Content uploaded by Jean-Francois Esculier
Author content
All content in this area was uploaded by Jean-Francois Esculier on May 01, 2020
Content may be subject to copyright.
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JSEN.2020.2982568, IEEE Sensors
Journal
IEEE SENSORS JOURNAL, VOL. XX, NO. XX, MONTH X, XXXX 1
XXXX-XXXX © XXXX IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Abstract—A growing number of people all over the world are running. Gathering in-field data with wearable sensors
is attractive for runners, clinicians and coaches to improve running performance, avoid injury or return to running
after an injury. However, it is yet to be proven that commercially available wearable sensors provide valid data. The
objective of this study was to assess the validity of five wearable sensors (Moov NowTM, MilestonePod, RunScribeTM,
TgForce and Zoi) to measure ground reaction force related metrics, step rate, foot strike pattern, and vertical
displacement of the center of mass during running. Concurrent/criterion validity was assessed against a laboratory-
based system using Pearson’s correlation coefficients and ANOVAs. Step rate measurement provided by all
wearable sensors was valid (all r>0.96 and p<0.001). Only Zoi provided valid vertical displacement of the center of
mass (r=0.81, p<0.001); only TgForce provided meaningful estimates of instantaneous vertical loading rate (r=0.76,
p<0.001); only MilestonePod could discriminate between a rear-, mid- and fore-foot strike pattern during running
(p<0.001). None of the wearable sensors was valid for estimating peak braking force. In conclusion, only a few
metrics provided by these commercially available wearable sensors are valid. Potential buyers should therefore be
aware of such limitations when monitoring running gait variables.
Index Terms— accelerometer; Bland-Altman; inertial measurement unit; impact force; kinematic.
I. INTRODUCTION
UNNING is a very popular activity that can greatly
improve individual health status [1]. Researchers have
This research was funded by The Running Clinic and the Ordre
Professionnel de la Physiothérapie du Québec (OPPQ) and supported
in part by the Sentinel North program of Université Laval (Canada First
Research Excellence Fund). B. PdF. is supported by a postdoctoral
fellowship from the Fonds de recherche du Québec—Santé (FRQS)
and the Sentinel North program of Université Laval. J-S.R. is
supported by a salary award from the Canadian Institutes of Health
Research (CIHR). J-F.E. is supported by a postdoctoral fellowship from
the CIHR.
B. Pairot de Fontenay is with the Centre for Interdisciplinary
Research in Rehabilitation and Social Integration, Quebec City,
Quebec, Canada G1M 2S8 and The Running Clinic, Lac Beauport,
Quebec, Canada (e-mail: benoit.pdf@fmed.ulaval.ca)
J. S. Roy is with the Centre for Interdisciplinary Research in
Rehabilitation and Social Integration, Quebec City, Quebec, Canada
G1M 2S8 and the Department of Rehabilitation, Faculty of Medicine,
Université Laval, Quebec City, Quebec, Canada G1V 0A6.
B. Dubois is with The Running Clinic, Lac Beauport, Quebec,
Canada.
L. Bouyer is with the Centre for Interdisciplinary Research in
Rehabilitation and Social Integration, Quebec City, Quebec, Canada
G1M 2S8 and the Department of Rehabilitation, Faculty of Medicine,
Université Laval, Quebec City, Quebec, Canada G1V 0A6.
J.F. Esculier is with the Department of Physical Therapy,
University of British Columbia, Vancouver, British Columbia, Canada
and The Running Clinic, Lac Beauport, Quebec, Canada.
been interested in better understanding the way humans run
for many years [2]–[4]. For healthcare professionals and
coaches alike, improving running gait is fundamental for
injury prevention [5], rehabilitation [6] and performance
improvement [7]. Thus, gathering in-field data could aid in
advancing knowledge and providing personalized advice to
runners aiming to improve their performance or avoiding
being sidelined due to injury.
A number of kinematic and kinetic variables have been
associated with running injuries. Impact characteristics, in
particular the vertical loading rate of the ground reaction force
during the stance phase, has been suggested as a key factor for
stress fractures and overall injuries [8], [9]. Importantly, gait
retraining to decrease vertical loading rate has shown
promising reductions in injury rates [10]. Although less
studied, high peak braking force of the ground reaction force
has also been identified as a potential contributor to injury risk
[11]. Step rate [12] and foot strike pattern [13] also represent
important variables as they have been linked with running
injuries. For runners seeking to improve performance, lesser
peak braking force [14] and vertical displacement of the center
of mass [15], and increased step rate [16] have been correlated
with better running economy. A non-rearfoot strike pattern has
Validating commercial wearable sensors for
running gait parameters estimation.
Pairot
de
Fontenay
B
,
Roy
JS
,
Du
bois
B
,
Bouyer
L
and
Esculier
JF
R
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JSEN.2020.2982568, IEEE Sensors
Journal
3 IEEE SENSORS JOURNAL, VOL. XX, NO. XX, MONTH X, XXXX
also been linked with performance in observational studies
[17]–[19].
Healthcare professionals and coaches wishing to analyze
or alter someone’s running gait need to gather objective data
from the targeted individuals. However, laboratory-based
motion analysis systems are expensive and not accessible to
most. Furthermore, they may not reflect a runner’s actual gait
pattern while running in their usual environment [20].
Wearable sensors represent a potential low-cost solution to
collect in-field data [21]. Accurate wearables could be used to
monitor the ability and compliance of runners when gait
modifications are recommended to reduce injury risk, return to
running after an injury or improve performance [22].
Current commercial devices usually include an inertial
measurement unit and use manufacturer proprietary
algorithms with specific combinations of data from
accelerometers, gyroscopes, and magnetometers. Some
manufacturers claim that their device provide metrics related
to vertical and antero-posterior ground reaction forces, step
rate, foot strike pattern and vertical displacement of the center
of mass, which may all be valuable for runners seeking to
improve running performance and avoid injuries.
However it is yet to be proven that these wearable
sensors offer valid data to runners, clinicians and coaches [23],
[24]. Sensor location, fixation methods, weight of the device
and both data collection/processing methods may influence
measurement of running gait characteristics and therefore
affect the quality of the extracted data [25], [26].
The main objective of this study was therefore to assess
the concurrent validity of several popular, commercially
available wearable sensors in estimating vertical ground
reaction force (vertical loading rates) during running against a
laboratory-based instrumented treadmill and motion capture
system (gold standard). The second aim was to assess
concurrent or criterion validity of the same wearable sensors
in estimating antero-posterior ground reaction force (peak
braking force), measuring step rate and vertical displacement
of the center of mass, and determining foot strike pattern. Our
hypothesis was that estimates of vertical loading rates and
peak braking force would be more accurate with devices
attached to the tibia, while foot strike pattern would be more
accurate with shoe-mounted sensors. We also hypothesized
that most wearable sensors would show good to excellent
estimates of step rate and vertical displacement of center of
mass.
II. MATERIALS AND METHODS
A. Population
A convenience sample of 32 healthy participants was
recruited through the electronic mailing list of employees and
students of Université Laval (Table 1). Participants were
excluded if they reported any lower extremity musculoskeletal
injury or surgery within 2 years prior to participation, or
neurological disorder that could interfere with the task. All
participants had to be physically active (Tegner score >5/10)
and aged 18 to 45 years old. Participation in the study was
voluntary and informed consent was provided by all
participants. This study was approved by the local ethics
committee.
TABLE 1
PARTICIPANTS’ CHARACTERISTICS
Characteristics
n (female, male)
32 (13, 19)
Age, years (x±sd)
27.0 ± 5.5
Height, cm (x±sd)
174.4 ± 8.5
Weight, kg (x±sd)
69.1 ± 11.4
Teg
ner score (x±sd
, range
)
6.
7
± 1.4
(
5
-
10)
B. Wearable Sensors
As vertical loading rate of impact is a potential key outcome
associated with running related injuries, we selected wearable
sensors that provided at least one measure of impact [8], [9].
Based on this criterion, we found five wearable sensors
available on the market when this study was designed: Moov
NowTM (Moov, San Mateo, California, USA), MilestonePod
(Milestone Sports, Long beach, California, USA),
RunScribeTM (Montara, California, USA), Zoi (Runteq,
Tampere, Finland) and TgForce (Kelsec Systems Inc.,
Montréal, Canada). Other parameters associated with running
injuries or running performance such as metrics related to
antero-posterior ground reaction force (peak braking force),
step rate, foot strike pattern and vertical displacement of the
center of mass were also collected when available. Detailed
characteristics including sensor placement of the selected
sensors are reported in Appendix 1.
C. Data Collection
1) Wearable sensors
Wearable sensors were placed according to the
manufacturers’ instructions (Appendix 1) and data were
collected with their dedicated commercial application (for
TgForce, a beta Windows version software was used to collect
data). Devices were not calibrated, as calibration is not a
necessary step as per manufacturers’ recommendations.
2) Lab-based System
Three-dimensional motion analysis was performed using
a 9-camera VICON motion capture system and VICON Nexus
software (VICON motion systems, CA, USA). Kinematic data
were collected at 200 Hz. Rigid triads of non-colinear
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JSEN.2020.2982568, IEEE Sensors
Journal
3 IEEE SENSORS JOURNAL, VOL. XX, NO. XX, MONTH X, XXXX
reflective markers were placed over the lumbosacral junction,
and on the lateral part of the foot, shank and thigh, bilaterally.
Triads attached to the thighs were made of custom-molded
thermoplastic and were secured with Velcro straps to
minimize movement artefacts. An additional 21 markers were
temporarily placed bilaterally over specific anatomical
landmarks for a standing calibration trial (anterior and
posterior tip of shoe, first and fifth metatarsal heads, medial
TABLE 2
EXTRACTED DATA CHARACTERISTICS FOR THE TESTED WEARABLE
SENSORS AND VALIDATION TESTING WITH LAB-BASED SYSTEM
Wearable
sensor
Extracted
data for
analysis
(averaged
over 1 min)
Validation testing with lab-
based system
Moov NowTM
(v5.2.4809.252)
- Step rate
- Impact
MoovNow_Step rate / Lab-
based_Step rate
MoovNow_Impact / Lab-
based_AVLR&IVLR
Milestone Pod
(v3.4.10)
- Foot strike
type (heel,
mid, toe)
- Step rate
- Rate of
impact (low,
mid, high)
MilestonePod_Footstrike / Lab-
based_Footstrike
MilestonePod_Step rate / Lab-
based_Step rate
MilestonePod _Rate of impact /
Lab-based_AVLR&IVLR
RunScribeTM
(v2.7.1)
- Step rate
- Shock
(combination
of impact
and braking
score)
- Impact Gs
- Braking Gs
- Foot strike
type
RunScribe_Step rate / Lab-
based_Step rate
RunScribe _Shock / Lab-
based_AVLR&IVLR
RunScribe _Impact-Gs / Lab-
based_AVLR&IVLR
RunScribe _Braking-Gs / Lab-
based_PeakBrakingForce
RunScribe_Footstrike / Lab-
based_FootAngle
Zoi
Runteq (2012-
2016)
- Step rate
- Braking
- Impact
- Bouncing
Zoi_Step rate / Lab-based_Step
rate
Zoi_Braking / Lab-
based_PeakBrakingForce
Zoi_Impact / Lab-
based_AVLR&IVLR
Zoi_Bouncing / Lab-
based_VertDispCM
Wearable
sensor
Extracted
data for
analysis
(averaged
over 1 min)
Validation testing with lab-
based system
TgForce
(v2.0.0.10)
- Step rate
- Max
vertical
acceleration
- Max
sagittal
acceleration
- Max vector
acceleration
TgForce_Step rate / Lab-
based_Step rate
TgForce_Z / Lab-based_
AVLR&IVLR
TgForce_X / Lab-
based_PeakBrakingForce
TgForce_3D / Lab-based_
AVLR&IVLR&PeakBrakingForce
AVLR: average vertical loading rate; IVLR: instantaneous vertical loading
rate; VertDispCM: vertical displacement of the center of mass.
and lateral malleoli, medial and lateral femoral epicondyles,
landmarks for a standing calibration trial (anterior and
posterior tip of shoe, first and fifth metatarsal heads, medial
and lateral malleoli, medial and lateral femoral epicondyles,
anterior superior iliac spine, iliac crest) prior to movement
data collection (Fig. 1). All markers were placed by the same
investigator who was an experienced physical therapist.
Simultaneously to three-dimensional motion analysis, and
after calibration, the ground reaction forces were collected at
1000 Hz using an instrumented treadmill with force plates
(Bertec, Columbus, OH, USA).
Fig. 1. Marker and triad placement for calibration.
D. Data Analysis
1) Wearable Sensors
Data were extracted from the different applications.
Running gait parameters associated with running injuries and
running performance were extracted (ground reaction force
[vertical and antero-posterior], step rate, foot strike pattern and
vertical displacement of the center of mass). Characteristics of
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JSEN.2020.2982568, IEEE Sensors
Journal
3 IEEE SENSORS JOURNAL, VOL. XX, NO. XX, MONTH X, XXXX
the extracted data for each wearable system are described in
Table 2.
2) Lab-based System
Concurrent or criterion validity of the wearable sensors
was evaluated using a gold standard laboratory-based
instrumented treadmill and motion capture system. Data
analyses were performed off-line using TheMotionMonitor
(Chicago, IL, USA) and a custom software written in Matlab
(MathWorks Inc., MA, USA). Three-dimensional hip, knee,
and ankle joint kinematics were calculated. Raw ground
reaction force data were filtered with a 50 Hz fourth order low
pass Butterworth filter. The vertical ground reaction force data
were then used to identify initial contact and toe off events
during running (20 N threshold).
Average and instantaneous vertical loading rates
(AVLR/IVLR) were calculated as described in Tenforde et al.
[27] to allow calculation in mid/forefoot strikers (no impact
peak). Briefly, the point of interest was determined as the first
point, above 75% of the participant’s body weight (bw) for
which the instantaneous slope of the vertical ground reaction
force was below 15 bw/s. AVLR was then calculated as the
average slope in the largest continuous region in the 20–80%
point of interest force region for which the slope was above
15% bw/s. IVLR was calculated as the peak slope between 20
and 100% of the force at the point of interest.
The step rate was calculated as the number of steps per
minute. The foot strike angle was calculated to determine the
foot strike pattern [28]. The foot strike angle was defined as
the foot angle at impact minus foot angle during static
calibration. Foot strike pattern was determined as followed:
midfoot strike = -1.68°< FSA < 8.08°, rearfoot strike = FSA >
8.08°, and forefoot strike = FSA < -1.68° [28].
Peak braking force was defined as the maximal posterior
force observed from foot strike to toe-off. Vertical
displacement of the center of mass was defined as the
difference between the highest and lowest position of the
center of mass for each stride. All variables were averaged
over one minute.
3) Statistical Analysis
Statistical analyses were conducted using the R software
(version R.2.7.2.; R Foundation for Statistical Computing,
Vienna, Austria). Pearson’s correlation coefficients were
calculated to determine the association between continuous
variables extracted by each wearable sensor and the lab-based
system. Associations were classified as negligible (0.0-0.3),
low (0.31-0.5), moderate (0.51- 0.7), good (0.71- 0.9), or
excellent (0.91-1.0) [29]. Good to excellent correlations were
considered clinically meaningful. Bland-Altman plots were
also computed when the variable collected by the wearable
sensor and the lab-based system were the same (step rate and
vertical displacement of the center of mass). Bias and 95%
limits of agreement were used to determine the accuracy of the
metrics provided by the wearable sensors.
Zoi provided a percent of rear-, mid-, and fore-foot strike
during the run. This format of data did not allow any statistical
analysis to test the validity of this metric.
To compare discrete and continuous variables, a one-way
ANOVA was performed (e.g. the three categories of
MilestonePod_Rate-of-impact to AVLR and IVLR). For all
tests, an alpha level of 0.05 was used for statistical
significance.
III. RESULTS
Data collected from all 32 participants were used for
RunScribeTM. However, due to technical issues with other
wearable sensors, data of 31 participants were analyzed for
MilestonePod and TgForce, 30 for Zoi and 25 for Moov
NowTM, respectively.
A. Step rate
Step rate measured by Moov NowTM, MilestonePod,
RunScribeTM, Zoi and TgForce showed excellent correlations
to the step rate obtained from the lab-based system (all r>0.96,
all p<0.001) (Table 3). According to Bland-Altman plots, bias
ranged from +0.93 to +4.51 and 95% limits of agreement from
±0.92 to ±6.13 steps per minute, respectively (Table 3).
B. Vertical Displacement of the Center of Mass
Zoi_Bouncing showed a good correlation to the vertical
displacement of the center of mass obtained with the lab-based
system (r=0.81, p<0.001) (Table 3). According to the Bland-
Altman plot, bias is -1.78cm and 95% limits of agreement
±1.50cm, respectively (Fig. 2, Table 3).
Fig. 2. Bland-Altman plots for vertical displacement of the
center of mass during running (cm). Difference
(Zoi_Bouncing - lab-based) versus average of values
measured by the lab-based system and Zoi_Bouncing.
Bold dotted line represents bias and dotted lines 95% limits
of agreement.
C. Foot Strike Pattern
Foot strike classification from MilestonePod
(MilestonePod_Footstrike [scored 1 to 3, and defined by the
manufacturer as heel, mid, toe, respectively]) could
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JSEN.2020.2982568, IEEE Sensors
Journal
3 IEEE SENSORS JOURNAL, VOL. XX, NO. XX, MONTH X, XXXX
significantly discriminate foot strike patterns when compared
with foot strike angles obtained from the lab-based system
(one-way ANOVA, p<0.001) (Fig. 3).
Fig. 3. MilestonePod_Footstrike (scored 1 to 3, and
defined by the manufacturer as heel, mid, toe,
respectively) according to foot strike angle (FSA: midfoot
strike = -1.68° < FSA < 8.08°, rearfoot strike = FSA >
8.08°, and forefoot strike = FSA < -1.68°) obtained from
the lab-based system. Grey dots represent a discrepancy
between data from Milestone-Pod and the lab-based
system. One-way ANOVA, p<0.001.
RunScribe_Foostrike was not correlated to the foot strike
angle obtained from the lab-based system (r=0.090, p=0.688)
(Table 3).
D. Vertical Ground Reaction Force
Rate of impact classification from MilestonePod
(MilestonePod_Rate of impact [scored 1 to 3, and defined by
the manufacturer as low, medium and high rate of impact,
respectively]) could not be determined from the lab-based
system measures of AVLR and IVLR (one-way ANOVA,
both p>0.05) (Fig. 4).
RunScribe_Schock and RunScribe_Impact-Gs were not
correlated to either AVLR (r=-0.421, p=0.992 and r=-0.133,
p=0.767, respectively) or IVLR (r=-0.312, p=0.959 and r=-
0.106, p=0.720, respectively) obtained from the lab-based
system (Table 3).
A low correlation was found between Moov-
Now_Impact and IVLR (r=0.374, p=0.033), but not AVLR
(r=0.289, p=0.080) obtained from the lab-based system (Table
3).
TgForce_Z showed a moderate correlation to AVLR
(r=0.681, p<0.001) and a good correlation to IVLR (r=0.758,
p<0.001) obtained from the lab-based system (Fig. 5 and
Table 3).
E. Antero-posterior Ground Reaction Force
RunScribe_Braking-Gs was not significantly correlated
to the peak braking force obtained from the lab-based system
(r=-0.376, p=0.983) (Table 3).
Fig. 4. MilestonePod_Rate-of-impact (scored 1 to 3, and
defined by the manufacturer as low, medium and high rate
of impact, respectively) according to both A) average
vertical loading rate (AVLR) and B) instantaneous vertical
loading rate (IVLR) obtained from the lab-based system.
Grey dots represent a discrepancy between data from
Milestone-Pod and the lab-based system. One-way
ANOVA, p>0.05.
Fig. 5. Comparison of instantaneous vertical loading rate
(IVLR) from the lab-based system and TgForce_Z
(r=0.758, p<0.001). bw: body weight
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JSEN.2020.2982568, IEEE Sensors
Journal
3 IEEE SENSORS JOURNAL, VOL. XX, NO. XX, MONTH X, XXXX
TgForce_X showed a low correlation (r=-0.470,
p=0.004) and TgForce_3D a moderate correlation (r=-0.532,
p=0.001) to peak braking force (Fig. 6 and Table 3).
Fig. 6. Peak braking force from the lab-based system according
to TgForce_3D (r=-0.532, p=0.001). bw: body weight
IV. DISCUSSION
The first aim of this study was to assess the concurrent
validity of several wearable sensors in measuring vertical
ground reaction force (vertical loading rates) during running
against a laboratory-based instrumented treadmill and motion
capture system. The second aim was to assess the concurrent
or criterion validity of the same wearable sensors at estimating
antero-posterior ground reaction force (peak braking force),
measuring step rate and vertical displacement of the center of
mass, and determining foot strike pattern. Five commercially
available wearable sensors were tested: MilestonePod, Moov
NowTM, TgForce, Zoi and RunScribeTM. Overall, measures
provided by the wearable sensors ranged from not correlated
to strongly correlated to the measurement from our lab-based
system, thereby partially confirming our hypotheses.
The five sensors provided different metrics that were
thought to represent vertical ground reaction forces. When
compared to actual vertical loading rates (average and
instantaneous), a gait parameter associated with running injury
and running injury and running injury prevention [8]–[10], the
correlations with these metrics ranged from none to high. This
confirmed our hypothesis that only devices attached to the
tibia (TgForce and Moov NowTM) showed significant
correlations with lab-based vertical loading rates. However,
TgForce was the only wearable sensor to be valid (r>0.70) in
estimating IVLR. In Hennig et al.’s study [30], peak tibial
acceleration was very strongly correlated to IVLR (r=0.98)
during running. The 3-gram accelerometer used in their study
was bone mounted on the distal medial end of the tibia. In
2003, Laughton et al. [31] showed significant correlations
between vertical loading rates and peak tibial acceleration
using a skin mounted accelerometer on the distal medial end
of the tibia (3.28 grams) during running. They also found an
influence of foot strike pattern with higher correlations in
forefoot strikers than rearfoot strikers for AVLR. More
recently, Ngoh et al. [32] validated the use of a neural network
TABLE 3
SUMMARY OF RESULTS FOR THE FIVE TESTED
WEARABLE SENSORS.
Lab-based Moov NowTM
S
tep rate
I
mp
a
ct
Step rate r=0.976, p<0.001
(
2.26±1.98
spm
)
AVLR
r=0.289,
p=0.080
IVLR
r=0.374,
p=0.033
Lab-based Milestone-Pod
S
tep r
ate
S
tep rate
r=
0.9
9
7, p<0.001
(
1.
59±1.44
spm
)
Lab-based
RunScribeTM
Step
r
ate
Shock Impact-
GS
Braking
-
GS
F
oot
st
r
ike
Step rate
r=0.999,
p<0.001
(1.12±0.
92
spm
)
AVLR
r= -0.421,
p
=0.992
r= -0.133,
p=0.767
IVLR
r= -0.312,
p=0.
959
r= -0.106,
p=0.720
PeakBraking
Force
r=-0.376,
p=0.983
Footstrike
Angle
r=0.090,
p=0.688
Lab-based Zoi
S
tep ra
te
Impact
Braking
Bouncing
Step rate
r=0.998,
p<0.001
(0.93±1.28
spm
)
AVLR
r=0.256,
p=0.086
IVLR
r=0.243,
p=0
.098
PeakBraking
Force
r= -0.202,
p=0.858
VertDisp
CM
r=0.813,
p<0.001
(
-
1
.
78
±
1.50
c
m
)
Lab-based TgForce
S
tep rate
Z
X
3
D
Step rate
r=0.955,
p<0.001
(4.5±6.13
spm
)
AVLR
r=0.681,
p<0.001
IVLR
r=0.758,
p<0.001
r=0.371,
p=0.020
PeakBraking
F
orce
r= -0.470,
p=0.004
r= -0.532,
p=0.001
Pearson’s correlation coefficient with the lab-based system, p-value, and
Bland-Altman bias and 95% limits of agreement for the five tested wearable
sensors. AVLR: average vertical loading rate; IVLR: instantaneous vertical
loading rate; spm: steps per minute.
model and accelerometer for estimating vertical ground
reaction force (cross-correlation coefficient > 0.99) during
running with a shoe mounted device in rearfoot strikers.
Location (e.g. shoe mounted), vibrations (attachment system
[e.g. shoelaces] and sensor weight [> 10g]) of the device and
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JSEN.2020.2982568, IEEE Sensors
Journal
3 IEEE SENSORS JOURNAL, VOL. XX, NO. XX, MONTH X, XXXX
foot strike pattern alter acceleration profile during running
[26], [31], [33]. These interferences may be compensated by
advanced data processing to provide valid estimates of ground
reaction forces [32] but might also explain the reported
differences between the different wearable sensors tested in
this study.
In agreement with our hypothesis, healthcare
professionals, coaches and runners can rely on step rate
measurements provided by all wearable sensors tested in this
study. Step rate is a common measure of running gait analysis.
It has been associated with both running injury and running
economy [12], [16]. Step rate calculation is based on foot
strike detection that is a simple measure to achieve with an
inertial measurement unit. Overall, wearable sensors slightly
over-estimated step rate on average by 2.08 with a mean error
(at 95%) of ±2.35 steps per minute (for a step rate measured at
170 by the lab-based system, wearables sensors will show on
average a step rate from 169.73 to 174.43 steps per minute).
For other running gait characteristics measured by the
wearable sensors (metrics related to antero-posterior ground
reaction force, vertical displacement of the center of mass and
foot strike pattern), concurrent or criterion validity strongly
depended of both the metrics and device.
To provide recommendations for potential buyers
(clinicians, coaches and runners), results for each wearable
sensor will be separately discussed below.
A. MilestonePod
MilestonePod is valid and accurate to measure step rate
during running. A potential important parameter to monitor
during running is the foot strike pattern as a non-rearfoot
strike has been associated with performance [17]–[19]. From
our results, this device seems to discriminate between a rear-,
mid- and forefoot strike pattern. Caution is warranted with this
result considering the low proportion of mid- and fore-foot
strikers in our sample (5 and 2 participants, respectively).
Further research is therefore required to confirm the validity of
this metrics. MilestonePod was not found to be valid in
estimating vertical loading rates. Therefore, potential buyers
may consider this sensor if their objective is to monitor step
rate and foot strike pattern only.
B. Moov NowTM
Moov NowT M is valid and accurate to measure step rate
during running. However, our results do not show that Moov
NowTM is valid to assess vertical ground reaction force during
running. Despite a tibial placement, it is possible that more
advanced signal processing is required to improve vertical
loading rate’s related metrics. Moov NowTM should be
considered if step rate is the only parameter to monitor.
C. TgForce
TgForce is valid to measure step rate during running
although this device is the least accurate of the five tested
sensors for this parameter. Interestingly, TgForce is the only
device providing metrics that were correlated to both vertical
and antero-posterior ground reaction forces. It uses
manufacturer proprietary algorithms with specific
combinations of data from accelerometers, gyroscopes, and
magnetometers. Unfortunately, the information about specific
models is proprietary and could not be shared with our
research team. TgForce is valid to estimate IVLR and the
coefficient of correlation was almost clinically meaningful for
AVLR (r=0.68). It is likely that both device location (distal
and medial tibial end) and attachment, and data processing
influence TgForce estimates of vertical loading rates during
running. However, the correlation coefficient for peak braking
force was too weak to consider TgForce valid for this metrics.
Between the five sensors tested in this study, potential buyers
aiming to monitor VLR should choose TgForce (TgForce_Z)
to get the best estimates (with a tight attachment).
D. Zoi
Zoi is valid and accurate to measure step rate and vertical
displacement of the center of mass during running. Vertical
displacement of the center of mass has been associated with
running economy [15] and may therefore be important to
monitor for runners seeking performance improvement. In
terms of ground reaction force, Zoi is not valid to estimate
vertical loading rates nor peak braking force. As discussed by
Wundersitz et al. [34], location of the device (chest) away
from the site of interest (foot contact) is likely to
introduce/amplify ground reaction force estimation errors. For
foot strike pattern, Zoi provides a percent of rear-, mid-, and
fore-foot strike during the run, however assessing the validity
of this measure with a statistical test was not feasible.
Potential buyers could consider Zoi if the objective is to
monitor step rate and vertical displacement of the center of
mass.
E. RunScribeTM
RunScribeTM is valid and accurate to measure step rate
during running. A recent study by Garcia-Pinillos et al. [35]
regarding the validity of RunScribeTM for measuring step rate
during running against a high-speed video analysis system
supports our results. Moreover, the authors found, in
agreement to our findings, that RunScribeTM slightly
overestimated the step rate. This sensor also provides a
measure of foot strike pattern on an interrupted scale from 0
(rearfoot strike) to 15 (forefoot strike). However, this metrics
is not correlated to the foot strike angle obtained from our lab-
based system. None of the metrics provided by RunScribeTM
(Impact Gs, Shock and Braking Gs) are valid to estimate both
vertical and antero-posterior ground reaction forces.
RunScribeTM should be considered by potential buyers only
when step rate is the variable of interest.
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JSEN.2020.2982568, IEEE Sensors
Journal
3 IEEE SENSORS JOURNAL, VOL. XX, NO. XX, MONTH X, XXXX
F. Strengths and Limitations of the Study
The main strength of this work was to test the
concurrent/criterion validity of multiple commercially
available wearable sensors for runners in the same study, by
an independently funded research team (all wearable sensors
were either bought or borrowed and returned to
manufacturers). Moreover, this is the first independent study
to test the validity of vertical and antero-posterior ground
reaction forces during running against a lab-based system.
The wearable sensors tested in this study were not
designed for research purposes, making it impossible to
automatically synch them with our lab-based system.
Therefore, we had to synchronize systems manually. Since we
used average metrics from one minute of running, we are
confident that the synchronization error is very small.
In our sample, 5 participants were mid-foot strikers and 2
were fore-foot strikers according to our lab-based system. This
small number of non-rearfoot strikers was therefore a
limitation to determine the validity of the wearable sensors to
discriminate between a rear-, mid- and fore-foot strike pattern.
More research is therefore required with more non-rearfoot
strikers.
V. CONCLUSION
Wearable sensors are attractive products to gather in-
field data on running gait. Concurrent or criterion validity of
MilestonePod, Moov NowTM, TgForce, Zoi and RunScribeTM
was assessed in this study against a laboratory-based
instrumented treadmill and motion capture system. Estimation
of ground reaction force (vertical and antero-posterior),
measurement of step rate and vertical displacement of the
center of mass, and foot strike pattern detection during
running were analyzed. Overall, runners, clinicians and
coaches must keep in mind that only a limited number of the
metrics provided by these commercially available wearable
sensors are actually valid. Our results showed that step rate
measurement is valid for all wearable sensors. TgForce was
the only device providing valid metrics of instantaneous
vertical loading rate. Only MilestonePod seemed to be valid to
discriminate between a rear-, mid- and fore-foot strike pattern.
Only Zoi was valid to estimate vertical displacement of the
center of mass. None of the wearable sensors was valid for
estimating peak braking force during running. Potential buyers
should therefore be aware of such limitations of wearable
sensors for monitoring running gait and choose their wearable
sensor according to the metrics they want to monitor.
ACKNOWLEDGMENT
We would like to thank all runners who participated in this
study.
REFERENCES
[1] L. C. Hespanhol Junior, J. D. Pillay, W. van
Mechelen, and E. Verhagen, “Meta-Analyses of the
Effects of Habitual Running on Indices of Health in
Physically Inactive Adults,” Sports Med. Auckl. Nz,
vol. 45, no. 10, pp. 1455–1468, 2015.
[2] C. L. Hamill, I. E. Clarke, E. G. Frederick, L. J.
Goodyear, and E. T. Howley, “EFFECTS OF
GRADE RUNNING ON KINEMATICS AND
IMPACT FORCE,” Med. Sci. Sports Exerc., vol. 16,
no. 2, p. 184, Apr. 1984.
[3] B. C. Heiderscheit, “Gait retraining for runners: in
search of the ideal,” J. Orthop. Sports Phys. Ther.,
vol. 41, no. 12, pp. 909–910, Dec. 2011.
[4] D. E. Lieberman, “What we can learn about
running from barefoot running: an evolutionary
medical perspective,” Exerc. Sport Sci. Rev., vol.
40, no. 2, pp. 63–72, Apr. 2012.
[5] C. Napier, C. K. Cochrane, J. E. Taunton, and M.
A. Hunt, “Gait modifications to change lower
extremity gait biomechanics in runners: a
systematic review,” Br. J. Sports Med., vol. 49, no.
21, pp. 1382–1388, Nov. 2015.
[6] C. D. Bowersock, R. W. Willy, P. DeVita, and J. D.
Willson, “Reduced step length reduces knee joint
contact forces during running following anterior
cruciate ligament reconstruction but does not alter
inter-limb asymmetry,” Clin. Biomech. Bristol Avon,
vol. 43, pp. 79–85, Feb. 2017.
[7] I. S. Moore, “Is There an Economical Running
Technique? A Review of Modifiable Biomechanical
Factors Affecting Running Economy,” Sports Med.
Auckl. Nz, vol. 46, pp. 793–807, 2016.
[8] A. A. Zadpoor and A. A. Nikooyan, “The
relationship between lower-extremity stress
fractures and the ground reaction force: a
systematic review,” Clin. Biomech. Bristol Avon,
vol. 26, no. 1, pp. 23–28, Jan. 2011.
[9] H. van der Worp, J. W. Vrielink, and S. W.
Bredeweg, “Do runners who suffer injuries have
higher vertical ground reaction forces than those
who remain injury-free? A systematic review and
meta-analysis,” Br. J. Sports Med., vol. 50, no. 8,
pp. 450–457, Apr. 2016.
[10] Z. Y. S. Chan et al., “Gait Retraining for the
Reduction of Injury Occurrence in Novice Distance
Runners: 1-Year Follow-up of a Randomized
Controlled Trial,” Am. J. Sports Med., vol. 46, no.
2, pp. 388–395, Feb. 2018.
[11] C. Napier, C. L. MacLean, J. Maurer, J. E.
Taunton, and M. A. Hunt, “Kinetic risk factors of
running-related injuries in female recreational
runners,” Scand. J. Med. Sci. Sports, vol. 28, no.
10, pp. 2164–2172, Oct. 2018.
[12] L. E. Luedke, B. C. Heiderscheit, D. S. B. Williams,
and M. J. Rauh, “Influence of Step Rate on Shin
Injury and Anterior Knee Pain in High School
Runners,” Med. Sci. Sports Exerc., vol. 48, no. 7,
pp. 1244–1250, 2016.
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JSEN.2020.2982568, IEEE Sensors
Journal
3 IEEE SENSORS JOURNAL, VOL. XX, NO. XX, MONTH X, XXXX
[13] A. I. Daoud, G. J. Geissler, F. Wang, J. Saretsky,
Y. A. Daoud, and D. E. Lieberman, “Foot strike and
injury rates in endurance runners: a retrospective
study.,” Med. Sci. Sports Exerc., vol. 44, no. 7, pp.
1325–1334, Jul. 2012.
[14] D. E. Lieberman, A. G. Warrener, J. Wang, and E.
R. Castillo, “Effects of stride frequency and foot
position at landing on braking force, hip torque,
impact peak force and the metabolic cost of
running in humans,” J. Exp. Biol., vol. 218, no. Pt
21, pp. 3406–3414, Nov. 2015.
[15] K. Halvorsen, M. Eriksson, and L. Gullstrand,
“Acute effects of reducing vertical displacement
and step frequency on running economy,” J.
Strength Cond. Res., vol. 26, no. 8, pp. 2065–
2070, Aug. 2012.
[16] T. J. Quinn, S. L. Dempsey, D. P. LaRoche, A. M.
Mackenzie, and S. B. Cook, “Step Frequency
Training Improves Running Economy in Well-
Trained Female Runners,” J. Strength Cond. Res.,
Jul. 2019.
[17] H. Hasegawa, T. Yamauchi, and W. J. Kraemer,
“Foot strike patterns of runners at the 15-km point
during an elite-level half marathon,” J. Strength
Cond. Res., vol. 21, no. 3, pp. 888–893, Aug.
2007.
[18] M. E. Kasmer, X.-C. Liu, K. G. Roberts, and J. M.
Valadao, “The Relationship of Foot Strike Pattern,
Shoe Type, and Performance in a 50-km Trail
Race,” J. Strength Cond. Res., vol. 30, no. 6, pp.
1633–1637, 2016.
[19] M. E. Kasmer, X.-C. Liu, K. G. Roberts, and J. M.
Valadao, “Foot-strike pattern and performance in a
marathon,” Int. J. Sports Physiol. Perform., vol. 8,
no. 3, pp. 286–292, May 2013.
[20] N. Chambon, N. Delattre, N. Guéguen, E. Berton,
and G. Rao, “Shoe drop has opposite influence on
running pattern when running overground or on a
treadmill,” Eur. J. Appl. Physiol., vol. 115, no. 5, pp.
911–918, May 2015.
[21] C. Napier, J.-F. Esculier, and M. A. Hunt, “Gait
retraining: out of the lab and onto the streets with
the benefit of wearables,” Br. J. Sports Med., vol.
51, no. 23, pp. 1642–1643, Dec. 2017.
[22] R. W. Willy, “Innovations and pitfalls in the use of
wearable devices in the prevention and
rehabilitation of running related injuries,” Phys.
Ther. Sport, vol. 29, no. Supplement C, pp. 26–33,
Jan. 2018.
[23] L. C. Benson, C. A. Clermont, E. Bošnjak, and R.
Ferber, “The use of wearable devices for walking
and running gait analysis outside of the lab: A
systematic review,” Gait Posture, vol. 63, pp. 124–
138, Jun. 2018.
[24] J. M. Peake, G. Kerr, and J. P. Sullivan, “A Critical
Review of Consumer Wearables, Mobile
Applications, and Equipment for Providing
Biofeedback, Monitoring Stress, and Sleep in
Physically Active Populations,” Front. Physiol., vol.
9, p. 743, 2018.
[25] A. G. Lucas-Cuevas, A. Encarnación-Martínez, A.
Camacho-García, S. Llana-Belloch, and P. Pérez-
Soriano, “The location of the tibial accelerometer
does influence impact acceleration parameters
during running,” J. Sports Sci., vol. 35, no. 17, pp.
1734–1738, Sep. 2017.
[26] K. R. Sheerin, D. Reid, and T. F. Besier, “The
measurement of tibial acceleration in runners—A
review of the factors that can affect tibial
acceleration during running and evidence-based
guidelines for its use,” Gait Posture, vol. 67, pp.
12–24, Jan. 2019.
[27] A. S. Tenforde, M. C. Ruder, S. T. Jamison, P. P.
Singh, and I. S. Davis, “Is symmetry of loading
improved for injured runners during novice barefoot
running?,” Gait Posture, vol. 62, pp. 317–320, Mar.
2018.
[28] A. R. Altman and I. S. Davis, “A kinematic method
for footstrike pattern detection in barefoot and shod
runners,” Gait Posture, vol. 35, no. 2, pp. 298–300,
Feb. 2012.
[29] M. M. Mukaka, “Statistics corner: A guide to
appropriate use of correlation coefficient in medical
research,” Malawi Med. J. J. Med. Assoc. Malawi,
vol. 24, no. 3, pp. 69–71, Sep. 2012.
[30] E. M. Hennig, T. L. Milani, and M. A. Lafortune,
“Use of Ground Reaction Force Parameters in
Predicting Peak Tibial Accelerations in Running,” J.
Appl. Biomech., vol. 9, no. 4, pp. 306–314, Nov.
1993.
[31] C. A. Laughton, I. M. Davis, and J. Hamill, “Effect
of Strike Pattern and Orthotic Intervention on Tibial
Shock during Running,” J. Appl. Biomech., vol. 19,
no. 2, pp. 153–168, May 2003.
[32] K. J.-H. Ngoh, D. Gouwanda, A. A. Gopalai, and Y.
Z. Chong, “Estimation of vertical ground reaction
force during running using neural network model
and uniaxial accelerometer,” J. Biomech., vol. 76,
pp. 269–273, Jul. 2018.
[33] M. Norris, R. Anderson, and I. C. Kenny, “Method
analysis of accelerometers and gyroscopes in
running gait: A systematic review,” Proc. Inst.
Mech. Eng. Part P J. Sports Eng. Technol., vol.
228, no. 1, pp. 3–15, Mar. 2014.
[34] D. W. T. Wundersitz, K. J. Netto, B. Aisbett, and P.
B. Gastin, “Validity of an upper-body-mounted
accelerometer to measure peak vertical and
resultant force during running and change-of-
direction tasks,” Sports Biomech., vol. 12, no. 4,
pp. 403–412, Nov. 2013.
[35] F. García-Pinillos et al., “Agreement between the
spatiotemporal gait parameters from two different
wearable devices and high-speed video analysis,”
PLoS ONE, vol. 14, no. 9, Sep. 2019.
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JSEN.2020.2982568, IEEE Sensors Journal
3 IEEE SENSORS JOURNAL, VOL. XX, NO. XX, MONTH X, XXXX
APPENDIX
Appendix 1: Specifications of the tested wearable sensors
Wearable sensor Device placement
(according to
manufacturer
instructions)
Power
supply
Start-stop Real-time
view
Smartphone
needed
during
running
?
Results view Data
exportation
Weight
(g)
Price
($US)
Moov NowTM
(v5.2.4809.252)
‘’outside of the ankle
and the loop end of
the band forward’’
CR 2032 Manual - Step rate
- Impact
Yes - Step rate (mean,
max)
- Impact (mean)
No 15.1 59.95
Milestone Pod
(v3.4.10)
Shoelaces
CR 2032 Auto No No - Foot strike pattern
(%heel, mid, toe)
- Step rate (mean,
max)
- Rate of impact
(%low, mid, high)
Yes
Score for each
minute:
- Step rate
- Rate of
impact (low,
mid, high)
- Foostrike
(heel, mid,
toe)
13.0 29.95
RunScribeTM
(v2.7.1)
Heel Mount
Rechargeable
battery
Auto /
Manual
- Impact Gs
- Braking Gs
- Footstrike
No - Step rate (mean)
- Shock
(mean)(combination
of impact and braking
score)
- Impact Gs (mean)
- Braking Gs (mean)
- Foot strike type
(mean)(from 0 [rear-
foot strike] to 15
[fore
-
foot strike])
Yes (score by
step)
15.0
(each
device)
249.00
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JSEN.2020.2982568, IEEE Sensors Journal
3 IEEE SENSORS JOURNAL, VOL. XX, NO. XX, MONTH X, XXXX
Zoi
Runteq (2012-
2016)
One pod on a chest
strap. One pod on the
shoelaces.
Rechargeable
battery
Manual - Step rate
(shoelaces)
- Impact
(chest)
- Bouncing
(chest)
- Braking
(
shoelaces
)
No - Step rate (mean)
- Braking (mean)
- Impact (mean)
- Bouncing (mean)
- Foot strike type
(%ball, middle, heel)
No 12.0
(each
device)
149.00
TgForce, Beta
version
(v2.0.0.10)
Medial end of tibia Rechargeable
battery
Manual,
Beta
version
- Step rate
- Max
vertical
acceleration
(Z)
- Max
transverse
acceleration
(Y)
- Max
sagittal
acceleration
(X)
- Max vector
acceleration
(3D)
Beta version - Step rate
- Max vertical
acceleration (Z)
- Max transverse
acceleration (Y)
- Max sagittal
acceleration (X)
- Max vector
acceleration (3D)
- by step 10.0 189.00
