2. 2
Problems of the New Age and the New World
Developed Countries
Elderly people - 44.7 M
(2013), 98M by 2060
Invasive and costly
diagnosis
One size fits all Diagnostic
/ Treatment protocols
Some diseases yet to
have a cure
Developing Countries
Capacity - not enough
doctors per patient
Reachability – specialized
primary care not available
Affordability - majority
cannot afford to pay the
cost
http://www.aoa.acl.gov/Aging_Statistics/index.aspx
3. 3
Diagnosis from Symptoms and Signs is still an Art based on aggregate rules -
“Diagnosis is the heart of the medical art”
Data-driven systems allows Diagnosis to be Evidence-based
than rule based – allows personalization and adaptation
From Illness to Wellness and From Rule to Evidence
Need to go towards Wellness Driven Models
All stakeholders incentivized to keep patients healthy
http://media.cagle.com/107/2012/09/21/119074_600.jpg
Illness Driven model incentivizes
people being sick
“The health care system is really designed to reward
you for being unhealthy. If you are a healthy person
and work hard to be healthy, there are no benefits.”
- Mike Huckabee
http://www.brainyquote.com/quotes/keywords/health_care.html#WmKeI72wL5Wg6wqG.99
http://www.greekmedicine.net/diagnosis/Introduction.html
4. 4
Data-driven Systems – Sensing and Analytics to the Help
24x7 remote
monitoring of
activity,
physiology and
pathology
Automated
generation of
alerts on
anomaly
Personalized
Prognosis and
risk profiling
Discovery of
new diagnosis
protocols
Reachable,
Affordable
Elderly /
Home Care
Personalized Predictive Maintenance of our Body using Internet-of-Things
Sensing Analytics
Reduce
Doctor Load,
Improve
Capacity
Towards
Wellness
Driven
Systems
Repeatable
Diagnostics
New Cures
5. 5
Internet-of-Things based Remote Sensing and Analytics System
Mobile phone as
medical gateway
TCS Connected
Universe
Platform
Web Request
Patient
Records
Social
Network
Healthcare
Portal
Expert Doctor
Wearables
Nearables – Mobiles,
Camera, 3D, Thermal, …..
Instruments
• Real-time View
• Alerts for Medical Emergency
• Analytics for Diagnostics / Prognostics
Rural Remote Healthcare –
Villages in Chhattisgarh, Gujarat
Home Monitoring –
Hospital in Bangalore
Elderly People Monitoring –
Pilot at Singapore
6. 7
Camera
• Heart-rate and Respiratory Rate
• Blood Pressure
Microphone
• Heart Sound
• Heart Rate Variability
Accelerometer,
Magnetometer, Gyroscope
• Step Count, Activity
• Fall Detection
Physiological Sensing on Mobile - Affordable
Heart Rate, BP, HRV, Respiratory Rate
Heart Sound
8. 9
Challenges and Results
Heart Rate
• Movement Artefacts
• Incorrect Placement / Obstruction of Blood Flow
BP
• Physical Modeling of Cardiovascular Systems
Activity / Fall Detection
• Orientation Correction
• False Positives from Normal Activity
Solution Accuracy
Heart Rate ~2 bpm
HRV (SDNN) 89%
BP 80% to 85%
Activity Classification (Static, Walking, Brisk Walking, Jogging) 90%
Step Count 95%
Fall Detection 99% Detection, 92% False
Alarm Removal
9. 10
Physiological Sensing using Nearable (Camera / RF) - Unobtrusive
http://www.extremetech.com/extreme/149623-mit-releases-open-source-software-that-reveals-invisible-motion-and-detail-in-video
Fadel Adib et.al., “Smart Homes that Monitor Breathing and Heart Rate”, CHI 2015, Seoul
10. 11
Real-time Alert for Anomalies
– when to go and see and doctor
Prognosis for CAD
– early filtering help for doctors for
prescribing more specific tests
Use case - Early Detection of Coronary Artery Disease (CAD)
By 2020, CAD will be the leading cause of death in Western and Asian countries
• 20-30% deaths in industrialized countries, 60% of world heart ailments from India
• CAD is a modern epidemic according to WHO
• Current method of 3D Angiography costly, obtrusive and harmful to health
Working with doctors at a Cardiac Specialty Hospital in Kolkata
Blood Pressure, Heart Rate, Blood Oxygen from Wearable / Mobile / Nearable
11. 12
Cognitive Computing and AI – the Future
Deep LearningDeep QA
http://www.cbsnews.com/news/jeopardy-winning-computer-now-using-its-brain-for-
science/
http://www.slate.com/blogs/future_tense/2012/06/27/google_computers_learn_t
o_identify_cats_on_youtube_in_artificial_intelligence_study.html
12. 13
CAD Alerting and Prognosis - Architecture
Live Patient Data (Sensing)
Stored Medical
Records
Knowledge Base
Reasoning
Alert Generation
Healthcare Portals,
Medical Books, Article
Diagnostic / Prognostic Support
Relevant Data
Evidence based
Learning Text Mining
Knowledge Access
Stream Handling
Anomaly
Detection
Other
Filters
Deductive Abductive Others
Entities
Rules
Relations
Multi-
variate
association
Rule mining
Deep
Learning
Cognitive
Computing
Available Dataset – MIMIC-II
– Waveform for 2500 patients matched with medical records - HR, BP, RR, SpO2
– Classified into approx. 700 CAD and 1200 non-CAD patients using ICD-9 codes
14. 15
CAD Detection – from PPG and Heart Sound
PPG Windows
extracted from 19
subjects (15
nonCAD and 4
CAD)
CAD
Predicted
nonCAD
Predicted
CAD
Diagnosed
91% 9%
nonCAD
Diagnosed
23% 77%
Deep Learning based Work under Progress
15. 16
Use Case - Tele-Rehabilitation for Stroke Patients
annual cost in EURO in European
economy: - twice the cost of cancer
798 billion
people worldwide need
rehabilitation services
do not receive rehabilitation
treatment after discharge
2/31 billion
RehabWeek conference 2015 by NeuroAtHome (http://www.neuroathome.net/p/home.html)
• Existing Quantitative Gait
Analysis systems costs approx.
Rs. 35 lakhs & not readily
available in the market.
• Expensive maintenance costs
• Difficult for patients to frequently
visit hospitals
16. 17
Solution Architecture
Left Heel: Line of Progression Right Heel: Line of Progression
Store Raw
Data
Patient’s
Exercise
Parameter
Patient
History
Extract
Parameters
Working with doctors at a Neuro-Speciality Hospital in Kolkata
17. 18
Results
Stride Length estimated using Kinect data and validated w.r.t GaitRite
Definition: distance between the heel-
points of 2 consecutive footprints of
same foot. In Fig 1.: Stride Length =
Distance Between Points A,B
Fig. 1.
Mean Absolute Deviation
between our estimated stride-
length and GaitRite
measurement is about
3.084cm.
Single Limb Standing – duration and Jitter Measurement
18. 19
Publications and Awards
1) "A Robust Heart Rate Detection using Smart-phone Video", in MobileHealth workshop of Mobihoc 2013
2) “UbiHeld - Ubiquitous Healthcare Monitoring System for Elderly and Chronic Patient”, in Recognize2Interact Workshop of
UbiComp 2013
3) “AcTrak - Unobtrusive Activity Detection and Step Counting using Smartphones”, Mobiquitous 2013
4) "Demo Abstract: HeartSense – Estimating Blood Pressure and ECG from Photoplethysmograph using Smart Phones", SenSys
2013, Italy, Rome. 11-15 Nov. 2013
5) “Improved heart rate detection using smart phone” In Proceedings of the 29th Annual ACM Symposium on Applied Computing
(ACM-SAC), 2014
6) "PhotoECG: Photoplethysmography to Estimate ECG Parameters", ICASSP 2014
7) "Smart Phone Based Blood Pressure Indicator", in MobileHealth workshop of Mobihoc 2014 11-Aug, 2014, Philadelphia, PA,
USA.
8) "Estimating Blood Pressure using Windkessel Model on Photoplethysmogram", 36th Annual International Conference of the
IEEE Engineering in Medicine and Biology Society (EMBC '14), Chicago, Illinois, USA on August 26-30, 2014.
9) "Effects of Fingertip Orientation and Flash Location in Smartphone Photoplethysmography", Third International Workshop on
Recent Advances in Medical Informatics (RAMI-2014), ICACCI 24-27 Sept. 2014, Delhi.
10)"HeartSense: Estimating Heart rate from Smartphone Photoplethysmogram using Adaptive Filter and Interpolation" in 1st
International Conference on IoT Technologies for HealthCare (HealthyIoT, IoT-360), 2014
11)"Demo Abstract: HeartSense: Smart Phones to Estimate Blood Pressure from Photoplethysmography" in 11th ACM
Conference on Embedded Networked Sensor Systems (SenSys 2014) – Best Demo Award
12)"HeartSense: Photoplethysmography to Estimate Physiological Vitals" in The 4th International Conference on the Internet of
Things, 2014
13)"Noise Cleaning and Gaussian Modeling of Smart Phone Photoplethysmogram to improve Blood Pressure Estimation“,
Presented in ICASSP 2015
14)“Novel Peak detection to estimate HRV using Smartphone Audio”, presented in Body Sensor Network (BSN) 2015
15)“Feasibility Analysis for Estimation of Blood Pressure and Heart Rate using A Smart Eye Wear”, Wearable workshop in Mobisys
2015
Aegis Graham Bell Award for Smart Healthcare