Development of a Miniaturized Motion Sensor for Tracking Warning

Signs of Low-back Pain

Jérôme Molimard

, Tristan Delettraz

and Etienne Ojardias

Mines Saint-Etienne, Univ. Lyon, Univ. Jean Monnet, INSERM, U 1059 Sainbiose,

Centre CIS, F - 42023 Saint-Etienne, France

Clinical Gerontology Department, University Hospital of Saint-Étienne, Saint-Étienne, France

Keywords: Low-Back Pain (LBP), Inertial Measurement Unit (IMU), Hip & Shoulder Dissociation, Lumbar Lordosis

Angle, Measuring Device.

Abstract: Low-back pain (LBP) is a widespread disease which can also be highly disabling, but physicians lack of

basic understanding and diagnosis tools. During this study, we have designed and built a new wearable

device capable of detecting features helpful in LBP follow-up while being non-invasive. The device has

been carefully validated, and shows good metrological features, with small noise level (σ = 1°) and no

observable drift. Two simple exercises were proposed to two young volunteers, one of them with LBP

history. These exercises are designed to target two characteristics: the lumbar lordosis angle and the hip &

shoulder dissociation. Even if no general rules can be extracted from this study, we have shown that Inertial

Measurement Units (IMU) are able to pick up those characteristics and the obtained values are meaningful

refereeing to LBP disease. Henceforth, we are confident in going to clinical studies to investigate the link

between back related feature and LBP, in particular the hip & shoulder dissociation which is poorly

documented.

1 INTRODUCTION

Low-back pain (LBP) is a widespread affliction in

most developed and industrialized countries. It is a

major disability factor both at work and in every-day

life (Bauer et al., 2017). Worldwide, it is the most

reported reason for seeking care from a primary care

physician (Traeger et al., 2017). LBP is accountable

for the most sick leaves and it touches 70% of

people at least once in his/her life (Koes et al.,

2006). Therefore, it costs the French government

more than one billion euros each year (Depont et al.,

2010). LBP care is difficult since about 90% of all

patients suffer “non-specific low-back pain” which

means that while the pain is apparent, its cause

remains unknown (Koes et al., 2006). Furthermore,

73% of patients experiencing LBP will go through

an other episode within a year (Koes et al., 2006)

which will usually be more painful than the last one

(Riihimäki et al., 1991).

Hence, a lot of scientific studies have been made

in order to find a way to better monitor the condition

of patients afflicted by LBP, among them the lumbar

lordosis angle (Evcik et al., 2003). Likewise, since

people afflicted by LBP may have trouble

performing normal muscle activities, like bending

their back sideways. The detection and

quantification of some abnormal minute back

movements, like the hip & shoulder dissociation

(Park et al., 2012), could establish a new method to

label the beginning of a serious illness. Both subjects

generated great interest in the medical community

(Baek et al., 2010).

Many methods to monitor those characteristics

have been put forward. For instance,

Electromyography (EMG) is the most popular

technique for muscle activity observation. EMGs

can measure various movements at a high rate, with

almost no added weight. Their most commonly

reported drawbacks are the difficulties to set up the

device and its sensors need to by applied sometimes

invasively under the skin of the patient (Butler et al.,

2010). Moreover, the results given by EMG still

need to be processed to get the desired displacement

or angle.

A popular alternative solution is the use of

optical motion capture system. Nevertheless, the

high cost of the cameras and markers may be a

Molimard, J., Delettraz, T. and Ojardias, E.

Development of a Miniaturized Motion Sensor for Tracking Warning Signs of Low-back Pain.

DOI: 10.5220/0010291701290134

In Proceedings of the 14th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2021) - Volume 1: BIODEVICES, pages 129-134

ISBN: 978-989-758-490-9

129

limiting factor to its usage, as well as the fact that it

can only handle movements contained in a closed

and limited area (Nakamoto et al., 2018).

Recently, strain sensors were successfully used

to monitor lumbar motion. They also can be made

wearable, unlike the previous two techniques.

Besides, they are lightweight and inexpensive,

which reinforce their portability (Nakamoto et al.,

2018). But their usage still is limited in space and

along one plane only. At the moment, an application

to accurately measure movements of the back from

top to bottom along three axes is still out of reach.

Last, inertial measurement units (IMUs) are

electronic devices composed of 2 or 3 sensors (3-

axes accelerometer, 3-axes gyroscope and optionally

3-axes magnetometer) that can report the

acceleration and orientation of one object using

Attitude and Heading Reference System (AHRS)

algorithm. They are small and can easily be

integrated in a wearable devices (Baek et al., 2010).

In addition, they can measure small motions and

rotations of the back along the axes with ease (Zhao

et al., 2017) and can be combined to tackle a larger

array of movements and be more accurate (Chhikara

et al., 2008). Last, the physical values obtained from

IMUs are very close to the displacements and angles

that are usually sought.

Many works emphased the real time features that

IMUs can provide in order to design a portable

measuring device to gather data on the everyday life

of an affected patient to better his treatment and

warn him of unsafe positions. Recently, Beange et

al. (2019) or Graham et al. (2020) proposed

applications of IMU to monition the spine in the

context of LBP.

The goal of this study is to provide physician a

measuring instrument that could be used to detect

and monitor back disorders. As such, the solution

must be low-cost, usable in a closed environment

without highly-dedicated technical skills. In the

following, we will present an IMU-based solution,

with detailed validation protocol, and give two first

application examples on movement tracking and hip

& shoulder dissociation.

2 MATERIAL AND METHODS

2.1 Device Elements

BackMonitor is an in-house system built on the

Feather M0 development board (Adafruit

Industries

, New York, USA), an Arduino

compatible processor that includes a Bluetooth Low

Energy (BLE) module. The sensor (BNO055 Bosch

Sensortec - Kusterdingen, Germany) is a 9 DOF

(accelerometer, gyroscope, and magnetometer) IMU

embedding an AHRS processing. Each sensor has a

size of 20mm×27mm×4mm, and weights 3 g; the

micro-controller board, can easily to be placed in the

trousers pocket (60mm×30mm×25mm, 30g).

2.2 Metrology

In the context of low-back pain monitoring, both

linear acceleration and Euler angles are useful,

respectively for hip & shoulder dissociation

detection and for the lumbar lordosis angle follow-

up. Those two quantities can be directly gathered

from the BNO055 (via an internal fusion algorithm).

BNO055 comes with autocalibration feature.

This process is a black box that ought to be verified

anyway. Acceleration can be 2-points checked easily

by using gravity, but rotation must be studied in

more details. Thus, a calibration protractor with

IMU holder was 3D printed to set angles with an

accuracy of 0.5°.

The angles were measured form each sensor and

in each axis between 0° and 165° every 5° for X and

Z directions and from 0° to 90° for Y direction

according to Euler angle definition. Moreover, the

board was initialized while being at 0°.

2.3 Assembly

The hardware layout for the simultaneous

measurement of data from two sensors is presented

in Fig.1, that shows the electronic circuit diagram.

The two BNO055s are connected to the Feather

M0 Bluefruit by I2C bus. Each BNO055 has a

specific address (respectively 0x29 and 0x28 from

left to right) depending on its ADR pin level.

Besides, a 110 mAh battery was followed by a

switch connected directly to the processor’s power.

It is necessary, since the goal is to conceive a

portable system. A changeover diode soldered onto

the board allows both the USB port and the battery

to be connected without any risks as the battery acts

as a backup power.

Three switches named A, B and C and a blue

LED were added for the user to select the reading

mode (hip & shoulder dissociation detection/lumbar

lordosis angle measure, serial/BLE connection, raw /

AHRS data).

BIODEVICES 2021 - 14th International Conference on Biomedical Electronics and Devices

130

Figure 1: Electrical circuit diagram.

Figure 2: Position of the IMUs for both hip & shoulder

dissociation detection (right) and the lumbar lordosis angle

measure (left).

2.4 Sensors Arrangement

The sensor position along the patient’s back has

been set according to physicians usual practice

(Fig.2). Each sensor is stuck on the patient’s skin

with bio-compatible double-sided adhesive tape.

The lumbar lordosis angle is defined as the angle

formed by the tangent to the superior plate of the

transitional thoraco-lumbar vertebra the most

inclined on the horizontal, usually L1, and the

tangent to the inferior plate L5. Thus, one IMU

should be put over the L1 vertebra and one over the

L5 vertebra. The lumbar lordosis angle θ

can be

defined as the difference between the Euler angles of

the IMUs along the Z axis θ

and θ

Detecting hip & shoulder dissociation is picking

up some conflicting accelerations between the upper

part of the back and the lower part of the back. As a

first suggestion, one IMU is placed over the highest

vertebra on the patient’s back T1, and one over the

lowest L5. The hip & shoulder dissociation will be

characterized as the phase difference between the

analytic signals obtained from IMU antero-posterior

accelerations A

and A

at T1 and L5.

2.5 Experimental Protocol

A first feasibility test is designed. The system is

tested on two volunteers. They are both youth

women with the same morphology, one with a

medical history of LBP and scoliosis (subject 1) and

the other one without any reported back-related issue

(subject 2).

Two simple exercises were proposed to the

subjects.

First, each subject was asked to sit down on a

chair for about 10 seconds, then, to stand up and to

stand still for the next 10 seconds. An object was

placed 50 cm in front of him and the subject was

asked to bend his back in order to pick it up. After

that, the patient stood up for 10 seconds once more

and then sat down for 10 seconds (Exercise 1).

Second, a time-up-and-go test is done: after 10

seconds on the chair, the patient was asked to walk

for about 3 m, turn back and then go back to the

chair and stay sit for 10 seconds (Exercise 2).

3 RESULTS

3.1 Metrology

Table 1 presents the slope and offset between the

angle given by the protractor and the measured one.

Most of the coefficients of determination R

are

higher that 0.99. The slope is close to 1 – and can be

corrected easily. X axis presents the higher

sensitivity drift for both sensors.

Table 1: Linearity parameters.

Sensor name Axis offset Slope R

0x28 X -0.5122 0.9312 0.9995

Y 1.2707 0.9844 0.9999

Z 1.6358 0.9879 0.9999

0x29 X -1.9851 0.9065 0.9973

Y 1.0071 0.9692 0.9998

Z 6.1829 0.9411 0.9993

Development of a Miniaturized Motion Sensor for Tracking Warning Signs of Low-back Pain

131

The residue between the best linear fit and the

experimental results are the same for 0x28 and 0x29.

For Y and Z directions, the residues are essentially

random. In the X direction, a 2

order bias appears

and can be easily corrected. In these conditions, the

standard deviation σ is less than 1° (Fig.3).

The IMUs did not show any measurable drift

over a time period of 10 min. Identically, the

calibration procedure was performed 10 days after

the fist measure, and results were in close

agreement. Therefore, it was concluded that

reproducibility was good enough and that this point

was not an issue.

Last, the BNO055 coming with an auto-

calibration feature running after every starting up, it

is worth checking the BON055 measurement

stability. Measurement at 0° and 75° were performed

while varying the starting position. For the Y and Z

direction, and for both IMUs, the returned value

barely changed with the starting position but it was

concluded that the X value was zeroed at the start

whatever the real starting angle. Hence, it is required

to build a stand for the IMUs so that the initial value

for the X axis to be stable for both sensors.

Figure 3: Residues for Euler angles (in degrees) on 0x28.

3.2 Feasability Tests

Fig. 4 presents the sensor 0x28 and 0x29

orientations versus time for the two subjects

performing Exercise 1. Sensors are represented by a

stick corresponding to their longitudinal axis (Y

The lumbar lordosis angle is calculated from the

angle over the Z axis as follow:

= −180° − θ

− θ

(1)

Finally, the lumbar lordosis angle corresponding

to each subject was obtained by averaging the values

while standing up.

Figure 4: Sensor orientation and lumbar lordosis angle for

subject 1 and 2 during Exercise 1.

For the hip & shoulder dissociation detection,

charts of linear accelerations along all axis have

been produced. The parts where the subject was sat

down were remove because they were irrelevant

(mostly null) and to better focus on the moving part

of the experiment. As the goal is to pick up an offset

between the two IMUs, analytical signals on the

linear accelerations for the Z axis were calculated

with OCTAVE (Eaton, 2019) in order to find raw

phase differences (Fig.5).

Figure 5: Hip & shoulder phase difference for antero-

posterior (Z) acceleration.

0 20406080100120140160180

-2

-1,5

-1

-0,5

0,5

1,5

2,5

Angl e (° )

Res i dues ( ° )

Subject 1

Subject 2

0x29

0x28

0x29

0x28

Subject 1

Subject 2

0°

180°

Antero-posterior phase difference (°)

0612

180

- 180

Time (s)

BIODEVICES 2021 - 14th International Conference on Biomedical Electronics and Devices

132

4 DISCUSSION

The metrological results showed that the sensor

arrangement gives reliable results, with a

repeatability σ = 1° and no observable drift.

A first test was done, with two women of

comparable age and morphology, one with LBP

history and the other without. Of course, these

results cannot have any statistical meaning;

nevertheless, they have a demonstrative interest.

During Exercise 1, the patient afflicted by LBP

(subject 1) has a mean lumbar lordosis angle in

stand-up phase of −19° and the other −48°. The

lumbar lordosis angle being considered as natural in

the range −45°±9°, subject 1 is out of the safe

interval, unlike subject 2 who is not afflicted by

LBP. Fig. 4 shows a difference between the lumbar

lordosis angle value in stationary stand-up phase

between the 2 subjects. While subject 2’s angle

varies from −25° to −60°, subject 1’s angle stays

always close to −20°. Subject 1’s movements appear

as more restricted in range than subject 2’s. More,

subject 1’s movements are slower than subject 2’s.

This is in agreement with previous works stating a

decrease both in speed and in range of motion for

LBP patients (Errabity et al., 2020).

Exercise 2 focuses on acceleration, and it is

possible to extract basic gait analysis information.

For example here, subject 1 was about 2s slower

than subject 2. But, much detailed observations can

be done: while subject 1’s steps keep a similar shape

and range through time, subject 2’s are fluctuating

through time. The pain might force subject 1 to limit

her walking strategy to few movements while

subject 2 can freely adapt her movements to the

current stance.

The hip & shoulder dissociation presented as a

phase difference on Fig.5 discriminates the two

subjects. Indeed, at the turning point (t ≈ 6.5 s), the

T1 and L5 vertebrae of subject 1 are nearly in

opposition of phase with the lower part of the back

lagging behind the upper part. This is not seen in

subject 2’s case as the phase when turning back (t ≈

5.5 s) is not much different than when subject 2 is

walking. Henceforth, the hip & shoulder dissociation

could be detected for the subject with LBP and not

for the healthy one using phase analysis.

5 CONCLUSION

During this study, we have designed and built a new

wearable device capable of detecting features

helpful in LBP follow-up while being non-invasive.

The metrological validation of BackMonitor

arrangement shows good features, with small noise

level (σ = 1°) and no observable drift.

Two simple exercises, one combining stand-up,

sit and bending movements, the other being a

classical time-up-and-go test, were proposed to two

young volunteers, one of them with a LBP history.

Signal was processed to extract the lordosis angle

and hip & shoulder dissociation. Even if no general

rules can be extracted from this study, we have

shown that IMUs are able to pick up those

characteristics and the obtained values are

meaningful refereeing to LBP disease.

Hence, we are confident in going to clinical

studies to elaborate the link between back related

feature and LBP, in particular the hip & shoulder

dissociation which is poorly documented.

REFERENCES

Bauer, C., Rast, F., Ernst, M., Meichtry, A., Kool, J.,

Rissanen, S., Suni, J., and Kankaanpää, M., 2017. The

effect of muscle fatigue and low back pain on lumbar

movement variability and complexity. Journal of

Electromyography and Kinesiology, 33, 94–102.

Traeger, A., Buchbinder, R., Harris, I., and Maher, C.,

2017. Diagnosis and management of low-back pain in

primary care. CMAJ, 189(45), E1386–E1395.

Koes, B., Van Tulder, M., and Thomas, S., 2006.

Diagnosis and treatment of low back pain. BMJ,

332(7555), 1430–1434.

Depont, F., Hunsche, E., Abouelfath, A., Diatta, T., Addra,

I., Grelaud, A., Lagnaoui, R., Molimard, M., and

Moore, N., 2010. Medical and non-medical direct

costs of chronic low back pain in patients consulting

primary care physicians in France. Fundamental &

clinical pharmacology, 24 (1), 101–108.

Riihimäki, H., 1991. Low-back pain, its origin and risk

indicators. Scandinavian journal of work, environment

& health, 81–90.

Evcik, D., and Yücel, A., 2003. Lumbar lordosis in acute

and chronic low back pain patients. Rheumatology

international, 23 (4), 163–165.

Park, W.-H., Kim, Y. H., Lee, T. R., and Sung, P. S., 2012.

Factors affecting shoulder–pelvic integration during

axial trunk rotation in subjects with recurrent low

back pain. European Spine Journal, 21 (7), 1316–

1323.

Baek, J., and Yun, B.-J., 2010. Posture monitoring system

for context awareness in mobile computing. IEEE

Transactions on instrumentation and measurement, 59

(6), 1589–1599.

Butler, H. L., Lariviere, C., Hubley-Kozey, C. L., and

Sullivan, M. J., 2010. Directed attention alters the

temporal activation patterns of back extensors during

Development of a Miniaturized Motion Sensor for Tracking Warning Signs of Low-back Pain

133

trunk flexion–extension in individuals with chronic low

back pain. European Spine Journal, 19 (9), 1508–

1516.

Nakamoto, H., Yamaji, T., Yamamoto, A., Ootaka, H.,

Bessho, Y., Kobayashi, F., and Ono, R., 2018.

Wearable lumbar-motion monitoring device with

stretchable strain sensors. Journal of Sensors. Article

ID 7480528, 7 pages.

Zhao, H., Wang, Z., Qiu, S., Shen, Y., and Wang, J., 2017.

IMU-based gait analysis for rehabilitation assessment

of patients with gait disorders. In 2017 4th

International Conference on Systems and Informatics

(ICSAI), IEEE, 622–626.

Chhikara, A., Rice, A., McGregor, A. H., and Bello, F.,

2008. Wearable device for monitoring disability

associated with low back pain. World, 10, p. 13.

Beange, K. H., Chan, A. D., Beaudette, S. M., Graham, R.

B., (2019). Concurrent validity of a wearable IMU for

objective assessments of functional movement quality

and control of the lumbar spine. Journal of

Biomechanics, 97, 109356.

Graham, R. B., Dupeyron, A., van Dieën, J. H., 2020.

Between-day reliability of IMU-derived spine control

metrics in patients with low back pain. Journal of

Biomechanics, 113, 110080.

Eaton, J.W., Bateman, D., Hauberg, S., Wehbring, R.,

2019. GNU Octave version 5.2.0 manual: a high-level

interactive language for numerical computations,

online reference: https://www.gnu.org/software/

octave/doc/v5.2.0/.

Errabity, A., Han, W.-S.,Bonnaire, R., Pannetier,, R.,

Molimard, J., 2020. Lumbar spine kinematics in

people with and without low back pain: a systematic

review. Applied Ergonomics, under submission.

BIODEVICES 2021 - 14th International Conference on Biomedical Electronics and Devices

134