We are generating 2.5 Billion GB of data every day. That's a lot of data! We will need super human expertise to make sense out of it. Well, that's exactly what AI can help us do it.
This talk is going to focus on:
i) What is AI?
ii) How AI can help with health care?
iii) How FHIR will help with the adoption of AI
iv) What are the next three steps for any health organization in order to adopt AI?
2. About Me
CEO @ API Garage
10+ years of Industry Experience
Technology Mentor @ Biomedical Zone ( Ryerson University &
St. Michael’s Hospital)
Founder of Applied AI Toronto Community
6. Big Data-berg
AI is required to
explore hidden
potential within
that data.
Only 0.5% of data is being used
for analytics. (Source: IDC)
Evolutionary Strategies
Deep Learning
Churn Prevention
Anomaly Detection
Recommendation Engine
Sentiment Analysis
Predictive Analysis
11. How to avoid?
Filter data before
training your models
Test locally, but use
the production data
Monitor usage, and
have a fallback
12. IBM & MD Anderson
● IBM Watson cognitive computing system for its mission to
eradicate cancer.
● Project cost $62M+ ($39.2 million to IBM and $21.2 million
to PriceWaterhouseCoopers), yet did not meet its goals.
● “The OEA(Oncology Expert Advisor) R&D project was a
success, and likely could have been deployed had MD
Anderson chosen to take it forward” -- IBM Officials
● The project is on hold. Requesting bids from other
contractors.
13. Involve I.T.
department from
the beginning.
Do not underestimate
the data cleaning and
integration work.
Do not over invest in
business before
technical validation.
How to avoid?
14. Google DeepMind with the NHS
● 1.6 million NHS patient records shared who visited Royal
Free Hospital in London
● Records included: name, address, abortion and HIV details
● Kidney Monitoring mobile app - “Streams”
● Although Google does not see the data
● Under investigation by National Data Guardian (NDG) for
illegal access to the data
● Expecting a verdict in next couple of weeks
15. Do not rush through paperwork in excitement. Most
importantly, get the user's’ consent. “Apologize
later” startup mentality does not work here!
How to avoid?
16. 16
Build AI use cases for
your organization
Use AI Design Sprint or many other
methodologies to build AI use cases.
1
Clean up and annotate
your data
Collect, clean, and enrich your data. If you
are working with a third party vendor make
sure legality is covered.
2
Start experimenting
Experiment internally, first. As leaders in
your firm, it’s your duty to change this.
3
Next
Three
Steps
17. 1. Build AI Use Cases
The sky is the limit
Anything you can image can be and
will be automated. Great quantity of
ideas. Quality of the ideas suffer.
One-second rule
Any task taking <= 1 second to
complete given all the information will
be automated. Brings practicality to
brainstorming.
Market research
Researching what competitors,
partners, and other similar companies
doing in the industry. Adopt anything
useful.
AI Design Sprint
Customer & data centric realistic
approach to exploring the data
possibility to solve real world
problems.
Project Requirements: Given ‘x’ data as the input, the model should be able to return ‘y’.
18. 2. Clean up and annotate your data
List data requirement for the use cases from step 1.
User permissions required on the data.
Clean up, annotate, and enhance your data.
Aggregate data in one place.
19. 3. Experimentation
It’s an iterative process.
User permissions required on the data.
Clean up, annotate, and enhance your data.
Aggregate data in one place.
20. @apigarage
THANK
YOU 20 Camden Street, Unit 200
Toronto, ON M5V 1V1
P : (647) 878 4217
E : cpatel@apigarage.com
w: www.apigarage.com