Artificial Intelligence as a
Disruptor
What’s coming and how AI can Boost
The Value of Lab Testing
Ajit Singh, Ph.D.
Partner, Artiman Ventures
Professor, School of Medicine
Stanford University
Executive War College
New Orleans, May 2nd 2018
five
vignettes
1 what can we learn from nature
2
myth vs reality: what is feasible today3
evolution of AI and transdisciplinarity
have you been stuck behind a Google car4
AI in medicine: when will the future arrive5
1 what can we learn from nature
myth vs reality: what is feasible today
evolution of AI and transdisciplinarity
have you been stuck behind a Google car
AI in medicine: when will the future arrive
2
3
4
5
lets take a short
journey
through a place of immense
diversity
a rich
diversityof life
where?
galapagos
why
right
temperature
1
good
mobility
2
natural mixing of
diversity
3
coral reefs
connectionsare made
failuresare discarded
quickly
nature’s
innovationplatform
diversity
human-made
innovationplatform
trans
disciplinarity
1
2
3
4
5
what can we learn from nature
myth vs reality: what is feasible today
evolution of AI and transdisciplinarity
have you been stuck behind a Google car
AI in medicine: when will the future arrive
Turing, A.M. (1950), Computing machinery and intelligence, Mind, Vol. 59, pp. 433-460
1950the Turing test
1951First chess-playing program developed by Dietrich Prinz at the University of Manchester. It ran on the Ferranti Mark I.
1955Allen Newell and Herbert Simon created the Logic Theorist that proved 38 of the first 52 theorems in Principia Mathematica
1956The term “Artificial Intelligence" was adopted at the Dartmouth Conference
implementationstwo prevailing
of artificial intelligence
algorithmssymbolic implementation of AI:
that manipulate symbols that represent concepts
(knowledge is hand-coded)
(also called expert systems)
1.
neural networksconnectionist implementation of AI
(knowledge is machine-learnt)
2.
Academic Disciplines relevant to AI.
Philosophy Logic, reasoning, foundations of learning, language, rationality.
Mathematics Algorithms, computation, (un)decidability, (in)tractability, probability.
Economics Utility, Decision theory
Neuroscience Neurons as information processing units.
Psychology How do people behave, perceive, process
Cognitive Science Knowledge Representation
Computer Engg. Computing, Storage, Communication
Control Systems Optimization of objective functions over time
Linguistics Knowledge representation, grammar
trans
disciplinarity
and…
data
(lots of it)
1
2
3
4
5
what can we learn from nature
myth vs reality: what is feasible today
evolution of AI and transdisciplinarity
have you been stuck behind a Google car
AI in medicine: when will the future arrive
questions let’s answer some illustrative
(these are illustrative, and not comprehensive)
human braincan we build hardware as complex as the
Q
brain10 12 neurons, 10 14 synapses, 10 -3 seconds cycle time
computers106 transistors per CPU, 103 – 104 CPUs, 10 9 bits of RAM, 10 - 8
seconds cycle time
YES computer systems with as many “neurons” as our brain, soon.
NO computer systems with interconnectivity of the brain any time soon
BUT building hardware will not achieve a computer behave like a brain!
recognize speechcan computer systems
Q
large vocabularies
speech has unclear word-boundaries but clear phoneme-boundaries
background noise, other speakers, accents, colds
CONTEXT plays a very important role
(“I would like some dream and sugar in my coffee”)
(“olive oil” is made from olives. “baby oil” is made for babies)
(“time flies like an arrow”)
YES CAN recognize individual words from small vocabulary (99% accuracy)
NO CANNOT “understand” speech (60% accuracy even in defined domains)
BUT building context in constrained domains will be possible in near future
see and perceivecan computer systems
Q
pattern recognition evolved over millions of years
vast variations in lighting conditions, occlusions, unknown objects
scene understanding evolved over millions of years
CONTEXT plays a very important role
YES to some kinds of objects in constrained environments (face recognition)
NO to general scene understanding in unconstrained environments
BUT solving this problem is central to building intelligent systems
1
2
3
4
5
what can we learn from nature
myth vs reality: what is feasible today
evolution of AI and transdisciplinarity
have you been stuck behind a Google car
AI in medicine: when will the future arrive
2 million milesand sensor data (camera, lidar, etc.) of
of Google caronly to reach the current capability
(the one you would not like to be stuck behind)
30 million milesto reach human-like capability
it will take sensor data from
30,000 milesthat human drivers are able to achieve in
why?
50+ million yearshumans are able to apply
of learning in pattern-matching
(that’s the power of evolution at work)
1
2
3
4
5
what can we learn from nature
myth vs reality: what is feasible today
evolution of AI and transdisciplinarity
have you been stuck behind a Google car
AI in medicine: when will the future arrive
breast cancerdiagnosis and therapy selection
let’s extrapolate the self-driving car learning to
correlated datawill we need
how many patients’ / healthy people’s
seven billion
(that’s the entire human population of planet earth)
great expectationsthis is a case of
(great, but unrealistic)
more importantly
replicatinghuman knowledge
simply
bad ideaby learning from humans is a
succeedwatson WILL
1 Optellum optellum.com Cancer diagnostics with AI and Big data
2 Quantitative Insights
quantinsights.com Machine learning based Computer aided diagnosis of breast cancer
3 Artificial Intelligence in Medicine
aim.ca AI based Software solution for cancer surveillance, prevention and research
4 GliaLab glialab.com Artificial Intelligence based health tech company
5 Magentiq Eye magentiq.com Computer vision technology in endoscopic procedures
6 Imagia Cybernetics imagia.com Medical image analysis using artificial intelligence for detecting cancer
7 PathAI pathai.com AI based solutions for clinical diagnosis
8 RadSupport radsupport.io Deep learning for mammography image analysis
9 Exact Imaging imagistxprostate.com
Prostate cancer imaging
10 Notable Labs notablelabs.com Precision medicine and clinical decision support solutions for Brain cancer
11 Freenome freenome.com Deep learning platform for detecting circulating cancer-related markers
12 Smart Health Care smarthealthcare.se Deep learning algorithms for cancer diagnostics.
13 LPixel lpixel.net Image analysis tools using artificial intelligence
14 ClearView Diagnostic
clearviewdiagnosticsinc.com
Computer-aided detection and diagnosis algorithms for breast cancer detection.
15 iSonoHealth isonohealth.com Self-monitoring Solution for Early Breast Cancer Detection
16 VisExcell visexcell.com Breast cancer risk assessment using AI
17 MobileODT mobileodt.com Smartphone-based diagnostic tools for the screening of cervical cancer
18 Entopsis entopsis.com Universal diagnostic platform
19 HaploX haplox.com NGS-based liquid biopsy tests for precision medicine
20 KMD Biomarkers kmdbio.com Urine-based metabolomics assay for determining cancer progression
21 Immortagen immortagen.com AI-based NGS solutions for cancer diagnostics
22 Maxwell MRI maxwellmri.com Imaging software solutions for cancer diagnostics.
23 AlgoRhythm Diagnostics
algorhythmdiagnostics.com
AI-based analysis tools for radiology images
24 Inform Genomics informgenomics.net Predictive pharmacogenomic platform for Oncology
25 CureMatch curematch.com Matches cancer patients with optimal Rx combinations
26 INTIO intio.us Image analysis for Interventional Oncology
27 Bering beringresearch.com Provides an Artificial Intelligence (AI) platform to identify complex patients and prioritizes actionable interventions
28 Deontics deontics.com SaaS based personalized CDSS using artificial Intelligence
29 Tempus tempus.com Oncology platform for sequencing, analytics and informatics
30 Avatar DSS avatardss.com Precision treatment support for cancer patients
31 Intervention Insights
interventioninsights.com
Evidence-based decisions support tool for molecular oncology
32 Syapse syapse.com Molecular profiling platform for diagnosing and treating patients
33 Oncora Medical oncoramedical.com Decision support tool for Radiation Oncologists
34 DiaScan diascan.co Software that analyzes CT scans for lung cancer using machine learning
35 LungDirect lungdirect.com LungDirect is Matrix Analytics’ end-to-end software suite that facilitates lung cancer screening data submission and optimizes patient care.
36 Medial EarlySign earlysign.com Develops a machine learning based decision support tool which enables personal and outcome-based interpretation of medical data.
37 N-of-One n-of-one.com Molecular profiling and analysis for hospitals
38 Feedback fbk.com Software solutions and workflows for medical image analysis
39 Hexalotus hexalotus.com 3D Medical Image Processing
40 Signifikance signifikance.com Application of genetic discoveries in clinical practice to enable Precision Medicine
41 Kuveda kuveda.com Provider of SaaS tools for analysis of cancer patients molecular profiles,
42 Turbine turbine.ai Design effective cancer combination therapies using artificial intelligence
43 Medint medint.io Medical Intelligence & personalized data platform to support medical decisons
44 Percans Oncology zkbymed.com Personalized treatment planning and services for cancer patients
45 Aindra aindra.in Developer of handheld devices with technology to detect and identify people
landscape of 40+ startups in pathology
modeling and capturing
EXISTINGknowledge and
making it available
modeling and learning
NET NEWknowledge to
improve human performance
dichotomy
modeling and capturing
EXISTINGknowledge and
making it available
successesarrhythmia recognition from electrocardiograms
coronary heart disease risk group detection
monitoring prescription of restricted use antibiotics
early melanoma diagnosis
breast cancer diagnosis
modeling and learning
NET NEWknowledge to
improve human performance
CELLWORKS
oncology therapy selection
GENXSYS
decision support for GP
SECONDOPINIONS.COM
radiology second opinion
promising
near termpredicting the
from?learning
1.
2.
3.
4.
5.
long termpredicting the
the future of the patient
Sample Market Signals
Networks & Sensors
Wearables
Learning System
Robotics
Sample Use Cases
Drivers of Disruption
Deloitte Cognitive Engagement – designed to boost patient involvement in care and expand the type of alerts and interactivity
New Ways to Engage
Tyto Care – handheld device that patients can use to self-examine their mouth, throat, eyes, heart, lungs, skin, and temperature
Tools for Self-Service
Ginger.io – aggregates cellphone data to monitor patient’ mental health and alert caregivers when symptoms are problematic
The Quantified Self
the future of care delivery
Sample Market Signals
Healogram – Mobile platform that helps providers remotely monitor patients post-surgical procedure
Where Care is Delivered
Retina Selfie – Retina self-imaging allows patients to monitor for diseases like multiple sclerosis and detect early warning signs
When Care is Delivered
Camera resolution
Robotics and NLP
Augmented Reality
Learning Systems
iDAvatars – Virtual avatar, Sophie, uses artificial intelligence and natural language processing to remotely monitor patients
Who Care is Delivered by
Sample Use Cases
Drivers of Disruption
the future of healthcare operations
Robotics
Learning Systems
Analytics
Communication Speed
Sample Use Cases
Aethon TUG Robots –Smart, autonomous robots substitute for the labor needed to transport materials & clinical supplies
Improved Productivity
Care at Hand – Helps to prevent readmissions by deploying nursing staff skills to maximum efficiency
Efficient Administration
Evena – Technician glasses provide high-definition, real-time images of vascular anatomy to enable fast, precise IV access
Workforce Augmentation
Drivers of Disruption Sample Market Signals
the future of precision medicine
Sample Market Signals
Genomics
Personalized Medicine
Nanotechnology
Sample Use Cases
Drivers of Disruption
LiverChip – Dynamic 3-D cell culture platform can exactly mimic the architecture and physiology of the human liver
New Treatment Methods
Human Longevity Inc. –building the world’s most comprehensive database on human genotypes
Preventative Measures
Wyss Institute –Harvard scientists built a nano-robot from designer DNA to deliver drug dosages to specific cell types
Increased Precision
the labwhat does it mean for
1. digitizing is key. start now
2. indexing is key. start now
3. storing correlated clinical data is key
4. start a pilot – coding, for example
5. start a pilot – chat-bot for counseling
the differences between us and them
emotion understanding consciousness
creativity
A.I. is the discipline of how to make computers do things at which, at the moment, people are better.
www.artiman.com