JANUARY/FEBRUARY 2010 190278-6648/10/$26.00 © 2010 IEEE

R ecent advances in the electronics industry and

wireless communication have enabled the evo-

lution of innovative application domains. Small-

er embedded processors and systems have allowed

a new level of mobile communication and interac-

tion in everyday life. In particular, the expansion of

broadband wireless services and the advancement

of handheld technology have allowed for real-time

patient monitoring in locations where not previous-

ly possible. Low-cost sensors and wireless systems

can now create a constantly vigilant and pervasive

monitoring capability at home, work, and in con-

ventional point-of-care environments (e.g., prima-

ry care physician offices, outpatient clinics, and

rehabilitation centers). A large research community

(e.g., the UCLA Wireless Health Institute) and a na-

scent industry is beginning to connect medical care

with technology developers, vendors of wireless and

sensing hardware systems, network service provid-

ers, and enterprise data management communities.

Wearable devices focusing on personal health, reha-

bilitation, and early disease detection are now being

prototyped. All of this has led to the new notion of

“wireless healthcare.”

A variety of applications lie within the wireless

healthcare category. Initially, wireless healthcare

(known previously as telehealth) mainly referred to

a remote consultation of physicians located in differ-

ent geographical locations for diagnosis and advice

on treatment. Later, with advances made in robotics Digital Object Identifier 10.1109/MPOT.2009.934698

© COMSTOCK

Providing a cushion for wireless healthcare application development

Yu Hu, Adam Stoelting, Yi-Tao Wang, Yi Zou, and Majid Sarrafzadeh

Authorized licensed use limited to: Yu Hu. Downloaded on January 18, 2010 at 23:51 from IEEE Xplore. Restrictions apply.

20 IEEE POTENTIALS

and high-speed communication, telesur-

gery applications emerged where a sur-

geon performs surgery on a patient when

the two are not physically in the same

location. Now, all aspects of healthcare

including physiological signal monitor-

ing, diagnosis, rehabilitation, treatment,

and surgical procedures utilize advanced

technologies and are generally referred

to as wireless healthcare. Telehealth sys-

tems are not only used to make health-

care applications available in remote

areas, such as homes, schools, nursing

homes, and military camps, but also in

ubiquitous infrastructures that can im-

prove the quality of overall healthcare.

Have a PoSeatTo demonstrate the system design and

the implementation of a typical wireless

healthcare application, we present the de-

velopment of the PoSeat, a smart seat

cushion for chronic back pain prevention.

As reported by CNN, back pain affects

80% of Americans at some time in their

lives, and it is the single most expensive

problem for working adults: 31% of all

workers compensation dollars are for

lower back pain, as reported by Herman

Miller Inc. It has been shown that

im proper seated posture is one of the

major causes of chronic back pain. The

factors that increase back pain risk include

the seated posture, the duration, and the

vibration strength of the seat. A low-cost

and portable system that automatically

detects these factors can help prevent

chronic back pain. For instance, PoSeat

can monitor a taxi driver’s seat to calcu-

late the accumulated vibration strength

and warn when he or she is at risk of suf-

fering back problems.

The functionality of our PoSeat system

is illustrated in Fig. 1. When the PoSeat

cushion is attached to any chair, it will

automatically collect various signals de-

scribing the user’s sitting behavior (i.e.,

postures and duration) and the ambient

environment (i.e., vibration strength).

These signals are sent to a portable

device (a cell phone, a PDA, or a net-

book) periodically. The portable device

analyzes these signals and sends the

user a warning in real time if an inap-

propriate posture (e.g., leaning left

for too long) is detected. In addition,

the portable device will automatically

synchronize with a remote database to

upload the user’s seated posture histo-

ry. An up-to-date summary of the user’s

seated posture is provided through an

online service. There, users can also see

a more comprehensive annual heath as-

sessment as reported by a personal

physician based on the historical record

from the database.

System designAs shown in Fig. 2, the system can

be divided into three parts: (a) on-cush-

ion circuitries, (b) the software in the

mobile device, and (c) the remote server

and database.

On-cushion circuitry designThe on-cushion circuitries are elec-

tronics attached inside the seat cushion

for signal collection and preprocessing.

An accelerometer is used to estimate the

vibration strength, and multiple pressure

sensors are used to detect the seated

posture of the user. The analog signals

gathered by these two types of sen-

sors are collected by a microprocessor

through data acquisition circuitry and an

analog-digital-converter (ADC). A timer

is set in the microprocessor in order to

periodically send out the sensor data.

The transmission of the data from on-

cushion circuitry to the mobile device is

completed by a Bluetooth module, which

communicates with the microprocessor

via a serial port.

Fig. 3 shows the implementation of

the on-cushion circuitry in Fig. 2. For the

rapid prototyping and proof-of-concept,

we assembled various off-the-shelf parts.

We used the embedded development kit

(including three PCB boards) from Cross-

bow Technology and six pressure sensors

from TecScan Systems. The on-cushion

circuitries are powered by two AA batter-

ies and, based on our experiments, can

last for a week in our system without any

power management schemes.

The placement of the six pressure

sensors is also shown in Fig. 3: four sen-

sors are placed in the back and two are

placed in the bottom. The accelerometer

is put in the back. The leaning angle of

the seat back can be analyzed based on

the readings from the four back sensors

and the accelerometer. In addition, the

Key Features of the PoSeat:

• Portable: deployable to various

types of chairs

• Cheaper: compared to buy an

advanced chair

• Easier to deploy: plug and play

• More intelligent: smart posture

detection

• More convenient: communication

with your cell phone

Health AssessmentService

Fig. 1 An overview of the PoSeat system.

On-Cushion

Circuitries

Accelerometer

Microprocessor

ADC

SerialController

BluetoothModule

Bluetooth

Connection

(a) (b) (c)

AnalysisProgram

Web ServiceClient

DataAccquisition

PressureSensors

Web ServiceServer

InternetMobile Device

Fig. 2 System design.

Authorized licensed use limited to: Yu Hu. Downloaded on January 18, 2010 at 23:51 from IEEE Xplore. Restrictions apply.

JANUARY/FEBRUARY 2010 21

vibration strength is analyzed based on

the frequency of the change of the ac-

celerometer reading. Note that the nu-

mber of pressure sensors used in our

system is based on multiple factors. First,

the maximum number of ADC channels

available for the microcontroller (MICA2)

is eight, and six pressure sensors and one

accelerometer take seven ADC channels.

This means that the addition of extra

pairs of pressure sensors will require the

addition of at least one microcontroller.

Multiple microcontrollers increase the

size and cost of the overall system as

well as the complexity of the system

design due to the necessary communica-

tion among them.

Second, our posture detection algo-

rithm is based on a support vector

machine. Increasing the number of pres-

sure sensors will increase the dimen-

sions (and therefore, the complexity) of

the learning model, which will in turn

increase the time for the model building

and may also actually decrease the accu-

racy of the posture analysis. We experi-

mented with multiple placement schemes

using up to 16 pressure sensors, but the

results showed that the six pressure sen-

sors were sufficient to achieve an ideal

accuracy using our posture analysis algo-

rithm. Intuitively, the placement of these

pressure sensors captures the most sen-

sitive observation points for posture

detection, and therefore filters out the

noise introduced by extra sensors.

The description of parts and the cost

of goods are shown in Table 1. The total

price for prototyping the on-cushion cir-

cuitries is US$460. In comparison, a

commercially available body pressure

measurement system (BPMS) developed

by TecScan Systems has an array of 42

by 48 pressure sensors, and it is sold for

over US$10,000.

The microprocessor (i.e., MICA2)

runs TinyOS, an embedded operating

system. Through TinyOS, we can easily

set the sampling rate (i.e., period to send

out a set of sensor readings) to 5 s. The

data sent via Bluetooth are seven two-

byte words, including 2 B for each pres-

sure sensor reading and 2 B for the

accelerometer reading. Note that smart

power management can be easily imple-

mented by dynamically changing this

sample rate based on the user’s

historical data while still maintaining

acceptable accuracy.

Mobile application developmentAs shown in Fig. 2, the mobile device

part of our system consists of two compo-

nents: a data analysis system and a client

program connecting the mobile device to

a remote database. Since the mobile

device communicates to the on-cushion

circuitries via Bluetooth, any handsets

(iPhone, Google G1, or PDA) with Blue-

tooth and Internet (Wifi, 3G, or EDGE)

access can be used in our system. In our

prototype implementation, a Windows

Mobile-powered HP PDA was chosen due

to the user-friendly development platform

provided by Microsoft Visual Studio 2005.

To analyze the seated posture of a

user, we developed a machine learning-

based classifier. To cope with the variety

of the decision-making parameters (the

body weight, the body height, and pos-

ture behaviors) of individual users, we

featured a user-dependent learning ap -

proach. Before using the system for the

first time, a user needs to perform a train-

ing process that takes two minutes to

complete. The user will be prompted to

sit in a few different postures for a certain

amount of time to collect training data

(see Fig. 4). After the training data is

Table 1. Description of parts and the cost of the goods for prototyping.

Part Name Description Price

Crossbow MTS310B Light, temperature, acoustic, sounder, dual-axis accelerometer and dual-axis mag sensor board

US$148

Crossbow MDA100CB Light, temperature, prototype area sensor/DAQ board

US$50

Crossbow MPR410 Atmel ATMega 128L Processor (430 MHz) and Atmel AT45DB41B nonvolatile memory

(512 KB), ADC, and serial port

US$79

Roving networks RN-24 Bluetooth module US$79

Tecscan sensors Six pressure sensors US$94

Seat cover A seat cushion US$10

Total price US$460

Smart power management can be easily implemented by dynamically changing this sample rate based on the user’s historical data while still maintaining acceptable accuracy.

Fle

xiF

orc

e A

201 S

ensor

Pre

ssure

Sensor

Pressure Sensors

Sensor Board: MTS310

Data Acquisition Board: MDA100

MICA2 + ADC + Serial Port

Bluetooth: RN-24

Fig. 3 Hardware assembly.

Authorized licensed use limited to: Yu Hu. Downloaded on January 18, 2010 at 23:51 from IEEE Xplore. Restrictions apply.

22 IEEE POTENTIALS

collected, our machine learning program

running in the mobile device will auto-

matically build a model based on support

vector machine (SVM), one of the most

popular supervised learning schemes.

The key parameters in this model are the

readings collected from the pressure sen-

sors and the accelerometer. The model

for a specific user is then stored in a file

(1 KB space is needed) in the mobile

device and can be loaded for future use.

After the SVM model is built, the pos-

ture detector works as follows. In order

to obtain a relatively accurate analysis on

the seated posture in a three-dimensional

space, we employ a hybrid classifier, as

shown in Fig. 5(c). Six classes can be

identified: leaning left, leaning right, lean-

ing back, leaning forward, sitting straight,

and no one sitting. Our classifier consists

of a decision tree for a coarse classifica-

tion of posture information on X-Z plane

and Y-Z plane. Each node of the decision

tree is a subclassifier based on the SVM

model. This subclassifier compares the

observation points against the SVM model

that we have built and returns a final

decision on a particular plane.

Based on the posture analysis results

obtained by the above detection algo-

rithm, a user’s posture history is com-

pared against posture history patterns

that are known to cause back problems

to identify potential risks (e.g., shifting

positions regularly or sitting in certain

postures for too long) that have been

predefined by a physician. The alarm

will be triggered if these problematic

posture patterns are found.

The mobile device will periodically

analyze the postures using the above

machine learning-based algorithm and

transmit stored posture readings to a Web

server when the mobile device is online.

If the mobile device is offline for long

periods of time, the device will limit the

size of stored readings by increasing the

granularity of each stored item. That is, it

will compute and store the average of

every n readings, where n increases the

longer the device is offline.

Remote server designThe communication between the mo-

bile device and the remote server uses

Web services, a communication frame-

work designed to support interoperable

machine-to-machine interaction over a

network. The communication between

the Web service server and client is

through Simple Object Access Protocol

(SOAP), which sends and receives XML

messages via HTTP. The server side

often provides a machine-readable de-

scription of the operations written in the

Web Services Description Language

(WSDL), and the client obtains the type

Fig. 4 The snapshot of the training program running in a PDA.

Z Z

X

X

Y

Y

65%

25%35%

75%

Left Right Upper Back Lower Back

SVM:X–Z

SVM:Y–ZReturn

Return Return

Return

StraightLean Back

Left/Right Straight No One Sitting

(a) (b) (c)

Fig. 5 A machine learning-based classifier for sitting posture analysis.

A user’s posture history is compared against posture history patterns that are known to cause back problems to identify potential risks that have been predefined by a physician.

Authorized licensed use limited to: Yu Hu. Downloaded on January 18, 2010 at 23:51 from IEEE Xplore. Restrictions apply.

JANUARY/FEBRUARY 2010 23

and function information by acquiring

the service description by the WSDL

from the server side.

In our development, we used Micro-

soft Visual Studio 2005 with ASP.NET

framework. The Microsoft .NET frame-

work provides Web services as an inte-

gral part of the architecture, making it a

simple platform to implement our proto-

type. A detailed documentation of the

.NET framework can be found in the

Microsoft Developer Network (MSDN).

The remote server for PoSeat can be

deployed on any Web server running

ASP.NET and Web services.

When the data from the mobile devices

is transmitted to a remote database

through the Web service, the server can

calculate statistics based on the quantita-

tive correlation between the sitting behav-

ior and various back pain problems. The

quantitative and qualitative results of the

analysis are stored separately and shown

to the user via a Web page. Fig. 6 shows

a screen snapshot of the Web page for the

health assessment of a particular user. It

gives the user an overall assessment and

analysis of his or her risk of suffering

from various back problems. In addition,

it compares the user to all other users in

the database in order to motivate him or

her to correct their problems.

Conclusion and future workWe have presented an infrastructure for

a typical wireless healthcare application—a

smart cushion for back pain prevention.

Many other interesting applications can

be developed based on similar frame-

works. For instance, the Nike�iPod sport

kit can be simply implemented by integrat-

ing the on-cushion circuitries into the insole

of the shoes. Due to the configurability of

the system, design issues such as power

and reliability can be optimized through

modifications of the data sampling rate,

communication frequency, and the analy-

sis algorithms running on the handset. In-

terested readers are referred to the Web

page of UCLA Wireless Health Institute at

www.wirelesshealth.ucla.edu.

Read more about it • UCLA Wireless Health Institute

(2009). [Online]. Available: http://www.

wirelesshealth.ucla.edu/

• F. Dabiri, T. Massey, H. Noshadi,

H. Hagopian, and M. Sarrafzadeh, “Light-

weight medical BodyNets,” in Proc. 2nd Int. Conf. Body Area Networks, Florence,

Italy, 2007, pp. 48–56.

• A. Vahdatpour, M. Sarrafzadeh,

W. Wu, L. Au, B. Jordan, T. Stathopou-

los, M. Batalin, W. Kaiser, M. Fang, and J.

Chodosh, “The SmartCane system: An as-

sistive device for geriatrics,” in Proc. 3rd Int. Conf. Body Area Networks (BodyNets), Tempe, Arizona, Mar. 2008, pp. 1–4.

• F. Dabiri, A. Vahdatpour, H. Noshadi,

H. Hagopian, and M. Sarrafzadeh, “Ubiqui-

tous personal assistive system for neuropa-

thy,” in Proc. 2nd Int. Workshop on Systems and Networking Support for Healthcare and Assisted Living Environments (Health-Net) and ACM MobiSys, Breckenridge, Col-

orado, July 2008, pp. 1–6.

• S. Rohman, C. Santarelli, J. K. Pol-

lard, M. E. Fry, A. Theodorou, and N.

Mohoboob, “Wireless and web-based

medical monitoring in the home,” Med. Inform. Internet Med., vol. 27, no. 3, pp.

217–219, 2002.

• D. Neild, J. T. Heatley, R. S. Kala-

wsky, and P. A. Bowman, “Sensor

networks for continuous health moni-

toring,” BT Technol. J., vol. 22, no. 3,

pp. 130–139, 2004.

• National Research Council. Mus-culoskeletal Disorders and the Work-place: Low Back and Upper Extremities. Washington, D.C.: National Academies

Press, 2001

• J. Hartvigsen, J. Hartvigsen, C. Le-

boeuf-Yde, S. Lings, and E.H. Corder., “Is

sitting-while-at-work associated with low

back pain? A systematic, critical literature

review,” Scand. J. Public Health, vol. 28,

no. 3, pp. 230–239, 2000.

• A. M. Lis, K. M. Black, H. Korn,

and M. Nordin, “Association between

sitting and occupational LBP,” Eur. Spine J., vol. 16, no. 2, pp. 283–298, 2007.

• N. Cristianini and J. Shawe-Taylor,

An Introduction to Support Vector Ma-chines and Other Kernel-Based Learning Methods. Cambridge, U. K.: Cambridge

Univ. Press, 2000.

About the authorsYu Hu (bryanhu@ece.ualberta.ca) is

an assistant professor in the Electrical

and Computer Engineering Department

at the University of Alberta, Canada. His

research interests include embedded and

reconfigurable computing.

Adam Stoelting (astoelting@ucla.edu)

is pursuing his M.S. in the Computer Sci-

ence Department at UCLA. His research

focuses on the development of wireless

healthcare applications.

Yi-Tao Wang (yitao@cs.ucla.edu) is

pursuing his Ph.D. in the Computer

Science Department at UCLA. His

research interests include the design,

development, and evaluation of sensor

networks.

Yi Zou (zouyi@cs.ucla.edu) is pursu-

ing his Ph.D. in the Computer Science

Department at UCLA. His research inter-

est is hardware acceleration of applica-

tion specific computation.

Majid Sarrafzadeh (majid@cs.ucla.

edu) is a professor in the Computer Sci-

ence Department at UCLA and codirec-

tor of the UCLA Wireless Health Institute.

His recent research interests are in the

areas of embedded and reconfigurable

computing, VLSI CAD, and design and

analysis of algorithms.

Fig. 6 Web page for the health assessment.

Authorized licensed use limited to: Yu Hu. Downloaded on January 18, 2010 at 23:51 from IEEE Xplore. Restrictions apply.