for Medical Systems What to get out of IoT and Process Data · for Medical Systems What to get out of IoT and Process Data Dr. Mirko Appel November 2016 Dr. Mirko Appel | HC SV CS

Unrestricted © Siemens Healthcare GmbH, 2016

Optimized Service Delivery for Medical Systems

What to get out of IoT and Process Data

Dr. Mirko Appel November 2016

Dr. Mirko Appel | HC SV CS SLM DGA AS

Page 2 | Unrestricted © Siemens Healthcare GmbH, 2016

Siemens Healthineers

2016-02-20Page 3 HC xxx – xxx



> €1 bnR&D spent

73countries

with direct presence

>209,000patients

every hour2

Biggestsupplier

of medtechinfrastructure

Worldmarketleader

in most businesses

> 45,000employees

1,474invention

disclosuresin 2015

> 70%of critical clinical

decisions are influenced by the

type of technology we provide1

~ €13 bnrevenue

Access for

1.08 bnpeople

in developing countries2

As your partner, we offer expertise and resources for your specific needs

1 AdvaMedDX, “A Policy Primer on Diagnostics”, June 2011, page 3

2 Siemens AG, “Sustainable healthcare strategy - Indicators in fiscal 2014”, page 3-4



Outcomes

Costs

Future

1956CLINISTIX − dry chemistry testing for glucose in urine

E.v. BehringW. C. Röntgen

1901 Nobel prize winners (Physics + Medicine)

1896 Industrially manufac-tured X-ray appliance for medical diagnostics

2011First integrated, simultaneous whole-body MRI and PET system

1998First Siemens track-based laboratory automation system

1967First real-time ultrasound scanner

1975First Siemens CT scanner

1983First Siemens MRI scanner

2008Robotic-assisted angiography system

2009Multi-modality 3D imaging network

2008 Digital radio-graphysystem with wireless flat-panel detector

2012Wireless transducers for ultrasound

2014“Free breathing”CT scanning with powerful dual X-ray sources and two detectors

Molecular DX

Digital Health Services

Enterprise Services

Advanced Therapies

1999First intuitive medical IT platform from Siemens

2001First PET/CT system from Siemens

2006Diagnostic analyzer integrating four tech-nologiesin one system

2005First Dual Source CT scanner

1957Fully automated discrete chemistry analyzer for whole blood or serum

1982First acridiniumester based chemilumin-escenceimmuno-assays

2015Wide-angle image acquisition breast tomosynthesis– Mammomat® Inspiration

2015First Twin RoboticX-ray scanner for enhanced patientcare and productivity

Built on over 120 years of dedication to innovation,we have a long history in enabling healthcare providers …

2016-02-20Page 5 HC xxx – xxx



Ultrasound

We enable real-time access to decision-critical information

Advanced Therapies

We enable advanced therapeutic procedures

• Cardiology• Interventional Radiology• Radiation Oncology• Surgery

*Incubated within Business Function Strategy & InnovationImage courtesy Diagnostic Imaging: CMRR, Minneapolis, MGH, BostonImage courtesy Advanced Therapies: IHU Strasbourg, France

... offering the broadest and deepest portfolio

Diagnostic Imaging

We help achieve highest diagnostic quality and efficiency

• Computed Tomography• Magnetic Resonance• Molecular Imaging• Radiography & Fluoroscopy• Imaging IT

Laboratory Diagnostics

We enable clinical and workflow excellence in the lab

• Chemistry, Automation & Immunoassay• Hemostasis, Hematology

& Specialty Business

• Molecular Diagnostics*

Point of Care

We provide critical patient information in-office and at the bedside

Services

We help achieve best institutional performance

• Customer Services• Digital Health Services• Enterprise Services & Solutions

Outcomes

Costs • Blood Gas• Diabetes• Urinalysis

• Cardiology• Radiology• Obstetrics & Gynecology



Siemens Healthineers Customer Care Program

The products/features and/or service offerings (here mentioned) are not commercially available in all countries and/or for all modalities. If the services are not marketed in countries due to regulatory or other reasons, the service offering cannot be guaranteed. Please contact your local Siemens organization for further details.

Management Services

Consulting and partnership for economically sustainable healthcare delivery business

User Services

Enabling users with expertise and efficiency in the long run

IT Services

Delivering comprehensive and professional IT services to optimally run healthcare delivery

UserServices

ManagementServices

ITServices

TechnischeServicesSystemServices

Siemens Remote Service

LifeNet

System Services

Proactively ensuring that medical systems operate at peak performance

Siemens

Integrated

Service

Management

Siemens

Utilization

Management

Education

Management

Check

End-to-end

Automation

Workflow

Solutions

Business Management

Healthcare IT Care

Education

Product /

Clinical

Training

Remote

Application

Services

Optimize

CARECRADLE

Efficiency

Consulting

System Care and Repair

Siemens

Shared

Services

Siemens

Protect

Plans

Siemens

Guardian

Program

Evolve

Program

Guardian

Program

incl.

TubeGuard

Guardian

Program

incl.

ImageGuard

Siemens IT Care PlanSiemens Performance

Plans

Evolve

Program

Education Plans PEP Connect

Admin

Plus



Together with SAS, we built up technology, processes and organization to implement business analytics

Service Process

Data

Data Storage and Aggregation

Consistent Analytical Toolset

Data Sources

Business AnalyticsPlatform

Grow & safeguard service business Increase operational efficiency Improve decision speed and quality

What will happen next? How

will trends continue? What

scenarios are likely?

Example: Prediction of

system failures

Why is it happening? What

are conclusions from

associations, clusters,

trends?

Example: Commonly

exchanged service parts

Where are patterns,

anomalies, or dependencies

in the data?

Example: Peaks of error

patterns across fleet

How do we do things

better? What is the best

decision for a complex

problem?

Example: Smarter exchange

of service parts

Business AnalyticsCompetence Groups & Key Users

Service Business Processes

InstalledBaseData

LogisticsData …

Installed Base

Data Mining Statistical AnalysisPredictive Modeling & Forecasting

Optimization

Data Exploration & Integration

DI S

tud

ioV

isu

al A

nal

ytic

s

Data Analysis

Ente

rpri

se G

uid

eEn

terp

rise

Min

er

Results Presentation & Distribution

Vis

ual

An

alyt

ics

Web

Rep

ort

Stu

dio

http://www.google.de/url?source=imgres&ct=tbn&q=http://www.safenet-inc.com/uploadedImages/Partners/logos/Teradata.jpg?n%3D9733&sa=X&ei=gw9TVeD2MMqLsgGjjoG4BA&ved=0CAUQ8wc&usg=AFQjCNHkg2nNDoXE73UKVrvVvUMXMAM06A

http://www.google.de/url?source=imgres&ct=tbn&q=http://www.safenet-inc.com/uploadedImages/Partners/logos/Teradata.jpg?n%3D9733&sa=X&ei=gw9TVeD2MMqLsgGjjoG4BA&ved=0CAUQ8wc&usg=AFQjCNHkg2nNDoXE73UKVrvVvUMXMAM06A

http://www.google.de/url?source=imgres&ct=tbn&q=http://airpair-blog.s3.amazonaws.com/wp-content/uploads/2014/03/hadoop-logo-Patrick-Lie-Hadoop-Expert-and-big-data.png&sa=X&ei=zw9TVeuyIImXsgGMkICwBA&ved=0CAUQ8wc&usg=AFQjCNG4CxuX7DkSSEm8HqV2DInr3td9lQ

http://www.google.de/url?source=imgres&ct=tbn&q=http://airpair-blog.s3.amazonaws.com/wp-content/uploads/2014/03/hadoop-logo-Patrick-Lie-Hadoop-Expert-and-big-data.png&sa=X&ei=zw9TVeuyIImXsgGMkICwBA&ved=0CAUQ8wc&usg=AFQjCNG4CxuX7DkSSEm8HqV2DInr3td9lQ

http://www.google.de/url?source=imgres&ct=tbn&q=http://cliparts.co/cliparts/di4/okn/di4okn95T.png&sa=X&ei=ghFTVejFA8SdsAHNxYC4BA&ved=0CAUQ8wc&usg=AFQjCNFb8eyUpFfndhtnp9ZNJUBQHHIRMg






132 countries

in

different parts on stock in Memphis

40,000over

35,000different parts on stock in Frankfurt

over

10,000different parts on stock in Singapore

over

14,000different parts on stock in Brussel

over

300customers are supplied with service parts

over

Service part delivery within 24 hours for about 98% of all orders.

Global availability of service parts



Service Manager monitors

the process and analyzes

the performance and

efficency

Monitoring & optimization

Returned parts aresent back for analysis in factory and previous service orders are analyzed for unused parts

Return appraisal

System is up and running

again after Customer

Service Engineer replaces

several parts at once

Fix on site

Service Operation

Supporter orders service

parts based on available

information & experience

Clarification and parts ordering

Evidence-based service parts usage –reduction of service parts tourism & consumption

1) NDF: No defect found | 2) TS: Troubleshooting

Return appraisal and troubleshooting data:

Part A: NDF1) 7% TS2): 19%Part B: NDF 52% TS: 43%Part C: NDF 20% TS: 4%

System is down & Customer calls. System sends error logs; symptoms often unclear

System down

Did I order the right service parts

Did I replace the right service parts

Is my process running efficiently

Data Driven Solution Approach

Collect, aggregate, and evaluate return appraisal and

troubleshooting information and provide it across

service delivery chain

Analyze parts consumption patterns and provide to PLM

Business Impact due to smarter part

exchange – most parts are included in service

contracts!



Service Operation Supporter

Did I order the right service parts?

Reconsider ordering of service parts with high NDF1) & TS2) rate

Customer Service Engineer

Did I replace the right service parts?

Exchange first service parts with lower NDF rate in case multiple parts have to be exchanged

Evidence-based service parts usageAnalytical results integration in Visual Analytics & SAP

Service Manager

Is my process running efficiently?

Analyze tree map by identifying service parts with high NDF-rate and high consumption




Customer Service Engineer receives decision support on mobile front end on-site

Evidence-based service part exchange

Customer Service Engineers (CSEs) receive NDF1) and TS2)

rates for selected service parts

Based on the provided statistical information CSEs decides which parts should be ordered and which parts should really be replaced




Dashboards enable Service Managers to optimize service process efficiency

Process performance optimization

Multiple dashboards provide in-depth information and role-specific views on service parts statistics.

Data is updated on a daily basis and enables the service managers to take corrective actions in a timely manner



Dashboards enable Service Managers to optimize service process through evidence-based data

SERVICE PART X20058%3000€

MATERIAL TEXT:QUANTITY_CONSUMEDNDF_RATE_WORLD[%]SERVICE_PARTS_PRICE



Dashboards enable Service Managers to optimize service process through country benchmarking



Predictive Service – Provide decision support to service operations process

5 to 10 TB of Historical Data Service Experience

Pre

cisi

on

Capture Rate

LOG

Files

Service

Data

Component failure coming up?

Predictive Events and Decision Support



Business Analytics Platform

Predictive ServiceHow do we build a predictive model?

day1 2 3 4 5 6 7 8 9

Data mining and

modeling for

specific system

component

Identify common failure patterns

Prediction model

applied in daily

scoring process

Model containing rules topredict future failures

Scan Start CTCT_IBY_349 ==> ABC_2999Scan Abort CT

HV_Drops > 4 ==> CT_DEF_3065

Scan Start CTXYZ_3028 ==> CT_XYZ_3044Scan Abort CT



2016-06-29 07:23:16 **_SCU921 LOG:

cran_a 0.0, rao_a 0.0, sid_a

2016-06-29 07:23:19 **_SCU924 LOG:

origin tbl, A 0.00, B 0.00

2016-06-29 07:23:19 **_SCU825 LOG:

Movemon

2016-06-29 07:24:02 **_ACU1158

LOG: fluoro footswitch pressed

2016-06-29 07:24:09 **_ACU1162

LOG, VID async param change

2016-06-29 07:25:45 **_ACU1153

LOG, Xray end of xray

2016-06-29 07:26:06 **_SYC38(Log)

Internal information

2016-06-29 07:27:11 **_KRC0Robotic

Stand Developer Info

2016-06-29 07:27:15 **_ANG255ANG

diagnostic info

Medical Systems

Worldwide base of installed systems provides technical data for service purposes

Secure connectivity through certified remote service infrastructure

System Log Data

System event-logs files, containing many different kinds of events, 20.000 to 100.000 lines per day and system

Predictive ServiceData assets

Service Notifications

Service process-data, telling us when system failures occur and which parts have been replaced



• We have built a feature-vector for every day, containing 7.000 attributes from the raw data.

• Good separation between healthy and faulty systems on the very day a failure occurs

• Separation becomes tougher the earlier we try to predict failure

Predictive ServiceVisualization of the logfile-feature-space

Day o

f system failu

re3

0 d

ays befo

re system failu

re



Predictive Service – Methodology 1Classical Data-Mining on the logfile-feature-space

• After visualization, we build models using classical data-mining methods such as logistic regression, decision tree or support vector machines.

• Models show good performance on common KPIs, e. g. lift, ROC/ AUC, etc.!

• BUT: performance requirements in service delivery process are more demanding:

• False alarms are very bad and therefore cause high penalties.

• We need alarms in the right time frame (several days before the failure), not too late and not too early.

• Thus: Applicability is limited, works for selected components and needs manual tuning

Capture Rate: How many failures do we predict?

Precision: How many alarms are correct?t



Predictive Service Methodology 2 – Prediction models based on system physics and expert knowledge

Example Use Case

• Magnetic resonance imaging systems require a coldhead, which is cooled by liquid helium.

• Coldhead failures usually result in the helium to “boil off”.

• To prevent this, “Magnet Monitor” provides predictions of cold head failures on a daily basis.

• Impact:

downtime reduction no helium loss

less effort for troubleshooting and cold head exchange



Predictive Service Methodology 2Model Generation & Hypothesis Verification

Model Generation

• Expert input provides thresholds for sensor-signals

• “Blocking conditions“ are defined to filter the original alarms in order to reduce false alarm rate

• Initial validation of generated alarms on historical data (hypothesis verification)

• Secondary validation of generated alarms on live data, while not actually using the alarms in service process (pilot phase)

• Final model captures more than 90% of cold head failures and generates no false alarms

Decreasing helium pressure in the compressor is one alert trigger



Predictive Service Methodology 2Complete process chain

Generate Reporting

import service datamatch alerts toservice data

compute report tables

send to Visual Analytics

Model Scoring

import sensordata

data cleaningfix missing data

data aggregation;generate ABT

Calculate blocking conditions

send alerts toservice center & reporting

calculateAlerts (Expert Parameters)

evaluateperformance



Initial predictive service applications for selected components online

Magnetic Resonance Tomography:

• Monitoring and failure prediction of “Cold Heads“ ensuring helium cooling

• Extends average coldhead lifetime in system –from time based to condition based

• Avoidance of helium loss and express service delivery

Angiography Systems:

• Monitoring & failure prediction of selected x-ray tubes

• Patterns identify critical tubes, experts review based on additional, not yet modelled parameters

• Tubes covered by service contract are replaced within one week in average

• Reduction of failure probability during intervention

Siemens Healthineers

Hospital/ Laboratory

!Analytics Infrastructure Scoring of prediction models against

log file data

Siemens

Remote

Service

Infrastructure

Notification toservice center in caseof model hit



Dr. Mirko AppelHead of Analytical Services

Services, Customer Services

Hartmannstr. 1691054 Erlangen

Germany

Phone: +49 (9131) 84-8257Mobile: +49 (1522) 2786903E-Mail: [email protected]

Contact

Engineering success. Pioneering healthcare.

Unrestricted © Siemens Healthcare GmbH, 2016

Now’s our time

to inspirethe future

of healthcare together

for Medical Systems What to get out of IoT and Process Data · for Medical Systems What to get out of IoT and Process Data Dr. Mirko Appel November 2016 Dr. Mirko Appel | HC SV CS

Documents