analysis in Bioinformatics
Elena Sügis
 
Text mining and deep learning for biomedicine
Text mining and deep learning for biomedicineText mining and deep learning for biomedicine
Text mining and deep learning for biomedicine
Zhiyong Lu, PhD FACMI
 
Big Data & ML for Clinical Data
Big Data & ML for Clinical DataBig Data & ML for Clinical Data
Big Data & ML for Clinical Data
Paul Agapow
 

Similar to How deep learning reshapes medicine (20)

Deep learning application to medical imaging: Perspectives as a physician
Deep learning application to medical imaging: Perspectives as a physicianDeep learning application to medical imaging: Perspectives as a physician
Deep learning application to medical imaging: Perspectives as a physician
 
Big Data in Biomedicine: Where is the NIH Headed
Big Data in Biomedicine: Where is the NIH HeadedBig Data in Biomedicine: Where is the NIH Headed
Big Data in Biomedicine: Where is the NIH Headed
 
Big Data and the Promise and Pitfalls when Applied to Disease Prevention and ...
Big Data and the Promise and Pitfalls when Applied to Disease Prevention and ...Big Data and the Promise and Pitfalls when Applied to Disease Prevention and ...
Big Data and the Promise and Pitfalls when Applied to Disease Prevention and ...
 
Large Language Models, No-Code, and Responsible AI - Trends in Applied NLP in...
Large Language Models, No-Code, and Responsible AI - Trends in Applied NLP in...Large Language Models, No-Code, and Responsible AI - Trends in Applied NLP in...
Large Language Models, No-Code, and Responsible AI - Trends in Applied NLP in...
 
Recent Advances in Deep Learning Techniques for Electronic Health Record
Recent Advances in Deep Learning Techniques for Electronic Health RecordRecent Advances in Deep Learning Techniques for Electronic Health Record
Recent Advances in Deep Learning Techniques for Electronic Health Record
 
tranSMART Community Meeting 5-7 Nov 13 - Session 2: Creating a Comprehensive ...
tranSMART Community Meeting 5-7 Nov 13 - Session 2: Creating a Comprehensive ...tranSMART Community Meeting 5-7 Nov 13 - Session 2: Creating a Comprehensive ...
tranSMART Community Meeting 5-7 Nov 13 - Session 2: Creating a Comprehensive ...
 
Talk on reproducibility in EEG research
Talk on reproducibility in EEG researchTalk on reproducibility in EEG research
Talk on reproducibility in EEG research
 
DEEP FACIAL DIAGNOSIS: DEEP TRANSFER LEARNING FROM FACE RECOGNITION TO FACIAL...
DEEP FACIAL DIAGNOSIS: DEEP TRANSFER LEARNING FROM FACE RECOGNITION TO FACIAL...DEEP FACIAL DIAGNOSIS: DEEP TRANSFER LEARNING FROM FACE RECOGNITION TO FACIAL...
DEEP FACIAL DIAGNOSIS: DEEP TRANSFER LEARNING FROM FACE RECOGNITION TO FACIAL...
 
2011 11 16 - Vreeman - Corralling Creativity with Standards
2011 11 16 - Vreeman - Corralling Creativity with Standards2011 11 16 - Vreeman - Corralling Creativity with Standards
2011 11 16 - Vreeman - Corralling Creativity with Standards
 
Machine Learning in Biology and Why It Doesn't Make Sense - Theo Knijnenburg,...
Machine Learning in Biology and Why It Doesn't Make Sense - Theo Knijnenburg,...Machine Learning in Biology and Why It Doesn't Make Sense - Theo Knijnenburg,...
Machine Learning in Biology and Why It Doesn't Make Sense - Theo Knijnenburg,...
 
IFMIA 2019 Plenary Talk : Deep Learning in Medicine; Engineers' Perspectives
IFMIA 2019 Plenary Talk : Deep Learning in Medicine; Engineers' PerspectivesIFMIA 2019 Plenary Talk : Deep Learning in Medicine; Engineers' Perspectives
IFMIA 2019 Plenary Talk : Deep Learning in Medicine; Engineers' Perspectives
 
Ccids 2019 cutting edges of ai technology in medicine
Ccids 2019 cutting edges of ai technology in medicineCcids 2019 cutting edges of ai technology in medicine
Ccids 2019 cutting edges of ai technology in medicine
 
14 00-20171207 rance-piv_c
14 00-20171207 rance-piv_c14 00-20171207 rance-piv_c
14 00-20171207 rance-piv_c
 
BiTeM / SIBTex @ TREC CDS 2014
BiTeM / SIBTex @ TREC CDS 2014BiTeM / SIBTex @ TREC CDS 2014
BiTeM / SIBTex @ TREC CDS 2014
 
Final_Presentation.pptx
Final_Presentation.pptxFinal_Presentation.pptx
Final_Presentation.pptx
 
인공지능 논문작성과 심사에관한요령
인공지능 논문작성과 심사에관한요령인공지능 논문작성과 심사에관한요령
인공지능 논문작성과 심사에관한요령
 
Machine learning in biology
Machine learning in biologyMachine learning in biology
Machine learning in biology
 
Basics of Data Analysis in Bioinformatics
Basics of Data Analysis in BioinformaticsBasics of Data Analysis in Bioinformatics
Basics of Data Analysis in Bioinformatics
 
Text mining and deep learning for biomedicine
Text mining and deep learning for biomedicineText mining and deep learning for biomedicine
Text mining and deep learning for biomedicine
 
Big Data & ML for Clinical Data
Big Data & ML for Clinical DataBig Data & ML for Clinical Data
Big Data & ML for Clinical Data
 

Recently uploaded

Couples presenting to the infertility clinic- Do they really have infertility...
Couples presenting to the infertility clinic- Do they really have infertility...Couples presenting to the infertility clinic- Do they really have infertility...
Couples presenting to the infertility clinic- Do they really have infertility...
Sujoy Dasgupta
 
Novas diretrizes da OMS para os cuidados perinatais de mais qualidade
Novas diretrizes da OMS para os cuidados perinatais de mais qualidadeNovas diretrizes da OMS para os cuidados perinatais de mais qualidade
Novas diretrizes da OMS para os cuidados perinatais de mais qualidade
Prof. Marcus Renato de Carvalho
 
Pharynx and Clinical Correlations BY Dr.Rabia Inam Gandapore.pptx
Pharynx and Clinical Correlations BY Dr.Rabia Inam Gandapore.pptxPharynx and Clinical Correlations BY Dr.Rabia Inam Gandapore.pptx
Pharynx and Clinical Correlations BY Dr.Rabia Inam Gandapore.pptx
Dr. Rabia Inam Gandapore
 
TEST BANK for Operations Management, 14th Edition by William J. Stevenson, Ve...
TEST BANK for Operations Management, 14th Edition by William J. Stevenson, Ve...TEST BANK for Operations Management, 14th Edition by William J. Stevenson, Ve...
TEST BANK for Operations Management, 14th Edition by William J. Stevenson, Ve...
kevinkariuki227
 
Maxilla, Mandible & Hyoid Bone & Clinical Correlations by Dr. RIG.pptx
Maxilla, Mandible & Hyoid Bone & Clinical Correlations by Dr. RIG.pptxMaxilla, Mandible & Hyoid Bone & Clinical Correlations by Dr. RIG.pptx
Maxilla, Mandible & Hyoid Bone & Clinical Correlations by Dr. RIG.pptx
Dr. Rabia Inam Gandapore
 
Phone Us ❤85270-49040❤ #ℂall #gIRLS In Surat By Surat @ℂall @Girls Hotel With...
Phone Us ❤85270-49040❤ #ℂall #gIRLS In Surat By Surat @ℂall @Girls Hotel With...Phone Us ❤85270-49040❤ #ℂall #gIRLS In Surat By Surat @ℂall @Girls Hotel With...
Phone Us ❤85270-49040❤ #ℂall #gIRLS In Surat By Surat @ℂall @Girls Hotel With...
Savita Shen $i11
 
Pulmonary Thromboembolism - etilogy, types, medical- Surgical and nursing man...
Pulmonary Thromboembolism - etilogy, types, medical- Surgical and nursing man...Pulmonary Thromboembolism - etilogy, types, medical- Surgical and nursing man...
Pulmonary Thromboembolism - etilogy, types, medical- Surgical and nursing man...
VarunMahajani
 
Physiology of Special Chemical Sensation of Taste
Physiology of Special Chemical Sensation of TastePhysiology of Special Chemical Sensation of Taste
Physiology of Special Chemical Sensation of Taste
MedicoseAcademics
 
Superficial & Deep Fascia of the NECK.pptx
Superficial & Deep Fascia of the NECK.pptxSuperficial & Deep Fascia of the NECK.pptx
Superficial & Deep Fascia of the NECK.pptx
Dr. Rabia Inam Gandapore
 
For Better Surat #ℂall #Girl Service ❤85270-49040❤ Surat #ℂall #Girls
For Better Surat #ℂall #Girl Service ❤85270-49040❤ Surat #ℂall #GirlsFor Better Surat #ℂall #Girl Service ❤85270-49040❤ Surat #ℂall #Girls
For Better Surat #ℂall #Girl Service ❤85270-49040❤ Surat #ℂall #Girls
Savita Shen $i11
 
NVBDCP.pptx Nation vector borne disease control program
NVBDCP.pptx Nation vector borne disease control programNVBDCP.pptx Nation vector borne disease control program
NVBDCP.pptx Nation vector borne disease control program
Sapna Thakur
 
Are There Any Natural Remedies To Treat Syphilis.pdf
Are There Any Natural Remedies To Treat Syphilis.pdfAre There Any Natural Remedies To Treat Syphilis.pdf
Are There Any Natural Remedies To Treat Syphilis.pdf
Little Cross Family Clinic
 
heat stroke and heat exhaustion in children
heat stroke and heat exhaustion in childrenheat stroke and heat exhaustion in children
heat stroke and heat exhaustion in children
SumeraAhmad5
 
Tom Selleck Health: A Comprehensive Look at the Iconic Actor’s Wellness Journey
Tom Selleck Health: A Comprehensive Look at the Iconic Actor’s Wellness JourneyTom Selleck Health: A Comprehensive Look at the Iconic Actor’s Wellness Journey
Tom Selleck Health: A Comprehensive Look at the Iconic Actor’s Wellness Journey
greendigital
 
Charaka Samhita Sutra sthana Chapter 15 Upakalpaniyaadhyaya
Charaka Samhita Sutra sthana Chapter 15 UpakalpaniyaadhyayaCharaka Samhita Sutra sthana Chapter 15 Upakalpaniyaadhyaya
Charaka Samhita Sutra sthana Chapter 15 Upakalpaniyaadhyaya
Dr KHALID B.M
 
Lung Cancer: Artificial Intelligence, Synergetics, Complex System Analysis, S...
Lung Cancer: Artificial Intelligence, Synergetics, Complex System Analysis, S...Lung Cancer: Artificial Intelligence, Synergetics, Complex System Analysis, S...
Lung Cancer: Artificial Intelligence, Synergetics, Complex System Analysis, S...
Oleg Kshivets
 
Cervical & Brachial Plexus By Dr. RIG.pptx
Cervical & Brachial Plexus By Dr. RIG.pptxCervical & Brachial Plexus By Dr. RIG.pptx
Cervical & Brachial Plexus By Dr. RIG.pptx
Dr. Rabia Inam Gandapore
 
ACUTE SCROTUM.....pdf. ACUTE SCROTAL CONDITIOND
ACUTE SCROTUM.....pdf. ACUTE SCROTAL CONDITIONDACUTE SCROTUM.....pdf. ACUTE SCROTAL CONDITIOND
ACUTE SCROTUM.....pdf. ACUTE SCROTAL CONDITIOND
DR SETH JOTHAM
 
How to Give Better Lectures: Some Tips for Doctors
How to Give Better Lectures: Some Tips for DoctorsHow to Give Better Lectures: Some Tips for Doctors
How to Give Better Lectures: Some Tips for Doctors
LanceCatedral
 
BRACHYTHERAPY OVERVIEW AND APPLICATORS
BRACHYTHERAPY OVERVIEW  AND  APPLICATORSBRACHYTHERAPY OVERVIEW  AND  APPLICATORS
BRACHYTHERAPY OVERVIEW AND APPLICATORS
Krishan Murari
 

Recently uploaded (20)

Couples presenting to the infertility clinic- Do they really have infertility...
Couples presenting to the infertility clinic- Do they really have infertility...Couples presenting to the infertility clinic- Do they really have infertility...
Couples presenting to the infertility clinic- Do they really have infertility...
 
Novas diretrizes da OMS para os cuidados perinatais de mais qualidade
Novas diretrizes da OMS para os cuidados perinatais de mais qualidadeNovas diretrizes da OMS para os cuidados perinatais de mais qualidade
Novas diretrizes da OMS para os cuidados perinatais de mais qualidade
 
Pharynx and Clinical Correlations BY Dr.Rabia Inam Gandapore.pptx
Pharynx and Clinical Correlations BY Dr.Rabia Inam Gandapore.pptxPharynx and Clinical Correlations BY Dr.Rabia Inam Gandapore.pptx
Pharynx and Clinical Correlations BY Dr.Rabia Inam Gandapore.pptx
 
TEST BANK for Operations Management, 14th Edition by William J. Stevenson, Ve...
TEST BANK for Operations Management, 14th Edition by William J. Stevenson, Ve...TEST BANK for Operations Management, 14th Edition by William J. Stevenson, Ve...
TEST BANK for Operations Management, 14th Edition by William J. Stevenson, Ve...
 
Maxilla, Mandible & Hyoid Bone & Clinical Correlations by Dr. RIG.pptx
Maxilla, Mandible & Hyoid Bone & Clinical Correlations by Dr. RIG.pptxMaxilla, Mandible & Hyoid Bone & Clinical Correlations by Dr. RIG.pptx
Maxilla, Mandible & Hyoid Bone & Clinical Correlations by Dr. RIG.pptx
 
Phone Us ❤85270-49040❤ #ℂall #gIRLS In Surat By Surat @ℂall @Girls Hotel With...
Phone Us ❤85270-49040❤ #ℂall #gIRLS In Surat By Surat @ℂall @Girls Hotel With...Phone Us ❤85270-49040❤ #ℂall #gIRLS In Surat By Surat @ℂall @Girls Hotel With...
Phone Us ❤85270-49040❤ #ℂall #gIRLS In Surat By Surat @ℂall @Girls Hotel With...
 
Pulmonary Thromboembolism - etilogy, types, medical- Surgical and nursing man...
Pulmonary Thromboembolism - etilogy, types, medical- Surgical and nursing man...Pulmonary Thromboembolism - etilogy, types, medical- Surgical and nursing man...
Pulmonary Thromboembolism - etilogy, types, medical- Surgical and nursing man...
 
Physiology of Special Chemical Sensation of Taste
Physiology of Special Chemical Sensation of TastePhysiology of Special Chemical Sensation of Taste
Physiology of Special Chemical Sensation of Taste
 
Superficial & Deep Fascia of the NECK.pptx
Superficial & Deep Fascia of the NECK.pptxSuperficial & Deep Fascia of the NECK.pptx
Superficial & Deep Fascia of the NECK.pptx
 
For Better Surat #ℂall #Girl Service ❤85270-49040❤ Surat #ℂall #Girls
For Better Surat #ℂall #Girl Service ❤85270-49040❤ Surat #ℂall #GirlsFor Better Surat #ℂall #Girl Service ❤85270-49040❤ Surat #ℂall #Girls
For Better Surat #ℂall #Girl Service ❤85270-49040❤ Surat #ℂall #Girls
 
NVBDCP.pptx Nation vector borne disease control program
NVBDCP.pptx Nation vector borne disease control programNVBDCP.pptx Nation vector borne disease control program
NVBDCP.pptx Nation vector borne disease control program
 
Are There Any Natural Remedies To Treat Syphilis.pdf
Are There Any Natural Remedies To Treat Syphilis.pdfAre There Any Natural Remedies To Treat Syphilis.pdf
Are There Any Natural Remedies To Treat Syphilis.pdf
 
heat stroke and heat exhaustion in children
heat stroke and heat exhaustion in childrenheat stroke and heat exhaustion in children
heat stroke and heat exhaustion in children
 
Tom Selleck Health: A Comprehensive Look at the Iconic Actor’s Wellness Journey
Tom Selleck Health: A Comprehensive Look at the Iconic Actor’s Wellness JourneyTom Selleck Health: A Comprehensive Look at the Iconic Actor’s Wellness Journey
Tom Selleck Health: A Comprehensive Look at the Iconic Actor’s Wellness Journey
 
Charaka Samhita Sutra sthana Chapter 15 Upakalpaniyaadhyaya
Charaka Samhita Sutra sthana Chapter 15 UpakalpaniyaadhyayaCharaka Samhita Sutra sthana Chapter 15 Upakalpaniyaadhyaya
Charaka Samhita Sutra sthana Chapter 15 Upakalpaniyaadhyaya
 
Lung Cancer: Artificial Intelligence, Synergetics, Complex System Analysis, S...
Lung Cancer: Artificial Intelligence, Synergetics, Complex System Analysis, S...Lung Cancer: Artificial Intelligence, Synergetics, Complex System Analysis, S...
Lung Cancer: Artificial Intelligence, Synergetics, Complex System Analysis, S...
 
Cervical & Brachial Plexus By Dr. RIG.pptx
Cervical & Brachial Plexus By Dr. RIG.pptxCervical & Brachial Plexus By Dr. RIG.pptx
Cervical & Brachial Plexus By Dr. RIG.pptx
 
ACUTE SCROTUM.....pdf. ACUTE SCROTAL CONDITIOND
ACUTE SCROTUM.....pdf. ACUTE SCROTAL CONDITIONDACUTE SCROTUM.....pdf. ACUTE SCROTAL CONDITIOND
ACUTE SCROTUM.....pdf. ACUTE SCROTAL CONDITIOND
 
How to Give Better Lectures: Some Tips for Doctors
How to Give Better Lectures: Some Tips for DoctorsHow to Give Better Lectures: Some Tips for Doctors
How to Give Better Lectures: Some Tips for Doctors
 
BRACHYTHERAPY OVERVIEW AND APPLICATORS
BRACHYTHERAPY OVERVIEW  AND  APPLICATORSBRACHYTHERAPY OVERVIEW  AND  APPLICATORS
BRACHYTHERAPY OVERVIEW AND APPLICATORS
 

How deep learning reshapes medicine

  • 1. How deep learning reshapes medicine Hongyoon Choi Department of Nuclear Medicine, Seoul National University Hospital
  • 2. Current Finding Hypermetabolic solid mass in the RLL (8.3), suggesting lung cancer Small LNs in mediastinal 4R, 7 without hypermetabolism. Otherwise, no abnormal hypermetabolic lesion suggesting metastasis.
  • 3. Our Future? Google’s plenary lecture at AACR 2018
  • 4. Our Future? Vinyals, Oriol, et al. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2015. Vision Deep CNN Language Generating RNN “A group of people shopping at an outdoor market. There are many vegetables at the fruit stands.” Demo: http://vqa.cloudcv.org/
  • 5. CONTENTS INTRODUCTION Era of Medical Big Data DEEP LEARNING IN MEDICINE Brief overview of Deep learning Real application to medicine PERSPECTIVES What to solve New roles as a physician
  • 6. Era of medical big data
  • 7. INTRODUCTION  Single target  Pharmaceutical companies  Systemic, multimodal, multidimensional  Tech companies Current Future
  • 10. INTRODUCTION Flood of data Wearable polysomnography Wearable EKG Blood glucose monitoring system
  • 11. INTRODUCTION Previous Near Future Use Herceptin! ? What is the best choice for chemotherapy?
  • 12. Integrating multi-omics Genome Phenome Proteome Imaging Metabolome Biosensor CNN for imaging RNN for health care records Multidimensional data clustering INTRODUCTION + Deep learning Tsunami in medical fields
  • 13. INTRODUCTION 2016 1 billion USD 2024 AI Industry in Healthcare (from Global Market Insights) First Target is Image…
  • 14. INTRODUCTION ROC curve - better than dermatologists Esteva, Andre, et al. Nature 2017 Computer tech papers invade to Nature/Cell/Science & NEJM/Lancet/JAMA
  • 15. INTRODUCTION Diabetic Retinopathy  Better or equivalent to ophthalmologists Normal DM Gulshan, Varun, et al. JAMA 2016 ChestXnet  Equivalent/Superior to radiologists (?) Pranav Rajpurkar, … Andrew Ng, 2017. Arxiv
  • 16. INTRODUCTION FDA approve a device for diagnosing diabetic retinopathy (2018.4) AI-aided system (CT angiography for large vessel occlusion) Year of AI invasion to clinic FDA approve OsteoDetect(2018.5)
  • 17. INTRODUCTION • Really replace? – Imaging specialists’ role – Find clinical unmet needs ?
  • 18. 3 min Overview of Deep learning
  • 19. 3 min overview of deep learning • Deep learning: y=f(x) where y: label. x: data f cat
  • 20. 3 min overview of deep learning • Conventional Perceptron Lesion size Circularity Hounsfield unit x1 x2 x3 Y w1 w2 w3 b0 Activation function Output = 1 if Y >0 Output = 0 if Y <0 Output 1: Malignancy 0: Benign Find optimized W for minimized error
  • 21. 3 min overview of deep learning • Limitation in previous perceptron 0 1 1 3 5 7 8 8 0 0 0 1 3 3 5 3 0 0 1 2 4 7 7 1 0 0 2 3 8 5 7 6 2 5 8 8 8 4 9 5 0 0 8 8 6 4 2 3 128x128 =16,384 Output - Require good manual features - Raw data  Too big. - More layers?  Difficulty in learning
  • 22. 3 min overview of deep learning • MLP to Deep learning - Require good manual features - Raw data  Too big. - More layers?  Difficulty in learning • Automatic feature extraction from raw data CNN and RNN • New methods for deep layer training New activation function & Stochastic gradient descent • Methods for reducing overfitting
  • 23. 3 min overview of deep learning • Convolutional Neural Network • Skim locally, instead of look all things
  • 24. 3 min overview of deep learning identify line / some texture identify head lights and wheels identify Car! Hierarchical Recogntion
  • 25. 3 min overview of deep learning • Convolutional Neural Network
  • 26. 3 min overview of deep learning • Convolutional Neural Network ImageNet Challenge Results 28.2% 2010 25.8% 2011 16.4% 2012 Shallow model AlexNet 11.7% 2013 6.7% 2014 3.57% 2015 GoogleNet ResNet 8-layers 22-layers 152-layers
  • 27. 3 min overview of deep learning • Recurrent Neural Network
  • 28. 3 min overview of deep learning • More modules to train deep layers Problem of Vanishing Gradient Solved by nonlinearity function Sigmoid  ReLU , tanh, ELU, Leaky ReLU Sigmoid ReLU
  • 29. 3 min overview of deep learning • More modules to reduce overfitting Problem of overfitting Apple
  • 30. 3 min overview of deep learning • More modules to reduce overfitting Dropout
  • 31. Overview of deep learning • Current Concept of Deep learning Deep layered neural network Output + Data type-specific layers Convolution Recurrent Modification for training + ReLU activation SGD training Dropout Batch normalization Variable Loss …
  • 32. Deep learning in Medicine Particularly for nuclear medicine
  • 33. Deep learning in Medicine DL for medical imaging: Supervised learning using CNN f cat f: CNN f Lung cancer Simple application of CNN for diagnosis
  • 34. Deep learning in Medicine Diabetic Retinopathy  Better or equivalent to ophthalmologists Normal DM Gulshan, Varun, et al. JAMA 2016 ChestXnet  Equivalent/Superior to radiologists (?) Pranav Rajpurkar, … Andrew Ng, 2017. Arxiv DL for medical imaging: Supervised learning using CNN
  • 35. Deep learning in Medicine DL for medical imaging: Supervised learning using CNN AD & NC MCI-converter & non-converter FDG and amyloid PET to predict future cognitive decline Choi H and Jin KH KSNM 2016; Behav Brain Res 2018
  • 36. Deep learning in Medicine DL for medical imaging: Supervised learning using CNN FDG and amyloid PET to predict future cognitive decline Accuracy AD vs. NC 96.0% 85.4% 80.7% Deep CNN FDG quantification AV45 quantification Accuracy MCI conversion 84.2% 75.4% 80.1%84.2% Deep CNN-based 82% Moraldi et al. 2016 (MRI+Clinical Variables) 78% Zhang et al. 2013 (FDG, MRI+Clinical) Feature Extraction + Machine learning 72% Shaffer et al. 2013 (FDG, MRI, CSFl)
  • 37. Deep learning in Medicine Output score measured by baseline PET & 3-year cognitive score f p(Alzheimer|X) Direct biomarker A single parameter Choi H and Jin KH KSNM 2016; Behav Brain Res 2018
  • 38. Deep learning in Medicine https://adfdgpet.appspot.com Online Demo Input file Web application Output: likelihood for AD & predicted cognitive score Output: Cognitive dysfunction-related map p(Alzheimer|X)
  • 39. Deep learning in Medicine SWEDD: Clinically PD, Image normal Expert1 Expert2 Conventional quantification SWEDD Deep learning- abnormal 2-year follow-up 80% Abnormal Deep learning : Redefine SWEDD diagnosis Choi, H., … ,Lee. D.S. NeuroImage: Clinical 2017 Dopamine transporter imaging for Parkinson’s disease diagnosis 3D CNN model normal Parkinson
  • 40. Deep learning in Medicine Laborious Work Replaced by DL: Segmentation Choi, H., & Jin, K. H. J Neurosci Methods 2016 de Brebisson, et al. CVPR 2015.
  • 41. Deep learning in Medicine Laborious Work Replaced by DL: Detection Liu Y, et al. Arxiv 2017
  • 42. Deep learning in Medicine Enhance Image Acquisition & Quality Dahl et al. Arxiv 2017 Standard dose 1/200 Dose 1/200 Dose + CNN
  • 43. Deep learning in Medicine Image Generation Cat Cat Common deep learning model Generative model z = f(x) where x: data, z: discriminative features f: classifier model x = g(z) where x: data, z: latent g: generation function
  • 44. Deep learning in Medicine Generative Adversarial Network z~N(0,1) G: generator Karras T, et al. Arxiv 2017 G: generator Isola P, et al. arxiv, 2016.
  • 45. Generative Adversarial Network Structural MR generation from PET Florbetapir PET Generator: U-net Skip connection Generated MR PETandgeneratedMRPETandrealMR Discriminator Real or Fake Generative Adversarial Networks for MR generation z G(z) z & G(z) z & x Choi H and Lee DS, J Nucl Med 2017. Deep learning in Medicine RealMRI Generated MRI 18F-Florbetapir PET
  • 46. Deep learning in Medicine PET-based normalization SUVR measurement MRI-based normalization SUVR measurement MR generation Generated MRI-based normalization transformation apply to PET SUVR measurement Choi H and Lee DS, J Nucl Med 2017. Generated-MRbased quantification PreviousPET-based quantification Gold standard Gold standard
  • 47. Deep learning in Medicine Conditional Generation Antipov G, Arxiv 2017
  • 48. Encoder Latent features Generator + Age Latent features + Age VAE model for brain PET generation  Brain metabolism aging movie Choi H,… Lee DS. Biorxiv 2017 Deep learning in Medicine Conditional Generation
  • 49. Choi H,… Lee DS. Biorxiv 2017 Estimating normal population distribution Deep learning in Medicine
  • 50. Deep learning in Medicine • Individual vs population – Aging of individual brain : Comparing with virtual population Population distribution of brain metabolism at each age by iterative generating from the VAE model Generator Latent features N(0,1) Random sampling from normal distribution + Age Age 50 55 60 65 70 75 • Evaluating individual brain’s aging compared with general population • Define ‘pathologic aging’ Choi H,… Lee DS. Biorxiv 2017
  • 51. Practical Issues & Perspectives
  • 52. Practical Issues Gap between Tech-initiated DL and Clinician’s need We just need BMI to know obesity!
  • 53. Practical Issues – Various purposes • More than simple diagnosis: Prognosis, Disease status monitor, Response prediction. – Ambiguous ground-truth • Diagnosis is not a simple classification Gap between Tech-initiated DL and Clinician’s need
  • 54. Practical Issues Clinical diagnosis: Spectrum instead of a clear-cut classification Blood pressure = a type of biomarker Cardiovascular eventrisk 140/90 HypertensionNormal Blood Pressure (one-dimensional) Quantitative biomarker reflecting prognostic outcome Direct Most diseases… Multiple domain and high-dimensional Not rely on single measurement
  • 55. Practical Issues SUVmax A value maximize information of whole voxels to represent patient’s prognosis What we expect: Deep learning as a ‘best parameter’ extractor from high-dimensional data Simple Diagnosis << Biomarker
  • 57. Practical Issues Where to get Hospital • Best source • Practical model validation: Real world problems • Ethical issues and relatively small data size Public database • Easy to get • Red ocean • Limitation in real-world validation
  • 58. Practical Issues • Public databases – NM images included • The Cancer Imaging Archives • PPMI (Parkinson), ABIDE (Autism), ADHD-200 (ADHD), OASIS (Brain MR), HCP (Brain MR) • ADNI (Alzheimer PET, MRI) – Chest X-ray8 dataset – DDSM (mammography) – Kaggle diabetic retinopathy – MICCAI brain tumor segmentation challenges
  • 59. Practical Issues Data Size – Does deep learning require big data? – Always more than 100,000 images? Answer : No Answer : Yes
  • 60. Practical Issues Data Size – Voxel-based training: Segmentation/Superresolution etc. – Image generation (Pix2Pix) :~100 3d volumes – Image augmentation : rotation/flipping (but, cautious) – Task and complexity dependent: AD vs normal was trained by ~300 cases.
  • 61. Practical Issues Image labels – Unlabeled data >> Labeled data • e.g. all FDG PET/CT >> Baseline breast cancer FDG PET/CT – Unbalanced label • e.g. Lymphoma PET >> Adrenocortical carcinoma PET  Importance of unsupervised & semi- supervised learning
  • 62. Practical Issues Tumor metabolism estimated by FDG PET image Gene networks Gene expression data with FDG PET image  ~ 40 pairs Gene expression data alone  ~ 1000 cases Image labels Example of semisupervised learning in practical problem Choi H and Na KJ. Theranostics 2018
  • 63. Practical Issues Expressiondata Tumor metabolism (maximum SUV) Encoder Decoder HiddenFeatures 626 64 1 Supervised Training (n = 20) Unsupervised Training (n = 226) Dimensions Example of semisupervised learning in practical problem Semisupervised learning - Supervised training combined with autoencoder Choi H and Na KJ. Theranostics 2018
  • 64. Practical Issues At report… - Lung cancer cannot be excluded. Rec> “Clinical correlation is recommended” Issues of Uncertainty Suggestive of cancer Probably, cancer Possibly, cancer Cancer cannot be excluded Recommend further exam Deep learning trained by Natural Images “Cat”
  • 65. Practical Issues Data distribution Dataset for model training/validation Alzheimer Normal Real data at Hospital Normal Alzheimer FTD DLB PSP Depression Real data at Community Normal • Will DL model work at real clinic? • How to deal with rare/unseen disease? Aged person with subjective cognitive decline
  • 66. Perspectives as a Physician  Medicine is too complex  Problem of responsibility  Too many empirical things  Didn’t you see AlphaGo?  Domination is already started  At least, many doctors will lose their job
  • 67. Perspectives as a Physician Tsunami in medical fields Deep Learning Big Data
  • 68. Perspectives as a Physician • Gaps between Tech. & Hosp. • Become an expert of data • Roles will be changed – Not just replace MD’s role – Laborious things replaced by AI – New information produce new jobs
  • 69. Perspectives as a Physician • Roles will be changed – Laborious things replaced by AI – New information produce new jobs • Performed by AI ~1 sec. • New information for medical decision (Cortical thickness, Tumor volume, etc as a clinical routine)
  • 70. Perspectives as a Physician Disruptive Innovation: Raw medical & healthcare data Diet + Previous Glucose Level Future Glucose Level & Scheduling Insulin Sugar.iq from Medtronic (FDA approved 2016.9)
  • 71. Perspectives as a Physician Disruptive Innovation: Raw medical & healthcare data
  • 72. Perspectives as a Physician Disruptive Innovation: Raw medical & healthcare data HTC DeepQ Tricoder Predicting PVC from daily EKG Diagnosis of otitis media
  • 73. Perspectives as a Physician • Accelerating changes to daily healthcare from hospital-based health Deep learning facilitates left-shifting
  • 74. Perspectives as a Physician Future medical decision ConcatenatingFeatures • Diagnosis • Management Plan AbnormalNormal Integrated biomarker based on DL RiskatDeath Human Cannot Do!
  • 75. Perspectives as a Physician Future medical decision Previous classification Single target Based on some receptors Breast cancer classification Current classification Multiple targets Based on ~50 transcripts
  • 76. Perspectives as a Physician Future medical decision Breast cancer classification Genome Phenome Proteome Imaging Metabolome Biosensor Unsupervised learning Integrative classification Future classification Multiomics targets Based on unsupervised learning
  • 77. Perspectives as a Physician Empirical -based Evidence -based Data -driven “약을 써보니 낫더라 “RCT를 해보니 이 약은 효과가 유의하게 있다 “모든 이 사람의 활동 데이터, Omics 데이터, 영상 정보 등을 통합하여 볼 때 이 약이 가장 적합할 것이다 Deep Learning

Editor's Notes

  1. 강의순서 Intro Intro of Intro: 조직학 + 최근구글비디오 – 괴리. 누가 biomedical field를 점령할 것인가. Data의 시대 각론 Deep learning이 어떤 역할 : Image data augmentation + Image diagnosis Omics data 등. 연속측정가능, raw data, easily acquisible data로 부터 의미 추출하기 매우 낮은 수준의 데이터를 어떻게 활용할 것인가. EMR data부터 기본 monitor장비에서의 데이터, 나아가서 wearable device. 파괴적 혁신을 일으키는 deep learning Biomarker와 data 통합에 대해. 미래의료와 딥러닝관련 앞으로 의사가 해야할 일? 무엇의 전문가가 될 것인가. 의사의 role.
  2. SO, I’ll briefly introduce current trends in medical fields in terms of data a science. I can darely say your role as an expert in data will substantially change medical environment. Next, I’ll introduce my recent researches and related researches related to deep learning-based biomarker. And, I’ll share some practical issues in deep learning and perspectives.
  3. Recently, many doctors’ intrests were totally changed. When we get some meeting with doctors, one of the common themes of chat is Artificial intelligence. I think now, it’s era of medical data scientists.
  4. Health care market is really big market. In US, healthcare market occupies 15% of GDP and in Korea, around 8% of GDP. Then, who gets this money? And who is the big brother in this market?. So far, in this market, ruler was pharmaceutical companies. They target a specific symptoms, specific disease and even more, specific molecules. But all of things are changed. Biomedical data are markedly increased and role of the analysis of these data is increased. Thus, tech companies are entering healthcare markets and dominating the market.
  5. For example, a representative AI company, deepmind already launches deepmind health and they have tried to develop medical AI applications. Google and google’s subcompanies such as verily have developed various biotechnologies based on IT. IBM Watson is another famous AI applciations in hospital. Apple also develops various healthcare devices such as apple watch combined with electrocardiogram. Even more, healthcare startups are rapidly increasing and some other hardware companies such as Samsung also entering into health care market.
  6. The core is flood of data. Previously, researches only focused on single molecule, single gene. What the role of a specific gene is. But, today, its thousand dollar genome era. Only 1000$ is required for analyzing personal genome.
  7. Furthermore, due to wwearble technology, medical data are collected from common life. Not from hospital. Iphone combined EKG was approved by FDA which enables routine checkup for general people. Continous blood glucose monitoring system can generate 24 hr, 365 days blood glucose data.
  8. Paroxysmal Supraventricular tachycardia
  9. This change was based on two factors. Increased data and appropriate algorith, deep learning. These synergistic integration of these two factors induce tsumani in medical fields which is inevitable.
  10. Besides, AI application to medical fields is also markedly increasing. Global market of AI industry is rapidly grown and will occupy around 25 billion dollars. And among various subfields in medical area, the first target is image analysis. As boom of deep learning was started from ImageNet challenge and many innovative deep learning models have been developed for image recognition and process, the first target of AI in medicine is medical imaging.
  11. This AI doctor seems to be becoming reality in recent research.. This nature paper was published by Stanford group. They used more than hundred thousands of skin lesion images and deep learning model, Inception, GoogleNet, to diagnose skin cancer. The performance was better than dermatologists and they develops prototype which can be embedded in smartphone app.
  12. For discriminating fundoscopic image, a job of ophthalmologists, DL model shows better performance for discriminating diabetic retinopathy. Recently another big guy of AI, Andrew Ng introduce Chest Xnet which interpret chest X-ray image equivalent to radiologists.
  13. For discriminating fundoscopic image, a job of ophthalmologists, DL model shows better performance for discriminating diabetic retinopathy. Recently another big guy of AI, Andrew Ng introduce Chest Xnet which interpret chest X-ray image equivalent to radiologists.
  14. Then, I’ll lose my job? My reading will be totally replaced by AI? I think DL application to medical imaging should not only focus on replacement of doctor. Deep learning can play important role in medicine and innovate and progress by solving clinical unmet needs. So, I want to introduce what clinician want to DL. And because of unique characteristic of medical imaging data, some practical issues in DL application exist. So I’ll show you in next part.
  15. So, the input of the model was both PET volumes and the output was probability of Alzheimer. This probability score can be used for predicting cognitive outcome in MCI subjects as well as discriminating AD. The y-axis represents cognitive score and high CDR means poor cognitive function. The output of the model of baseline image was correlated with future cognitive score changes.
  16. This 3D CNN model was embedded into web application. If you have some FDG brain PET images, entering the image can produce the probability of Alzheimer and the cognitive dysfunction-related map like this.
  17. So, from supervised learning, I can develop a deep CNN model that differentiate PD from normal with better performance than me, and my colleagues. Furthermore ,a specific disease entity exists in PD, scans without evidence of dopaminergic deficit. This entitiy is a Parkinson’s disease but don’t show abnormality in Dopmaine imaging. This abnormality has been determined by human reading, but from our DL model, we found that some of SWEDD patients were not really SWEDD as they have imaging abnormality.
  18. This process was facilitated by deep learning segmentation. This was my study, one of the first study that specific brain region segmentation using deep learning.
  19. This process was facilitated by deep learning segmentation. This was my study, one of the first study that specific brain region segmentation using deep learning.
  20. This process was facilitated by deep learning segmentation. This was my study, one of the first study that specific brain region segmentation using deep learning.
  21. I can summarize in a word what clinician want for DL is a biomarker. Technologist-initiated DL model focuses on above or human level image interpretation. Thus, sort of arxiv papers chase better performance in disease lesion detection, segmentation and classification. But really, clinical want to know whether the subject die or not and if then, when will die. And what the proper treatment is. This answer is directly related to biomarker.
  22. This gap between technicians and clinicians are started from these issues. Medical imaging has various purposes, not only for simple diagnosis. Clinician acquire medical imaging and exams to know patients’ outcome and to monitor disease status during treatment. Furthermore, diagnosis is not clear-cut as ImageNet challenge classification.
  23. Clinical diagnosis is spectrum instead of clear cut classes. For example, hypertension is defined by higher systolic blood pressure 140 or higher diastolic blood pressure 90. But Blood pressure is a spectrum. Just a criteria was used to determine initiation of antihypertensive treatment. Because this point is related to increased risk of cardiovascular disorders. Clincian want a parameter, blood pressure, instead of definitve diagnosis of hypertension. That is, blood pressure is a one dimensional parameter, quantitative biomarker that reflects prognostic outcome in terms of cardiovascular disorders. However, many disorders are defined by multiple domains and highdimensional data instead of single parameter such as blood pressure.
  24. For example, brain diseases are diagnosed by symptoms, brain imagng, lab test and drug responses. Doctor makes decision qualitatively and empirically by considering patients’ predicted outcome. What clinician want to deep learning or AI is ‘summarized biomarker’ which reflects these high-dimensional and multimodal data. Thus, instead of simple diagnosis, DL shoud focus on generating a simple feature score, biomarker, which reflect patients’ outcome like blood pressure.
  25. Confused…