Informatics in disease management: What will the future bring?

Mike Hogarth, MD, FACMI, FACP
Mike Hogarth, MD, FACMI, FACPClinical Research Information Officer at University of California San Diego Health en University of California San Diego Health
Informatics in
Disease
Management:
What will the
Future bring?
Michael Hogarth, MD, FACMI, FACP
Clinical Research Information Officer, and
Director of Biomedical Informatics, Altman
Clinical Translational Research Institute
University of California San Diego Health
Informatics and Disease Management -
here comes machine learning (ML) and
artificial intelligence (AI)!
FYI: We have been using “machine learning”
for over 200 years
http://www.andreykurenkov.com/writing/ai/a-brief-history-of-neural-nets-and-deep-learning/
So what’s new?
• most AI/ML algorithms are not new
• why is “machine learning” and “deep learning” a new revolution
in the use of AI?
https://mc.ai/awesome-ai-the-guide-to-master-artificial-intelligence/
Key Informatics Trends in 2019...
1. HITECH and EHR adoption now at
over 90%
– has made massive amounts of
healthcare data (“dirta”)
available
– fuel for “machine learning”
based predictive systems
2. High performance computing is
highly commoditized
– Commercial clouds - “data center
at your fingertips”
– example: 11 machine cluster for
6 hours = $3.96 (Jan 2019)
– cheap computing to generate
machine learning models
3. Machine learning engines and
algorithms are commoditized
(ML as a service)
The commercial attraction to ML/AI in Healthcare
https://www.healthcarebusinessinsights.com/blog/information-technology/growing-value-ai-machine-learning-healthcare/
Why is healthcare such an attractive target for
optimization?
Big tech sees significant opportunities
https://healthitanalytics.com
Research into ML/AI “models” for various tasks
-- a very hot area!
Yes, there is “magical thinking” in this space...
• MD Anderson partnered
with IBM in 2012 to help
oncologists with diagnoses
and choosing therapeutic
courses
• IBM Watson Manager -
“training alongside doctors
to do what they can’t”
• Feb 2017 - project canceled
after MD Anderson had
already paid IBM over $39M
(originally priced at
$2.4M)!!!
But don’t throw out the baby with the bath water...
Using ML on 36 cases evenly
split between AML and
normal, an algorithm
correctly distinguished
disease progression in 90% of
cases of relapsing AML, and
was 100% correct in
distinguishing between AML
and normal bone marrow
How does machine learning work?
https://www.edureka.co/blog/what-is-machine-learning/
Machine Learning Approaches
https://www.kdnuggets.com/2017/11/3-different-types-machine-learning.html
ML Basics: Supervised
• training data set
• data used to “train the
model”
• Once ‘trained’, the
model can make
predictions or suggest
decisions with new
data inputs
https://bigdata-madesimple.com/machine-learning-explained-understanding-supervised-unsupervised-and-reinforcement-learning/
ML Basics: Unsupervised
• The ML model
‘learns’ through self-
finding structures or
associations in the
data
• Will automatically
assemble “clusters”
based on variables
available in the data
set
https://bigdata-madesimple.com/machine-learning-explained-understanding-supervised-unsupervised-and-reinforcement-learning/
ML Basics: Reinforcement Learning
• Paradigm is a “software
agent” that attempts to
achieve as close to the
“goal state” as possible.
• Goal state = best outcome
• Follows a concept of “hit
and trial” where the agent
is rewarded or penalized
(using points) for a
particular action or
answer.
• It attempts to ‘learn’ using
hit/trial and can then
predict with new data
https://bigdata-madesimple.com/machine-learning-explained-understanding-supervised-unsupervised-and-reinforcement-learning/
Artificial Neural Networks (ANNs)
• AI system loosely based on biological
neural networks in brains
• not an algorithm - its a framework for
machine learning algorithms that work
by using multi-weighted inputs across
multiple collaborating nodes to process
data inputs
• can be supervised or unsupervised ML
• first neural networks were implemented
in circuits
– neurophysiologist Warren McCulloch and
mathematician Walter Pitts wrote paper
theorizing how neurons might work in a
network with weighted inputs
– first tested in circuitry - Mark I Perceptron
– Software based NN’s became known as
“artificial neural networks”
Mark I Perceptron with patchboard
to connect nodes, potentiometers as
weights
“Deep” Learning
What is “deep learning”?
● uses a multi-layered “artificial neural
network”
● first layer (captures raw data -- input
layer)
● “hidden layers” -- integrate the values of
the raw input nodes to derive
“transformed features”
● predicts outcomes or classifies based on
transformed features
“Convolutional” Neural Networks (CNN)
• a type of multi-layered (deep) ANN designed for recognition of visual
features
• originally devised in 1988 by Yann LeCun at AT&T Labs to recognize
handwritten digits
• inspired by the connectivity pattern seen between neurons in the
visual cortex
• Convolution layers apply a convolution operation to the input, passing
result to next layer.
• Convolution operation emulates response of an individual neuron to
visual stimuli. Pooling coalesces outputs from one layer into a single
neuron in the next layer
• significantly resurging interest in CNNs due to:
– availability of labeled digital image sets (large training sets)
– high performance computing with graphical processing units
(GPU) necessary for the significant computation required
Machine Learning (ML) image
interpretation - supervised learning
Convolutional (deep) Neural Network and Dermatology
• trained a CNN with 1.29M
images of 2,032 different
conditions from 18
clinician-curated
repositories
• tested CNN based classifier
on 376 cases, head-to-head
with 21 board-certified
dermatologists on two
scenarios with real biopsy-
proven lesions
– differentiating benign
seborrheic keratoses from
keratinocyte carcinomas
(n=135)
– differentiating malignant
melanoma from benign nevi
(n=130;n=111)
• Results: CNN performance
was as good as the
dermatologists
(all AUCs over 0.91)!
Convolutional (deep) Neural Network and Echocardiography
“view classification”
• trained a CNN with 267
transthoracic echo studies
(200,000 images) to
differentiate 15 different
standard echo views (12
video, 3 still image views -
parasternal long axis, right
ventricular inflow, short
axis at mitral level, etc..)
• tested with 27 echo studies
(20,000 images)
– video views -- 97.8%
accurate
– still images - 91.7%
accurate
• 4 board certified
echocardiographers on the
27 echo studies --- 79.4%
average accuracy
Deep Learning and Interpretation of Retinal Images for
diabetic retinopathy
• trained a ML systems
with 494,661 retinal
images
• validated (tested) using
dataset of 71,896
images from 14,880
patients
• detection of vision-
threatening diabetic
retinopathy: AUC 0.958
(95% CI)
• detection of referable
diabetic retinopathy:
AUC=0.936(95% CI)
Reinforcement Learning in Sepsis - the “AI Clinician”
• developed an AI Clinician computational
model using reinforcement learning to
suggest optimal treatment of sepsis
regarding fluid resuscitation and pressor
administration
• AI clinician’s goal - maximize the
probability of a good outcome
• Developed using two large ICU databases
– MIMIC III and eICU Research
Institute Database (eRI)
– 17,083 admissions from MIMIC
– 79,073 admissions from eRI
• Performance:
– “the value of the AI Clinician’s
selected treatment is, on average,
reliably higher than human clinicians
– Validation done on a separate
dataset of cases.
– Found that cases where the
clinician action most closely
matched what AI Clinician’s would
have done had the lowest
mortality!
Artificial Intelligence Sepsis Expert (AISE)
• Nemati and colleagues at
Emory trained and validated on
31,000 Emory ICU admissions
– used 60 features: ECG, BP, HR,
MAP, SpO2, temp, GCS, FiO2,
Chem7 analytes, ABG, lactate,
etc..
• AISE predicted sepsis an
average of 4-8hrs in advance
(AUC of 0.85) of non-assisted
recognition of impending
sepsis
(using Sepsis 3 definition)
Regulating AI-based Clinical Decision Support -
is the FDA ready?
2017 FDA guidance on decision support
software systems
• Clinical Decision Support (CDS) vs.
Patient Decision Support (PDS)
• With guidance from 21st Century Cures
Act, FDA introduced new guidance to
make it easier for software developers to
obtain clearance (to spark innovation)
– Low risk =(weight management,
mindfulness tools, etc.. not under
FDA oversight
– Products intended for the
diagnosis, cure, mitigation,
prevention, or treatment of a
condition will fall under FDA
oversight
– Software intended to process or
analyze medical images, signals
from diagnostic devices (ECG, EEG)
will continue under FDA oversight
Recent FDA Decisions for AI Based Systems
- De Novo: regulatory pathway for low-to-moderate risk
devices of a new type for which there is no legally
marketed predicate device with substantial equivalence
- 510(k): a pre-market submission to FDA to demonstrate
the device is at least a safe and effective (equivalent) to
a legally marketed device that is not subject to
premarket approval
- Class II: moderate-to-high
risk device
- Special controls:
- performance
standards
- post-market
surveillance
- special labeling
This has now also captured the attention of the
legal profession...
https://www.americanbar.org/groups/health_law/publications/aba_health_esource/2016-2017/october2017/machinelearning/
The future of disease management
could be bright...
https://ai100.stanford.edu/sites/g/files/sbiybj9861/f/ai100report10032016fnl_
singles.pdf
or it could go horribly wrong...
http://blog.petrieflom.law.harvard.edu/2017/10/26/
One thing is certain, AI has become an
integral part of society -- it is here to stay
Domino’s Pizza delivered by autonomous driving vehicle
A common question:
“When will we be “replaced” by smart machines?”
Surgeon
Nephrologist
Questions?
Tioga Pass, Yosemite
1 de 34

Recomendados

EHR v2.0: Optimizing Usability and Utility por
EHR v2.0: Optimizing Usability and UtilityEHR v2.0: Optimizing Usability and Utility
EHR v2.0: Optimizing Usability and UtilityMike Hogarth, MD, FACMI, FACP
2.8K vistas34 diapositivas
Data Quality in Healthcare: An Important Challenge por
Data Quality in Healthcare: An Important ChallengeData Quality in Healthcare: An Important Challenge
Data Quality in Healthcare: An Important ChallengeMike Hogarth, MD, FACMI, FACP
658 vistas29 diapositivas
Cerner ppt por
Cerner pptCerner ppt
Cerner pptCHIRANTAN BOSE MD.,IFCAP.
1.8K vistas26 diapositivas
Connected Health & Me - Matic Meglic - Nov 24th 2014 por
Connected Health & Me - Matic Meglic - Nov 24th 2014Connected Health & Me - Matic Meglic - Nov 24th 2014
Connected Health & Me - Matic Meglic - Nov 24th 2014ipposi
272 vistas37 diapositivas
Big Data and Smart Healthcare por
Big Data and Smart Healthcare Big Data and Smart Healthcare
Big Data and Smart Healthcare Sujan Perera
2.6K vistas15 diapositivas
Machine Learning and Prediction in Medicine por
Machine Learning and Prediction in MedicineMachine Learning and Prediction in Medicine
Machine Learning and Prediction in MedicineChad You
78 vistas16 diapositivas

Más contenido relacionado

La actualidad más candente

AI in Healthcare por
AI in HealthcareAI in Healthcare
AI in HealthcarePaul Agapow
151 vistas33 diapositivas
The Life-Changing Impact of AI in Healthcare por
The Life-Changing Impact of AI in HealthcareThe Life-Changing Impact of AI in Healthcare
The Life-Changing Impact of AI in HealthcareKalin Hitrov
135 vistas55 diapositivas
2015 iHT2 Health IT Beverly Hills Summit por
2015 iHT2 Health IT Beverly Hills Summit 2015 iHT2 Health IT Beverly Hills Summit
2015 iHT2 Health IT Beverly Hills Summit Health IT Conference – iHT2
740 vistas45 diapositivas
Big Data in Healthcare: Hype and Hope on the Path to Personalized Medicine por
Big Data in Healthcare: Hype and Hope on the Path to Personalized MedicineBig Data in Healthcare: Hype and Hope on the Path to Personalized Medicine
Big Data in Healthcare: Hype and Hope on the Path to Personalized MedicineNew York eHealth Collaborative
7K vistas36 diapositivas
Clinical Analytics por
Clinical AnalyticsClinical Analytics
Clinical AnalyticsMichael Bice
2.7K vistas30 diapositivas
Introduction To Medical Data por
Introduction To Medical DataIntroduction To Medical Data
Introduction To Medical DataDr Neelesh Bhandari
3.6K vistas19 diapositivas

La actualidad más candente(20)

AI in Healthcare por Paul Agapow
AI in HealthcareAI in Healthcare
AI in Healthcare
Paul Agapow151 vistas
The Life-Changing Impact of AI in Healthcare por Kalin Hitrov
The Life-Changing Impact of AI in HealthcareThe Life-Changing Impact of AI in Healthcare
The Life-Changing Impact of AI in Healthcare
Kalin Hitrov135 vistas
Clinical Analytics por Michael Bice
Clinical AnalyticsClinical Analytics
Clinical Analytics
Michael Bice2.7K vistas
Machine learning in health data analytics and pharmacovigilance por Revathi Boyina
Machine learning in health data analytics and pharmacovigilanceMachine learning in health data analytics and pharmacovigilance
Machine learning in health data analytics and pharmacovigilance
Revathi Boyina175 vistas
Using Big Data to Personalize the Healthcare Experience in Cancer, Genomics a... por DrBonnie360
Using Big Data to Personalize the Healthcare Experience in Cancer, Genomics a...Using Big Data to Personalize the Healthcare Experience in Cancer, Genomics a...
Using Big Data to Personalize the Healthcare Experience in Cancer, Genomics a...
DrBonnie3604K vistas
Accure ai healthcare offering v4 por Accureinc
Accure ai healthcare offering v4Accure ai healthcare offering v4
Accure ai healthcare offering v4
Accureinc315 vistas
Data Mining in Health Care por ShahDhruv21
Data Mining in Health CareData Mining in Health Care
Data Mining in Health Care
ShahDhruv211.4K vistas
Role of Big Data in Medical Diagnostics por Nishant Agarwal
Role of Big Data in Medical DiagnosticsRole of Big Data in Medical Diagnostics
Role of Big Data in Medical Diagnostics
Nishant Agarwal1.4K vistas
IBM Watson for Healthcare por IBM_CH
IBM Watson for HealthcareIBM Watson for Healthcare
IBM Watson for Healthcare
IBM_CH1.3K vistas
Big Data Provides Opportunities, Challenges and a Better Future in Health and... por Cirdan
Big Data Provides Opportunities, Challenges and a Better Future in Health and...Big Data Provides Opportunities, Challenges and a Better Future in Health and...
Big Data Provides Opportunities, Challenges and a Better Future in Health and...
Cirdan276 vistas
The data explosion along the care cycle (Dell Healthcare) por Eric Van 't Hoff
The data explosion along the care cycle (Dell Healthcare)The data explosion along the care cycle (Dell Healthcare)
The data explosion along the care cycle (Dell Healthcare)
Eric Van 't Hoff2.8K vistas
Importance of patient voice por Repustate
Importance of patient voiceImportance of patient voice
Importance of patient voice
Repustate69 vistas

Similar a Informatics in disease management: What will the future bring?

AI-powered Medical Imaging Analysis for Precision Medicine por
AI-powered Medical Imaging Analysis for Precision MedicineAI-powered Medical Imaging Analysis for Precision Medicine
AI-powered Medical Imaging Analysis for Precision MedicineSean Yu
26 vistas34 diapositivas
Image Analytics In Healthcare por
Image Analytics In HealthcareImage Analytics In Healthcare
Image Analytics In HealthcareAlgoAnalytics Financial Consultancy Pvt. Ltd.
2.3K vistas16 diapositivas
Presentación del nodo Valenciano en Bonn en el comité de Euro-BioImaging por
Presentación del nodo Valenciano en Bonn en el comité de Euro-BioImagingPresentación del nodo Valenciano en Bonn en el comité de Euro-BioImaging
Presentación del nodo Valenciano en Bonn en el comité de Euro-BioImagingmaigva
805 vistas12 diapositivas
인공지능 논문작성과 심사에관한요령 por
인공지능 논문작성과 심사에관한요령인공지능 논문작성과 심사에관한요령
인공지능 논문작성과 심사에관한요령Namkug Kim
2.5K vistas92 diapositivas
Randy Goebel for the KIEF 2018. FROM DATA TO ECONOMIC VALUE por
Randy Goebel for the KIEF 2018. FROM DATA TO ECONOMIC VALUERandy Goebel for the KIEF 2018. FROM DATA TO ECONOMIC VALUE
Randy Goebel for the KIEF 2018. FROM DATA TO ECONOMIC VALUEKyiv International Economic Forum
74 vistas13 diapositivas
Intelligent data analysis for medicinal diagnosis por
Intelligent data analysis for medicinal diagnosisIntelligent data analysis for medicinal diagnosis
Intelligent data analysis for medicinal diagnosisIRJET Journal
34 vistas14 diapositivas

Similar a Informatics in disease management: What will the future bring?(20)

AI-powered Medical Imaging Analysis for Precision Medicine por Sean Yu
AI-powered Medical Imaging Analysis for Precision MedicineAI-powered Medical Imaging Analysis for Precision Medicine
AI-powered Medical Imaging Analysis for Precision Medicine
Sean Yu26 vistas
Presentación del nodo Valenciano en Bonn en el comité de Euro-BioImaging por maigva
Presentación del nodo Valenciano en Bonn en el comité de Euro-BioImagingPresentación del nodo Valenciano en Bonn en el comité de Euro-BioImaging
Presentación del nodo Valenciano en Bonn en el comité de Euro-BioImaging
maigva805 vistas
인공지능 논문작성과 심사에관한요령 por Namkug Kim
인공지능 논문작성과 심사에관한요령인공지능 논문작성과 심사에관한요령
인공지능 논문작성과 심사에관한요령
Namkug Kim2.5K vistas
Intelligent data analysis for medicinal diagnosis por IRJET Journal
Intelligent data analysis for medicinal diagnosisIntelligent data analysis for medicinal diagnosis
Intelligent data analysis for medicinal diagnosis
IRJET Journal34 vistas
DESIGN AND IMPLEMENTATION OF CARDIAC DISEASE USING NAIVE BAYES TECHNIQUE por IRJET Journal
DESIGN AND IMPLEMENTATION OF CARDIAC DISEASE USING NAIVE BAYES TECHNIQUEDESIGN AND IMPLEMENTATION OF CARDIAC DISEASE USING NAIVE BAYES TECHNIQUE
DESIGN AND IMPLEMENTATION OF CARDIAC DISEASE USING NAIVE BAYES TECHNIQUE
IRJET Journal9 vistas
The XNAT imaging informatics platform por imgcommcall
The XNAT imaging informatics platformThe XNAT imaging informatics platform
The XNAT imaging informatics platform
imgcommcall1.4K vistas
Artificial Intelligence for Automated Decision Support Project por Valerii Klymchuk
Artificial Intelligence for Automated Decision Support ProjectArtificial Intelligence for Automated Decision Support Project
Artificial Intelligence for Automated Decision Support Project
Valerii Klymchuk3.1K vistas
AI IN PATH final PPT.pptx por DivyaGaurav4
AI IN PATH final PPT.pptxAI IN PATH final PPT.pptx
AI IN PATH final PPT.pptx
DivyaGaurav4632 vistas
Decision Support System for clinical practice created on the basis of the Un... por blejyants
Decision Support System for clinical practice created on the basis of  the Un...Decision Support System for clinical practice created on the basis of  the Un...
Decision Support System for clinical practice created on the basis of the Un...
blejyants1.6K vistas
Bimcv labman eu-bi por maigva
Bimcv labman eu-biBimcv labman eu-bi
Bimcv labman eu-bi
maigva1.2K vistas
Machine Learning in Modern Medicine with Erin LeDell at Stanford Med por Sri Ambati
Machine Learning in Modern Medicine with Erin LeDell at Stanford MedMachine Learning in Modern Medicine with Erin LeDell at Stanford Med
Machine Learning in Modern Medicine with Erin LeDell at Stanford Med
Sri Ambati5.2K vistas
[Review] High-performance medicine: the convergence of human and artificial i... por Dongmin Choi
[Review] High-performance medicine: the convergence of human and artificial i...[Review] High-performance medicine: the convergence of human and artificial i...
[Review] High-performance medicine: the convergence of human and artificial i...
Dongmin Choi1.2K vistas
Thesis presentation por Aras Masood
Thesis presentationThesis presentation
Thesis presentation
Aras Masood82 vistas
Bimcv eub heidelberg por maigva
Bimcv eub heidelbergBimcv eub heidelberg
Bimcv eub heidelberg
maigva802 vistas
Week 11 12 chap11 c-2 por Zahir Reza
Week 11 12 chap11 c-2Week 11 12 chap11 c-2
Week 11 12 chap11 c-2
Zahir Reza274 vistas
Health Care Application using Machine Learning and Deep Learning por IRJET Journal
Health Care Application using Machine Learning and Deep LearningHealth Care Application using Machine Learning and Deep Learning
Health Care Application using Machine Learning and Deep Learning
IRJET Journal11 vistas
Big Data and Artificial Intelligence por Kamarul Imran
Big Data and Artificial Intelligence Big Data and Artificial Intelligence
Big Data and Artificial Intelligence
Kamarul Imran141 vistas

Más de Mike Hogarth, MD, FACMI, FACP

Big Data in Clinical Research por
Big Data in Clinical ResearchBig Data in Clinical Research
Big Data in Clinical ResearchMike Hogarth, MD, FACMI, FACP
39 vistas50 diapositivas
Informatics and the merging of research and quality measures with bedside care por
Informatics and the merging of research and quality measures with bedside careInformatics and the merging of research and quality measures with bedside care
Informatics and the merging of research and quality measures with bedside careMike Hogarth, MD, FACMI, FACP
134 vistas58 diapositivas
Keep us safe: An overview of US public health informatics systems and archite... por
Keep us safe: An overview of US public health informatics systems and archite...Keep us safe: An overview of US public health informatics systems and archite...
Keep us safe: An overview of US public health informatics systems and archite...Mike Hogarth, MD, FACMI, FACP
785 vistas92 diapositivas
Taking Quantum Computing for a Spin: What is Imaginary and What is Real? por
Taking Quantum Computing for a Spin: What is Imaginary and What is Real?Taking Quantum Computing for a Spin: What is Imaginary and What is Real?
Taking Quantum Computing for a Spin: What is Imaginary and What is Real?Mike Hogarth, MD, FACMI, FACP
501 vistas64 diapositivas
Linking Electronic Patient Records and Death Records: Challenges and Opportun... por
Linking Electronic Patient Records and Death Records: Challenges and Opportun...Linking Electronic Patient Records and Death Records: Challenges and Opportun...
Linking Electronic Patient Records and Death Records: Challenges and Opportun...Mike Hogarth, MD, FACMI, FACP
216 vistas32 diapositivas
The OneSource Initiative: An Approach to Structured Sourcing of Key Clinical ... por
The OneSource Initiative: An Approach to Structured Sourcing of Key Clinical ...The OneSource Initiative: An Approach to Structured Sourcing of Key Clinical ...
The OneSource Initiative: An Approach to Structured Sourcing of Key Clinical ...Mike Hogarth, MD, FACMI, FACP
280 vistas49 diapositivas

Más de Mike Hogarth, MD, FACMI, FACP(15)

Último

Breast Ductography.pptx por
Breast Ductography.pptxBreast Ductography.pptx
Breast Ductography.pptxPeerzadaJunaidUlIsla
45 vistas18 diapositivas
Asthalin Inhaler (Generic Albuterol Sulfate Inhaler) por
Asthalin Inhaler (Generic Albuterol Sulfate Inhaler) Asthalin Inhaler (Generic Albuterol Sulfate Inhaler)
Asthalin Inhaler (Generic Albuterol Sulfate Inhaler) The Swiss Pharmacy
16 vistas20 diapositivas
Epilepsy and Anti epileptic drugs por
Epilepsy and Anti epileptic drugsEpilepsy and Anti epileptic drugs
Epilepsy and Anti epileptic drugsA. Gowtham Sashtha
25 vistas42 diapositivas
INTRODUCTION TO PHARMACEUTICAL VALIDATION SCOPE and MERITS OF VALIDATION.pptx por
INTRODUCTION TO PHARMACEUTICAL VALIDATION SCOPE and MERITS OF VALIDATION.pptxINTRODUCTION TO PHARMACEUTICAL VALIDATION SCOPE and MERITS OF VALIDATION.pptx
INTRODUCTION TO PHARMACEUTICAL VALIDATION SCOPE and MERITS OF VALIDATION.pptxABG
121 vistas40 diapositivas
ICH AND WHO GUIDELINES FOR VALIDATION OF EQUIPMENTS.pptx por
ICH AND WHO GUIDELINES FOR VALIDATION OF EQUIPMENTS.pptxICH AND WHO GUIDELINES FOR VALIDATION OF EQUIPMENTS.pptx
ICH AND WHO GUIDELINES FOR VALIDATION OF EQUIPMENTS.pptxABG
69 vistas45 diapositivas
Mental Health with Chronic Illness.pptx por
Mental Health with Chronic Illness.pptxMental Health with Chronic Illness.pptx
Mental Health with Chronic Illness.pptxScleroderma Foundation of Greater Chicago
16 vistas16 diapositivas

Último(20)

Asthalin Inhaler (Generic Albuterol Sulfate Inhaler) por The Swiss Pharmacy
Asthalin Inhaler (Generic Albuterol Sulfate Inhaler) Asthalin Inhaler (Generic Albuterol Sulfate Inhaler)
Asthalin Inhaler (Generic Albuterol Sulfate Inhaler)
The Swiss Pharmacy16 vistas
INTRODUCTION TO PHARMACEUTICAL VALIDATION SCOPE and MERITS OF VALIDATION.pptx por ABG
INTRODUCTION TO PHARMACEUTICAL VALIDATION SCOPE and MERITS OF VALIDATION.pptxINTRODUCTION TO PHARMACEUTICAL VALIDATION SCOPE and MERITS OF VALIDATION.pptx
INTRODUCTION TO PHARMACEUTICAL VALIDATION SCOPE and MERITS OF VALIDATION.pptx
ABG121 vistas
ICH AND WHO GUIDELINES FOR VALIDATION OF EQUIPMENTS.pptx por ABG
ICH AND WHO GUIDELINES FOR VALIDATION OF EQUIPMENTS.pptxICH AND WHO GUIDELINES FOR VALIDATION OF EQUIPMENTS.pptx
ICH AND WHO GUIDELINES FOR VALIDATION OF EQUIPMENTS.pptx
ABG69 vistas
Gastro-retentive drug delivery systems.pptx por ABG
Gastro-retentive drug delivery systems.pptxGastro-retentive drug delivery systems.pptx
Gastro-retentive drug delivery systems.pptx
ABG238 vistas
The Art of naming drugs.pptx por DanaKarem1
The Art of naming drugs.pptxThe Art of naming drugs.pptx
The Art of naming drugs.pptx
DanaKarem127 vistas
Fetal and Neonatal Circulation - MBBS, Gandhi medical College Hyderabad por Swetha rani Savala
Fetal and Neonatal Circulation - MBBS, Gandhi medical College Hyderabad Fetal and Neonatal Circulation - MBBS, Gandhi medical College Hyderabad
Fetal and Neonatal Circulation - MBBS, Gandhi medical College Hyderabad
Swetha rani Savala23 vistas

Informatics in disease management: What will the future bring?

  • 1. Informatics in Disease Management: What will the Future bring? Michael Hogarth, MD, FACMI, FACP Clinical Research Information Officer, and Director of Biomedical Informatics, Altman Clinical Translational Research Institute University of California San Diego Health
  • 2. Informatics and Disease Management - here comes machine learning (ML) and artificial intelligence (AI)!
  • 3. FYI: We have been using “machine learning” for over 200 years http://www.andreykurenkov.com/writing/ai/a-brief-history-of-neural-nets-and-deep-learning/
  • 4. So what’s new? • most AI/ML algorithms are not new • why is “machine learning” and “deep learning” a new revolution in the use of AI? https://mc.ai/awesome-ai-the-guide-to-master-artificial-intelligence/
  • 5. Key Informatics Trends in 2019... 1. HITECH and EHR adoption now at over 90% – has made massive amounts of healthcare data (“dirta”) available – fuel for “machine learning” based predictive systems 2. High performance computing is highly commoditized – Commercial clouds - “data center at your fingertips” – example: 11 machine cluster for 6 hours = $3.96 (Jan 2019) – cheap computing to generate machine learning models 3. Machine learning engines and algorithms are commoditized (ML as a service)
  • 6. The commercial attraction to ML/AI in Healthcare https://www.healthcarebusinessinsights.com/blog/information-technology/growing-value-ai-machine-learning-healthcare/
  • 7. Why is healthcare such an attractive target for optimization?
  • 8. Big tech sees significant opportunities https://healthitanalytics.com
  • 9. Research into ML/AI “models” for various tasks -- a very hot area!
  • 10. Yes, there is “magical thinking” in this space... • MD Anderson partnered with IBM in 2012 to help oncologists with diagnoses and choosing therapeutic courses • IBM Watson Manager - “training alongside doctors to do what they can’t” • Feb 2017 - project canceled after MD Anderson had already paid IBM over $39M (originally priced at $2.4M)!!!
  • 11. But don’t throw out the baby with the bath water... Using ML on 36 cases evenly split between AML and normal, an algorithm correctly distinguished disease progression in 90% of cases of relapsing AML, and was 100% correct in distinguishing between AML and normal bone marrow
  • 12. How does machine learning work? https://www.edureka.co/blog/what-is-machine-learning/
  • 14. ML Basics: Supervised • training data set • data used to “train the model” • Once ‘trained’, the model can make predictions or suggest decisions with new data inputs https://bigdata-madesimple.com/machine-learning-explained-understanding-supervised-unsupervised-and-reinforcement-learning/
  • 15. ML Basics: Unsupervised • The ML model ‘learns’ through self- finding structures or associations in the data • Will automatically assemble “clusters” based on variables available in the data set https://bigdata-madesimple.com/machine-learning-explained-understanding-supervised-unsupervised-and-reinforcement-learning/
  • 16. ML Basics: Reinforcement Learning • Paradigm is a “software agent” that attempts to achieve as close to the “goal state” as possible. • Goal state = best outcome • Follows a concept of “hit and trial” where the agent is rewarded or penalized (using points) for a particular action or answer. • It attempts to ‘learn’ using hit/trial and can then predict with new data https://bigdata-madesimple.com/machine-learning-explained-understanding-supervised-unsupervised-and-reinforcement-learning/
  • 17. Artificial Neural Networks (ANNs) • AI system loosely based on biological neural networks in brains • not an algorithm - its a framework for machine learning algorithms that work by using multi-weighted inputs across multiple collaborating nodes to process data inputs • can be supervised or unsupervised ML • first neural networks were implemented in circuits – neurophysiologist Warren McCulloch and mathematician Walter Pitts wrote paper theorizing how neurons might work in a network with weighted inputs – first tested in circuitry - Mark I Perceptron – Software based NN’s became known as “artificial neural networks” Mark I Perceptron with patchboard to connect nodes, potentiometers as weights
  • 18. “Deep” Learning What is “deep learning”? ● uses a multi-layered “artificial neural network” ● first layer (captures raw data -- input layer) ● “hidden layers” -- integrate the values of the raw input nodes to derive “transformed features” ● predicts outcomes or classifies based on transformed features
  • 19. “Convolutional” Neural Networks (CNN) • a type of multi-layered (deep) ANN designed for recognition of visual features • originally devised in 1988 by Yann LeCun at AT&T Labs to recognize handwritten digits • inspired by the connectivity pattern seen between neurons in the visual cortex • Convolution layers apply a convolution operation to the input, passing result to next layer. • Convolution operation emulates response of an individual neuron to visual stimuli. Pooling coalesces outputs from one layer into a single neuron in the next layer • significantly resurging interest in CNNs due to: – availability of labeled digital image sets (large training sets) – high performance computing with graphical processing units (GPU) necessary for the significant computation required
  • 20. Machine Learning (ML) image interpretation - supervised learning
  • 21. Convolutional (deep) Neural Network and Dermatology • trained a CNN with 1.29M images of 2,032 different conditions from 18 clinician-curated repositories • tested CNN based classifier on 376 cases, head-to-head with 21 board-certified dermatologists on two scenarios with real biopsy- proven lesions – differentiating benign seborrheic keratoses from keratinocyte carcinomas (n=135) – differentiating malignant melanoma from benign nevi (n=130;n=111) • Results: CNN performance was as good as the dermatologists (all AUCs over 0.91)!
  • 22. Convolutional (deep) Neural Network and Echocardiography “view classification” • trained a CNN with 267 transthoracic echo studies (200,000 images) to differentiate 15 different standard echo views (12 video, 3 still image views - parasternal long axis, right ventricular inflow, short axis at mitral level, etc..) • tested with 27 echo studies (20,000 images) – video views -- 97.8% accurate – still images - 91.7% accurate • 4 board certified echocardiographers on the 27 echo studies --- 79.4% average accuracy
  • 23. Deep Learning and Interpretation of Retinal Images for diabetic retinopathy • trained a ML systems with 494,661 retinal images • validated (tested) using dataset of 71,896 images from 14,880 patients • detection of vision- threatening diabetic retinopathy: AUC 0.958 (95% CI) • detection of referable diabetic retinopathy: AUC=0.936(95% CI)
  • 24. Reinforcement Learning in Sepsis - the “AI Clinician” • developed an AI Clinician computational model using reinforcement learning to suggest optimal treatment of sepsis regarding fluid resuscitation and pressor administration • AI clinician’s goal - maximize the probability of a good outcome • Developed using two large ICU databases – MIMIC III and eICU Research Institute Database (eRI) – 17,083 admissions from MIMIC – 79,073 admissions from eRI • Performance: – “the value of the AI Clinician’s selected treatment is, on average, reliably higher than human clinicians – Validation done on a separate dataset of cases. – Found that cases where the clinician action most closely matched what AI Clinician’s would have done had the lowest mortality!
  • 25. Artificial Intelligence Sepsis Expert (AISE) • Nemati and colleagues at Emory trained and validated on 31,000 Emory ICU admissions – used 60 features: ECG, BP, HR, MAP, SpO2, temp, GCS, FiO2, Chem7 analytes, ABG, lactate, etc.. • AISE predicted sepsis an average of 4-8hrs in advance (AUC of 0.85) of non-assisted recognition of impending sepsis (using Sepsis 3 definition)
  • 26. Regulating AI-based Clinical Decision Support - is the FDA ready?
  • 27. 2017 FDA guidance on decision support software systems • Clinical Decision Support (CDS) vs. Patient Decision Support (PDS) • With guidance from 21st Century Cures Act, FDA introduced new guidance to make it easier for software developers to obtain clearance (to spark innovation) – Low risk =(weight management, mindfulness tools, etc.. not under FDA oversight – Products intended for the diagnosis, cure, mitigation, prevention, or treatment of a condition will fall under FDA oversight – Software intended to process or analyze medical images, signals from diagnostic devices (ECG, EEG) will continue under FDA oversight
  • 28. Recent FDA Decisions for AI Based Systems - De Novo: regulatory pathway for low-to-moderate risk devices of a new type for which there is no legally marketed predicate device with substantial equivalence - 510(k): a pre-market submission to FDA to demonstrate the device is at least a safe and effective (equivalent) to a legally marketed device that is not subject to premarket approval - Class II: moderate-to-high risk device - Special controls: - performance standards - post-market surveillance - special labeling
  • 29. This has now also captured the attention of the legal profession... https://www.americanbar.org/groups/health_law/publications/aba_health_esource/2016-2017/october2017/machinelearning/
  • 30. The future of disease management could be bright... https://ai100.stanford.edu/sites/g/files/sbiybj9861/f/ai100report10032016fnl_ singles.pdf
  • 31. or it could go horribly wrong... http://blog.petrieflom.law.harvard.edu/2017/10/26/
  • 32. One thing is certain, AI has become an integral part of society -- it is here to stay Domino’s Pizza delivered by autonomous driving vehicle
  • 33. A common question: “When will we be “replaced” by smart machines?” Surgeon Nephrologist