BACKTRAINER

Computer-aided Therapy System with Augmented Feedback for the Lower Back

Dominique Brodbeck, Markus Degen, Michael Stanimirov

School of Life Sciences, University of Applied Sciences Northwestern Switzerland, Muttenz, Switzerland

Jan Kool

, Mandy Scheermesser

, Peter Oesch

, Cornelia Neuhaus

Department of Health, Zurich University of Applied Sciences, Winterthur, Switzerland

Clinic Valens, Center for rehabilitation, Valens, Switzerland

University Children’s Hospital UKBB Basel, Basel, Switzerland

Keywords:

Back pain, Physiotherapy, Computer-aided therapy, Augmented feedback, Ambient visualization.

Abstract:

Low back pain is an important problem in industrialized countries. Two key factors limit the effectiveness of

physiotherapy: low compliance of patients with repetitive movement exercises, and inadequate awareness of

patients of their own posture. The Backtrainer system addresses these problems by real-time monitoring of

the spine position, by providing a framework for most common physiotherapy exercises for the low back, and

by providing feedback to patients in a motivating way. A minimal sensor conﬁguration was identiﬁed as two

inertial sensors that measure the orientation of the lower back at two points with three degrees of freedom.

The software was designed as a ﬂexible platform to experiment with different hardware, and with various

feedback modalities. Basic exercises for two types of movements are provided: mobilizing and stabilizing.

We developed visual feedback - abstract as well as in the form of a virtual reality game - and complemented

the on-screen graphics with an ambient feedback device. The system was evaluated during ﬁve weeks in a

rehabilitation clinic with 26 patients and 15 physiotherapists. Subjective satisfaction of subjects was good,

and we interpret the results as encouraging indication for the adoption of such a therapy support system by

both patients and therapists.

1 INTRODUCTION

Low back pain (LBP) is a very frequent condition in

industrialized countries leading to high burden to the

health care system (van Tulder and Koes, 2002; van

Tulder and Koes, 1995). In the majority of cases,

no patho-anatomical causes for the complaints are

present and they are classiﬁed as non-speciﬁc LBP

(Deyo and Weinstein, 2001). Active and supervised

movement exercises are effective in reducing pain and

restoring function in non-speciﬁc LBP (Abenhaim L.,

e. a. 2000; Hayden et al., 2005). However, there

are several factors that limit the effectiveness of such

exercises and lead to poor therapy outcomes:

• proprioception is inadequate for the lumbar spine

• movement exercises are difﬁcult to learn

• exercises require a lot of repetition to be effective

• appropriate feedback requires continuous pres-

ence of a physical therapist

Thus, the main problems are insufﬁcient patient

motivation to comply with exercise regimes, as well

as the inability of the patients to exercise indepen-

dently. The aim of the Backtrainer project is to ad-

dress these limitations of current conservative ther-

apy, by automatically monitoring movement exercises

in real time, generating a motivating, game-like visual

feedback, and storing patients’ performance data for

later assessment.

The system has to be easy to use and simple

enough, in order for it to be adopted by physical ther-

apists, and so that it can be used by patients at home.

Although monitoring of back movements and pos-

tures in laboratory settings, but also at the workplace,

has gained a lot of attention in the past, there are no

technical solutions available that would fulﬁll these

requirements.

This leads to the following research questions:

• What is a minimal sensor conﬁguration that still

produces enough data, in order to generate valid

feedback for movement exercises?

Brodbeck D., Degen M., Stanimirov M., Kool J., Scheermesser M., Oesch P. and Neuhaus C. (2009).

BACKTRAINER - Computer-aided Therapy System with Augmented Feedback for the Lower Back.

In Proceedings of the International Conference on Health Informatics, pages 66-73

DOI: 10.5220/0001550400660073

 SciTePress

• What basic set of exercises needs to be supported

by the software to cover most common therapy

needs?

• In what form can the software provide intuitive

feedback about posture to focus patients’ attention

on the aspects that are relevant for the task?

• How can the software motivate patients to use the

system?

The problem was approached in three steps: (i)

identify the minimal sensor conﬁguration, (ii) build a

prototype system, and (iii) evaluate the system in a

controlled clinical setting.

2 BACKGROUND

There have been numerous research projects address-

ing the measurement of the kinematics of the spine.

High-end systems usually use optical sensors, with

either passive or active markers that are glued onto

the skin. These systems have a very high precision,

often in the sub-millimeter range, and are used by

the video-game and movie industry for motion cap-

turing and animation generation. Other measurement

setups consist of sensors based on ultrasonic waves

to measure distances, electro magnetic tracking sys-

tems (Van Herp et al., 2000; Jordan, 2001), resistance

strain gauges, optical ﬁbres (Dunne et al., 2006), or

sensors that use a combination of accelerometer, com-

pass, and gyroscope to determine the orientation of

the sensor (Lee and Laprade, 2003).

Each sensor technology has its advantages but also

its drawbacks. Optical systems can often capture only

in a small area, and there must be an unobstructed

line of sight from the camera to the markers. Systems

based on electromagnetism can be inﬂuenced by the

environment (i.e. training machines built of iron or

steel).

Research has also been done to compare the ac-

curacy of skin mounted (glued) sensors with radio-

graphs or magnetic resonance imaging used to iden-

tify the actual positions of vertebrae (Yang et al.,

2005; Mrl and Blickhan, 2006). The results show that

positions and motions of the skin markers can be used

as an estimate for the calculation of the position and

orientation of the underlying vertebrae. It should also

be noted that the goal of our system is not diagno-

sis, but to support physiotherapists whose work is also

based on surface observations.

Some recent publications focus on camera based

systems. In (Engels and Leloup, 2006) a system based

on low cost cameras has been described. This system

works with infrared cameras and a headband, with

mounted reﬂectors that can be used to recognize and

identify different sitting postures. Although the low

cost camera approach seems promising, it is not ap-

plicable in our situation because of the line of sight

problem, which applies to other optical systems as

well.

The therapy system that was proposed in (Sucar

et al., 2008) also uses cameras to track gestures from

stroke patients and, in addition, provides augmented

feedback in a game-like setup. Another promising

approach, using multi-modal feedback in neural re-

habilitation, is described in (Huang et al., 2005). This

system uses visual as well as auditory means for feed-

back on functional, task-oriented exercises.

3 IDENTIFICATION OF SENSOR

CONFIGURATION

In order to satisfy the requirement of a simple and

easy to handle system, we had to identify the mini-

mal conﬁguration of sensors that would still produce

enough data in order to classify movement quality and

performance, and to generate valid feedback. The hy-

pothesis was that it is sufﬁcient to measure the ori-

entation of the lumbar spine at two points with three

degrees of freedom.

We used an optical motion capturing system (Op-

totrak Certus from NDI) with 22 infrared LEDs posi-

tioned on the subjects’ back (see Figure 1). The pre-

cision of this system is in the sub-millimeter range,

and data was collected with a sampling rate of 30

Hertz. In addition to the telemetry data, we captured

the movement of the subjects on video at a rate of 15

frames per second. This provided us with a setup that

is sufﬁciently overdetermined, to allow us to simulate

and virtually evaluate many different potential sensor

conﬁgurations.

Figure 1: 22 markers were positioned in the region of the

lumbar spine (LS), and tracked with a high-precision optical

motion capturing system, in order to analyze the motion of

the lower back during standard movement tasks.

BACKTRAINER - Computer-aided Therapy System with Augmented Feedback for the Lower Back

Then we identiﬁed a set of movement tasks that

are commonly used for the management or in the di-

agnosis of low back pain:

1. Posture Correction (while seated)

2. Range of Motion (ﬂexion/extension)

3. Range of Motion (lateral ﬂexion)

4. Range of Motion (rotation)

5. Stabilization of lumbar spine during knee exten-

sion

6. Lifting test (light load: 7.5 kg)

7. Lifting test (heavy load: 7.5-45 kg individually)

We recruited 22 healthy subjects between the ages

of 18 to 55 years, and asked them to perform the

above sequence of tasks. This produced a total of 2.4

hours of video synchronized with 5.6 million 3D po-

sitions for all of the markers. Standard statistical tools

are too limited to explore and analyze this large body

of information. We therefore developed a highly in-

teractive visualization application that would allow us

to visually explore the data, and experiment with dif-

ferent scenarios.

The visualization application consists of multiple

coordinated views that simultaneously show:

• x, y, z positions of all the markers projected onto

the three planes of the body (sagittal, coronal,

transversal)

• 3D view of the marker positions in space

• synchronized video frames

• missing values (due to line of sight problems)

• derived values (e.g. spatial angles between pairs

of markers, distances)

Figure 2: The interactive visualization application uses mul-

tiple coordinated views to show different aspects, in order

to explore and analyze all the data that was recorded.

The software allows to plug in any number of al-

gorithms that produce derived values from selected

marker positions. We developed algorithms to mea-

sure angles and distances, as well as various pro-

jection methods onto the body planes. The videos

were examined visually by physiotherapy experts,

and marked up at points in time where subjects lost

their ability to stabilize the lumbar spine during the

exercises.

This analysis revealed the following results:

• The shape of the lumbar spine can be quanti-

ﬁed by measuring the angle between vertebrae

Th12/L1 and L5/S1.

• The difference between correct, stable movements

and unstable, potentially dangerous movements

can be identiﬁed in the data, and corresponds to

visual assessment by physical therapists.

• The necessary data to evaluate the quality of

movement can be acquired by the use of skin sur-

face sensors.

Based on these results we conﬁrmed our hypothe-

sis that for our intended goal to support therapy of low

back pain, it is sufﬁcient to measure the orientation of

the lumbar spine at two points, with three degrees of

freedom.

4 THE BACKTRAINER SYSTEM

4.1 System Overview

The Backtrainer prototype consists of two inertial

sensor modules, capable of measuring the three rota-

tional degrees of freedom. Each of the two modules is

positioned on the patients’ back using an elastic band

(Figure 8). Motion data from the two sensor modules

are transmitted to a PC.

On the PC, a therapy software receives the signals

and reconstructs the movements of the lumbar spine.

The software further consists of a patient database,

and a set of movement exercises that can be conﬁg-

ured for the individual patient. The software supports

the therapist in instructing complex movements, and

allows the patient to exercise independently in a moti-

vating, game-like environment, and document therapy

activities and progress.

Figure 3 shows an architectural overview of all

components of the system.

HEALTHINF 2009 - International Conference on Health Informatics

Figure 3: Overview of the Backtrainer system. Hardware,

exercise logic, and feedback are abstracted into separate

layers to guarantee high ﬂexibility and extensibility.

4.2 Inertial Sensors

Of the sensing technologies discussed in section 2,

we chose to use inertial sensors, because of their low

cost and simplicity of use. We ﬁrst considered to de-

velop a sensor based on accelerometers, gyroscopes,

and magnetometers on our own, but recently, many

commercial sensors of this type that match our speci-

ﬁcations have become available, and therefore we de-

cided against it.

We used sensors from two different manufactur-

ers. One was a wireless system (InertiaCube3 from

Intersense), and another one was a system where the

sensors are connected to the PC with USB cables

(MotionNode from GLI Interactive). We abstracted

the interfacing of the sensors in the software, so that

we are able to switch systems easily.

4.3 Software

The software is separated into several layers to guar-

antee high ﬂexibility and extensibility (Figure 3). The

core application layer is responsible for the hardware

abstraction, as well as basic patient, exercise and ther-

apy management. The exercises are abstracted into a

separate layer, which makes it possible to easily add

any number of exercises to the system. This exercise

Figure 4: Use cases resulting from the task analysis.

logic layer takes care of conﬁguration, generation of

feedback, and determination of the exercise success

level. Rendering and interaction are done in the feed-

back layer, to support the use of different feedback

modalities.

In order to design the user interaction, a task anal-

ysis was performed based on scenarios of future ther-

apy sessions. The resulting use cases are summarized

in Figure 4.

Analysis of the use cases and iterative prototyping

together with therapists, lead to three areas of inter-

action that were then implemented in the following

main screens:

Device: Management of the hardware (Initialization,

calibration, monitoring).

Personae: Management and selection of therapist

and patients, monitor therapy (exercise perfor-

mances, access to historical data).

Exercises: Selection and performance of exercises

(conﬁguration, personalization, feedback).

These three screens reproduce the main work ﬂow

executed in a therapy session:

1. Initialize the device and make sure the sensors are

mounted correctly and deliver signals.

2. Select therapist and patient, review previous ses-

sions and decide on the exercises to be performed

in the current session.

3. Select, perform and evaluate exercises.

4.4 Exercises

The exercises are the main concept within the Back-

trainer system, they target the training area and can be

divided into the two groups of mobilizing and stabi-

lizing exercises.

Mobilization Exercises

The aim of mobilization exercises is to restore the

range of motion of the patient. The physiotherapist,

BACKTRAINER - Computer-aided Therapy System with Augmented Feedback for the Lower Back

together with the patient, set range limits which en-

close the required movement range to achieve the

treatment goal. This range is visualized to the pa-

tient by a white ball moving within the predeﬁned

range limits. The ball turns its color to green if the

wanted limits are reached and to red if these are ex-

ceeded. This information assures him/her that mo-

tions within this range will be most effective. This is

important, because it prevents patients to be overcau-

tious or overambitious, which would result in a lower

success rate of the therapy. The success level of a

particular performance of the mobilizing exercise is

deﬁned as the ratio of number of times that the limit

was reached to the total number of attempts.

Figure 5 shows the application window with the

mobilizing exercise selected. The slider and buttons

on the right allow the adjustment of exercise parame-

ters.

Figure 5: Mobilizing feedback.

Stabilization Exercises

The goal of stabilization exercises is to hold the lum-

bar spine in a given stable position, while perform-

ing movements such as ”squats”, lifting weight, or

changing from sitting to standing. For this type of

exercise, it is important to provide the patient with

an augmented feedback of the posture of the lumbar

spine, as proprioception of this region is typically low.

The metaphor of a ”green range” was introduced,

meaning that motions within this range are perfectly

tolerable. This range is adjustable and allows to de-

ﬁne the level of difﬁculty for the exercise.

Our ﬁrst approach for the visualization of the lum-

bar spine posture used a comic like stick ﬁgure that

had a bendable spine. Tests showed however that this

approach worked only, if the subject was positioned

exactly as the ﬁgure on the screen (i.e. standing up-

right). In other situations (e.g. sitting, kneeling), this

concrete depiction turned out to be more confusing

than helpful.

We therefore replaced the ﬁgure by a more ab-

stract visualization of a sphere balancing on a curved

convex surface. Figure 6 shows these two approaches

side by side. When the patient leaves the ”green

range”, then the ball slides down on one side of the

surface, and changes its color from yellow gradually

to red, depending on how much the current measured

angle is away from the green angle.

Figure 6: Stabilizing feedback: The ﬁrst version (left) used

a ﬁgure like feedback but was then replaced by a more ab-

stract visualization using a ball balanced on a bump.

Game Exercise

As described in section 1, movement exercises with

many repetitions are key to therapy effectiveness. Pa-

tient compliance with repetitive movement exercise

regimes is problematic though. We developed a sim-

ple game with the aim of enhancing motivation and

compliance.

In the game, the patient controls a bat at the bot-

tom of the window (Figure 7). The bat can be moved

from left to right, according to the difference in ro-

tational angles between the upper and lower sensor

modules. Which of the three rotational planes should

be used for the mapping, can be freely chosen by the

therapist, depending on the therapy goal. The task in

this game is to catch the balls that are rolling from the

back toward the front at randomly chosen offsets from

the center line. The ratio of caught vs. missed balls

is displayed as a score, and the ﬁnal score is recorded

as the success level of this exercise in the patients’

therapy history.

This simple game is implemented in the Back-

trainer software using basic ”OpenGL” commands,

Figure 7: Game feedback: A ball (grey) is rolling from the

back toward the player and has to be caught by a bat (blue).

HEALTHINF 2009 - International Conference on Health Informatics

but the software design explicitly addresses the pos-

sibility to integrate more sophisticated games that can

be based on so called game-engines (Figure 3). Game

engines facilitate the implementation of 3D games

providing elaborate functions for the realization of

virtual worlds, avatars and leveling systems.

Lightbulb Feedback

The movements that a patient executes while perform-

ing a movement exercise often involve a rotation of

the line of vision of the patient (e.g. rotation of the

upper body in the coronal plane), or the line of vision

is not oriented horizontally (e.g. patient lying on the

chest). In such situations, it is unpractical to provide

a visual feedback on a computer screen with a ﬁxed

position. It would be helpful to provide feedback that

is not directional in nature, but embedded in the envi-

ronment in an ambient way.

Auditory feedback is one possibility. However,

physiotherapy for low back patients is often per-

formed in clinical therapy settings, where many pa-

tients exercise in the same room at the same time. In

such a setting, auditory feedback can be distracting

and confusing.

Research in the ﬁeld of human computer interac-

tion suggests the use of ”ambient displays”. Ambient

displays are abstract, peripheral displays that visual-

ize information on the periphery of a user’s attention

(Mankoff et al., 2003). This approach can also be

adapted to physiotherapy, to let the patient concen-

trate on the exercises and nevertheless perceive feed-

back about the movements.

For the Backtrainer system, we built a very simple

but effective component to generate an ”ambient feed-

back” using a modiﬁed color changing LED Light-

bulb (Figure 8). The Lightbulb was instrumented with

a USB-interface in order to be able to change the color

from software for any combination of the two primary

colors red and green.

The Lightbulb matches the ball metaphor that is

used for the feedback shown in the software on the

computer screen. Some exercises can be performed

with only the Lightbulb feedback (e.g. stabilizing

exercise) while others (e.g. game exercise) use the

Lightbulb as an additional feedback modality. Apart

from being an ambient feedback system for the pa-

tients, the bulb can also be used by physiotherapists

coaching several patients at the same time. The thera-

pist can observe the emitted color from a distance and

intervene to adapt exercise settings, i.e. if there are

too many ”misses” while playing the game exercise.

Figure 8 shows a therapy situation with the Back-

trainer mounted on the patients back, the LED Light-

bulb and the physiotherapist. The patient is perform-

Figure 8: The Backtrainer system in a therapy situation.

ing a stabilizing exercise (weight lifting while stabi-

lizing the lumbar spine).

5 CLINICAL EVALUATION

In order to evaluate the Backtrainer system, we per-

formed an exploratory study in a controlled clinical

setting. The goal of this study was to use the sys-

tem under realistic conditions, in order to evaluate

the practicality in various situations. In particular the

goals were:

• Evaluation of the practicality (expenditure of

time, handling, ease of use) and application in a

therapeutic setting.

• Evaluation of acceptance of the system by thera-

pists and patients.

• Evaluation of the feedback produced by the sys-

tem.

The study was performed in the rehabilitation

clinic Valens, Switzerland. 15 physical therapists

were given a short introduction to the Backtrainer

system. After this introduction, the therapists were

free to use the system at their discretion during

their regular therapy sessions with patients that suf-

fer from chronic back pain or that have undergone

back surgery. Both, patients and therapists were asked

to ﬁll out questionnaires asking about their subjective

satisfaction with the system. In addition, the software

is equipped with extensive logging to provide objec-

tive data about the use of the system.

The study was planned for a duration of ﬁve

weeks. We report preliminary results after 80% of

this time has elapsed. The following discussion of the

results focuses on those aspects that relate to the soft-

ware of the Backtrainer system. In depth analysis of

the overall study is subject of further research, once

the study is completed.

BACKTRAINER - Computer-aided Therapy System with Augmented Feedback for the Lower Back

During the period under investigation (21 days),

the system has been in use between 1 and 2 hours per

day. 26 patients have performed therapy sessions with

the Backtrainer. They have performed a total of 248

exercises distributed as follows: mobilizing (17%),

stabilizing (51%), game (32%). The average dura-

tion of an exercise was about 2 minutes with the 90%

quantile at 5 minutes (8 minutes for the game exer-

cise). Since the game exercise can be considered a

mobilizing exercise, the distribution between the two

types of exercises is just about half and half.

We have received ﬁlled-out questionnaires from

18 patients and 8 therapists up to this point. The num-

bers are not complete due to the fact that the study

is still on-going, and questionnaires will typically be

ﬁlled out toward the end. Nevertheless, the numbers

are sufﬁciently large that they do provide an indica-

tion for the important trends.

Tables 1, 2, and 3 show the answers to selected

questions. In general, the feedback provided by the

Backtrainer is considered helpful by both patients and

therapists, and it matches the visual observations of

the therapists. Patients generally indicate that when

training individually, it is more fun to train with the

Backtrainer. We will have to further investigate into

the reason for the few less favorable answers though.

Therapists indicate that they can use the Backtrainer

to measure therapy progress. The evaluation of the

software by the therapists is favorable.

Table 1: Patient questionnaires (n = 18). Encoding: 4 Com-

pletely agree, 3 Agree somewhat, 2 Disagree somewhat, 1

Completely disagree.

Encoded Answers

[# answers]

Statistics

Question 4 3 2 1 ¯x σ

Feedback from

the Backtrainer

was helpful

12 6 0 0 3.7 0.5

Feedback is easy

to understand

13 5 0 0 3.7 0.4

Makes it easier to

perform exercises

on your own

8 7 0 2 3.2 0.9

Independent

training is more

fun with the

Backtrainer

8 7 2 1 3.2 0.9

Table 2: Physical therapist questionnaires (n = 8). Encod-

ing: 4 Completely agree, 3 Agree somewhat, 2 Disagree

somewhat, 1 Completely disagree.

Encoded Answers

[# answers]

Statistics

Question 4 3 2 1 ¯x σ

Feedback from

the Backtrainer

was helpful for

patients

4 4 0 0 3.5 0.5

The feedback

matched my

observations

4 2 1 0 3.4 0.7

It is possible to

measure therapy

progress with the

Backtrainer

4 3 1 0 3.4 0.7

Patients are

motivated to use

5 3 0 0 3.6 0.5

Table 3: Software evaluation by physical therapists (n = 8).

Encoding: 4 Completely agree, 3 Agree somewhat, 2 Dis-

agree somewhat, 1 Completely disagree.

Encoded Answers

[# answers]

Statistics

Question 4 3 2 1 ¯x σ

Overall

impression is

very good

0 8 0 0 3.0 0.0

The software

fulﬁlls its task

4 4 0 0 3.5 0.5

The software

matches my

expectations and

habits

2 5 1 0 3.1 0.6

6 CONCLUSIONS AND FUTURE

WORK

We have developed a system to support physiotherapy

of low back pain. We found that this is possible by

measuring the orientation of the lumbar spine at two

points, with three degrees of freedom. The dynamic

behavior and accuracy of commercially available in-

ertial sensors are good enough for this application.

The layered software architecture that we de-

veloped has proven effective in integrating differ-

ent hardware systems, and providing the ﬂexibility

needed for prototyping. The separation of code for

the exercise concept into core, logic, and feedback is

useful for providing various feedback modalities in a

HEALTHINF 2009 - International Conference on Health Informatics

modular way.

The task analysis and the division of the interface

into three areas, resulted in a system that was easy

to use and matched the workﬂow of a typical therapy

session. The distribution of the exercise types per-

formed during the clinical evaluation shows that the

distinction between mobilizing and stabilizing move-

ments is fundamental and well reﬂected in practice.

The abstract visual feedback that we designed was

considered helpful. Ambient feedback in the form

of the Lightbulb proved to be a very useful addition

to the computer screen in a real-life therapy setting.

With regard to feedback and motivation, the study

only provided some ﬁrst hints though. Participants

liked the game and the feedback, but there is fur-

ther systematic investigation needed to answer our re-

search questions in this area.

The usage patterns and the answers from the ques-

tionnaires from the clinical evaluation provide a stable

foundation for the further development of the Back-

trainer system. Since therapists did not have to follow

a ﬁxed protocol, but were free to use the Backtrainer

when they saw a need, we interpret the numbers that

we found as encouraging indication for the adoption

of such a therapy support system.

The above results suggest future work for the elab-

oration of the system in several areas:

• Evaluate other feedback modalities (e.g. audi-

tory, tactile), and other ambient devices (e.g. light

emitting ﬂoor panels). Also the use of wearable

3D-Displays (Eye-goggles) will be evaluated.

• Explore a telemedical scenario in which the ex-

ercises performed by the patient at home can be

evaluated by geographically distant physiothera-

pists to provide guidance for the patients.

• Evaluate ”virtual reality” game-like feedback

modalities to further raise motivational factors.

ACKNOWLEDGEMENTS

This work was supported by funding from the Swiss

Innovation Promotion Agency CTI. The authors

would like to thank Hocoma AG for their support.

REFERENCES

Abenhaim L., e. a. (2000). The role of activity in the ther-

apeutic management of back pain: Report of the Paris

International Task Force on Back Pain. Spine, 25(4

Suppl):1S33S.

Deyo, R. and Weinstein, J. (2001). Low back pain. N Engl

J Med, 344(5):36370.

Dunne, L. E., Walsh, P., Smyth, B., and Caulﬁeld, B.

(2006). Design and evaluation of a wearable optical

sensor for monitoring seated spinal posture. In ISWC,

pages 6568.

Hayden J.A., M.W. van Tulder, G. T. (2005). Systematic re-

view: strategies for using exercise therapy to improve

outcomes in chronic low back pain. Ann Intern Med,

142(9):77685.

Huang, H., Ingalls, T., Olson, L., Ganley, K., Rikakis, T.,

and He, J. (2005). Interactive multimodal biofeedback

for task-oriented neural rehabilitation. Engineering

in Medicine and Biology Society, 2005. IEEE-EMBS

2005. 27th Annual International Conference of the,

pages 25472550.

Jordan, K., Dziedzic, K., Mullis, R., Dawes, P. T., and

Jones, P. W. (2001). The development of threedimen-

sional range of motion measurement systems for clin-

ical practice. Rheumatology, 40(10):10811084.

L. Engels, T. Leloup. T, N. W. (2006). Imaging technolo-

gies for avoiding back pain at work. In In proceedings

First Symposium of the IEEE/EMBS Benelux Chapter,

pages 235238.

Mankoff, J., Dey, A. K., Hsieh, G., Kientz, J., Lederer, S.,

and Ames, M. (2003). Heuristic evaluation of ambi-

ent displays. In CHI 03: Proceedings of the SIGCHI

conference on Human factors in computing systems,

pages 169176, New York, NY, USA. ACM.

Mrl F., Blickhan R. (2006). Three-dimensional relation of

skin markers to lumbar vertebrae of healthy subjects in

different postures measured by open MRI. European

Spine Journal, 15(6):74251.

Raymond Y.W. Lee, Judi Laprade, E. H. F. (2003). A re-

altime gyroscopic system for three-dimensional mea-

surement of lumbar spine motion. Medical Engineer-

ing & Physics, 16(5):817824.

Sucar, L. E., Leder, R. S., Reinkensmeyer, D. J., Hernndez,

J., Azcrate, G., Casteeda, N., and Saucedo, P. (2008).

Gesture therapy - a low-cost vision-based system for

rehabilitation after stroke. In Azevedo, L. and Lon-

dral, A. R., editors, HEALTHINF (2), pages 107111.

INSTICC - Institute for Systems and Technologies of

Information, Control and Communication.

Van Herp, G., Rowe, P., Salter, P., and Paul, J. P. (2000).

Three-dimensional lumbar spinal kinematics: a study

of range of movement in 100 healthy subjects aged 20

to 60+ years. Rheumatology, 39 (12):13371340.

van Tulder M., B. Koes, C. B. (2002). Low back pain. Best

Pract Res Clin Rheumatol, 16(5):76175.

van Tulder M.W., B.W. Koes, L. B. (1995). A cost-

oﬁllness study of back pain in The Netherlands. Pain,

62(2):23340.

Yang, Z., Grifﬁth, J., Leung, P., Pope, M., Sun, L., and Lee,

R. (2005). The accuracy of surface measurement for

motion analysis of osteoporotic thoracolumbar spine.

Engineering in Medicine and Biology Society, 2005.

IEEE-EMBS 2005. 27th Annual International Confer-

ence of the, pages 68716874.

BACKTRAINER - Computer-aided Therapy System with Augmented Feedback for the Lower Back