Validated Assessment of Gait Sub-Phase Durations in Older Adults
using an Accelerometer-based Ambulatory System
Mohamed Boutaayamou
1,2
, Sophie Gillain
3
, Cédric Schwartz
1
, Vincent Denoël
1
, Jean Petermans
3
,
Jean-Louis Croisier
1
, Bénédicte Forthomme
1
, Jacques G. Verly
2
,
Olivier Brüls
1
and Gaëtan Garraux
4
1
Laboratory of Human Motion Analysis, University of Liège (ULiège), Liège, Belgium
2
INTELSIG Laboratory, Department of Electrical Engineering and Computer Science, ULiège, Liège, Belgium
3
Geriatric Department, University Hospital Center of Liège, Liège, Belgium
4
GIGA - CRC In vivo Imaging, ULiège, Liège, Belgium
Keywords: Validated Extraction, Refined Gait Parameters, Gait Sub-Phase Durations, Foot-Worn Accelerometers, Older
Adults, Signal Processing.
Abstract: Validated extraction of gait sub-phase durations using an ambulatory accelerometer-based system is a current
unmet need to quantify subtle changes during the walking of older adults. In this paper, we describe (1) a
signal processing algorithm to automatically extract not only durations of stride, stance, swing, and double
support phases, but also durations of sub-phases that refine the stance and swing phases from foot-worn
accelerometer signals in comfortable walking of older adults, and (2) the validation of this extraction using
reference data provided by a gold standard system. The results show that we achieve a high agreement
between our method and the reference method in the extraction of (1) the temporal gait events involved in the
estimation of the phase/sub-phase durations, namely heel strike (HS), toe strike (TS), toe-off (TO), maximum
of heel clearance (MHC), and maximum of toe clearance (MTC), with an accuracy and precision that range
from ‒3.6 ms to 4.0 ms, and 6.5 ms to 12.0 ms, respectively, and (2) the gait phase/sub-phase durations,
namely stride, stance, swing, double support phases, and HS to TS, TO to MHC, MHC to MTC, and MTC to
HS sub-phases, with an accuracy and precision that range from ‒4 ms to 5 ms, and 9 ms to 15 ms, respectively,
in comfortable walking of a thirty-eight older adults ( (mean ± standard deviation) 71.0 ± 4.1 years old). This
demonstrates that the developed accelerometer-based algorithm can extract validated temporal gait events and
phase/sub-phase durations, in comfortable walking of older adults, with a promising degree of
accuracy/precision compared to reference data, warranting further studies.
1 INTRODUCTION
Accelerometer-based systems have been used as a
reliable solution for the human gait analysis (e.g.,
Moe-Nilssen et al., 2004; Hartmann et al., 2009;
Rueterbories et al., 2010). Their hardware part has the
advantage to include low-cost, small, and lightweight
accelerometer units with an easy handling and
generally low power consumption. The use of these
accelerometer-based systems is particularly relevant
for the gait analysis of older adults considering the
growing interest of using the gait pattern as a marker
of risk of negative clinical outcomes or as a marker of
robustness (e.g., Gillain et al., 2015). However, there
is a current unmet need in terms of the extraction of
gait sub-phases that allow the partitioning of the gait
cycle into refined parameters, such as the swing sub-
phases. These refined gait parameters could have an
advantage in quantifying subtle changes during the
walking of older adults. Indeed, an increased step
variability has been reported to be linked to a higher
fall-risk or fall history (Hausdorff et al., 2001; Allali
et al., 2017).
In this context, we developed a signal processing
algorithm to automatically extract validated gait
events, namely heel strike (HS), toe strike (TS), and
toe-off (TO), from three-axis accelerometer signals
measured at the level of the heel and toe of the right
and left feet during the walking of young and healthy
subjects (Boutaayamou et al., 2015). This algorithm
uses a segmentation method that roughly detects
relevant signal sub-regions (Boutaayamou et al.,
2017a). Gait events are further extracted with high
accuracy and precision in these signal sub-regions.
248
Boutaayamou, M., Gillain, S., Schwartz, C., Denoël, V., Petermans, J., Croisier, J-L., Forthomme, B., Verly, J., Brüls, O. and Garraux, G.
Validated Assessment of Gait Sub-Phase Durations in Older Adults using an Accelerometer-based Ambulatory System.
DOI: 10.5220/0006716202480255
In Proceedings of the 11th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2018) - Volume 4: BIOSIGNALS, pages 248-255
ISBN: 978-989-758-279-0
Copyright © 2018 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved
In this paper, we extend and modify this algorithm
to automatically extract (1) times of occurrence of
HS, TS, TO, and newly considered gait events,
namely maximum of heel clearance (MHC) and
maximum toe clearance (MTC), and (2) durations of
stride, stance, swing, and double support phases, and
durations of sub-phases that refine the stance and
swing phases from foot-worn accelerometer signals
in comfortable walking of older adults. In addition,
we consider a stride-by-stride validation of this
extraction using reference gait events and gait
phase/sub-phase durations provided by a reference
kinematic method (used as gold standard).
2 METHODS
2.1 Participants and Gait Setting
Volunteers, who were included in a two-years
prospective research for the Gait Analysis and Brain
Imagery (GABI) study, participated to the walking
tests considered in this paper (Gillain et al., 2017).
Briefly, the goal of the GABI study is to highlight the
gait parameters associated with the fall risk in the
community of old people without fall history.
Inclusion criteria included: being at least 65 years old,
living independently at home, being able to reach the
motion analysis laboratory, and being able to sign
inform consent. Exclusion criteria included: fall
history in the previous year, the need of walking aids,
gait disorders, and/or an increased fall risk related to
a neurological or osteo-articular disease (e.g.,
Parkinson disease, polyneuropathy, stroke, lumbar
conflict, etc.), dementia, recent hip or knee prosthesis
(≤ 1 year), musculoskeletal pain during walking, an
acute respiratory or cardiac illness (< 6 month), a
recent hospitalization (< 3 month), non-treated or
insufficiently treated co-morbidities (e.g. HTA,
diabetes, etc.), and a cardiac pacing. The local ethic
committee of the University hospital of Liège (CHU
Liège, Belgium) approved the protocol and all
participants signed informed consent.
In the context of the present study, gait signals
were recorded during comfortable walking speed of
thirty-eight older adults (21 women and 17 men), with
(mean ± standard deviation) age = 71.0 ± 4.1 years;
height = 166 ± 11 cm; weight = 71 ± 15 kg, body
mass index = 25,6 ± 3.8 kg/m².
All participants were equipped with four small
three-axis accelerometer units (2 cm x 1 cm x 0.5 cm;
range ±12 g). These four accelerometer units were
directly attached to the heel and toe of each shoe. Our
accelerometer ambulatory system synchronously
recorded gait data at 200 Hz from these four
accelerometer units. A detailed description of the
ambulatory accelerometer system is given in
(Boutaayamou et al., 2017a). The participants wore
their own regular shoes and were also equipped with
four active markers. Each marker was attached on
each accelerometer unit, i.e., the four markers were
also attached to the shoes at the level of the heel and
toe. A four-camera Codamotion
system (Charnwood
Dynamics; UK) recorded gait data from these active
markers at 200 Hz, during 60 seconds for each gait
test. The participants were asked to walk along a track
in a wide, clear, and straight hallway, at their
preferred, self-selected usual speed and by looking
forward to the walking direction. Each participant
walked a total distance of 99 m following the
trajectories shown in Figure 1. We consider here only
gait data that were recorded according to straight
walking lines, i.e., during non-turning walking
episodes. All the walking tests were performed at the
Laboratory of Human Motion Analysis of the
University of Liège, Belgium.
Figure 1: Experimental setting illustrating the walking
trajectories with a total distance of 99 m. This total distance
must be covered by each participant during a gait test in a
comfortable walking speed.
2.2 Algorithm Development and
Validation Method
In order to accurately and precisely quantify the
durations of the stance and swing phases and their
associated sub-phases, during a gait cycle (i.e., the
duration of a stride phase), it is important to extract,
during the same gait cycle, accurate and precise
moments of gait events involved in the estimation of
these phase/sub-phase durations.
The proposed extraction algorithm uses
distinctive and remarkable features on both
longitudinal and antero-posterior accelerations of the
heel and toe for each foot. Depending on the nature of
these features, a suitable method is employed to
accurately and precisely extract gait events of
interest. For clarity, we consider only one foot for the
description of the algorithm. The algorithm would be
applied in the same way for the left and right foot. We
2
1
3
4
The four cameras of the kinematic reference system
: distance = 19.5 m.
: distance = 30 m.
: distance = 30 m.
: distance = 19.5 m.
3
4
1
2
Validated Assessment of Gait Sub-Phase Durations in Older Adults using an Accelerometer-based Ambulatory System
249
consider, hereafter, sagittal (heel/toe) accelerations of
each foot to identify the times of occurrence of the
gait events, namely HS
accel
, TS
accel
, TO
accel
, MHC
accel
,
and MTC
accel
. The subscript accel refers to our
method. All data were analyzed using Matlab R2009b
(MathWorks, USA).
2.2.1 Extraction of HS, TS, TO, MHC, and
MTC from Accelerometer Data
In the present study, we adapt the method described
in (Boutaayamou et al., 2015) to extract TO from the
vertical heel acceleration (Figure (1-bottom)). HS and
TS were extracted from the vertical heel acceleration
(Figure (1-bottom)) and vertical toe acceleration
(Figure (2-bottom)), respectively; the detailed
description of their extraction in the walking of older
adults is beyond the scope of the present paper and
will be considered in a future paper. Rather, we
describe the newly developed method for the
extraction of the gait events for refining the swing
phase, namely MHC and MTC.
The algorithm extracts first (in this order) HS, TS,
TO, and MTC before it extracts MHC:
• The time of the maximum of the toe clearance
event: MTC
accel
.
MTC
accel
is defined as the moment when the toe
accelerometer reaches its maximum position during
the swing phase. We consider distinctive vertical toe
acceleration features that indicate where MTC can be
found in the time domain.
As MTC
accel
occurs after TO and before the heel
strike of the next stride, denoted by HS
2accel
, we seek
MTC
accel
in the segment [TO
accel
+ 0.4*(HS
2accel
–
TO
accel
), HS
2accel
]. MTC
accel
is automatically extracted
in the vertical toe acceleration restricted to this
segment. The resulting local signal is then filtered
with a 4
th
-order zero-lag Butterworth low-pass filter
(cutoff frequency = 7 Hz). We then detect the local
minimum, t
min
, in this filtered signal.
We consider a second local segment defined from
the restriction of the vertical toe acceleration to the
interval [TO
accel
, t
min
+ 0.4*(HS
2accel
– TO
accel
)]. This
local segment is filtered with a 4
th
-order zero-lag
Butterworth low-pass filter (cutoff frequency = 11
Hz). Based on the resulting local filtered signal, we
define a remarkable point, t
cz
, that corresponds to the
time when a zero crossing of this resulting local
filtered signal occurs before t
min
. It is then assumed
that MTC
accel
is the time t
cz
+0.75*(t
min
– t
cz
).
• The time of the maximum of the heel clearance
event: MHC
accel
.
MHC
accel
is defined as the moment when the
maximum clearance between the heel accelerometer
and the ground is achieved during the swing phase. In
contrast to (Boutaayamou et al., 2017a), where
MHC
accel
event was extracted from the vertical heel
acceleration, we consider distinctive vertical toe
acceleration features that indicate where MHC
accel
can
be found in the time domain. MHC
accel
uses the
previously extracted t
cz
and TO
accel
, and it is assumed
that MHC
accel
is the time TO
accel
+ 0.18*(TO – t
cz
).
2.2.2 Extraction of HS, TS, TO, MHC,
and MTC from Kinematic System
Data
Reference gait events, i.e., HS
ref
, TS
ref
, TO
ref
, MHC
ref
,
and MTC
ref
were extracted from the vertical
coordinates of the left/right heel and toe markers
(gold standard) during consecutive strides to validate,
on a stride-by-stride basis, the considered left/right
gait events and phase/sub-phase durations (Figure 2).
The subscript ref refers to the reference method. More
details about the extraction of these reference data are
given in (Boutaayamou et al., 2015).
2.2.3 Extraction of Gait Phase/Sub-Phase
Durations
Left/right temporal gait phases, such as durations of
left/right stance, swing, stride, and double support
phases, are calculated on the basis of the previous gait
events as follows:
• Left stride duration (time between two consecutive
left HSs)
Left stride = HS
left
(i+1) – HS
left
(i).
• Right stride duration (time between two
consecutive right HSs)
Right stride = HS
right
(i+1) – HS
right
(i).
• Left stance duration (time between left HS (HS
left
)
and left TO (TO
left
) during stride i)
Left stance = TO
left
(i) – HS
left
(i).
• Right stance duration (time between right HS
(HS
right
) and right TO (TO
right
) during stride i)
Right stance = TO
right
(i) – HS
right
(i).
• Left swing duration (time between HS
left
of
stride i+1 and TO
left
of stride i)
Left swing = HS
left
(i+1) – TO
left
(i).
• Right swing duration (time between HS
right
of
stride i+1 and TO
right
of stride i)
Right swing = HS
right
(i+1) – TO
right
(i).
BIOSIGNALS 2018 - 11th International Conference on Bio-inspired Systems and Signal Processing
250
(1)
(2)
Figure 2: Left/right reference gait events, i.e., (1-top) heel strike (HS
ref
) and toe-off (TO
ref
), and (2-top) toe-strike (TS
ref
),
maximum of heel clearance (MHC
ref
), and maximum of toe clearance (MTC
ref
) were extracted from the vertical coordinates
of left/right heel and toe markers (gold standard). Left/right accelerometer gait events, i.e., (1-bottom) HS
accel
and TO
accel
, and
(2-bottom) TS
accel
, MHC
accel
, and MTC
accel
were extracted from left/right vertical heel and toe accelerations (our
accelerometer system). These gait events are shown on each signal to illustrate the stride-by-stride validation method.
• Left double support duration (time between TO
left
and HS
leftright
during stride i)
Left double support = TO
left
(i) – HS
right
(i).
• Right double support duration (time between
TO
right
and HS
left
during stride i)
Right double support = TO
right
(i) – HS
left
(i).
We also use the gait events TS, MHC, and MTC
to calculate the left/right gait sub-phase durations as:
• Left HS2TS duration (time between left TS (TS
left
)
and HS
left
of stride i)
Left HS2TS = TS
left
(i) – HS
left
(i).
• Right HS2TS sub-phase duration (time between
right TS (TS
right
) and HS
right
of stride i)
Right HS2TS = TS
right
(i) – HS
right
(i).
• Left TO2MHC sub-phase duration (time between
left MHC (MHC
left
) and TO
left
during stride i)
Left TO2MHC = MHC
left
(i) – TO
left
(i).
• Right TO2MHC sub-phase duration (time between
right MHC (MHC
right
) and TO
right
during stride i)
Right TO2MHC = MHC
right
(i) – TO
right
(i).
• Left MHC2MTC sub-phase duration (time
between MHC
left
and left MTC (MTC
left
) of stride i)
Left MHC2MTC = MTC
left
(i) – MHC
left
(i).
• Right MHC2MTC sub-phase duration (time
between MHC
right
and right MTC (MTC
right
) of
stride i)
Right MHC2MTC = MTC
right
(i) – MHC
right
(i).
• Left MTC2HS sub-phase duration (time between
HS
left
and MTC
left
of stride i)
Left MTC2HS = HS
left
(i) – MTC
left
(i).
• Right MTC2HS sub-phase duration (time between
HS
right
and MTC
right
of stride i)
Right MTC2HS = HS
right
(i) – MTC
right
(i).
2.2.4 Evaluation Method
We evaluated the level of agreement between our
method and the reference method by quantifying, on
a stride-by-stride basis, (1) the accuracy and precision
in the extraction of the gait events, and (2) the mean
error and absolute error in the extraction of the
phase/sub-phase durations.
Accuracy and precision were computed as the
mean and standard deviation (std. dev.), respectively,
of the differences between the gait events for each
stride, i.e., HS
accel
– HS
ref
, TS
accel
– TS
ref
,
TO
accel
– TO
ref
, MHC
accel
– MHC
ref
, and
MTC
accel
– MTC
ref
.
The mean error and the absolute error were
calculated as the mean and std. dev. of the differences
between the phase/sub-phases durations from our
method and those from the gold standard, and the
mean and std. dev. of absolute values of these
differences, respectively. Bland-Altman plots were
also created to evaluate the difference (1) between the
extracted gait events, and (2) between the phase/sub-
phases durations from our method and those from the
reference method.
3 RESULTS
3.1 Validated Extraction of HS, TS,
TO, MHC, and MTC in
Comfortable Walking of Older
Adults
Table 1 shows the stride-by-stride validation results
of the extraction of the gait event timings, i.e., HS,
TS, TO, MHC, and MTC in comfortable walking of
older adults (mean walking speed = 1.324 m/s). The
Validated Assessment of Gait Sub-Phase Durations in Older Adults using an Accelerometer-based Ambulatory System
251
Table 1: Stride-by-stride validation results of the five gait events detection in comfortable walking of older adults (mean
walking speed = 1.324 m/s). These results are given as the accuracy (mean of the differences), the precision (std. dev. of the
differences), limits of agreement, 95% confidence interval (CI) of the differences, and 95% CI of the lower and upper limits
of agreement.
Accuracy (ms)
(precision (ms))
Limits of agreements
(ms)
95% CI of the
differences (ms)
95% CI of the upper
limits (ms)
No. of
events
HS
0.8 (12.0)
[−22.7 24.3]
[ −0.2 1.8]
[22.5 26.0]
540
TS
−3.6 (9.6)
[−22.5 15.3]
[ −4.4 − 2.8]
[13.9 16.8]
517
TO
−0.1 (7.0)
[−13.9 13.6]
[−0.7 0.4]
[12.7 14.6]
636
MHC
4.0 (8.8)
[−13.2 21.1]
[ 3.3 4.6]
[20.0 22.3]
705
MTC
0.3 (6.5)
[−12.5 13.1]
[−0.2 0.8]
[12.2 13.9]
681
(1)
(2)
(3)
(4)
(5)
Figure 3: Bland‒Altman plot results of the extracted gait events, i.e., (1) HS, (2) TS, (3) TO, (4) MHC, and (5) MTC, measured
using our method and the reference method, with mean (dash-dotted line in the middle) of differences
HS
accel
– HS
ref
, TS
accel
– TS
ref
, TO
accel
– TO
ref
, MHC
accel
– MHC
ref
, and MTC
accel
– MTC
ref
. 95% of these differences are
between the lines ± 1.96 std. dev. (dashed lines). (+) and (o) refer to gait events measured at the left foot and those measured
at the right foot, respectively.
accuracy and precision of gait events detection ranged
from −3.6 ms to 4.0 ms, and 6.5 ms to 12.0 ms,
respectively. Given the sampling frequency of 200 Hz
of the recorded heel and toe accelerations for both
feet, the accuracy and the precision of detection are
less than the durations of 1 sampling period (i.e.,
5 ms) and 3 sampling periods (i.e., 15 ms),
respectively.
Figure 3 shows the Bland-Altman plot results for
the extracted gait events. These plots show small
mean differences between the accelerometer-based
algorithm extraction and the reference method in
accordance with the accuracy of detection provided in
Table 1. In addition, the limits of agreement (i.e.,
mean ± 1.96 std. dev.) and their associated 95%
confidence interval (CI) exhibit small variations in
the times of the gait events (Table 1).
3.2 Validated Extraction of the Gait
Phase/Sub-Phase Durations in
Comfortable Walking of Older
Adults
Table 2 shows the results of the comparison between
the values of the left/right gait phase/sub-phase
durations obtained by our accelerometer-based
algorithm and those obtained by the reference
method. These phase/sub-phase durations could be
estimated with a mean absolute error less than 11 ms.
Bland–Altman plots show a mean difference between
our method and the reference method of 0 ms (95%
CI, −28 ms to 29 ms) for stride time, of 0 ms (95%
CI, −28 ms to 27 ms) for stance time, of 0 ms (95%
CI, −27 ms to 28 ms) for swing time, of 0 ms (95%
CI, −28 ms to 27 ms) for double support duration, of
−3 ms (95% CI, −31 ms to 24 ms) for HS2TS sub-
phase duration, 5 ms (95% CI, −31 ms to 24 ms) for
BIOSIGNALS 2018 - 11th International Conference on Bio-inspired Systems and Signal Processing
252
Table 2: Values of left (L)/right (R) gait phase/sub-phase durations extracted by our method are compared to those extracted
by a reference optoelectronic method, Codamotion, in comfortable walking of older adults (n = 38, 71.0 ± 4.1 years old). This
comparison is given as the mean of differences (mean error) and mean of absolute differences (abs. error) between these
values.
Gait phase/sub-phase
durations
Side
Accelerometers
data (ms)
Codamotion
data (ms)
Mean error
(ms)
Abs. error
(ms)
No. of
parameters
Stride time
L & R
1044 ± 78
1044 ± 79
0 ± 15
11 ± 10
373
Stance time
L & R
661 ± 58
662 ± 61
0 ± 14
11 ± 9
440
Swing time
L & R
388 ± 27
388 ± 28
0 ± 14
11 ± 9
497
Double support
L & R
136 ± 24
136 ± 26
0 ± 14
11 ± 9
437
HS2TS sub-phase
L & R
79 ± 11
82 ± 16
−3 ± 14
11 ± 10
410
TO2MHC sub-phase
L & R
53 ± 5
48 ± 10
5 ± 10
8 ± 7
607
MHC2MTC sub-phase
L & R
300 ± 22
303 ± 22
−4 ± 9
8 ± 7
582
MTC2HS sub-phase
L & R
36 ± 12
37 ± 16
0 ± 13
10 ± 8
518
TO2MHC sub-phase duration, −4 ms (95% CI, −
23 ms to 14 ms) for MHC2MTC sub-duration, and of
0 ms (95% CI, −26 ms to 25 ms) for MTC2HS sub-
phase duration (Figure 4).
4 DISCUSSION
We have presented in this paper an ad hoc algorithm
for the extraction of the durations of (1) the left/right
stride, stance and swing phases, and (2) the left/right
sub-phases refining the left/right stance and swing
phases during non-turning, overground walking
episodes in older adults, from left/right sagittal heel
and toe accelerations.
This algorithm takes advantage of existing
remarkable features in the recorded accelerometers
data to detect the gait events from relevant local
acceleration signals. The validation of the extraction
of the gait events and associated gait phase/sub-phase
durations was carried out in comfortable walking of
older adults (n=38). The experimental results show a
good agreement between our algorithm and the
reference method provided by a kinematic system
(gold standard), and demonstrate an accurate and
precise detection of HS, TS, TO, MHC, and MTC. In
addition, our algorithm extracts the durations of
associated gait phases/sub-phases with a good
accuracy and precision.
Table 3 shows an overview of related work that
reported an accuracy and precision of the extraction
of gait phase durations in comfortable walking of
older adults. Compared to stride, stance, and swing
times calculated in (Rampp et al., 2015) (i.e., 2 ms ±
68 ms, 9 ms ± 69 ms, and 8 ms ± 45 ms,
respectively), the accuracy and precision are
improved in our method (i.e., 0 ms ± 15 ms, 0 ms ±
14 ms, and 0 ms ± 14 ms, respectively). Better
accuracy and precision in stance and swing times are
also found in our method compared to (Trojaniello et
al., 2014) (i.e., 10 ms ± 19 ms and 9 ms ± 19 ms,
respectively). Moreover, the accuracy and precision
of the stride time in (Trojaniello et al., 2014) (i.e.,
0 ms ± 14 ms) are similar to our results. The absolute
error in the extraction of the stride time is also
improved in our method (i.e., 11 ms ±
10 ms) compared to results reported in (Micó-
Amigo et al., 2016) (i.e., 21 ms ± 12 ms).
The presented algorithm has the major advantage
to quantify gait sub-phase durations that have a clear
significance to clinical practitioners, since the
estimation of these gait sub-phase durations is based
on fundamental events of walking. Moreover, this
algorithm allows a stride-by-stride extraction which
may be relevant for the gait analysis of some specific
population such as Parkinson’s disease patients who
experience freezing of gait, a sudden and brief
episodic alteration of strides regulation. Moreover,
the high precision achieved in the extraction of the
gait phase/sub-phase durations promises excellent
results in case of tracking the subtle decline/changing
in these durations in older adults. This algorithm
could be thus relevant for characterizing, e.g., the
progression of a neurological disease, and for an early
prediction of, e.g., elderly falls.
This algorithm used the cutoff frequencies of
7 Hz and 11 Hz for the MTCaccel extraction from
recorded data at 200 Hz (Sec. 2.2.1); these cutoff
frequencies should then be adapted in the case of
lower/higher sample rate. In addition, we defined
empirically the intervals [TO
accel
, t
min
+ 0.4*(HS
2accel
– TO
accel
)] and [TO
accel
+ 0.4*(HS
2accel
– TO
accel
),
HS
2accel
], and the times TO
accel
+ 0.18*(TO – t
cz
) and
t
cz
+0.75*(t
min
– t
cz
) for the detection of MHC
accel
and
MTC
accel
in comfortable walking older adults (Sec.
2.2.1). These intervals and times would require
further investigations in the case of slow and fast
Validated Assessment of Gait Sub-Phase Durations in Older Adults using an Accelerometer-based Ambulatory System
253
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
Figure 4: Bland‒Altman plot results of durations of (1) stride phase, (2) stance phase, (3) stance phase, (4) double support
phase, (5) HS2TS sub-phase, (6) TO2MHC sub-phase, (7) MHC2MTC sub-phase, and (8) MTC2HS sub-phase extracted
during consecutive strides by our method and the gold standard method in comfortable walking of older adults. (+) and (o)
refer to left and right time-related gait phases/sub-phases, respectively.
walking speeds and in the case of pathological gait
patterns. Moreover, the algorithm is valid in case of a
heel strike at initial contact during walking, but might
be modified to be more flexible to take into account
situations where the heel strike (or other events) is
missing (e.g., in case of toe landing at initial contact)
such as in running or in some pathological conditions.
5 CONCLUSION
We presented and validated on a stride-by-stride basis
an ad hoc signal processing algorithm that extracts
durations of (1) the left/right stride, stance, swing, and
double support phases, and (2) the left/right sub-
phases that refine the left/right stance and swing
phases in comfortable walking of older adults (21
women and 17 men, 71.0 ± 4.1 years old), using an
ambulatory foot-worn accelerometer system. The
algorithm was tested against a reference kinematic
system (used as gold standard) and yielded (1) an
accuracy and precision that range from ‒3.6 ms to 4.0
ms, and 6.5 ms to 12.0 ms, respectively, for the
extraction of left/right HS, TS, TO, MHC, and MTC,
and (2) an accuracy and precision that range from ‒
4 ms to 5 ms, and 9 ms to 15 ms, respectively, for the
estimation of durations of left/right stride, stance,
swing, and double support phases, and of left/right
HS2TS, TO2MHC, MHC2MTC, and MTC2HS sub-
phases.
To the best of our knowledge, this is the first study
that demonstrates a good validation accuracy and
precision in the extraction of sub-phase durations
refining the stride phase duration during comfortable
walking of older adults and using an ambulatory foot-
worn accelerometer system.
In a future work, we plan to investigate (1) the
effect of the walking speed on the extraction accuracy
and precision of the aforementioned gait events and
phase/sub-phase durations in older adults, (2) the
capability of those gait phase/sub-phase durations to
differentiate elderly fallers from elderly non-fallers
using, e.g., classification models, (3) the application
of the proposed algorithm to the study of pathological
gait (e.g., gait of patients with Parkinson’s disease),
(4) the extension of this algorithm to deal with the
BIOSIGNALS 2018 - 11th International Conference on Bio-inspired Systems and Signal Processing
254
Table 3: Related work: accuracy and precision of the extraction of gait phase durations in older adults using inertial sensors.
Subjects
Diagnose
Gait phase durations
Mean error (ms)
Abs. error
(ms)
Micó-Amigo et al., 2016
20 elderly
healthy
Step/stride time
NA
21 ± 12
Rampp et al., 2015
101 elderly
geriatric
Stride time
2 ± 68
29 ± 62
Stance time
9 ± 69
33 ± 61
Swing time
8 ± 45
25 ± 38
Trojaniello et al., 2014
10 elderly
healthy
Stride time
0 ± 14
10 ± 10
Stance time
10 ± 19
22 ± 28
Swing time
9 ± 19
22 ± 27
NA: not available.
turning walking episodes, and (5) the extraction of
spatial gait parameters from the heel and toe
accelerations, taking advantage from the proposed
algorithm that enables splitting the gait cycle time
into small time intervals and thus the drift from
successive integration in these small intervals could
be minimized. In this context, i.e., the extraction of
spatial gait parameters from accelerometer data, we
obtained promising preliminary results reported in
(Boutaayamou et al., 2017b).
ACKNOWLEDGEMENTS
The authors would like to thank all the participants
who accepted to participate as volunteers to the
walking tests of the present study.
REFERENCES
Allali, G., Launay, C.P., Blumen, H.M., et al., (2017). Falls,
cognitive impairment, and gait performance: results
from the good initiative. J. of the American Medical
Directors Association, 18(4):335–40.
Boutaayamou, M., Schwartz, C., Stamatakis, J., et al.,
(2015). Development and validation of an
accelerometer-based method for quantifying gait
events. Medical Engineering & Physics, 37:226–232.
Boutaayamou, M., Denoël, V., Brüls, O., et al., (2017a).
Algorithm for temporal gait analysis using wireless
foot-mounted accelerometers. In: Fred A., Gamboa H.
(eds) Biomedical Engineering Systems and
Technologies. Communications in Computer and
Information Science, Springer, 690:236–254.
Boutaayamou, M., Schwartz, C., Denoël, V., Croisier, J.-
L., Verly, J.G., Garraux, G., and Brüls, O., (2017b). A
novel accelerometer-based method for stride length
estimation. In: 39
th
Annual International Conference of
IEEE Engineering in Medicine and Biology Society,
Jeju, South Korea.
Gillain, S., Dramé, M., Lekeu, F., et al., (2015). Gait speed
or gait variability, which one to use as a marker of risk
to develop Alzheimer disease? A pilot study. Aging
Clinical and Experimental Research, 28(2):249–255.
Gillain, S., Boutaayamou, M., Dardenne, N., et al., (2017).
Data set of healthy old people assessed for three
walking conditions using accelerometric and opto-
electronic methods. Aging Clinical and Experimental
Research, 1–9.
Hartmann, A., Luzi, S., Murer, K., de Bie, R.A., de Bruin,
E.D., (2009). Concurrent validity of a trunk tri-axial
accelerometer system for gait analysis in older adults.
Gait & Posture, 29(3):444–448.
Hausdorff, J.M., Rios, D.A., and Edelberg, H.K., (2001).
Gait variability and fall risk in community-living older
adults: a 1-year prospective study. Archives of Physical
Medicine and Rehabilitation, 82.
Micó-Amigo, M.E., Kingma, I., Ainsworth, E., et al.,
(2016). A novel accelerometry-based algorithm for the
detection of step durations over short episodes of gait in
healthy elderly. J. NeuroEngineering and
Rehabilitation, 13:38.
Moe-Nilssen, R., and Helbostad, J.L., (2004). Estimation of
gait cycle characteristics by trunk accelerometry.
Journal of Biomechanics, 37(1):121–126.
Rampp, A., Barth, J., Schülein, S., Gaßmann, et al. (2015).
Inertial sensor-based stride parameter calculation from
gait sequences in geriatric patients. IEEE Transactions
on Biomedical Engineering, 62(4):1089–1097.
Rueterbories, J., Spaich., E.G., Larsen, B., and Andersen
O.K., (2010). Methods for gait event detection and
analysis in ambulatory systems. Medical Engineering
and Physics, 32(6):545–552.
Trojaniello, D., Cereatti, A., Pelosin, et al., (2014).
Estimation of step-by-step spatio-temporal parameters
of normal and impaired gait using shank-mounted
magneto-inertial sensors: application to elderly,
hemiparetic, parkinsonian and choreic gait. Journal of
Neuroengineering & Rehabilitation, 11:152.
Validated Assessment of Gait Sub-Phase Durations in Older Adults using an Accelerometer-based Ambulatory System
255