[PDF] - Outline Medical applications using sensor networks Design PDF Document

SLIDE 1

Robust Wireless Clinical Monitoring for Ambulatory Patients

Octav Chipara

Outline

Medical applications using sensor networks
Design requirements for wireless clinical monitoring
Existing approach to wireless clinical monitoring
Initial prototype
4 - bit link estimator
CTP
Impact of mobility
Initial prototype revisited
Multi-user study

Medical applications using WSNs Assisted living - crowded space

behavior memory fall prevention

AlarmNet, UVA
Lace,Rochester
Smart Homes
Intel
Honeywell
Motorola
Philips
AlarmNet, UVA
Lace, Rochester

vitals

more projects at http://www.agingtech.org/techdemo.aspx

Smart homes Disaster recovery problem

Best practice:
use color-coded tags to indicate severity
constant reassessment of patients (5-15 mins)
verbal coordination between different organization (e.g., first

responders, hospitals, etc.)

Issues with current practice
limitations of colored tags
the tag may not accurately reflect a patients condition
unclear how to prioritize patients with same color
no room to leave notes
hard to coordinate activities between rescue workers and

hospitals, ambulances, etc.

SLIDE 2

Wireless clinical monitoring

Early detection is fundamental to reducing mortality and length of

stay in hospitals (e.g., in respiratory/cardiac arrests, septic shock)

Changes in vital signs may indicate clinical deterioration
automatic early detection systems are in the works
The accuracy and sensitivity of these systems depends on having

up-to-date vital sign information

Requirements

Designed for non-ICU settings
inexpensive
reliable
works for ambulatory patients
lifetime of 3-5 days (avg. hospitalization duration)
continuos monitoring once per minute
low maintenance cost
Integrate with electronic patient records to enable automatic early

detection records

Existing solutions

GE, Phillips,CISCO provide wireless medical telemetry systems and

have limited success in cardiology departments; However,

closed systems
high cost of infrastructure, relies on LAN
ad hoc deployment of nodes
high maintenance cost

what wireless technology should you pick?

Wireless devices

Device Passive RFID WLAN 802.15.4 Bluetooth (class 1) Bluetooth (class 3) Frequency 125 KHz 2.4-2.5 GHz 2.4 GHz 2.4 GHz 2.4 GHz Power 2-4W 100mW 1mW 100mW 1mW

Selecting a wireless technology

Criteria:
Low-power operation (3 - 5 day requirement, small packets)
Small bandwidth requirements
Minimize electro-magnetic interference
Degree of miniaturization (weight)
Inexpensive radios
Our choice: TelosB with 802.15.4

SLIDE 3

hardware

A wireless pulse-oximeter

system architecture

9

System components System components

Base-station node
has access to wired network
powered
integrates with electronic patient records
Relay nodes
deploy to ensure coverage
impractical to change batteries => powered
no access to wired network => reduce cost
Patient nodes
integrates sensing modalities with TelosB
battery operated

Wireless pulse-oximeter Relay node

10

System architecture System architecture

initial software prototype

SLIDE 4

Software prototype

Patient node:
OxylinkSensorC - device driver for pulse-oximeter
PacketLoggingC - logs received/transmitted packets
CollectionC - implements CTP
PatientAppC - implements the actual patient application
Base-station & relay nodes:
CollectionC
PacketLoggingC

4-bit link estimator

Link Quality Estimation

Identify good links
ETX: Expected Transmission Count

[Mobicom 2003]

TX ReTX ACK A B

1 ETX(L) = PRR(f) * PRR(b)

22

ETX Estimation Example

1.8 Beacons ETX Estimate (alpha = 0.8) 2.0 1.0 t1 t3 1.83 1.0 t2 3.0 2.04

State of the art

Not all information used
Coupled designs
Physical layer (LQI)
Coupled implementation

Network Layer

LE

24

Scope

Identify the information

different layers of the stack can provide

Define a narrow interface

between the layers and the link estimator

Describe an accurate and

efficient estimator implemented using the four bit interface

SLIDE 5

25

Layers and Information

Better estimator with information from different layers?
Physical Layer
Packet decoding quality
Link Layer
Packet Acknowledgements
Network Layer
Relative importance of links

Network Layer

LE

26

PHY Info Not Sufficient

Network Layer

LE

Network Layer

LE

Unacked PRR LQI

27

Physical Layer

Decoding Quality
Agile
Free
Asymmetric (receive) quality
Radio-specific
Examples
LQI, RSSI, SNR

Network Layer

LE

28

Link Layer

Outcome of unicast packet transmission
Higher quality links
Successful TX
Successful ACK reception
Example
EAR [Mobicom 2006]

A B DATA ACK

Network Layer

LE

29

SRC DST A

Network Layer

LE

30

Interface Details WHITE

Packets on this channel experience few errors

ACK

A packet transmission

n this link was

acknowledged

PIN

Keep this link in the table

COMPARE

Is this a useful link?

SLIDE 6

collection tree routing

Collection Tree Routing

Link dynamics: uses 4-bit link estimator
Control plane
Data plane

Control plane

Responsible for link quality estimation and route selection
ETX is used to path quality
uses both data and beacons for estimation
Route updates are disseminated in beacons
beacons transmitted on a variable timer
the beacon period is reset when an inconsistency is detected
the beacon period increased if route costs remain stable

Route selection

We strive to select the route with minimum ETX
switch routes only if its ETX is > 1 better than current route
it is dangerous to route on a path for which we have a little

statistical data

Loops are detected using feedback from the data plane

Data plane

Responsible for forwarding packets
Issues:
self-interference - a child’s transmission may interfere with its

parent’s transmission

duplicate suppression - result of lost ACKs
loop detection
ETX should monotonously decrease as a packet is forwarded

to the sink

long routes
duplicates

Is CTP suited for mobile users?

SLIDE 7

11

Impact of mobility on reliability Impact of mobility on reliability

Initial prototype used Collection Tree Protocol (CTP)
de facto data collection protocol in TinyOS
Reliability components
Setup:
1 volunteer
data rate: 1 reading/min

Data delivery to first relay Data delivery to the base station first-hop reliability relay reliability end-to-end reliability

Experimental setup

12

CTP under normal activity CTP under normal activity

Normal activity experiment showed
most packet losses occur on the first hop
packet drops tend to follow user movement
Hypothesis:

poor routing table management => use of outdated entries end-to-end reliability: 82.39% first-hop reliability: 85.38% relay reliability: 96.49%

13

DRAP DRAP

Approach - separate the routing problem in two parts
first-hop delivery: DRAP handles it
relay delivery: CTP handles it
Advantages of this approach:
reliable data collection from mobile users without changes to routing
reduces footprint and bandwidth overhead on patient nodes
Challenge: optimize for no mobility react quickly to mobility
no mobility => feedback from physical & link layer
mobility => n broken links

Efficient one-hop collection

14

DRAP’s reliability DRAP’s reliability

DRAP

end-to-end reliability: 99.33%

ne-hop reliability: 100%

relay reliability: 99.33%

CTP

end-to-end reliability: 82.39%

ne-hop reliability: 85.38%

relay reliability: 96.49%

SLIDE 8

15

Multi-user study Multi-user study

Seven health volunteers for 1 hour, normal activity
Reliability: 96.15% ~ 100%
Duty cycles:
0.2% -- no movement
2.0% -- frequent movement

44

References

For more details you may read:
O. Chipara, C. Brooks, S. Bhattacharya, C. Lu, R. Chamberlain, G.-
C. Roman, and T.C. Bailey, Reliable Data Collection from Mobile

Users for Real-Time Clinical Monitoring, Technical Report WUCSE-2008-25, Washington University.

The project described was supported by Award Number

UL1RR024992 from the National Center For Research Resources. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Center for Research Resources or the National Institutes of Health.