Call Girls Jp Nagar Just Call 7001305949 Top Class Call Girl Service Available
Â
Summit2013 ho-jin choi - summit2013
1. 18 July 2013
Ho-Jin Choi
Dept. of Computer Science, KAIST
Systems Biomedical Informatics Research
Center (SBI-NCRC), SNU College of Medicine
Personalized Context-Aware Health Avatar in
Smart Phone Environment
The 2013 STI Semantic Summit
Suzdal, Russia, July 17-19, 2013
1
2. Outline
ï SBI-NCRC â A brief introduction
ï Research Topics
ï Activity Recognition for Personalized Life-Care
ï Healthcare Service Framework for Continuous Context
Monitoring
ï Text Mining for Extracting Knowledge from Web Contents
ï Personalized Bio and Medical Data Analysis
2
4. SBI-NCRC
ï NCRC (National Core Research Center)
ï Government initiative to support interdisciplinary research & education
ï Since 2004, one or two centers newly selected each year
ï Funding scale, 2 million USD/year * 6.5 years
ï Systems Biomedical Informatics (SBI) Research Center
ï An NCRC established jointly by SNU Hospital and KAIST Computer Science
ï Born in September 2010
ï 24 professors/researchers participating from 4 organizations
ï SNUH, KAIST, Ajou University, ETRI
ï Goals for SBI-NCRC
ï To define and realize âDigital Selfâ or âHealth Avatarâ prescriptive medicine
ï To integrate clinical information and bio-information using IT
ï To launch an interdisciplinary program in Biomedical Informatics
ï To collaborate towards Joint KAIST-SNUH BIT Campus in Inchon area
4
5. 4P Medicine
ï Preventive medicine
ï Predictive medicine
ï Personalized medicine
ï Participatory medicine
Tests for early detection
Risk evaluation Prevention
Targeted
monitoring
Diagnosis Treatment
Results
monitoring
Caring Diseases Caring Health
5
7. Teaming
7
Group/Project Title PIâs Major
Group 1
Foundation Self: Molecular and Cellular Foundations
of Self
SNUH, Ajou U.
Project 1-1
Genomic Self: Translational Bioinformatics for Genomic
Health and Molecular Medicine
Psychiatry(1), Surgery(1),
Bioinformatics(1), Statistics(1)
Project 1-2
Phenomic Self: Data and Measurement driven
Discovery and Understanding of Human Disorders
Pathology(2), Bioinformatics(1)
Project 1-3
Physiomic Self: Multi-scale Modeling of Physical and
Physiological Systems of Human Body
Biomedical Engineering(2),
Neurosurgery(1), General
Practice(1)
Group 2 Simulated Self: Individuated Second Self SNUH, KAIST
Project 2-1
Semantic Self: Ontological Representation and
Engineering of Health Avatar
Nursing Informatics(2), Internal
Medicine(1), Pathology(1)
Project 2-2
Augmented Self: Multi-modal Assessment and
Treatment to Retain and Enhance Human Performance
NLP(1), Graphics(1), Image
Processing(1), Psychiatry(1)
Group 3
Ubiquitous Self: Life Logs and Distributed
Collaborations
KAIST, ETRI
Project 3-1
Distributed Self: Customized and Context-aware
Healthcare Service Agents in Smart Phone Environment
AI(1), Software Engineering(1),
Bioinformatics(1)
Project 3-2
Connected Self: Life Logs and Stream-Type Data Mining
for Health Protection
Information Systems(1), Data
Mining(1)
9. 9
ï Target healthcare domains
ï Obesity, diabetes, dementia
ï On-going research topics
ï Activity Recognition for Personalized Life-Care
(Prof. Ho-Jin Choi)
ï Healthcare Service Framework for Continuous
Context Monitoring (Prof. Jun-Hwa Song)
ï Text Mining for Extracting Knowledge from Web
Contents (Prof. Key-Sun Choi)
ï Personalized Bio and Medical Data Analysis (Prof.
Gwan-Su Yi)
Research Topics
10. Activity Recognition for Personalized
Life-Care
Prof. Ho-Jin Choi
Dept. of Computer Science
KAIST
10
11. Multi-Sensor Surveillance for Elderly Care
11
âPatient #1234 is
in a risky situationâ
Data observed from microphones helps
the system detect the potentially risky
situations .
The agent estimates patient #1234âs behaviors.
When preliminary conditions of
dangerous situations are occurred
to the patient, the agent alarms to
the caregiver.
12. 12
Activity Recognition from Video Image with Depth Sensor
ï Action Recognition with Automatically Detected Essential Body Joints
13. Technologies Involved
Understand Image Data
- RGB images (camera) and
depth images (depth sensor)
are sent to the system
- System then do
-Find a patient in a scene
-Track the patient
-Understand behaviors of
the patient
â Issues to challenge
- The level of complexity
of scenes and behaviors
- Scenes may contain
various objects and
backgrounds
- Human-behaviors should
be understood as much
as possible.
13
Understand Audio Data
-Audio data (microphones)
is sent to the system
-System then do
-Detect abnormal sounds
â Issues to challenge
- How accurate the system
detects abnormal sounds
Detect Risky Situation
-After analyzing data from
various sensors, the system
determines whether the
situation is potentially risky
-System constructs a
database for predefined
risky situations
-For every situation, the
system calculates the
likelihood of being risky
-If the likelihood scores
more than a threshold, it
alarms to the caregiver
â Issues to challenge
-How well the system
constructs the database
-The accuracy of likelihoods
Find Patientâs Location
-Smartphone gives and
receives various signals to
update patientâs geographic
information
â Issues to challenge
- How accurately the system
locates the patient
14. Wrist-Type Device Based Human Behavior Recognition
14
ï Mediated Interface for human-robot interaction
âŠ.
Health Care
Care Services
Raw data
(Behavior pattern,
Vital Signal, etc)
Old People
How to get âRaw Dataâ
From Old People?
Robots
Care-giver
Activity, Gesture,
Vital signal,
Location,
Identification(MI: Mediated Interface)
Ex: Watch, Ring
Robots
Care-giver
Elderly Care Services Using Robots
Suggestion
Fall
Detection
Wandering
Monitoring
Location
Monitoring
Care Services
15. Wrist-Type Device Based Human Behavior Recognition
15
ï Wrist-type and waist-type monitors
MCU Cortex-M3 (STM32F100)
RF(Zigbee) CC2520
Sensors 3-axis accelerometer(LIS331DLH)
3-axis gyro (L3G4200D)
Temperature/humidity (SHT21)
Brightness (TCS3414CS)
IR Photodiode (TSOP85238)
Emergency
button
1ê° (front side)
Memory card MicroSD
Battery [Li-Ion 600mAh]
Recharger External rechager
Strap Wrist: nato band
Waist: elastic belt
16. Lifestyle Manager Using SNS and Activity Recognition
16
Life-style patterns
Clinical history
Genetic information
Server
Smartphone
users
Lifestyle
ranking
Life
log
Lifestyle
disease risk
Default
behavior
registration
Location - time
elapse
threshold
Localization by
Wifi signal
17. 17
ï Analysis of life log and SNS
ï Lifestyle = Eating habit(timing and food types) + CAR(Circadian activity rhythm)
Server
Many smart phone users
Location dimension
Sleep : My room, Park, Motel
Rest : TV room, Living room, Lounge
Work or study : Work place, Study place
Enjoy : Shopping place, Cultural place, Attractions
Usual food : Restaurant
Exercise : Exercise place
Religious activities : Church, Buddhist temple
Fast food or snack : Fast food place, Mc. Donald,
Convenience store
Sugar-sweetened beverage : Cafeteria,
Convenience store
Smoking : Smoking place, Convenience store,
Alcohol : Alcohol place, Bar
Drug : Drug store
UNKNOWN
1. Lifestyle
recommendation
2. Measurement of
Lifestyle Metric
Data matrix
Tries for usersâ visiting patterns on
location & time dimension
People
Healthcare organizationsProactive/Reactive Services
Lifestyle Manager Using SNS and Activity Recognition
18. Lifestyle Manager Using SNS and Activity Recognition
18
- Home, Hotel
- Smoking place
- Drinking place
-Working place, Studying place
- Restaurant
- Cafeteria, Coffee shop
- Exercising place
- Religious place
-Attractions, Shopping place,
Cultural place, Enjoy place
- Hospital, Pharmacy
- Unknown
Location manager
- Facebook :
contains more
daily lives than
others
( e.g.,
âI ate a hamburger,
so cool.â 11:45AM )
Social Network
Services
-Walking
- Exercise / Sport
- Running
- Riding an automobile
- Riding a bicycle
- No activity
Activity Recognition
-Accelerometer
- Illuminance sensor
-WiFi
- GPS
Sensor Handler
- Name
- SNS information
-Address ( GPS )
- BMI
- âŠ
User profile
- Lifelog
- Carlorie counting
(day, week, âŠ)
- Lifestyle disease risk
Service Provider
20. 20
Problem Document1
Solution Document1
Original Document1
Original Document2
Original Document3
Original Document4
Original Document10
...
Problem
Document10. . .
Solution
Document10. . .
Split to problem and
solution documents
Topic Mining for Problems and Solutions
22. Healthcare Service Framework for
Continuous Context Monitoring
Prof. Jun-Hwa Song
Dept. of Computer Science
KAIST
22
23. Context-Aware Healthcare Service Scenarios
23
ï Example Scenarios
ï Obesity monitoring
ï Continuously monitors peopleâs activity level
and consumed calories, and suggests proper
exercises to the people.
ï Elderly people monitoring
ï Continuously monitors an elderly peopleâs
emergency situation such as slipping down on
a wet floor, and expedites an emergency call.
ï Cardiac patient monitoring
ï Continuously monitors a cardiac patientâs ECG,
and expedites an emergency call.
24. DietSense: Smartphone-Based Diet Monitoring for Enhancing
Obesity Self-Care
24
Comparing diet and physical activity
Monitoring Physical ActivityCapturing Diet
Calories consumed
from food
Calories burned during
physical activity
camera
microphone
accelerometer
25. Activity Log
Smartphone
1. Collecting activity data from the patient
2. Training ML algorithms for analyzing activity patterns
3. Figuring out the right does time without interrupting
the current work activity
4. Notify the does time and subsequent reminder
MedicineTaker
Motion Sensor
Activity Log
Place A Place A
Task 1 Task 2 Task 3
Action 1-1
Action 1-2
Action 1-3 Action 2-1
Action 2-2
Action 3-1
Type 1
Type 2
Smart Alarming for Long-Term Medicine Adherence
25
26. Continuous Context Monitoring
26
ï Continuous monitoring of userâs context
ï A key building block for personal context-aware applications
ï Often requires complex, multi-step, continuous processing for multiple devices
ï E.g., Running situation -> sensing in three 3-axis accelerometers, FFT processing,
recognition
Location-based
Services
HealthMonitoring
U-Trainer
U-Secretary
U-Reminder
Dietdiary
U-Learning
Behaviorcorrection
S
F C
S
S
F
F
C
SS
F
C S
S
F
C
S
F
C
Sensing
Feature extraction
Context recognition
PAN-scale dynamic
distributed computing
platform
Context monitoring
(e.g., sensing, feature extraction,
recognition)
Application logic
Location-based
Services
HealthMonitoring
U-Trainer
U-Secretary
U-Reminder
Dietdiary
U-Learning
Behaviorcorrection
A A
A App logic
27. Mobile Healthcare Service Framework
27
ï Develop a healthcare
service framework
ï To support multiple and long
running healthcare services with
highly scarce and dynamic
resources
ï Efficient resource utilization
ï Longer lasting operation (and
service) under highly scarce
resource situation
ï Quick and efficient abnormal
situation detection
ï Seamless (stable) operation
even under high resource
dynamics
ï Challenges
ï Limited battery power due to
mobility
ï Scarce computing resources of
mobile devices
ï Dynamic join/leave of
heterogeneous sensors
ï Multiple healthcare services
share highly limited resources
Resource Manager
Policy
Manager
Energy Manager
Sensor Broker
Sensor Detector
Communication
Manager
⊠âŠHeartbeat monitoringFall monitoring
Healthcare services
API
Message
Interpreter
Sensor Data Processor
Resource
Monitor
Network protocols (e.g., ZigBee, BT)
Health Context Monitor
Sensor data
Requests
Sensor availability/status
Results
Sensor detection/control, Data/status report
Sensors in BAN/PAN
Application Broker
Application
Interface
Result Manager Message Parser
Diet diary
GPSBVP/GSR Accelerometers
âŠâŠ
Anomaly Detector
Feature Extractor
Resource Coordinator
28. Healthtopia â Healthcare Platform
ï Providing API
for utilizing
health sensors
ï Saving power
consumption
for concurrent
multiple apps
28
29. Text Mining for Extracting Knowledge
from Web Contents
Prof. Key-Sun Choi
Dept. of Computer Science
KAIST
29
30. Food Ingredients and Recipe Advice for Controlling Obesity
Web Environment
Ontology
Mobile Environment
Recipe
Extraction
Web /Wikipedia
Automatic
User Experience
Extraction
Target Food/Dish
Recognition
Manual
User Experience
Input
Web Log / SNS
Equipment
How-to
Food/Dish
Restaurant
DB
Scenario 2.
New Recipe (Low calories)
Suggestion w/ same Ingredients
Scenario 1.
Food-NutritionAssociationVisualization
Nutrition
Ingredients
Food-Nutrition
Extraction
30
31. Mining Connections between Multiple Sources
31
Literature Web Clinical Data
HeterogeneousTextual Sources
-Textbook
- PubMed
- Blogs
-Wikipedia
- Personal health record
ï Medical information sources
ï Literature contents affect the Web contents
ï As background factual knowledge
ï Web contents have other benefits
ï Wide coverage
ï Huge collaborators (confidence)
ï Aggregating information from multiple sources
ï Analysis of trend evolving on literature/Web to identify factors that will improve the
quality of patient care
ï Reliability: Literature > News > Web (Wikipedia, blog, SNS)
ï Accessibility: Web â„ News > Literature
32. Detecting MeSH Keywords from Web Pages
32
ï Medical Subject Headings (MeSH)
ï NLM controlled vocabulary thesaurus used for indexing articles for PubMed
ï Tree structure (http://www.nlm.nih.gov/mesh/trees.html)
ï Provide an efficient way of accessing and organizing biomedical information
ï Examples of MeSH Headings
ï Body Weight, Kidney, Dental Cavity Preparation, Self Medication, Brain Edema
Extracting Candidates
Matching
Obesity is a medical condition in which excess body fat has a
ccumulated to the extent that it may have an adverse effect on
health, leading to reduced life expectancy and/or increased he
alth problems.[1][2] Body mass index (BMI), a measurement
which compares weight and height, defines people as overweig
ht (pre-obese) if their BMI is between 25 and 30 kg/m2, and o
bese when it is greater than 30 kg/m2.
âą Hyperinked terms are extracted as term candidatesLanguage Handling
33. Detecting MeSH Keywords from Web Pages
33
Extracting Candidates
Matching
Obesity
Medical
condition
Body
fat
Body Mass Index
weight
Dieting
Obesity
Medical
condition
Body
fat
Body Mass Index
Body
Weight
Diet
Link
Structure
MeSH
term
Language Handling
ï Language Handling
ï Polysemy and homonymy problem
36. Personalized Bio and Medical Data Analysis
Prof. Gwan-SuYi
Dept. of Bio and Brain Engineering
KAIST
36
37. Personalized Diseases Risk Analysis
User
Agent
Disease risk
Prediction model
Personal genome
Data processing model
Drug response
Prediction model
Disease risk info.
(SNP-Disease)
Drug response info.
(SNP-Drug)
Personal genome
info.
Personal sequence
data
New info. on
disease risk
New info. on drug
response
Update Update
request
result
Personalized
disease risk
Genome
profile
Personalized drug
response
Storage
Build database
Personalized Personalized
Drug
response
Diseases
risk
Obesity, Diabetes Obesity, Diabetes
37
38. Constructing Databases for Diseases Risk and Drug
Response
38
184(Type I Diabetes), 203(Type II Diabetes), 82(Obesity) entries for diseases related SNP markers
collected
228 drugs, 830 SNP markers, 1341entries for drug-SNP related information collected
Diseases
Drugs
Diseases risk info.
SNP ID
Gene
Gene Region (Locus)
Risk Allele
Odds Ratio
P-value
Reference
âŠ
PharmGKB
Drug Bank
Drug response info.
SNP ID
Gene
Gene Region (Locus)
Drug
Condition
Reference
âŠ
Integrated database for
diseases risks and drug responses
Public database
Drug related info.
23andMe
Navigenics
Pathway Genomics
Gene sequencing
service drug related
info.
23andMe
Navigenics
deCODEme
Gene sequencing
service GWAS info.
HugeNavigator
GAD
NCBI (HapMap &
NHGRI catalog)
Public database
GWAS info.
39. Developing Methods for Analyzing Diseases Risk and
Drug Response
39
Diseases
OMIM
PharmGKB
DrugBank
Drugs
PharmGKB
DrugBank
SNPs
dbSNP
HapMap
Genome type
WTCCC
Genome body
UCSC
Ensembl
AceView
Genome
Entrez
Bio. pathway
KEGG
Reactome
NCI pathway
Panther
SNPs
dbSNP
HapMap
Genome type
WTCCC
Biological information
SNP
SNP
Analysis tech. for diseases risk
Analysis tech. for drug response
Obesity, Diabetes
âą Extraction of drug response related SNPâs
âą Drug targeting and biological pathway based function analysis
âą Drug response prediction
âą Obesity (Diabetes) related SNP or SNP combinations info.
âą Genomic and biological pathway based function analysis
âą Diseases risk prediction
40. Service Platform for Personalized Information about
Diseases Risk and Drug Response
Agent
User
Plug-ins from
life-logging team
Diseases risk
prediction model
Personal genome info.
data processing model
Drug response
prediction model
Diseases risk info.
(SNP-Disease)
Drug response info.
(SNP-Drug)
Diseases DrugsGenome
body
SNP PathwayGenome
type
Life-logging
database
Personal genome
info.
40