1. Use of Artificial Intelligence in
AAOCA
Jai Nahar, MD, MBA
Associate Professor of Pediatrics
The GeorgeWashington University School of
Medicine and Health Sciences
Division of Cardiology
Children’s National Medical Center
Washington, DC
Congenital Heart Surgeons Society
Fall work weekend
Nov. 2017,Toronto
3. Agenda
1. AAOCA: Background, problem, questions to be addressed
2. Current Opportunity
3. Introduction to AI in medicine
4. PotentialApplication ofAI in AAOCA
5. Challenges
6. Future directions
4. AAOCA: Background
• Second leading cause of SCD in young athletes in USA
• Care of these patients involve three important challenges
1. Diagnosis: incidental diagnosis, lack of symptoms, initial presentation may
be SCD or sudden cardiac arrest
2. Risk Stratification: phenotypic heterogeneity
3. Ideal management
Risk
Stratification
Diagnosis
AAOCA
Ideal
Management
5. AAOCA:What causes the Critical Lethal event?
Critical Event :
Coronary ischemia
and lethal
Ventricular
Tachyarrhythmia
Decreased
coronary perfusion
due to Mechanical
factors, altered
blood flow
dynamics
Electrically unstable
myocardial
substrate
Unknown
Factors
?
• Acute angle take off and
Kinking of coronary artery at
its origin
• Flap like closure of abnormal
slit like coronary orifice
• Compression of anomalous
coronary artery between
Aorta and Pulmonary artery
during exercise
• Spasm of anomalous coronary
artery possibly secondary to
endothelial injury
Prior intermittent ischemia
Patchy Myocardial Fibrosis
• Local myocardial
metabolic factors
• Channelopathies
• Genetic
predisposition to
arrhythmia
Basso et al. (2000). Clinical profile of congenital coronary artery
anomalies with origin from the wrong aortic sinus leading to sudden
death in young competitive athletes. Journal of the American College of
Cardiology, 35(6), 1493-1501.
6. Problem
There are knowledge gaps/unknown variables which need to be filled/uncovered
to refine our risk stratification for AAOCA patients who are at highest risk of
sudden death
• Asymptomatic (hidden)
• Gray zone (symptoms of ? significance)
7. Task: Making the invisible, visible
AAOCA: Myocardial
ischemia, malignant
Ventricular
arrhythmia, sudden
death
Unknown/Invisible?
• Genetic make up
• Channelopathies
• Local myocardial/metabolic
factors
• Etc ??????
8. Questions
1. How can we uncover currently unknown phenotypes (cluster of variables with
non linear relation) which indicate high risk of sudden cardiac death?
2. How to develop good predictive models for risk stratification and prompt
detection of high risk AAOCA ?
9. Call for Action:Two step approach
Uncover unknown high risk
groups
Develop refined risk
stratification models
Deep
Phenotyping
10. Agenda
1. Define the problem, and questions to be addressed
2. Current Opportunity
3. Introduction to AI in medicine
4. Potential Application of AI to AAOCA
5. Challenges
6. Future directions
11. Current Opportunity : make the invisible, visible
• Data: CHSS AAOCA registry
• Artificial Intelligence Methods
Power of
Data
Leverage
AI
Refined risk
stratification
and patient
management
12. Dawn of New Era of
Augmented Intelligence
Physician and AI ( Human/Machine) synergy for facilitating better
• Diagnosis
• Disease management
• Clinical Decisions
Physicians Machines
Platform for
Precision Medicine
13. Agenda
1. Define the problem, and questions to be addressed
2. Current Opportunity
3. Introduction to AI in medicine
4. Potential Application of AI to AAOCA
5. Challenges
6. Future directions
14. Medical Intelligence: AI in Medicine
Big Data and Artificial Intelligence in Pediatric Cardiology
Anthony C. Chang, MD, MBA, MPH
16. Machine Learning
Andrew Ng
“Is the Science of getting computers to learn, without being explicitly programmed”
Full spectrum application in CV medicine
Machine
Learning
Diagnosis/Risk
stratification
Treatment
Prevention
Research
18. Supervised Learning
• Goal is to predict a known output orTarget
• Algorithm is taught with right answers (labels) for examples used in
training data set.
• Can be categorized as:
Classification problem: predicting categories, discrete value output
(0,1 etc.)
Regression problem: predict continuous values
Anomaly detection problem: predict unusual pattern
21. Unsupervised Learning
• Goal is to learn the intrinsic structure within data.
• No outputs to predict
• Task is to find hidden pattern/structure in data,
without human feedback
Cluster2
Cluster3
Cluster
1
22. Unsupervised Learning Application
Identify novel disease mechanisms, genotypes,
phenotypes: Filling the knowledge gaps
Identify Novel
Disease
Mechanisms
New
paths/approach
to therapy
Precision
Medicine
initiative
23. Unsupervised Learning Algorithms
• Clustering algorithms: K-Means, Hierarchical clustering
Used to cluster unlabeled data into different groups
Used when no obvious natural grouping
• Association rule- learning algorithm:
Help to uncover relationships between seemingly unrelated data items
• PrincipalComponent Analysis
• Sparse Coding
25. K-Means clustering Algorithms
(left) K-means in 2d. (right) K-means in 3d. You have
to imagine k-means in 4d.
http://stanford.edu/~cpiech/cs221/handouts/kmeans.html
26. Reinforcement learning
Reinforcement learning led to AlphaGo’s stunning victory over a human Go
champion
https://www.technologyreview.com/s/603501/10-breakthrough-technologies-2017-reinforcement-
learning/
Reinforcement learning is
learning by trial-and-error,
solely from rewards or
punishments.
https://deepmind.com/blog/deep-reinforcement-
learning/
Can be viewed as hybrid
of supervised and
unsupervised learning
27. Artificial Neural Network (ANN)
ANN are modeled after human neurons
• Nodes are like neurons
• Input layer: input data/ predictor variables/ features
• Hidden layer: processing of input
• Output layer:Target (prediction of class or value)
Lee, J.-G., Jun, S., Cho, Y.-W., Lee, H., Kim, G. B., Seo, J. B., & Kim, N. (2017). Deep Learning in Medical Imaging: General
Overview. Korean Journal of Radiology, 18(4), 570–584. http://doi.org/10.3348/kjr.2017.18.4.570
28. Deep Learning
• Part of Machine Learning
• Uses multiple layers of ANN
• Mimics the working of human brain
Lee, J.-G., Jun, S., Cho, Y.-W., Lee, H., Kim, G. B., Seo, J. B., & Kim, N. (2017). Deep Learning in Medical Imaging: General
Overview. Korean Journal of Radiology, 18(4), 570–584. http://doi.org/10.3348/kjr.2017.18.4.570
29. Deep Learning
Three types of Network:
1. Deep Neural Networks: Google’s Deep mind Alpha Go
2. Recurrent Neural Networks (RNN): natural language
processing, handwriting and speech recognition
3. Convolutional Neural Networks (CNN): Computer
vision, image recognition, CV imaging
30. Cognitive Computing
Systems mimicking human cognition, help
replicate human capabilities across the spectrum of sensory
perception, deduction, reasoning, learning and knowledge.
31. Upcoming inWorld of AI
• Quantum Computing: accelerate the data processing speed
• Neuromorphic chips
• Partnership with other technologies
AI and AR
AI and Robotics
32. Agenda
1. AAOCA: Background, problem, questions to be addressed
2. Current Opportunity
3. Introduction to AI in medicine
4. Potential Application of AI in AAOCA
5. Challenges
6. Future directions
33. First : A Machine Learning overview
Machine Learning in Medicine
Rahul C. Deo
Circulation. 2015;132:1920-1930, originally published November 16, 2015
https://doi.org/10.1161/CIRCULATIONAHA.115.001593
A: Matrix representation of the supervised
and unsupervised learning problem.
B: Decision trees map features to
outcome (Supervised Learning)
C: Neural networks predict outcome based
on transformed representations of
features. (Supervised Learning)
D:The k-nearest neighbor algorithm
assigns class based on the values of the
most similar training examples.
(Supervised Learning)
34. Phenomapping
Use of unsupervised machine learning in Phenotypic
classification of a heterogeneous clinical syndrome into discrete
phenogroups
PG1
PG2
PG3
PG4
PG5
Phenomapping
35. Phenotypic classification (Phenomapping)
A: Phenotype heat map of HFpEF.
B: BIC analysis for the identification of the
optimal number of phenotypic clusters
(pheno groups).
C: Kaplan–Meier curves
Survival free of cardiovascular (CV)
hospitalization or death stratified by
pheno groups.
Machine Learning in Medicine
Rahul C. Deo
Circulation. 2015;132:1920-1930
36. Stepwise approach for AAOCA
Detection of
High Risk
Cases
Develop
Supervised
model for
disease
prediction
Novel
phenogroup
selection using
Unsupervised
learning
37. Proposed Framework for AAOCA
Demographic
Data
Stress
lab/Nuclear
Medicine
Data
EP data
Surgical
data
Cath data
Imaging
data
Clinical
Data
Unsupervised learning based
Pheno-groups identification
Supervised Learning
based Model for
Risk stratification
High
Medium
Low
Expert
Knowledge
38. Challenges
• Deep learning:
Problem of model overfitting
Need for large data sets, multi Institutional collaboration
Setting a neural network is time consuming
• Need for sharing of machine learning expertise
• Need for Novel informative features to build improved models
• Need for refined biomarkers to access myocardial status
39. Future Directions
Seamless data collaboration across institutions, and enhanced Computer- Human
synergy lead Augmented intelligence will refine the path of precisionCardiovascular
Medicine.
Precision
Medicine
Big Data
Collaboration
Augmented
Intelligence
(Human/Machine
Partnership)
40. Conclusion
To optimally harness the power of Big Data and AI in
Medicine we need Multi Institutional collaboration
between Physicians, Data Scientists and Machine
Learning Experts.
42. References
1. Basso, C., Maron, B. J., Corrado, D., & Thiene, G. (2000). Clinical profile of congenital coronary artery anomalies with origin from the wrong aortic
sinus leading to sudden death in young competitive athletes. Journal of the American College of Cardiology, 35(6), 1493-1501.
2. Betancur, J., Otaki, Y., Motwani, M., Fish, M. B., Lemley, M., Dey, D., . . . Sharir, T. (2017). Prognostic value of combined clinical and myocardial
perfusion imaging data using machine learning. JACC: Cardiovascular Imaging,
3. Brothers, J. A. (2017). Introduction to anomalous aortic origin of a coronary artery. Congenital Heart Disease, 12(5), 600-602. doi:10.1111/chd.12497
4. Chang, A. C. (2016). Big data in medicine: The upcoming artificial intelligence. Progress in Pediatric Cardiology, 43, 91-94.
5. Chang, A. P., & Musen, M. (2012). Artificial Intelligence in Pediatric Cardiology: An Innovative Transformation in Patient Care, Clinical Research, and
Medical Education. Congenital Cardiology, 10, 1-9
6. Deo, R. C. (2015). Machine learning in medicine. Circulation, 132(20), 1920-1930. doi:10.1161/CIRCULATIONAHA.115.001593 [doi]
7. Krittanawong, C., Zhang, H., Wang, Z., Aydar, M., & Kitai, T. (2017). Artificial intelligence in precision cardiovascular medicine. Journal of the
American College of Cardiology, 69(21), 2657-2664.
8. Lee, J., Jun, S., Cho, Y., Lee, H., Kim, G. B., Seo, J. B., & Kim, N. (2017). Deep learning in medical imaging: General overview. Korean Journal of
Radiology, 18(4), 570-584.
9. Maron, B. J., Doerer, J. J., Haas, T. S., Tierney, D. M., & Mueller, F. O. (2009). Sudden deaths in young competitive athletes: Analysis of 1866 deaths
in the united states, 1980-2006. Circulation, 119(8), 1085-1092. doi:10.1161/CIRCULATIONAHA.108.804617 [doi]
10. Shah, S. J., Katz, D. H., Selvaraj, S., Burke, M. A., Yancy, C. W., Gheorghiade, M., . . . Deo, R. C. (2015). Phenomapping for novel classification of
heart failure with preserved ejection fraction. Circulation, 131(3), 269-279. doi:10.1161/CIRCULATIONAHA.114.010637 [doi]
11. Van Hare, G. F., Ackerman, M. J., Evangelista, J. A., Kovacs, R. J., Myerburg, R. J., Shafer, K. M., . . . American Heart Association Electrocardiography
and Arrhythmias Committee of Council on Clinical Cardiology, Council on Cardiovascular Disease in Young, Council on Cardiovascular and Stroke
Nursing, Council on Functional Genomics and Translational Biology, and American College of Cardiology. (2015). Eligibility and disqualification
recommendations for competitive athletes with cardiovascular abnormalities: Task force 4: Congenital heart disease: A scientific statement from
the american heart association and american college of cardiology. Circulation, 132(22), e281-91. doi:10.1161/CIR.0000000000000240 [doi]
Two step approach
Using unsupervised learning, uncover unknown high risk patterns, groups within the data
Evaluate their performance in subsequent supervised learning tasks (how useful these new patterns are to AAOCA). This can help in developing refined risk stratification models, facilitating precision medicine
Krittanawong, C., Zhang, H., Wang, Z., Aydar, M., & Kitai, T. (2017). Artificial intelligence in precision cardiovascular medicine. Journal of the American College of Cardiology, 69(21), 2657-2664.
This algorithm consist of a target / outcome variable (or dependent variable) which is to be predicted from a given set of predictors (independent variables). Using these set of variables, we generate a function that map inputs to desired outputs. The training process continues until the model achieves a desired level of accuracy on the training data. Examples of Supervised Learning: Regression, Decision Tree, Random Forest, KNN, Logistic Regression etc.
Supervised learning is so named because the data scientist acts as a guide to teach the algorithm what conclusions it should come up with. It’s similar to the way a child might learn arithmetic from a teacher. Supervised learning requires that the algorithm’s possible outputs are already known and that the data used to train the algorithm is already labeled with correct answers. For example, a classification algorithm will learn to identify animals after being trained on a dataset of images that are properly labeled with the species of the animal and some identifying characteristics.
Su[prevised learning algorithms have been applied to problems in prediction, diagnosis and treatment of CVD
Prediction accuracy depends on:
Algorithm used
Dataset
hypothesis
K means clustering: Data is clustered in K classes by K-Means Algorithm
Hierarchical clustering: can be two types
Agglomerative approach: Bottom up approach
Divisive: Top down approach
https://www.youtube.com/watch?v=RD0nNK51Fp8 (Stanford site)
An ANN is comprised of a network of artificial neurons (also known as "nodes"). These nodes are connected to each other, and the strength of their connections to one another is assigned a value based on their strength: inhibition (maximum being -1.0) or excitation (maximum being +1.0). If the value of the connection is high, then it indicates that there is a strong connection. Within each node's design, a transfer function is built in. There are three types of neurons in an ANN, input nodes, hidden nodes, and output nodes
Three types of deep learning
http://researcher.watson.ibm.com/researcher/view_group.php?id=4384
Medical Sieve is an ambitious long-term exploratory grand challenge project to build a next generation cognitive assistant with advanced multimodal analytics, clinical knowledge and reasoning capabilities that is qualified to assist in clinical decision making in radiology and cardiology. It will exhibit a deep understanding of diseases and their interpretation in multiple modalities (X-ray, Ultrasound, CT, MRI, PET, Clinical text) covering various radiology and cardiology specialties. The project aims at producing a sieve that filters essential clinical and diagnostic imaging information to form anomaly-driven summaries and recommendations that tremendously reduce the viewing load of clinicians without negatively impacting diagnosis.
http://circ.ahajournals.org/content/132/20/1920.short
Machine learning overview. A, Matrix representation of the supervised and unsupervised learning problem. We are interested in developing a model for predicting myocardial infarction (MI). For training data, we have patients, each characterized by an outcome (positive or negative training examples), denoted by the circle in the right-hand column, and by values of predictive features, as well, denoted by blue to red coloring of squares. We seek to build a model to predict outcome by using some combination of features. Multiple types of functions can be used for mapping features to outcome (B through D). Machine learning algorithms are used to find optimal values of free parameters in the model to minimize training error as judged by the difference between predicted values from our model and actual values. In the unsupervised learning problem, we are ignoring the outcome column and grouping together patients based on similarities in the values of their features. B, Decision trees map features to outcome. At each node or branch point, training examples are partitioned based on the value of a particular feature. Additional branches are introduced with the goal of completely separating positive and negative training examples. C, Neural networks predict outcome based on transformed representations of features. A hidden layer of nodes integrates the value of multiple input nodes (raw features) to derive transformed features. The output node then uses values of these transformed features in a model to predict outcome. D, The k-nearest neighbor algorithm assigns class based on the values of the most similar training examples. The distance between patients is computed based on comparing multidimensional vectors of feature values. In this case, where there are only 2 features, if we consider the outcome class of the 3 nearest neighbors, the unknown data instance would be assigned a “no MI” class. LDL indicates low-density lipoprotein; and MI, myocardial infarction.
Application of unsupervised learning to HFpEF. A, Phenotype heat map of HFpEF. Columns represent individual study participants; rows represent individual features. B, Bayesian information criterion analysis for the identification of the optimal number of phenotypic clusters (pheno groups). C, Survival free of cardiovascular (CV) hospitalization or death stratified by phenotypic cluster. Kaplan–Meier curves for the combined outcome of heart failure hospitalization, cardiovascular hospitalization, or death stratified by phenotypic cluster
AAOCA Data: Multidimensional with Phenotypic Heterogeneity