resilient healthcare by it support• in this presentation, we propose a resilient health care...
TRANSCRIPT
Resilient Healthcare
by IT support
IDU-D 2018
OrangeTechLab inc.
Jun Miyazaki Ph.D.
Introduction
• To be ready for unexpected incidents of pandemic or paramedic situations in developing countries
• Because these situations will affect and be spread globally to all over the world.
• In this presentation, we propose a resilient health care approach utilizing several IT technologies.
• IT technologies are statistical analysis, process mining, and finally AI based analysis.
Rise of Resilient Systems and engineering
• Resilient Systems and Engineering has been
started since early 21st centuries.
• The objective of the systems and engineering is
how we can manage both predictable and
unpredictable incidents in social activities.
• Based on unpredictable incidents like Tsunami
we Japanese experienced,
• Japan has more paid attention to Resilient
Systems concept
• The system can be capable even unexpected
disaster incidents may occur in the future.
3
The requirement for Resilient Healthcare in
developing countries
• Resilient Healthcare is one of resilient systems• predicted and unpredicted medical incidents and processes
are the problem domain.
• In developing countries, resilient healthcare is
very important• pandemics or severe paramedic situations may occur
• these threats are not only for the original countries but also
may be big threats for rest of the world.
• we propose ITU-D, how we can manage these
threats by applying resilient healthcare systems.
4
Resilient Healthcare as Resilient Engineering
There two levels of resilient safeties:
Safety-1: (Avoid failure processes, and keep successful
processes)
• Key principle of Resilient Healthcare is “to keep the
successful processes as much as possible”, rather
“to salvage failure processes” and avoid failure
processes.
Safety-2: (Predictive safety management)
• The activities are coordinated to keep the processes
to be succeed than to be failed by the human experts’
efforts, and predictive safety management is required.
5
Three Phase approach
• First phase
• Gather several predicted and unpredicted examples in the past
healthcare facts from every ITU-D related countries
• Data will be formalized to be a resilient example database.
• How we can set up anonymity is the challenge at this stage.
• Second phase• Analyze explicit and implicit processes in these descriptions, by
manually and semi automatically as possible.
• Third phase• Introduce statistical analysis, process mining utilizing machine
learning such as temporal deep learning technologies to analyze
both explicit and hidden processes in the database.
• Find what aspects will be the key feature for resilient healthcare
• Automatic prediction and unexpected process discoveries.
Three Phase approach
• First phase
1. Gather several predicted and unpredicted
examples in the past healthcare facts from
every ITU-D related countries
2. Data will be formalized to be a resilient
example database.
3. How we can set up anonymity is the
challenge at this stage.
Fig1. Phase1: Gathering Text Information
Raw data
Raw process
descriptions
in country A
Data Factory
Raw data
Raw process
descriptions
in country i
Raw data
Raw process
descriptions
in country n
Resilient Healthcare
data base
Data Tools
Data Tools for multi-kind Big Data
IoT censors data acquisition(censor data tools)
Scraping tools(in inter / intra)
Normalization for data schema
Anonymizetools
Legal(user agreement)
Data cleansing(noise reduction)
Training data tools
Video & imageprocessing
Fig2. Phase1: data tools
Three Phase approach
• Second phase
• Analyze explicit and implicit processes
in these descriptions, by manually and
semi automatically as possible.
Site Name 凡例
AAA hospital proces description
level1 Level3 Level3paramedic ①Process Design
②process
⑤information
③element
④medicine
G
e
n
e
r
a
l
a
f
f
a
i
r
P
a
t
i
e
n
t
f
i
e
l
d
a
t
t
n
d
c
l
e
a
r
k
H
R
N
u
e
s
e
h
o
s
p
i
t
a
l
D
o
c
t
o
r
start/stop process description connectorSys 外部IF
案件情報送付
情報受領 ゙案件情報送付
Fi le Server
データダウンロード
面談調整
メール
プリントアウト
情報受領
指示書
検討 面談設定
案件情報確認 面談参加
面談実施
Fig4. Phase2: Process Description
Video cameras
in
in fields j
Video cameras
in
field m
Video in fields (airports, hospitals, etc.)
(Existing security cameras can be used)
Data Factory
Resilient Healthcare
data base
Data Tools
(Video&image)
On line
Off line
Fig3. Phase1: Video in the fields
Three Phase approach
• Third phase
1. Introduce statistical analysis, process mining
utilizing machine learning such as temporal
deep learning technologies to analyze both
explicit and hidden processes in the
database.
2. Find what aspects will be the key feature for
resilient healthcare
3. Automatic prediction and unexpected
process discoveries.
Deep Learning
Training
(temporal neural
Network
analysis)
Detect and Prediction
• Face and Face expression
analysis
(Gaze, Paralysis, Fever)
• Gaits, Posture
Symptom of disease
Paralysis
Depression
Lack of sleep, etc.
• Analyze videos
and predict
situations in
pandemic and
paramedic cases
• Support Safty-1
and Safty-2 in
resilient healthcare
Data Factory
Data
Tools
Fig5. Phase3: AI processing for video data
Resilient
Healthcare
Videos
Learning
Factory
Medical application1 : Rehabilitation
• Take videos for patients and analyze the status• Manage past videos then propose and how to do rehabilitation• This can be applied to sport rehabilitation too
Analyze a person who is parallelized left hand side
Discussions and further issues
• How we can find hidden processes in resilient operations
• Multi-disciplinary approach is needed
Process Mining
• Process Mining is still in more research phase than the other technologies (machine learning and deep learning).
• But through this proposed activity, many resilient processes are gathered and well manually described in phase 1 and 2
• this condition will give chances to apply deep learning technology (especially temporal deep learning technology, such as RNN, LSTM etc.
• to process mining, as generally deep leaning needs more trained data for its better learning.
Fig6. Process Mining in Resilient healthcare
Resilient Data Base
(Event logs
and
Existing Process
Descriptions)
Resilient
Processes
Discovery
Conformance
Enhancement
IT systems
(Electric Medical Record
systems,
Mail systems,
Accounting systems
Etc.
Real World
(Pandemic &
Paramedic)support
Requirements
And designModeling
And Analysis
Ref) Replace Routine works into office software robotsEx) monthly review support with RPA (Robotic Process Automation) :
• Every month, people have to create KPI graph from several fragments of
excels which are downloaded by un-flexible intra-web systems
• Replace routine operations into RPA (software robots)
19
Intra
systemsCSV
files
excel
Excel macro
Files by each
Org.
Cut&Paste
into
slides
reportsMgt
Meeting
With stakeholders
Improvement process
KPI
visualization
Back logs
Discussions and further issues
How we can find hidden processes
in resilient operations
Multi-disciplinary approach is
needed
1.Domain
Knowledge• Healthcare
• Data Source
• Semantics
of Data
• Legal
3.Data Science• Data Model
• Analytics
Framework
• Statistical model
• Machine Learning
• Deep Learning
4.Systems
Architecture• Building System
Architecture
• Rapid
Dashboard
5.Consultation for AI/IoT & Data Science
Fig7: Multi-Disciplinary Business Architecture
2.Resilient
Engineering• Process
description
• Safty-1
• Safty2
Conclusion
We proposed following phases
• Phase1: Gathering process data to analyse resilient
healthcare in daily activities.
• Phase2: Describe both successful processes and failure
processes
• Phase3: Analyze described processes, discover hidden
processes, and predict critical processes
According to these phases, we will be able to support resilient
healthcare to be more systematic and semi-automatic manner.
We would like to invite ITU-D related organizations to join this
proposed activity.
22
Appendix
23
• Take videos for paralyzed elder patients and support
• Quantize the paralysis by wireframe
• Feedback them to doctors, patients and family
Wireframe model analysis to video
High level
analysis
Feed back to
Patients,
elders
High level analysis is subscribed as patent from OrangeTechLab
Application 3: support elders
Visualize paralysis by wireframe
Gaits Analytics Systems Image
deep
learning
a)Gaitsraw data /
footage
b)Traineddata
Data Factoryc) Training byFast AI machines
Trainednetwork
d) Create compacttrained network(distillation)e)Use compact inference engine smart phones
e) re-enforcement learningBy capturing from smart phones
f) Big data for application domainsEx appl. domains) Medical, sports, etc.
Inference on
Smart phones
- Subscription biz models for several gaits, posture, action analysis -