Agreement between inertial measurement unit and optoelectronic system to measure postural sway

Abstract

BACKGROUND:

Optoelectronic systems and force platforms represent the gold standard for postural sway assessment, but pose disadvantages in terms of equipment, cost and preparation time.

OBJECTIVE:

Wearable inertial measurement units (IMUs) have been proposed to overcome these issues, but have never been compared to an optoelectronic system. The study aim was therefore to investigate agreement between inertial measurement unit and optoelectronic system in postural sway assessment.

METHODS:

Thirty healthy volunteers performed four balance tasks. IMU was placed on the sacrum (S2) with a retroreflective marker over the sensor and subjects’ performance was simultaneously recorded by both systems. Total (TOT), anterior-posterior (AP) and medial-lateral (ML) length of trace, range, speed, root mean squared (RMS), and confidence ellipse were computed.

RESULTS:

ICCs revealed excellent correlations for Length-TOT, Length-AP and Speed-AP, good correlation for Length-ML, Speed-ML, Confidence Ellipse, Range-AP and RMS-AP, and moderate correlation for range-ML and RMS-ML. Bland-Altman plot showed greater estimation for Length-TOT, Length-AP, Speed-AP, confidence ellipse and RMS-AP using optoelectronic system, and for Length-ML, Range-AP, Range-ML, Speed-ML, RMS-ML using IMU. Both systems revealed the same differences among tasks.

CONCLUSION:

The excellent to good agreement of IMU for length of trace and speed parameters and its user-friendly application suggest its potential implementations in clinical practice.

Keywords

Optoelectronic system Inertial Measurement Unit postural sway

1. Introduction

Postural sway assessment provides information about balance and risk of fall. Optoelectronic systems and force platforms represent the gold standard for postural sway assessment, allowing for estimation of centre of mass and centre of pressure displacements [1, 2]. A simplified approach to assess postural sway consists of one marker placed on the sacrum (S2), whose displacement correlates with those of centre of mass [3]. However, this approach requires an expensive laboratory, equipped with an optoelectronic system.

The use of wearable Inertial Measurement Units (IMUs) have been proposed as alternative method to optoelectronic system to evaluate postural sway [4]. IMUs are low-cost, user-friendly and portable devices, which can be easily used in different environments. These sensors, placed at the sacrum, have been largely adopted in assessing postural sway of the human body during standing tasks [5, 6]. However, to the best of our knowledge, no studies have compared IMU accuracy in the measurement of postural sway during standing tasks through a comparison with an optoelectronic system (gold standard). The study aim was therefore to evaluate the agreement between an IMU and an optoelectronic system in the measurement of postural sway.

2. Methods

2.1 Participants and tasks

Thirty young healthy volunteers (between 18 and 30 years old) were enrolled. All participants signed a written informed consent form. The study was approved by the Ethical Committee for Human Investigations of Humanitas Research Hospital (n. CLF18/03). Participants were asked to perform four tasks consisting of bipodalic stance with open (OE) and closed (CE) eyes and monopodalic stance on the right (RM) and left (LM) legs. Each task lasted 1 minute, followed by 5 minutes of rest. Tasks were executed with arms along the side and, during the bipodalic tasks, feet were placed parallel (20 cm of distance). During OE, RM and LM tasks, subjects were asked to fix a point on a wall in front of them at two meters distance.

2.2 Movement acquisition

Acquisitions were conducted with participants in underwear. The wearable IMU (BTS G-Walk, BTS S.p.A., Italy) consisted of a sensor fusion technology with four integrated inertial platforms containing a tri-axial accelerometer, gyroscope and magnetometer. The IMU was attached to the lower back at the level of the second sacral vertebrae (S2). One spherical retroreflective marker (diameter 10 mm) was fixed over the IMU. Optoelectronic system data was recorded using eight infrared cameras (Smart-DX, BTS S.p.A., Italy). IMU data were streamed via Bluetooth ${}^{\@setsize{\scriptsize}{9.5pt}{\viiipt}{\@viiipt}\textregistered}$ 3.0 to a PC using a custom application (G-Studio, BTS S.p.A., Italy). Signals were sampled at 100 Hz and two systems were synchronized through a trigger box.

2.3 Data processing

Postural sway was quantified considering IMU and the trajectory of the marker placed over it. Data were low-pass filtered at 5 Hz using a sixth order non-phase distorting digital filter with Butterworth coefficients. In addition, accelerometer data were detrended to remove baseline drift using a linear least squares algorithm [6]. IMU gave information about the medial-lateral, anterior-posterior and vertical acceleration. The amount of displacement was calculated using a trigonometric approach: from the three components of the acceleration ( $a_{x}$ , $a_{y}$ , $a_{z}$ ), the magnitude of their resultant A was computed. A, $a_{x}$ , $a_{y}$ and $a_{z}$ were used to estimate the cosine directions $\cos_{\alpha}$ , $\cos_{\beta}$ , $\cos_{\gamma}$ , describing the trigonometric relation between the acceleration component and the acceleration resultant. With the knowledge of the cosine directions and the height between the floor and the sensor, the IMU and marker displacements were computed according to the following equation [7]:

$\displaystyle A=\sqrt{a_{x}^{2}+a_{y}^{2}+a_{z}^{2}}$ $\displaystyle D=-d_{x}/\cos_{\alpha}$ $\displaystyle d_{y}=D/\cos_{\beta}$ $\displaystyle d_{z}=D/\cos_{y}$

where $\cos_{\alpha}=a_{x}/A$ , $\cos_{\beta}=a_{y}/A$ , $\cos_{\gamma}=a_{z}/A$ .

The following parameters of postural sway were calculated: 1) Total (TOT), anterior-posterior (AP) and medial-lateral (ML) length of trace; 2) Range in AP and ML directions, quantified as distance between the maximum and minimum displacement; 3) Speed in AP and ML directions, calculated by dividing displacement by task duration; 4) Confidence ellipse, quantified as the 95% of the total area covered in the AP and ML directions using an ellipse to fit data; 5) Root mean squared (RMS) in AP and ML directions, quantified as the square root of the mean squares. Data analysis was performed using Matlab R2018a (Mathworks, Natick, MA, USA).

Table 1
ICCs between optoelectronic system and IMU for all parameters. Data are reported as median and range

	Optoelectronic system		IMU		ICC
Length-TOT (mm)	322.2	(120.7–977.2)	368.2	(156.5–969.2)	0.949
Length-AP (mm)	211.9	(92.4–563.7)	220.3	(84.5–678.5)	0.912
Length-ML (mm)	207.8	(52.9–759.7)	201.5	(73.7–552.3)	0.879
Range-AP (mm)	31.5	(10.9–85.5)	37.1	(16.2–180.9)	0.759
Range-ML (mm)	19.2	(4.0–139.6)	18.2	(4.04–176.1)	0.583
Speed-AP (mm/s)	3.5	(1.5–9.4)	3.7	(1.4–11.3)	0.912
Speed-ML (mm/s)	3.5	(0.9–12.7)	3.4	(1.2–9.2)	0.879
Ellipse (mm ${}^{2}$ )	345.4	(62.3–2704.0)	456.0	(34.4–2743.0)	0.898
RMS-AP (mm)	7.5	(3.0–32.7)	6.6	(1.8–15.9)	0.760
RMS-ML (mm)	2.4	(0.7–9.1)	3.8	(0.8–12.4)	0.701

TOT: total, AP: anterior-posterior, ML: medial-lateral, IMU: inertial measurement unit, ICC: Intraclass correlation coefficient.

2.4 Statistical analysis

Data of the four tasks were gathered and being not normally distributed, were described as median and range. The agreement between IMU and optoelectronic system was analysed using the Intraclass Correlation Coefficient (ICC) interpreted as follows: values greater than 0.9 indicated excellent reliability, between 0.75 and 0.9 good reliability, between 0.5 and 0.75 moderate reliability and less than 0.5 poor reliability [8]. In addition, Bland and Altman plots were calculated for each parameter and Mann-Whitney test was used to assess differences between the two systems for each parameter. Finally, differences among the conditions OE, CE and monopodalic standing (RM $+$ LM) were investigated with the Kruskal-Wallis test and in case of significance Bonferroni post-hoc test was used. Data were analyzed using SPSS Statistics 20.0 for Windows and statistical significance was set at $\alpha=$ 0.05.

Figure 1.

Bland and Altman plots for each parameter, with differences and the 95% limits of agreement. In parenthesis the $p$ -values of the differences between the two systems are reported, which are significant for RMS and Range-AP. Plots shows increased lack of agreement for higher sway.

3. Results

All 120 trials were used for the comparison between IMU and optoelectronic system. ICCs showed three excellent, four good and two moderate correlations (Table 1). Figure 1 shows Bland-Altman plots for each parameter, reporting mean differences, 95% limit of agreement and $p$ -value of the comparison between the two systems. Positive mean differences suggest greater estimation of optoelectronic system (Length-TOT, Length-AP, Speed-AP, Ellipse, RMS-AP), whereas negative values means greater estimation of IMU (Length-ML, Range-AP, Range-ML, Speed-ML, RMS-ML). $P$ -values showed significant differences between the two systems for Range-AP and RMS-AP. When considering OE, OC and monopodalic standing (RM $+$ LM) conditions, the same differences were reported by optoelectronic system and IMU (Table 2).

Table 2
Differences among the four tasks (open eyes, closed eyes, left and right single-leg stance) for all parameters. Data are shown as median and range. Both IMU and optoelectronic system detect the same differences among the three conditions

		OE ( $n=$ 30)		CE ( $n=$ 30)		SLS ( $n=$ 55)		$p$ -value
Length-TOT (mm)	OS	180.1	(92.4–227.3)	233.3	(146.0–332.6)	523.0	(345.3–977.2)	$<$ 0.001 ${}^{\S\S{\dagger}{\dagger}}$
	IMU	241.2	(175.6–367.8)	282.1	(156.5–407.1)	484.0	(290.9–969.2)	$<$ 0.001 ${}^{\S\S{\dagger}{\dagger}}$
Length-AP (mm)	OS	128.3	(92.4–227.3)	151.0	(99.1–228.9)	323.7	(205.5–563.7)	$<$ 0.001 ${}^{\S\S{\dagger}{\dagger}}$
	IMU	179.5	(115.6–275.4)	195.4	(84.5–334.8)	314.1	(188.0–678.5)	$<$ 0.001 ${}^{\S\S{\dagger}{\dagger}}$
Length-ML (mm)	OS	112.0	(52.9–207.8)	136.4	(61.0–219.3)	342.7	(203.9–756.7)	$<$ 0.001 ${}^{\S\S{\dagger}{\dagger}}$
	IMU	128.6	(73.7–201.5)	135.8	(78.4–267.8)	292.3	(153.9–552.3)	$<$ 0.001 ${}^{\S\S{\dagger}{\dagger}}$
Range-AP (mm)	OS	25.1	(10.9–42.1)	20.6	(12.7–39.6)	40.8	(26.7–85.5)	$<$ 0.001 ${}^{\S\S{\dagger}{\dagger}}$
	IMU	26.6	(16.2–74.3)	25.8	(17.4–72.5)	53.3	(29.6–180.9)	$<$ 0.001 ${}^{\S\S{\dagger}{\dagger}}$
Range-ML (mm)	OS	13.0	(5.1–27.4)	14.2	(4.0–26.6)	25.9	(13.3–139.6)	$<$ 0.001 ${}^{\S\S{\dagger}{\dagger}}$
	IMU	11.6	(9.3–14.3)	11.9	(4.0–25.8)	32.3	(13.6–176.1)	$<$ 0.001 ${}^{\S\S{\dagger}{\dagger}}$
Speed-AP (mm/s)	OS	2.1	(1.5–3.8)	2.5	(1.6–3.8)	5.4	(3.4–9.4)	$<$ 0.001 ${}^{\S\S{\dagger}{\dagger}}$
	IMU	3.0	(1.9–4.6)	3.3	(1.4–5.6)	5.2	(3.1–11.3)	$<$ 0.001 ${}^{\S\S{\dagger}{\dagger}}$
Speed-ML (mm/s)	OS	1.9	(0.9–3.5)	2.3	(1.0–3.7)	5.7	(3.4–12.7)	$<$ 0.001 ${}^{\S\S{\dagger}{\dagger}}$
	IMU	2.1	(1.2–3.4)	2.3	(1.3–4.5)	4.9	(2.6–9.2)	$<$ 0.001 ${}^{\S\S{\dagger}{\dagger}}$
Ellipse (mm ${}^{2}$ )	OS	182.4	(68.1–767.6)	178.5	(62.2–578.1)	642.9	(187.7–2704.0)	0.003 ${}^{\S\S{\dagger}{\dagger}}$
	IMU	212.3	(46.9–675.9)	183.9	(34.4–686.8)	704.3	(333.9–2743.0)	$<$ 0.001 ${}^{\S\S{\dagger}{\dagger}}$
RMS-AP (mm)	OS	6.1	(3.0–14.7)	5.0	(3.0–9.8)	9.6	(6.0–32.7)	$<$ 0.001 ${}^{\S\S{\dagger}{\dagger}}$
	IMU	5.4	(1.8–12.7)	3.9	(2.3–8.8)	5.0	(2.4 – 12.4)	$<$ 0.001 ${}^{\S\S{\dagger}{\dagger}}$
RMS ML (mm)	OS	2.0	(0.7–3.8)	1.9	(0.8–4.8)	3.4	(1.5–9.2)	$<$ 0.001 ${}^{\S\S{\dagger}{\dagger}}$
	IMU	2.7	(0.9–6.6)	2.6	(0.8–5.7)	9.1	(5.0–15.9)	$<$ 0.001 ${}^{\S\S{\dagger}{\dagger}}$

OE: open eyes, CE: closed eyes, SLS: single-leg stance, TOT: total, AP: anterior-posterior, ML: medial-lateral, OS: optoelectronic system, IMU: inertial measurement unit. OE vs CE ${}^{\ast}p<$ 0.05 ${}^{\ast\ast}p<$ 0.001; OE vs SLS ${}^{\S}p<$ 0.05 ${}^{\S\S}p<$ 0.001; CE vs SLS ${}^{{\dagger}}p<$ 0.05 ${}^{{\dagger}{\dagger}}p<$ 0.001.

4. Conclusions

The purpose of the study was to analyse agreement between IMU and optoelectronic system in the measurement of postural sway. In this study, tasks were selected to explore the agreement in a broad range of displacements.

ICCs showed that agreement varies from moderate to excellent, but Range-AP and Range-ML have not high ICC values. Moreover, a significant difference between the systems was also found for Range-AP. It is reasonable to speculate that range parameters might be sensitive to outliers due to IMU artefacts, rather than length or speed, where artefacts were more attenuated. In fact, the best agreements were found for Length-TOT length, Speed-ML and Speed-AP, whereas Length-ML and Length-AP showed good agreement. Moreover, Bland and Altman plots seems to show greater disagreement with higher abscissa values, representing wider sway, probably because the algorithm for the calculation of displacements of the IMU considers a fix distance between the floor and the device. This condition is lost proportionally to increase of body sway, depending on sway angle [9].

In conclusion, IMU showed a variable agreement with optoelectronic system (gold standard). The excellent to good agreement of the IMU for length of trace and speed and its user-friendly application suggest an implementation of this device in clinical and sportive practice.

Footnotes

Acknowledgments

The authors thank Dr. Emanuela Morenghi for her suggestions on data analysis and Patricia Taylor for the English language revision.

Conflict of interest

None to report.

References

Paillard

Noé

. Techniques and Methods for Testing the Postural Function in Healthy and Pathological Subjects. Biomed Res Int. 2015; 891390. doi: 10.1155/2015/891390.

De Leva

. Adjustments to Zatsiorsky-Seluyanov’s segment inertia parameters. J Biomech. 1996; 29: 1223–30. doi: 10.1016/0021-9290(95)00178-6.

Saini

Kerrigan

Thirunarayan

Duff-Raffaele

. The vertical displacement of the center of mass during walking: a comparison of four measurement methods. J Biomech Eng. 1998; 120(1): 133–9. doi: 10.1115/1.2834293.

Chiari

Dozza

Cappello

Horak

Macellari

Giansanti

. Audio-biofeedback for balance improvement: an accelerometry-based system. IEEE Trans Biomed Eng. 2005; 52: 2108–11. doi: 10.1109/TBME.2005.857673.

Mancini

Horak

Zampieri

Carlson-Kuhta

Nutt

Chiari

. Trunk accelerometry reveals postural instability in untreated Parkinson’s disease. Parkinsonism Relat Disord. 2001; 17: 557–62. doi: 10.1016/j.parkreldis.2011.05.010.

Doheny

McGrath

Greene

Walsh

McKeown

Cunningham

Crosby

Kenny

Caulfield

. Displacement of centre of mass during quiet standing assessed using accelerometry in order fallers and non-fallers. Conf Proc IEEE Eng Med Biol Soc. 2012; 3300-3. doi: 10.1109/EMBC.2012.6346670.

Mayagoitia

Lötters

Veltink

Hermens

. Standing balance evaluation using a triaxial accelerometer. Gait Posture. 2002; 16: 55–9. doi: 10.1016/S0966-6362(01)00199-0.

Koo

. A guideline of selecting and reporting intraclass correlation coefficient for reliability research. J Chiropr Med. 2016; 15(2): 155–63. doi: 10.1016/j.jcm.2016.02.012.

Fok

Lee

Vette

Masani

. Kinematic error magnitude in the single-mass inverted pendulum model of human standing posture. Gait Posture. 2018; 63: 23–6. doi: 10.1016/j.gaitpost.2018.04.021.