Anda di halaman 1dari 5

Journal of Applied Biomechanics, 2016, 32, 306  -310
© 2016 Human Kinetics, Inc. TECHNICAL NOTE

Validation of a Torso-Mounted Accelerometer for Measures

of Vertical Oscillation and Ground Contact Time During
Treadmill Running
Ricky Watari,1 Blayne Hettinga,1,2 Sean Osis,1,2 and Reed Ferber1,2
1University of Calgary; 2Running Injury Clinic

The purpose of this study was to validate measures of vertical oscillation (VO) and ground contact time (GCT) derived from a
commercially-available, torso-mounted accelerometer compared with single marker kinematics and kinetic ground reaction force
(GRF) data. Twenty-two semi-elite runners ran on an instrumented treadmill while GRF data (1000 Hz) and three-dimensional
kinematics (200 Hz) were collected for 60 s across 5 different running speeds ranging from 2.7 to 3.9 m/s. Measurement agree-
ment was assessed by Bland-Altman plots with 95% limits of agreement and by concordance correlation coefficient (CCC). The
accelerometer had excellent CCC agreement (> 0.97) with marker kinematics, but only moderate agreement, and overestimated
measures between 16.27 mm to 17.56 mm compared with GRF VO measures. The GCT measures from the accelerometer had
very good CCC agreement with GRF data, with less than 6 ms of mean bias at higher speeds. These results indicate a torso-
mounted accelerometer provides valid and accurate measures of torso-segment VO, but both a marker placed on the torso and
the accelerometer yield systematic overestimations of center of mass VO. Measures of GCT from the accelerometer are valid
when compared with GRF data, particularly at faster running speeds.

Keywords: vertical displacement, ground contact time, validity, treadmill running

Biomechanical factors account for a substantial proportion no research has been conducted to determine if such a device can
of variation in running economy,1,2 and among them vertical provide valid measures of GCT.
oscillation (VO) of the center of mass (CoM) and ground contact Therefore, the aim of this study was to test the validity of
time (GCT) are considered to be important elements.3–5 The gold measures of VO and GCT derived from a commercially-available
standard method for measuring VO is the kinetic method, in which torso-mounted accelerometer device. The GCT measures were com-
CoM displacement is calculated by the double integration of pared between the device and values derived from GRF, while VO
acceleration data derived from the ground reaction force (GRF).6–8 measures were compared between the device and 2 other methods:
However, this method requires research laboratory equipment that (1) the kinematic position of a single retro-reflective marker placed
is not easily accessible. In contrast, accelerometers can be used as directly on the device and (2) the CoM VO calculated by double-
a wearable device to facilitate the data collection process and as an integration of GRF data.
inexpensive and accessible method for the estimation of VO. For
example, a double integration of the acceleration provided by an
inertial sensor placed on the sacrum or the lumbar spine has been Methods
used to estimate the motion of CoM.9,10 However, little research Participants
concerning the validity of these devices during running is available.
Similar to measures of VO, gold standard methods for measur- An a priori power analysis (α = .05; β = 0.20; limits of agreement
ing GCT typically involve force plate GRF data or pressure-sensing aspiration 10 ms of GCT and 0.3 cm of VO, based on pilot data)
insoles,11–13 which again depend on a laboratory setting and/or revealed a minimum sample size of 18 subjects to reach statisti-
expensive equipment. Although accelerometers have also been used cal power. Therefore, 22 semi-elite runners were recruited for this
for gait event detection,14,15 the typical approach utilizes indepen- investigation (14 men and 8 women, age = 28.2 ± 10.1 y, height
dent sensors on each lower limb, which increases the cost and may = 1.73 ± 0.75 m, mass = 65.4 ± 8.1 kg). The inclusion criteria for
interfere with the activity of interest.16 A single torso-mounted accel- participation involved running at least 30 minutes per day a mini-
erometer could be less intrusive and more cost-effective, however mum of 3 days per week, free of pain at the time of testing, and
to be able to run comfortably on a treadmill at a speed of 3.9 m/s.
The University of Calgary Conjoint Health Research Ethics Board
approved this study and informed consent was obtained from all
subjects before participation.
Ricky Watari, Blayne Hettinga, Sean Osis, and Reed Ferber are with the
Faculty of Kinesiology, University of Calgary, Calgary, Alberta, Canada.
Blayne Hettinga, Sean Osis, and Reed Ferber are also with the Running
Injury Clinic, Calgary, Alberta, Canada. Reed Ferber is also with the Faculty A heart rate monitor with a built-in accelerometer (Forerunner 620,
of Nursing, University of Calgary, Calgary, Alberta, Canada. Address author Garmin International Inc., Olathe, KS) was attached to a strap on the
correspondence to Reed Ferber at torso of the runner, placed near the xiphoid process of the sternum,
Torso-Mounted Accelerometer Validation   307

onto which a spherical retro-reflective marker was affixed (14-mm all 3 methods (device, kinematic, and kinetic), and the GCT data
diameter, Mocap Solutions, Huntington Beach, CA). The acceler- between the device and kinetic methods. The CCC represents the
ometer unit was attached to a strap with 2 metal snap buttons, and a LoA considering both a location-shift parameter and a scale-shift
double-face adhesive tape was applied to the back of the equipment parameter that measures the difference between the slope and
to improve fixation to the participant’s skin. The participants were perfect correlation (45° angle). The strength of agreement was
asked to run on an instrumented treadmill (Bertec, Columbus, OH) classified as poor (CCC ≤ 0.40), moderate (CCC 0.41–0.60), good
with GRF data collected at a sampling rate of 1000 Hz, while the (CCC 0.61–0.80), very good (CCC 0.81–0.90), or excellent (CCC
three-dimensional (3D) kinematics of the marker were collected for > 0.91).19 The CCC was calculated independently for each running
60 seconds at 200 Hz using an 8-camera Vicon MX3 (Vicon Motion speed condition to evaluate the effect of speed on the measure-
System, Oxford, UK) motion capture system. ments. To further explore potential speed-related trends, Pearson
The subjects ran at 5 different speeds (2.7, 3.0, 3.3, 3.6, and correlation tests (P < .001) were performed to determine the linear
3.9 m/s) at a zero incline, with the order of speeds randomized. relationship between the difference and average of 2 methods used.
These speeds were defined so that they included a range of running In cases where the correlation was significant, the average difference
paces for recreational runners performing a half-marathon race. For of 2 methods was used to determine the mean bias, and the LoA
acclimation to the treadmill, each participant started with a 3-minute was calculated using the standard deviation of the residuals.20 All
walk, followed by 30 seconds of running at each speed condition statistical analyses and kinetic and kinematic data processing were
before data acquisition. A relatively short acclimation time was performed using custom-written software in MATLAB 8.3 (The
chosen to prevent fatigue of the participants during the experiment. MathWorks Inc., Natick, MA).
All participants wore the same type of running shoe (Pegasus, Nike,
Downloaded by W VA Univ Library on 09/16/16, Volume 32, Article Number 3

Beaverton, OR) to standardize footwear condition.

Data Processing and Analysis The comparison between the device and kinematic VO data indi-
The accelerometer under validation (“device” approach) contained cated they were closely related. The analysis revealed an excellent
built-in algorithms to calculate and output measures of VO and GCT agreement (CCC range 0.96–0.98), displaying less than 2.89 mm
from the acceleration data, and the output was sent to a wristwatch, of mean bias (LoA range 7.61–8.72 mm) (Table 1). On the other
via a 2.4 GHz ANT+ wireless communication system, which was hand, both the device and kinematic methods overestimated VO as
later downloaded onto a computer. compared with the gold standard kinetic calculation. Specifically, the
The VO and GCT data derived from the force plate measure- device overestimated VO, on average and across running speeds, by
ments (“kinetic” approach) were considered the gold standard between 16.15 mm and 17.56 mm (LoA range 17.97–21.43 mm) as
method for the current study. Measures of VO were calculated using compared with the kinetic method, with only moderate CCC values
methods adapted from Gutierrez-Farewik et al,8 in which accel- (range 0.38–0.55) on average (Table 1). The kinematic method
eration of the CoM was calculated from raw GRF data by double also overestimated VO, on average and across running speed, by
integrating the acceleration data over a single step. Force data were between 18.69 mm and 20.24 mm (LoA range 19.03–23.04 mm)
filtered using a low-pass fourth-order Butterworth filter with a 50 as compared with the kinetic method with poor-to-moderate CCC
Hz cut-off; after the first integration, a fourth-order Butterworth values (range 0.32–0.48) (Table 1).
high-pass filter with 0.5 Hz cut-off was applied to remove signal There was a positive correlation between the difference and
wander. GCT was measured as the period when vertical GRF data mean of VO estimations of both the device and kinematic methods
were above a threshold of 20 N and was measured in milliseconds as compared with the kinetic method in the Bland-Altman plots. This
(ms). Finally, the displacement of the accelerometer was measured correlation was similar across all speeds (0.65 < r < .70) (Figure 1).
using the raw marker kinematic data (“kinematic” approach), after The device underestimated measures of GCT, as compared
being filtered using a band-pass second-order recursive Butterworth with the kinetic method, with a mean bias less than 6 milliseconds
filter with a 0.5 Hz to 10 Hz pass-band. for the top 3 speeds (3.3, 3.6, and 3.9 m/s), and very good agree-
The validity of measurements was assessed using Bland- ment (CCC range 0.82–0.87). Larger differences were evident at
Altman plots with 95% limits of agreement (LoA; mean difference 2.7 m/s and 3.0 m/s, reaching 10.14 milliseconds and 17.02 mil-
of 2 methods ± 1.96 SD)17 and by concordance correlation coeffi- liseconds, respectively, with good agreement (CCC 0.76 and 0.69,
cients (CCC),18 with pairwise comparisons of the VO data between respectively; Figure 2).

Table 1  Mean differences (95% limits of agreement [LoA]) and the concordance correlation coefficient (CCC) for
comparisons between the different methods for measures of vertical oscillation and ground contact time
Vertical Oscillation (mm) Ground Contact Time (ms)
Device—Kinetic Kinematic—Kinetic Device—Kinematic Device—Kinetic
Running Speed Mean (LoA) CCC Mean (LoA) CCC Mean (LoA) CCC Mean (LoA) CCC
2.7 m/s 17.56 (17.97) 0.42 19.85 (19.04) 0.37 –2.29 (7.61) 0.98 –17.02 (43.68) 0.69
3.0 m/s 17.33 (18.15) 0.38 19.86 (19.03) 0.32 –2.53 (8.72) 0.97 –10.14 (45.71) 0.77
3.3 m/s 17.35 (18.22) 0.50 20.24 (20.03) 0.40 –2.89 (8.16) 0.98 –5.82 (43.93) 0.87
3.6 m/s 16.68 (19.14) 0.47 19.34 (21.39) 0.43 –2.66 (8.49) 0.97 –2.59 (38.72) 0.83
3.9 m/s 16.15 (21.43) 0.55 18.69 (23.04) 0.49 –2.42 (7.66) 0.98 –1.40 (44.02) 0.84

JAB Vol. 32, No. 3, 2016

Downloaded by W VA Univ Library on 09/16/16, Volume 32, Article Number 3

Figure 1 — Bland-Altman plots of vertical oscillation (VO) at a representative speed of 3.6 m/s. The difference in VO measured by 2 methods for each
subject is plotted against the average VO from 2 methods, with the mean bias represented by the straight line and the 95% limits of agreement of the
difference between methods represented by the shaded region. Comparisons are presented between (a) device and kinetic, (b) kinematic and kinetic,
and (c) device and kinematic.

Figure 2 — Bland-Altman plots comparing the device to the kinetic calculation of ground contact time (GCT) across 5 speeds. The difference between
GCT for each subject is plotted against the average GCT from both the device and kinetic. The average difference between device and kinetic is plotted
with the straight line and the 95% limits of agreement of the difference is plotted with the shaded region.

308 JAB Vol. 32, No. 3, 2016

Torso-Mounted Accelerometer Validation   309

Discussion decreased the LoA range (15.06 mm at 3.0 m/s to 17.97 mm at 3.9
m/s) (Figure 3); and improved the agreement between measures
The purpose of the current study was to test the validity of measures (CCC between 0.91–0.93). However, this correction must be
of VO and GCT derived from a commercially-available torso- used with caution, since there is a risk of overfitting in the linear
mounted accelerometer device. Overall, the device and kinematic regression, and further investigation is needed to fully validate this
methods demonstrated excellent agreement, with small measure- correction.
ment bias and a narrow LoA range. However, both of these methods The agreement of GCT measures between the device and
overestimated measures of whole-body CoM vertical oscillation, kinetic approaches was very good for the 3.3 m/s to 3.9 m/s run-
as calculated using the gold standard kinetic method. This overes- ning speeds. However, the device underestimates GCT by a slightly
timation bias seems reasonable considering that both the device larger margin at slower running speeds. To our knowledge, only
and kinematic methods represent measures from a single segment one other study has investigated the validity of GCT measures for
of the body, which results in the overestimation of CoM VO when a torso-mounted accelerometer device.22 These authors reported
compared with a more comprehensive multisegment approach.7 In errors greater than 100 milliseconds, far in excess to the results of
addition, the mean bias and LoA range of the device and kinematic the current study, and intraclass correlation coefficients near 0.50,
methods are consistent across all tested running speeds, which sug- for a device attached to the pelvis. The device in the current study
gests that external impact forces have little influence over measures exhibited a maximum average error of 17 milliseconds, which is
of VO.21 However, the measurement bias of the device and kinematic comparable to previous studies that have used a shank-mounted
methods seems to increase with greater values of VO in a linear accelerometer, and also reported a similar effect of speed.23 This
manner (Figure 1). This linear relationship should be taken into level of precision is also in line with modern techniques for estimat-
Downloaded by W VA Univ Library on 09/16/16, Volume 32, Article Number 3

consideration when using VO measures derived from the torso, and ing ground contact from joint kinematics.24
a simple linear correction could be implemented to the algorithm to We conclude that the device used in the current study is a viable
provide greater accuracy over a range of VO measures. alternative to expensive, specialized equipment for measuring cer-
Based on the data from this study, we suggest a linear correc- tain biomechanical variables in the field. However, the measures
tion of VO measures, considering a multivariate linear regression provided by the device for torso-segment vertical oscillation yield
using both the device output and the running speed, which results systematic overestimations of CoM vertical displacement when
in the following equation: compared with the gold standard kinetic method. Therefore, we
Vertical oscillation = 16.21 + 0.658 × Device measure + propose a linear correction for VO measures, which significantly
decreased the mean bias, decreased the LoA range, and improved
0.198 × Speed
the agreement between measures. Finally, measures of GCT by
This correction significantly decreased the mean bias, with the the device present very good agreement with the kinetic approach,
highest difference at 3.6 m/s, with 0.10 mm of mean bias; slightly and are considered a valid characterization of average GCT for

Figure 3 — Bland-Altman plots of corrected vertical oscillation (VO) across 5 speeds. The difference in VO for each subject is plotted against the
average VO from the device, after a linear correction, and the kinetic method. The average difference between measures is plotted with the straight line
and the 95% limits of agreement of the difference is represented by the shaded region.

JAB Vol. 32, No. 3, 2016

310  Watari et al

a runner, particularly at faster running speeds. Regardless, the 11. Zeni JA, Jr, Richards JG, Higginson JS. Two simple methods for
generalization of these results should be done with caution, since determining gait events during treadmill and overground walking
other aspects could also influence the measurements, such as other using kinematic data. Gait Posture. 2008;27(4):710–714. PubMed
running modalities and differences in anthropometric variables (eg, doi:10.1016/j.gaitpost.2007.07.007
body mass index). 12. Hansen AH, Childress DS, Meier MR. A simple method for deter-
mination of gait events. J Biomech. 2002;35(1):135–138. PubMed
Acknowledgment doi:10.1016/S0021-9290(01)00174-9
13. Hausdorff JM, Ladin Z, Wei JY. Footswitch system for measurement
We thank the Ministry of Education of Brazil (CAPES Foundation; Grant of the temporal parameters of gait. J Biomech. 1995;28(3):347–351.
no. #9408134) and Alberta Innovates: Health Solution (AI:HS) for finan- PubMed doi:10.1016/0021-9290(94)00074-E
cial support. 14. Sabatini AM, Martelloni C, Scapellato S, Cavallo F. Assessment of
walking features from foot inertial sensing. IEEE Trans Biomed Eng.
2005;52(3):486–494. PubMed doi:10.1109/TBME.2004.840727
References 15. Jasiewicz JM, Allum JHJ, Middleton JW, et al. Gait event detec-
1. Kyrolainen H, Belli A, Komi PV. Biomechanical factors affecting tion using linear accelerometers or angular velocity transducers
running economy. Med Sci Sports Exerc. 2001;33(8):1330–1337. in able-bodied and spinal-cord injured individuals. Gait Posture.
PubMed doi:10.1097/00005768-200108000-00014 2006;24(4):502–509. PubMed doi:10.1016/j.gaitpost.2005.12.017
2. Williams KR, Cavanagh PR. Relationship between distance running 16. Gonzalez RC, Lopez AM, Rodriguez-Uria J, Alvarez D, Alvarez
mechanics, running economy, and performance. J Appl Physiol. JC. Real-time gait event detection for normal subjects from lower
Downloaded by W VA Univ Library on 09/16/16, Volume 32, Article Number 3

1987;63(3):1236–1245. PubMed trunk accelerations. Gait Posture. 2010;31(3):322–325. PubMed

3. Gullstrand L, Halvorsen K, Tinmark F, Eriksson M, Nilsson J. Mea- doi:10.1016/j.gaitpost.2009.11.014
surements of vertical displacement in running, a methodological 17. Bland JM, Altman DG. Statistical methods for assessing agree-
comparison. Gait Posture. 2009;30(1):71–75. PubMed doi:10.1016/j. ment between two methods of clinical measurement. Lancet.
gaitpost.2009.03.001 1986;1(8476):307–310. PubMed doi:10.1016/S0140-6736(86)90837-8
4. Di Michele R, Merni F. The concurrent effects of strike pattern 18. Lin LI. A concordance correlation-coefficient to evaluate reproducibil-
and ground-contact time on running economy. J Sci Med Sport. ity. Biometrics. 1989;45(1):255–268. PubMed doi:10.2307/2532051
2014;17(4):414–418. PubMed doi:10.1016/j.jsams.2013.05.012 19. Ferber R, Kendall KD, McElroy L. Normative and critical criteria
5. Chapman RF, Laymon AS, Wilhite DP, McKenzie JM, Tanner DA, for iliotibial band and iliopsoas muscle flexibility. J Athl Train.
Stager JM. Ground contact time as an indicator of metabolic cost in 2010;45(4):344–348. PubMed doi:10.4085/1062-6050-45.4.344
elite distance runners. Med Sci Sports Exerc. 2012;44(5):917–925. 20. Bland JM, Altman DG. Measuring agreement in method compari-
PubMed doi:10.1249/MSS.0b013e3182400520 son studies. Stat Methods Med Res. 1999;8(2):135–160. PubMed
6. Thirunarayan MA, Kerrigan DC, Rabuffetti M, Croce UD, Saini M. doi:10.1191/096228099673819272
Comparison of three methods of estimating vertical displacement 21. Nigg BM, Bahlsen HA, Luethi SM, Stokes S. The influence of
of center of mass during level walking in patients. Gait Posture. running velocity and midsole hardness on external impact forces
1996;4:306–314. doi:10.1016/0966-6362(95)01058-0 in heel-toe running. J Biomech. 1987;20(10):951–959. PubMed
7. Gard SA, Miff SC, Kuo AD. Comparison of kinematic and kinetic doi:10.1016/0021-9290(87)90324-1
methods for computing the vertical motion of the body center of 22. Gouttebarge V, Wolfard R, Griek N, de Ruiter CJ, Boschman JS, van
mass during walking. Hum Mov Sci. 2004;22(6):597–610. PubMed Dieen JH. Reproducibility and Validity of the Myotest for Measuring
doi:10.1016/j.humov.2003.11.002 Step Frequency and Ground Contact Time in Recreational Runners. J
8. Gutierrez-Farewik EM, Bartonek A, Saraste H. Comparison and evalu- Hum Kinet. 2015;45(1):19–26. PubMed doi:10.1515/hukin-2015-0003
ation of two common methods to measure center of mass displacement 23. Purcell B, Channells J, James D, Barrett R. Use of accelerometers
in three dimensions during gait. Hum Mov Sci. 2006;25(2):238–256. during running. Paper presented at: BioMEMS and Nanotechnology
PubMed doi:10.1016/j.humov.2005.11.001 II 2006; Brisbane, Australia.
9. Moe-Nilssen R. A new method for evaluating motor control in gait 24. Osis ST, Hettinga BA, Leitch J, Ferber R. Predicting timing of foot
under real-life environmental conditions. Part 1: The instrument. strike during running, independent of striking technique, using princi-
Clin Biomech (Bristol, Avon). 1998;13(4-5):320–327. PubMed pal component analysis of joint angles. J Biomech. 2014;47(11):2786–
doi:10.1016/S0268-0033(98)00089-8 2789. PubMed doi:10.1016/j.jbiomech.2014.06.009
10. Moe-Nilssen R. Test-retest reliability of trunk accelerometry during
standing and walking. Arch Phys Med Rehabil. 1998;79(11):1377–
1385. PubMed doi:10.1016/S0003-9993(98)90231-3

JAB Vol. 32, No. 3, 2016