http://informahealthcare.

com/jmt
ISSN: 0309-1902 (print), 1464-522X (electronic)
J Med Eng Technol, 2013; 37(8): 498510
! 2013 Informa UK Ltd. DOI: 10.3109/03091902.2013.837529

INNOVATION

Assessing post-anterior cruciate ligament reconstruction ambulation

using wireless wearable integrated sensors
1

Faculty of Science, Universiti Brunei Darussalam, Jalan Tungku Link, Gadong BE1410, Brunei Darussalam and 2Sports Medicine & Research Centre,
Hassan Bolkiah National Stadium, Berakas, Brunei Darussalam
Abstract

Keywords

A hardware/software co-design for assessing post-Anterior Cruciate Ligament (ACL) reconstruction ambulation is presented. The knee kinematics and neuromuscular data during
walking (26 km h1) have been acquired using wireless wearable motion and electromyography (EMG) sensors, respectively. These signals were integrated by superimposition and mixed
signals processing techniques in order to provide visual analyses of bio-signals and
identification of the recovery progress of subjects. Monitoring overlapped signals simultaneously helps in detecting variability and correlation of knee joint dynamics and muscles
activities for an individual subject as well as for a group. The recovery stages of subjects have
been identified based on combined features (knee flexion/extension and EMG signals) using an
adaptive neuro-fuzzy inference system (ANFIS). The proposed system has been validated for
28 test subjects (healthy and ACL-reconstructed). Results of ANFIS showed that the ambulation
data can be used to distinguish subjects at different levels of recuperation after ACL
reconstruction.

Anterior cruciate ligament (ACL),

electromyography (EMG), knee kinematics,
recovery, motion sensors

1. Introduction
Anterior Cruciate Ligament (ACL) rupture leads to various
short- and long-term impacts on the injured subjects including
dynamic knee joint instability, neuromuscular impairments,
loss of proprioception, cartilage degeneration and early onset
of osteoarthritis [14]. In order to restore the knee kinematics,
kinetics and neurophysiologic dysfunction, ACL reconstruction is generally recommended. However, recent clinical
studies indicate the occurrence of cartilage degeneration
and early progression of osteoarthritis in subjects having
ACL reconstruction, which suggests incomplete restoration
of normal kinematics and neuromuscular function after ACL
surgery and a lack of proper monitoring of rehabilitation
progress [510].
The alterations in biomechanical features of the knee
during walking following ACL reconstruction have been
evaluated in different studies. Reduced knee flexion angles/
moments at mid-stance, extension deficit during swing phase,
gait variability and muscles weakness have been noticed
in ACL reconstructed (ACL-R) subjects even after 12 years
of surgery [1015]. Thus, in order to re-establish knee joint
and neuromuscular control and avoiding or minimizing postoperative complications, a careful monitoring of recovery
process after ACL surgery becomes crucial.

*Corresponding author. Email: arosha.senanayake@ubd.edu.bn

History
Received 8 February 2013
Revised 5 August 2013
Accepted 20 August 2013

In current clinical settings, during the rehabilitation period,

the subjects perform different exercises and the recovery
progress is evaluated based upon the either subjective or
partially objective tests/scores [1621]. An additional supplementary method for assessing the recovery is to provide
real-time or offline biofeedback to the clinicians and subjects.
The kinematics and surface electromyography (S-EMG)
signals have been found effective in providing a biofeedback
to physiatrists, clinicians and subjects during the post-ACL
surgery recovery period [2224]. In most of the previous
studies, optical motion capture and hard-wired electromyography (EMG) systems have been used to capture the human
motion parameters [5,10,11,15,25]. Although these systems
can be used in clinical applications, they have certain
drawbacks. Optical motion capture systems are usually
expensive, sensitive to calibration and reflections and bulky
[26,27]. The time required to set up the experiment and
data processing is long while using these systems. Similarly
hard-wired EMG systems are also less portable and they can
be used only in clinical settings or labs, allowing limited
movements to the subjects.
Micro-Electro-Mechanical-Systems (MEMS) and other
wireless sensors have already been proven as an alternative
mechanism for accurately measuring different bio-signals
[26]. Accelerometers, angular rate gyroscopes and magnetometers have been successfully used for observing the
kinematics changes after knee surgery [2830]. However,
the rehabilitation after ACL reconstruction not only requires

20
13

J Med Eng Technol Downloaded from informahealthcare.com by University Brunei Darussalam on 10/30/13
For personal use only.

S. M. N. Arosha Senanayake*1, Owais Ahmed Malik1, Pg. Mohammad Iskandar1, and Dansih Zaheer2

J Med Eng Technol Downloaded from informahealthcare.com by University Brunei Darussalam on 10/30/13
For personal use only.

DOI: 10.3109/03091902.2013.837529

monitoring of individual signals but it also needs to observe

the correlation and impact of these signals on each other
as the EMG signals influence the kinematics variability.
Moreover, it has been found that the movements of contralateral limb are also affected after ACL injury/surgery so a
comparison of parameters for both limbs can also be useful
in recovery progress monitoring [31]. There have been few
studies where efforts have been made to develop a rehabilitation monitoring system providing a visual analysis of
individual and overlapped bio-signals and classification of
ambulation patterns during the ACL rehabilitation period
using multiple integrated signals [3234]. These studies
provide either only one type of parameter for assessing
the recovery progress or have limited functionality when
multiple signals are combined. Data from wireless bodymounted MEMS and EMG sensors can be integrated to
design a portable system for visualizing superimposed biosignals during the post-ACL reconstruction recovery process.
In addition, the parameters can be integrated for identifying
the stage of recovery of ACL-R subjects using intelligent
techniques [35].
This paper proposes a hardware/software co-design for
assessing post-ACL reconstruction ambulation. The knee
kinematics and neuromuscular data during walking have
been acquired through wireless body-mounted motion and
EMG sensors, respectively. These signals were integrated
by superimposition and mixed signals processing techniques
in order to provide visual analyses of bio-signals and
identification of recovery progress of subjects. Due to the
non-stationary and time-varying nature of EMG signals, timefrequency features were also extracted from knee extensor and
flexor muscles using a continuous wavelet transform (CWT),
in addition to other time and frequency features. The wavelet
analysis is a quite suitable method for processing bio-signals
such as EMG, because it uses a multi-resolution analysis
technique for computing the transform of the segments of the

Assessing post-anterior cruciate ligament

499

signals with different resolutions for each segment [36,37].

This is useful as the distribution of intervals for high and
low frequency components are different in these bio-signals compared to other types of signals. The variations
in kinematics and EMG signals for individuals make the
recovery assessment task challenging. In order to model such
inputs, the adaptive neuro-fuzzy inference system (ANFIS)
can be more useful as it combines the neural network learning
approach and fuzzy logic based system to model the nonlinear functions. It can effectively identify the stochastic
changes in bio-signals like EMG and can also deal with
the impreciseness in measurements and variations due to
subjects physiological conditions [38,39]. A pattern set of
features from knee flexion/extension and EMG signals (time,
frequency and time-frequency) was prepared to train the
ANFIS for identifying the recovery progress of ACL-R
subjects. The intention of designing this system is to help in
reducing duration and cost of recovery and improving the
rehabilitation process by providing accurate and timely
information about the athletes knee functionality.

2. Methodology
2.1. System architecture
The hardware/software co-design of the proposed system is
shown in Figure 1. The recovery assessment system consisted
of two major hardware components which are (1) KinetiSense
(ClevMed. Inc, Cleveland, OH) with four wireless microelectro-mechanical motion sensor units and (2) a wireless
BioCapture (ClevMed. Inc) surface electromyography monitoring unit with electrodes. The body-mounted motion sensors
were used to measure the subjects lower extremity motion
during walking in terms of angular rate and linear acceleration. These angular rates and linear acceleration were then
used to compute the knee joint movements in the sagittal
plane. The electromyography sensors provided the muscle

Figure 1. Hardware/software co-design with information flow from wireless wearable sensors to software components for providing recovery
classification and visual biofeedback analysis for post-ACL reconstruction recovery.

J Med Eng Technol Downloaded from informahealthcare.com by University Brunei Darussalam on 10/30/13
For personal use only.

500

S. M. N. A. Senanayake et al.

recruitment pattern for knee extensors and flexors. The data

from these two hardware components were wirelessly transferred to the workstation for providing biofeedback (visual
analyses and recovery stage identification) to the physiatrists
and subjects. Data logging from all sensors (motion and EMG)
was carried out by transmitting sensors data using two basestations connected to the laptop. After data transmission from
different sensors, the processing of all data was done by using
custom developed software components designed in MATLAB
7.0. These software components were used to integrate
kinematic and EMG signals for preparing a pattern set of
features. This pattern set helped in designing an adaptive
neuro-fuzzy inference system (ANFIS) for intelligently classifying the recovery progress of ACL-R subjects. A visual
biofeedback module has also been developed which displays
individual and multiple superimposed signals simultaneously
for identifying the variations in ambulation patterns for healthy
and ACL reconstructed at different stages. Thus, an intelligent
components-based software system was developed and verified
for ACL recovery monitoring.
2.2. Subjects
Twenty-eight subjects (eight healthy males and 20 unilateral
ACL reconstructed) were recruited from the sports medicine
and research centre (Ministry of Sports), Universiti Brunei
Darussalam and performance optimization centre (Ministry
of Defence), Brunei Darussalam. The healthy subjects (Group
A) were a mean age of 28.3 (SD 4.15) years, mean height
167.2 (SD 4.10) cm and mean weight 68.5 (SD 12.88) kg.
ACL reconstructed subjects were further divided into two
groups (Group B: 10 subjects within 6 months of ACL
surgery; and Group C: 10 subjects with 1 or more years
of ACL surgery) based on their current rehabilitation stage.
For Group B, the mean age, mean height and mean weight
were 29.50 (SD 4.10) years, 173.5 (SD 5.5) cm and 67.5
(SD 12.5) kg, respectively. For Group C, the mean age,
mean height and mean weight were 31.5 (SD 4.5) years,
171.40 (SD 11.00) cm and 72.35 (SD 15.25) kg, respectively. All subjects read and signed an informed consent
form and ethical procedures were carried out according to
the guidelines approved by the Graduate Research Office and
Ethics Committee at Universiti Brunei Darussalam.
2.3. Experimental set-up
In order to test the proposed system, each subject was set up
with four motion sensors attached to his/her right thigh,
right shank, left thigh and left shank using flexible Velcro
straps and adhesive medical tape to note the angular rate and
accelerations of lower limb extremities. These sensors were
attached to the command module worn by subjects using
a waist-belt. The surface EMG signals were recorded by
placing foam snap electrodes on four different knee extensor
and flexor muscles, namely vastus medialis (VM), vastus
lateralis (VL), semitendinosus (ST) and biceps femoris (BF)
on both legs of the subjects. The EMG electrodes wires were
connected to a BioRadio module worn by the subjects using a
waist-belt. For skin preparation and electrodes placement,
SENIAM EMG guidelines were followed [40]. The EMG and
kinematics data were simultaneously recorded for subjects

J Med Eng Technol, 2013; 37(8): 498510

walking at different speeds (26 km h1) on a treadmill.

Different walking speeds were chosen to find out their impact
on changes in kinematics and neuromuscular signals and
accuracy of recovery classification. The walking duration
was set to 3035 s to collect data for multiple gait cycles from
each subject in order to take care of variability in data
and generating the data-set containing the knee dynamics and
neuromuscular signals for at least 20 gait cycles.
2.4. Data acquisition
The data acquisition begins after setting up the sensors as
described in the experimental set-up section. The KinetiSense
and BioCapture software were started simultaneously to
record knee kinematics and neuromuscular signals. For each
session, the required speed was set up on the treadmill by
the subject and, when he/she started walking, the data
collection commands were initiated on both software. The
3-D angular rates and linear accelerations were recorded
using KinetiSense software and the muscles activities and
2-D acceleration were noted using BioCapture software.
The data transfers between command module/BioRadio and
base station were performed by using a 2.4 GHz wireless link.
Upon completion of each session, the data from KinetiSense
and BioCapture software systems were exported to MATLAB
for further processing (filtering, knee angle computation,
EMG rectification, signals synchronization, etc.).
2.5. Data pre-processing
2.5.1. Knee angle computation
The knee angle measurements were obtained from each
motion sensor unit placed on the thigh and shank segments of
both legs. The sensors were aligned to provide knee angle
about the sagittal plane using angular rate about the z-axis
(Figure 2). The angular rate and acceleration measurements
obtained from the motion sensors were low-pass filtered using
a 6th order Butterworth filter to avoid noise due to motion
before computing the orientations. With respect to the
placement of each motion sensor, measurements for zeroreferencing were obtained prior to starting the experiment
(actual motion) when the subjects were in an upright position.
These measurements were then subtracted from each angular
rate during the experiment. The knee flexion/extension
measurements were calculated by applying trapezoidal integration to the angular rates from motion sensors fixed on
thighs and shanks of the subjects. The drift in measurements
from the gyroscope was corrected by fusing orientation
measurements from both the 3-D MEMS gyroscope and
accelerometer available in the motion sensors. For fusing
both signals and knee angle estimation, the complementary
filter was used due to its simplicity and reliability [29,41].
2.5.2. EMG processing
In order to remove noise and motion artifacts, the raw EMG
signals for four lower extremity muscles were band-pass
filtered and processed as per the standards [42]. For
generating the envelops, the EMG signals for each muscle
were first rectified and then band-pass filtered using 4th order
Butterworth filter.

J Med Eng Technol Downloaded from informahealthcare.com by University Brunei Darussalam on 10/30/13
For personal use only.

DOI: 10.3109/03091902.2013.837529

Assessing post-anterior cruciate ligament

501

Figure 2. Placement of motion and EMG sensors on both lower limbs of a human body to collect data for knee joint and neuromuscular movements.
(a) Front view, (b) back view.

2.6. Data integration and fusion

The integration of kinematics and neuromuscular data required
the synchronization of both bio-signals as these signals were
different in nature and recorded at different sampling rates by
using two devices. Before synchronization, re-sampling of both
kinematics and EMG signals was done in order to match their
sampling rates (the motion data was up-sampled by a factor of
15 and EMG data was up-sampled by a factor of 2). Then, the
synchronization of signals was done by detecting the gait cycle
for each subject during walking at different speeds. In order to
detect gait cycle, the heel strike (HS) event was identified from
motion data by using shank sagittal angular velocity. The HS
was identified by examining the timing characteristics of the
angular velocity and determining the two minima on either side
of a peak in velocity curve where every second minimum
indicates the HS event [43]. The HS event in EMG data was
identified by using antero-posterior acceleration from a 2-D
accelerometer available in BioCapture (see Figure 3) [44].
Once a gait cycle was detected, the superimposition of knee
kinematics and EMG signals of different muscles was done and
the required features were extracted from selective phases of
each gait cycle (see section 2.7).
The estimated knee orientation in the sagittal plane and
EMG envelops from different lower limb muscles were
superimposed to observe the changes in both type of signals
simultaneously. This overlapping of data allowed one to
monitor the knee flexion/extension, variation in muscles

strength, activation timings and durations for different

phases of each gait cycle for ACL reconstructed and healthy
legs (see Figure 4). This process avoids separate monitoring
of each of these signals and provides combined signals while
an athlete is performing a rehabilitation/sports activity.
2.7. Feature extraction and selection
The extraction of salient features from athletes motion is
an important step for collecting relevant data to identify
the inter- and intra-subject variability and recovery classification during the rehabilitation period. These features are
used by an intelligent rehabilitation analyzer as training and
testing data for determining the status of recuperation.
The kinematics and neuromuscular signals change in the
course of the rehabilitation process during different activities
performed by a subject. When a subject walks, his/her
muscles contract or extend and the knee angle changes
accordingly during each gait cycle. The knee angle and
activation timings, duration and strength of lower extremity
muscles vary in different phases of a gait cycle. The vastus
medialis, vastus lateralis, semitendinosus and biceps femoris
are mostly active during Load Response (LR), Mid Stance
(MSt) and Terminal Swing (TSw) phases [45]. In this study,
the kinematics and EMG features were extracted from all
phases of the gait cycle and then the selection of features was
done only for the relevant phases. The percentages to identify
each phase of the gait cycle were calculated based on the

502

S. M. N. A. Senanayake et al.

J Med Eng Technol, 2013; 37(8): 498510

Magnitude (Knee Angle, Angular Rate,Acceleration)

70
60
50
40
30
20
10
0
10
20

3.4

3.5

3.6

3.7

3.8

3.9

4.1

4.2

Figure 3. Identification of heel strike during ambulation with knee angle changes (---) using shank angular rate () from motion sensors and anteriorposterior acceleration ( ___ ) from BioCapture system to synchronize kinematics and EMG data for signals integration and overlapping.
80

70

50

40

30
0.02
20

EMG for Vastus Lateralis (mV)

0.04

60
Knee Flexion/Extension (degrees)

J Med Eng Technol Downloaded from informahealthcare.com by University Brunei Darussalam on 10/30/13
For personal use only.

4.3
x 104

Sampling Rate

10

10
15

16

17

18

19

20

21

22

23

24

0
25

Time (sec)

Figure 4. Superimposition of knee flexion/extension ( ___ ) and vastus lateralis EMG envelop ( ) for a healthy subject during walking at a speed
of 3 km h1 to represent the correlation between both signals.

existing literature [45]. MSt was further divided into two halves
to identify the joints and muscles activity in more detail. For
kinematics data, the average knee flexion/extension during
each selected phase was calculated. For EMG data classification, in previous studies, various kinds of features have been
used in time and frequency domains such as mean absolute
value, spectral analysis, zero-crossing and auto-regressive
coefficients, etc. In this study, a combination of time, frequency
and time-frequency EMG features has been used. The root
mean square (RMS) value for filtered EMG data was used as a
time-domain feature for each selected gait phase (1).
s
Z T2
1
EMGRMS
EMGt2 dt
1
T2  T1 T1
For time-frequency analysis, CWT has been chosen as it
has been proved effective in classification of non-stationary

signals like neuromuscular data [46,47]. CWT of an EMG

signal EMG(t) is defined in equation (2), where s represents
the scale parameter,  represents the translation diameter of
time shifting and the basis function * is obtained by scaling
the mother wavelet at time  and scale s.
Z 1
t   
1
EMGt 
dt
2
CWTEMG s,  p
s
s 1
The MATLAB software was used to obtain the wavelet coefficients for EMG signals by choosing Morlet
mother wavelet (after testing different types of mother
wavelets) and scales from 1256 [47]. Using CWT analysis
of EMG signals for the above-mentioned gait phases,
three features were selected: maximum absolute value of
wavelet coefficients, minimum absolute value of wavelet
coefficients and mean frequency (MNF). For frequency

Assessing post-anterior cruciate ligament

DOI: 10.3109/03091902.2013.837529

(a)

503

70
60

Knee Angle (degrees)

50
40
30
20
10
0
10

10

20

30

40

50

60

70

80

90

100

60

70

80

90

100

(b)

70
60
50

Knee Angle (degrees)

J Med Eng Technol Downloaded from informahealthcare.com by University Brunei Darussalam on 10/30/13
For personal use only.

Gait Cycle (%)

40
30
20
10
0
10

10

20

30

40

50
Gait Cycle (%)

Figure 5. Knee flexion/extension variations in the (a) ACL-R and (b) ACL-I leg of a subject at 3 km h1 speed after 2 months of surgery ( ___ )
average, (---) lower limit, (- - -) upper limit representing restricted extension during 2050% of the gait cycle.

fi and power spectrum Pi, the MNF was calculated as

follows:
Pn
fi Pi
EMGMNF Pim
3
n
im Pi
2.8. Intelligent recovery analysis
The recovery analysis of the subjects was done by using
ANFIS. The ANFIS is a fuzzy Sugeno model that adapts the
membership function parameters using neural network and
learns from the given data set [48]. In our study, the ANFIS
consisted of 68 inputs (knee flexion/extension and four EMG
features for four muscles for selected gait phases: 1  4 4 
4  4 68) and a single output (recovery class/status). The
system adjusted the membership function (m) parameters
based on the given data and the number of rules and output
of fuzzy rules was minimized.

speeds can be visualized using the devised hardware/software

co-design system. The variations in knee flexion/extension
of ACL-R and ACL-Intact (ACL-I) legs can be compared
at different speeds for a subject. Figure 5(a) represents
the average knee flexion/extension and its upper/lower
limits during multiple gait cycles of an ACL-R subject after
2 months of surgery. During the single limb support phase
(2050% of gait cycle) the ACL-R limb exhibits restricted
extension. ACL injury affects the contralateral limb, also
as shown in Figure 5(b), where the knee flexion/extension of
ACL-I limb of the same subject also deviates from normal
flexion/extension during the single limb support phase.
These deviations are also visible for higher speed walking
(Figure 6). The inter-subject comparison can also be done to
view the differences of kinematics among subjects at various
stages of recovery (Figure 7). There are noticeable differences
among subjects at different recovery stages during 2050%
of the gait cycle.

3. Results
3.1. Biofeedback: Visual analyses
3.1.1. Intra- and inter-subjects variations in knee
flexion/extension
Intra- and inter-subjects variations in knee flexion/extension
of both legs for a selected number of gait cycles at different

3.1.2. Intra- and inter-subjects comparison of muscle

characteristics within same/different leg(s)
Intra- and inter-subjects comparison of muscle characteristics
(activation timings, duration and strength) within the same/
different leg(s) for a selected number of gait cycles at
different speeds can also be visualized using the developed

504

S. M. N. A. Senanayake et al.

(a)

J Med Eng Technol, 2013; 37(8): 498510

70
60

Knee Angle (degrees)

50
40
30
20
10
0
10

10

20

30

40

50

60

70

80

90

100

60

70

80

90

100

(b)

70
60

Knee Angle (degrees)

50
40
30
20
10
0
10

10

20

30

40

50
Gait Cycle (%)

Figure 6. Knee flexion/extension variations in the (a) ACL-R and (b) ACL-I leg of a subject at 4 km h1 speed after 2 months of surgery ( ___ )
average, (---) lower limit, (- - -) upper limit representing restricted extension during 2050% of the gait cycle.
70
60
50
Knee Angle (degrees)

J Med Eng Technol Downloaded from informahealthcare.com by University Brunei Darussalam on 10/30/13
For personal use only.

Gait Cycle (%)

40
30
20
10
0
10

10

20

30

40

50

60

70

80

90

100

Gait Cycle (%)

Figure 7. Knee flexion/extension variations in the subjects at 4 km h1 speed of a healthy subject ( ___ ) average, after 1 year of surgery (- - -),
after 2 months of surgery (---), representing the visible differences during 2050% of gait cycle and maximum peak flexion values/timings.

system. The variations in muscle strength and activation

timings between ACL-R and ACL-I legs and within ACL-R/
ACL-I can be compared at different speeds for a single
subject for quadriceps and hamstrings. In order to compare
the muscle strength, the EMG signals from muscles are mean
normalized. Figure 8 shows the similarities in the activation
timings of VL and VM for a subject after 1 year of ACL
reconstruction, but the mean muscles strength varies, which

suggests that the vastus medialis is still weak in ACL-R leg

even after 1 year of surgery. Similarly, in Figure 9, lower
values of VM for the ACL-R leg as compared to VM of the
ACL-I leg demonstrate the poor recovery of muscle strength
in the reconstructed leg after surgery. This poor muscle
strength also results in variations in knee angle. On the other
hand, Figure 10 illustrates the similarities in the strength and
activations timings of the bicep femoris muscle for a healthy

Assessing post-anterior cruciate ligament

DOI: 10.3109/03091902.2013.837529

505

800
700

% of Mean EMG

600
500
400
300
200
100
0
3.5

4.5

5.5

6.5

7.5

Figure 8. Comparison of muscle percentage of mean strength and activation for vastus lateralis ( ) and vastus medialis () in the ACL-R leg
of a subject after 1 year of surgery at 4 km h1 speed representing weak vastus medialis.
300

% of Mean EMG

250

200

150

100

50

3.5

4.5

5.5
x 104

Sampling Rate

Figure 9. Comparison of percentage of mean strength and activation for vastus medialis muscles of ACL-I leg ( ) and ACL-R leg ( ___ ) of a
subject after 1 year of surgery at 4 km h1 speed, representing lower mean strength for vastus medialis in ACL-R leg.
1000
900
800

% of Mean EMG

J Med Eng Technol Downloaded from informahealthcare.com by University Brunei Darussalam on 10/30/13
For personal use only.

x 104

Sampling Rate

700
600
500
400
300
200
100
0
15

15.5

16

16.5

17

17.5

18

18.5

19

19.5

20

Time (sec)

Figure 10. Comparison of percentage of mean strength and activation for biceps femoris muscles of ACL-I leg ( ) and ACL-R leg ( ___ ) of a
subject after 1 year of surgery at 5 km h1 speed, representing more or less similar strength for muscles in both legs.

subject. Likewise, inter-subject comparisons are also provided

to view the differences of muscles characteristics among
subjects at various stages of recovery.
3.1.3. Superimposition of knee flexion/extension and
EMG envelops of different muscles
Superimposition of knee flexion/extension and EMG envelops of different muscles can be done in order to examine
the co-ordination and correlation between kinematics and

neuromuscular signals of a subject. Superimposition of

knee flexion/extension and vastus lateralis EMG envelop for
a healthy subject during walking at a speed of 3 km h1 is
shown in Figure 4. The vastus lateralis and vastus medialis
help in extending the knee joint and by overlapping their
EMG and corresponding knee flexion/extension, the contribution of these muscles on gait patterns can be determined.
In a normal healthy leg (Figure 4), during most of the gait
cycles the knee is properly extended due to good vastus
lateralis strength. However, during the third gait cycle

506

S. M. N. A. Senanayake et al.
(a)

J Med Eng Technol, 2013; 37(8): 498510

60

300

40

250
30
200
20
150
10

% of Mean EMG

Knee Extension/Flexion (degrees)

50

100

50

10
25

25.5

26

26.5

27

27.5

Time (sec)

40
300
200

20

% of Mean EMG

Knee Extension/Flexion (degrees)

J Med Eng Technol Downloaded from informahealthcare.com by University Brunei Darussalam on 10/30/13
For personal use only.

(b) 60

100
0

24.5

25

25.5

26

26.5

27

100
27.5

Time (sec)

Figure 11. Activation timings and percentage of mean strengths of vastus lateralis ( ) of (a) ACL-I and (b) ACL-R leg vs knee flexion/extension
( ___ ) of a subject after 2 months of surgery showing (a) visible and (b) low contraction of vastus lateralis during terminal swing, load response and
mid stance phases affecting the knee flexion/extension movements.

(1718 s) the low strength of the vastus lateralis causes

restricted extension in the knee. The activation timings and
strengths of vastus lateralis of ACL-I and ACL-R legs
overlapped with corresponding knee flexion/extension of a
subject approximately after 2 months of surgery are shown in
Figures 11(a) and (b), respectively. These figures illustrate
an observable difference in the knee angle in terms of
extension during the early stage of gait cycle of a subject after
2 months of surgery. The strength of vastus lateralis muscles
varies in the healthy and operated on leg for this subject,
which causes restricted extension during all gait cycles.
Similar overlapping can be plotted for hamstrings also.
The normal muscle recruitment of ST, BF and VL and
corresponding knee joint angle of controls can be visualized
in Figure 12(ac). Thus, the overlapping of kinematics and
EMG signals from different muscles can be used as a tool
to identify the normal/abnormal walking patterns and the
contribution of these muscles in generating those patterns.
3.2. Statistical analysis
The visual biofeedback (VBF) helped the physiotherapists in
identifying the exact muscles and gait phases for ACL-R
subjects for carrying out focused training and exercises in
order to restore the normal knee kinematics and required
muscles strength for these subjects. In order to evaluate the
effectiveness of visual biofeedback, the recruited ACL-R
subjects in Group B were randomly assigned to either of two
groups: with VBF (n 5) or without VBF (n 5). Similarly,
subjects in Group C were also randomized into one of the two

groups: with VBF (n 5) or without VBF (n 5). Following

the first session of collecting and monitoring data, all of the
ACL-R subjects were tested again after a period of 4 weeks
to find the effect of visual biofeedback. The average knee
extension angle during terminal stance, peak knee flexion and
normalized peak values for vastus lateralis, vastus medialis
and hamstring muscles strength were noted during ambulation at different speeds. The differences between groups
were tested using an independent sample Students t-test with
p50.05 considered as a significant value. For Group B,
significant differences (p50.05) were observed in knee
extension during terminal stance and peak knee flexion for
subjects having VBF as an assistive tool as compared to
subjects without VBF. Moreover, significant differences
(p50.05) were also noted in normalized peak values for
vastus lateralis and vastus medialis muscles. However, no
significant differences were noted in both groups (with VBF
and without VBF) for normalized peak values for hamstring
muscles strength. For Group C, significant differences
(p50.05) were observed only for the average knee extension
angle during terminal stance and normalized peak values for
vastus medialis strength in subjects having VBF as compared
to subjects without VBF.
3.3. Recovery status assessment
Based on the collected kinematics and EMG features, the
ANFIS classifiers were designed for different walking speeds
of the subjects. Each classifier was designed with a pattern set
of 68 selected features as input vectors from walking data and

Assessing post-anterior cruciate ligament

DOI: 10.3109/03091902.2013.837529
70

1000

60
50
40
500

30
20

% of Mean EMG

Knee Flexion/Extension (degrees)

(a)

507

10
0
10
15

15.5

16

16.5

17

17.5

18

18.5

19

0
19.5

Time (sec)

60
400

50
40
30
20

200

% of Mean EMG

Knee Flexion/Extension (degrees)

70

10
0
10
17.5

18

18.5

19

19.5

20

20.5

0
21

Time (sec)

(c) 70
60
400

50
40
30
20

200

% of Mean EMG

Knee Flexion/Extension (degrees)

J Med Eng Technol Downloaded from informahealthcare.com by University Brunei Darussalam on 10/30/13
For personal use only.

(b)

10
0
18

18.5

19

19.5

20

20.5

21

Time (sec)

Figure 12. Activation timings and strengths of (a) biceps femoris, (b) semitendinosus and (c) vastus lateralis ( ) vs knee flexion/extension (____) of
a healthy subject at a speed of 3 km h1.

its respective group (Group A, Group B and Group C) as

the target output. A sample set of 150 such patterns
(50 from each group) was prepared to train the ANFIS for
each speed. The testing data consisted of 30 samples (10
samples from each group) for verifying the classification
accuracy for each trained ANFIS network. Each ANFIS
was trained for 100 epochs and the step size had an initial
value of 0.01. At the end of 100 epochs the convergence
(mean square) error was found as 3.047  107, 7.749  107,
2.289  106, 6.449  107 and 2.942  107 for 2 km h  1,
3 km h1, 4 km h1, 5 km h1 and 6 km h1 speeds, respectively. The convergence of error for subjects walking at
2 km h1 is shown in Figure 13. Similar convergence
patterns were observed for other walking speeds. In order to
evaluate the performance of each ANFIS, a confusion
matrix was computed for all the data and the specificity and

sensitivity for each class for each speed was determined.

The performance evaluation of all systems is presented in
Table 1.

4. Discussion
The use of wireless body mounted sensors and the integration
of knee kinematics and neuromuscular signals have been
found helpful in assessing the recovery progress of ACL
reconstructed subjects. The monitoring of bio-signals and
their superimposition provide the co-ordination of knee
kinematics and muscles movements during ambulation. The
effect of each muscle, causing the variability of gait patterns,
can be detected for individual subjects using the proposed
system. Moreover, this study also suggests that the recovery
stage of subjects can be accurately identified based on the

508

S. M. N. A. Senanayake et al.
4

J Med Eng Technol, 2013; 37(8): 498510

x 107

3.9

J Med Eng Technol Downloaded from informahealthcare.com by University Brunei Darussalam on 10/30/13
For personal use only.

Root Mean Squared Error (RMSE)

3.8
3.7
3.6
3.5
3.4
3.3
3.2
3.1
3

10

20

30

40

50

60

70

80

90

100

Epoch

Figure 13. The curve of network convergence for 2 km h1 ANFIS.

Table 1. ANFIS performance evaluationsensitivity, specificity and
F-measure for different walking speeds.

Walking speed
2 km h1

3 km h1

4 km h1

5 km h1

6 km h1

Class
Group
Group
Group
Group
Group
Group
Group
Group
Group
Group
Group
Group
Group
Group
Group

A
B
C
A
B
C
A
B
C
A
B
C
A
B
C

Specificity
(%)

Sensitivity
(%)

F-measure
(%)

96.15%
88.46%
92.59%
100.00%
95.65%
95.45%
94.12%
95.45%
95.00%
93.75%
100.00%
92.86%
100.00%
100.00%
92.31%

80.00%
80.00%
77.78%
80.00%
100.00%
87.50%
80.00%
80.00%
85.71%
100.00%
83.33%
90.00%
100.00%
88.89%
100.00%

84.21%
76.20%
77.78%
88.89%
93.34%
87.50%
84.21%
80.00%
85.71%
94.11%
90.90%
90.00%
100.00%
94.11%
94.11%

features extracted from the kinematics and neuromuscular

data recorded during walking activity.
The biofeedback can help in identifying the muscles and
knee joint abnormality during ambulation in each phase of a
gait cycle. Once this abnormality/irregularity is recognized
then targeted and timely corrective measures can be taken.
Variations in knee angle between the ACL-R and ACL-I leg
can be monitored and the impact of ACL injury on the
contralateral leg can also be examined. The visual analyses
facilitates in identifying intra- and inter-subjects variations
of kinematics and neuromuscular signals during different
experiments. The superimposed/overlapped bio-signals can be
used to find the relationship between muscle recruitment and
knee joint movements. These relationships assist in detecting
the muscles which are causing the changes in the knee
flexion/extension. Further, it can also be used to visualize
the performance in different activities during rehabilitation

by observing different muscles activations. The statistical

analysis shows that visual biofeedback is useful in improving
the average knee extension, peak knee flexion and strength
of knee extensors. The differences in mean values of these
parameters for controlled and experimental groups suggest
that further investigations can be done in order to find
the impact of visual biofeedback during rehabilitation
training. The activation timings and duration properties of
muscles, tested for normal walking for healthy subjects by the
system, coincide with the previously reported research and
standards [45].
The restoration of normal gait patterns is an important
measure to evaluate the post-ACL reconstructed subjects
during convalescence. The persisted alterations in gait
parameters can result in cartilage degenerations and osteoarthritis [68]. Walking is one of the most common activities
performed by human beings and any changes in knee
dynamics during this activity would definitely have longterm effects. Based on the integrated kinematics and neuromuscular data for walking, the use of an adaptive neuro-fuzzy
intelligent system has been found quite effective in analysing
and classifying the recovery stage. Due to the fuzzy system,
the ANFIS takes care of the variability in the kinematics and
EMG data. The ANFIS performance analysis shows that the
system performs well in classification for higher walking
speeds, as during these speeds the knee and muscles
movements become more identifiable. This suggests that
the consideration of other dynamic activities like jogging
and running can be useful in differentiating the healthy and
ACL-R limbs.
Although the classification performance of ANFIS is
satisfactory, the system has been tested for a limited number
of patients (available from sports medicine and research
centre and performance optimization centre, Brunei
Darussalam). In order to further test the system for clinical
applications, more subjects with additional parameters would
be appropriate. The effects of gender, type of protocols

Assessing post-anterior cruciate ligament

DOI: 10.3109/03091902.2013.837529

followed by subjects, type of graft used and other physical

parameters were not considered in the design of the study.
ACL injury not only affects the knee movements in the
sagittal plane, but it also alters knee dynamics in another two
planes (rotation and abduction/adduction) [10]. The inclusion
of these kinematics movements in addition to existing
parameters can also be done to enhance the recovery
assessment performed by the developed system. Further, the
effects of other muscles on recovery assessment and design of
ANFIS can also be investigated.

3.

4.

5.

J Med Eng Technol Downloaded from informahealthcare.com by University Brunei Darussalam on 10/30/13
For personal use only.

5. Conclusions and future work

This paper introduced a novel approach for ACL recovery
monitoring based on integration and superimposition of knee
kinematics and neuromuscular parameters. The proposed
intelligent integrated system records the knee movements and
EMG data using wearable sensors and then performs recovery
analysis after extracting relevant features using a continuous
wavelet transform. Four EMG features (RMS value, mean
frequency, maximum and minimum wavelet coefficients) and
one kinematics feature (average knee flexion/extension) are
extracted during selected phases of a gait cycle to prepare
the patterns set for training and testing the ANFIS. This
system can be used as an assistive tool in conjunction with the
existing rehabilitation monitoring mechanisms. It enables the
clinicians, trainers and physiotherapists to objectively monitor
the rehabilitation progress of athletes during different convalescence stages. Further testing of the system with inclusion
of other activities (balance testing, one leg jumping, etc.)
and additional EMG features will be done in future as part
of ongoing research.

6.

7.

8.
9.

10.

11.

12.

Acknowledgement
This work is supported by the University Research Council
(URC) grant scheme at the Universiti Brunei Darussalam
under the grant No: UBD/PNC2/2/RG/1(195). with the title
Integrated Motion Analysis System (IMAS). The authors
appreciate the sports medicine centre at ministry of sports and
performance optimization centre at ministry of defence,
Brunei Darussalam for providing Brunei national athletes as
test subjects who had undergone the rehabilitation process
due to ACL surgeries as well as healthy test subjects involved
for non-invasive rehabilitation experiments. Further, the
authors also acknowledge the support provided by Mr
Illepurma Ranasinghe (head physical strength and conditioning) and Ms Maria Leah (physiotherapist) from the performance optimization centre at the Ministry of Defence.

Declaration of interest

13.

14.

15.

16.
17.
18.

The authors report no conflicts of interest. The authors alone

are responsible for the content and writing of the paper.

References
1. Murrell, G.A., Maddali, S., Horovitz, L., Oakley, S.P., and Warren,
R.F., 2001, The effects of time course after anterior cruciate
ligament injury in correlation with meniscal and cartilage loss.
American Journal of Sports Medicine, 29, 914.
2. Reider, B., Arcand, M.A., Diehl, L.H., Mroczek, K., Abulencia, A.,
Stroud, C.C., Palm, M., Gilbertson, J., and Staszak, P., 2003,

19.
20.

509

Proprioception of the knee before and after anterior cruciate

ligament reconstruction. Arthroscopy: The Journal of Arthroscopic
& Related Surgery, 19, 212.
Devita, P., Hortobagyi, T., Barrier, J., Torry, M., Glover, K.,
Speroni, D., and Money, J., Mahar, M., 1997, Gait adaptations
before and after anterior cruciate ligament reconstruction surgery.
Medicine and science in sports and exercise. Medicine & Science in
Sports & Exercise, 29, 853859.
Potter, H.G., Jain, S.K., Ma, Y., Black, B.R., Fung, S., and
Lyman, S., 2011, Cartilage injury after acute, isolated anterior
cruciate ligament tear: Immediate and longitudinal effect with
clinical/MRI follow-up. American Journal of Sports Medicine, 40,
276285.
Hall, M., Stevermer, C.A., and Gillette, J.C., 2012, Gait analysis
post anterior cruciate ligament reconstruction: Knee osteoarthritis
perspective. Gait & Posture, 36, 5660.
Jin, A., Lynch, J., Scholes, C., Li, Q., Coolican, M., and Parker, D.,
2012, Osteoarthritis following anterior cruciate ligament
reconstruction. Journal of Bone & Joint Surgery, 94-B (Suppl
XLI), 125.
Haughom, B., Schairer, W., Souza, R.B., Carpenter, D., Ma, C.B.,
and Li, X., 2012, Abnormal tibiofemoral kinematics following ACL
reconstruction are associated with early cartilage matrix degeneration measured by MRI T1rho. The Knee, 19, 482487.
Wong, J.M., Khan, T., Jayadev, C.S., Khan, W., and Johnstone, D.,
2012, Anterior cruciate ligament rupture and osteoarthritis
progression. Open Orthopaedics Journal, 6, 295300.
Papannagari, R., Gill, T.J., Defrate, L.E., Moses, J.M.,
Petruska, A.J., and Li, G., 2006, In vivo kinematics of the
knee after anterior cruciate ligament reconstruction: A clinical
and functional evaluation. American Journal of Sports Medicine,
34, 20062012.
Gao, B., and Zheng, N.N., 2010, Alterations in threedimensional joint kinematics of anterior cruciate ligament-deficient
and -reconstructed knees during walking. Clinical Biomechanics,
25, 222229.
Moraiti, C.O., Stergiou, N., Vasiliadis, H.S., Motsis, E., and
Georgoulis, A., 2010, Anterior cruciate ligament reconstruction
results in alterations in gait variability. Gait & Posture, 32,
169175.
Moraiti, C.O., Stergiou, N., Ristanis, S., Vasiliadis, H.S., Patras, K.,
Lee, C., and Georgoulis, A.D., 2009, The effect of anterior cruciate
ligament reconstruction on stride-to-stride variability. Arthroscopy,
25, 742749.
Shi, D.L., Wang, Y.B., and Ai, Z.S., 2010, Effect of anterior
cruciate ligament reconstruction on biomechanical features of knee
in level walking: A meta-analysis. Chinese Medical Journal
(England), 123, 31373142.
Timoney, J.M., Inman, W.S., Quesada, P.M., Sharkey, P.F.,
Barrack, R.L., Skinner, H.B., and Alexander, A.H., 1993,
Return of normal gait patterns after anterior cruciate ligament
reconstruction. American Journal of Sports Medicine, 21, 887889.
Lewek, M., Rudolph, K., Axe, M., and Snyder-Mackler, L., 2002,
The effect of insufficient quadriceps strength on gait after anterior
cruciate ligament reconstruction. Clinical Biomechanics (Bristol,
Avon), 17, 5663.
Gogia, P.P., Braatz, J.H., Rose, S.J., and Norton, B.J., 1987,
Reliability and validity of goniometric measurements at the knee.
Physical Therapy, 67, 192195.
Miller, P.J., 1985, Assessment of joint motion. In: Rothstein JM,
editor. Measurement in Physical Therapy (New York: Churchill
Livingstone). pp. 103136.
Fiebert, I., Gresley, J., Hoffman, S., and Kunkel, K., 1994,
Comparative measurements of anterior tibial translation using the
KT-1000 knee arthrometer with the leg in neutral, internal rotation,
and external rotation. Journal of Orthopaedic & Sports Physical
Therapy, 19, 331334.
Hanten, W.P., and Pace, M.B., 1987, Reliability of measuring
anterior laxity of the knee joint using a knee ligament arthrometer.
Physical Therapy, 67, 357359.
Highgenboten, C.L., Jackson, A., Meske, N.B., 1989,
Genucom, KT-1000, and Stryker knee laxity measuring device
comparisons. Device reproducibility and interdevice comparison
in asymptomatic subjects. American Journal of Sports Medicine,
17, 743746.

J Med Eng Technol Downloaded from informahealthcare.com by University Brunei Darussalam on 10/30/13
For personal use only.

510

S. M. N. A. Senanayake et al.

21. Shaw, T., Chipchase, L.S., and Williams, M.T., 2004, A users guide
to outcome measurement following ACL reconstruction. Physical
Therapy in Sport, 5, 5767.
22. Akkaya, N., Ardic, F., Ozgen, M., Akkaya, S, Sahin, F., and Kilic,
A., 2012, Efficacy of electromyographic biofeedback and electrical
stimulation following arthroscopic partial meniscectomy: A
randomized controlled trial. Clinical Rehabilitation, 26, 224236.
23. Boucher, T., Wang, S., Trudelle-Jackson, E., and Olson, S., 2009,
Effectiveness of surface electromyographic biofeedback-triggered
neuromuscular electrical stimulation on knee rehabilitation.
North American Journal of Sports Physical Therapy, 4, 100109.
24. Feller, J., Hoser, C., and Webster, K., 2000, EMG biofeedback
assisted KT-1000 evaluation of anterior tibial displacement. Knee
Surgery, Sports Traumatology Arthroscopy, 8, 132136.
25. Scanlan, S.F., Chaudhari, A.M., Dyrby, C.O., and Andriacchi, T.P.,
2010, Differences in tibial rotation during walking in ACL
reconstructed and healthy contralateral knees. Journal of
Biomechanics, 43, 18171822.
26. Mayagoitia, R.E., Nene, A.V., and Veltink, P.H., 2002,
Accelerometer and rate gyroscope measurement of kinematics:
An inexpensive alternative to optical motion analysis systems.
Journal of Biomechanics, 35, 537542.
27. Gouwanda, D., and Senanayake, S.M.N.A., 2008, Emerging trends
of body-mounted sensors in sports and human gait analysis. In:
Abu Osman NA, Ibrahim F, Wan Abas WAB, Abdul Rahman HS,
Ting H-N, Magjarevic R, editors. IFMBE Proceedings (Berlin:
Springer), pp. 715718.
28. Atallah, L., Jones, G.G., Ali, R., Leong, J.J.H., Lo, B., and Yang,
G.-Z., 2011, Observing recovery from knee-replacement surgery by
using wearable sensors. Proceedings of the 2011 International
Conference on Body Sensor Networks: IEEE Computer Society,
2934.
29. Ayoade, M., Morton, L., and Baillie, L., 2011, Investigating
the feasibility of a wireless motion capture system to aid in the
rehabilitation of total knee replacement patients. Paper appeared in
5th International Conference on
Pervasive Computing
Technologies for Healthcare (PervasiveHealth), 2011 (Dublin,
Ireland), pp. 404407.
30. Saito, H., Watanabe, T., and Arifin, A., 2009, Ankle and knee joint
angle measurements during gait with wearable sensor system for
rehabilitation. IFMBE Proceedings, 25, 506509.
31. Hewett, T.E., Di Stasi, S.L., and Myer, G.D., 2013, Current
concepts for injury prevention in athletes after anterior cruciate
ligament reconstruction. American Journal of Sports Medicine, 41,
216224.
32. Leporace, G., Batista, L.A., Muniz, A.M., Zeitoune, G., Luciano,
T., and Metsavaht, L., Nadal, J., 2012, Classification of gait
kinematics of anterior cruciate ligament reconstructed subjects
using principal component analysis and regressions modelling.
Conference Proceedings IEEE Engineering in Medicine & Biology
Society, 2012, 65146517.
33. Toffola, L.D., Patel, S., Ozsecen, M.Y., Ramachandran, R., and
Bonato, P., 2012, A wearable system for long-term monitoring of
knee kinematics. International Conference on Biomedical and
Health Informatics (BHI), IEEE-EMBS (Shenzhen, China), 57
January 2012, pp. 188191.

J Med Eng Technol, 2013; 37(8): 498510

34. Christanell, F., Hoser, C., Huber, R., Fink, C., and Luomajoki, H.,
2012, The influence of electromyographic biofeedback therapy on
knee extension following anterior cruciate ligament reconstruction:
A randomized controlled trial. Sports Medicine, Arthroscopy,
Rehabilitation, Therapy & Technology, 4, 41.
35. Senanayake, S.M.N.A., Malik, O.A., Iskandar, P.M., and Zaheer,
D., 2012, A hybrid intelligent system for recovery and performance
evaluation after anterior cruciate ligament injury. 12th International
Conference on Intelligent Systems Design and Applications (ISDA),
2012, IEEE. pp. 986991.
36. Englehart, K., Hudgins, B., Parker, P.A., and Stevenson, M., 1999,
Classification of the myoelectric signal using time-frequency based
representations. Medical Engineering & Physics, 21, 431438.
37. Kilby, J., Mawston, G., Gholam Hosseini, H., 2007, Continuous
wavelet transform analysis for the classification of surface electromyography signals. In: Magjarevic R, Nagel JHD, editors. World
Congress on Medical Physics and Biomedical Engineering 2006
SE 246. 14th edn. IFMBE Proceedings: Springer Berlin
Heidelberg DA 2007/01/01; 2007. p 10341036 LA English.
38. Khezri, M., and Jahed, M., 2011, A neuro-fuzzy inference
system for sEMG-based identification of hand motion commands.
IEEE Transactions on Industrial Electronics, 58, 19521960.
39. Kocer, S., 2010, Classification of EMG signals using neuro-fuzzy
system and diagnosis of neuromuscular diseases. Journal of
Medical Systems, 34, 321329.
40. Hermens, H.J., Freriks, B., Disselhorst-Klug, C., and Rau, G., 2000,
Development of recommendations for SEMG sensors and sensor
placement procedures. Journal of Electromyography Kinesiology,
10, 361374.
41. Madgwick, S., 2010, An efficient orientation filter for inertial and
inertial/magnetic sensor arrays. Bristol, UK: Department of
Mechanical Engineering, University of Bristol.
42. Merletti, R., 1999, Standards for reporting EMG data. Journal of
Electromyography and Kinesiology, 9, IIIIV.
43. Jasiewicz, J.M., Allum, J.H.J., Middleton, J.W., Barriskill, A.,
Condie, P., Purcell, B., and Li, R.C.T., 2006, Gait event detection
using linear accelerometers or angular velocity transducers in ablebodied and spinal-cord injured individuals. Gait & Posture, 24,
502509.
44. Evans, A.L., Duncan, G., and Gilchrist, W., 1991, Recording
accelerations in body movements. Medical & Biological
Engineering & Computing, 29, 102104.
45. Perry, J., 1992, Gait analysis: Normal and pathological function
(NJ, USA: SLACK Inc.).
46. Ismail, A.R., and Asfour, S.S., 1998, Continuous wavelet transform
application to EMG signals during human gait. Conference Record
of the Thirty-Second Asilomar Conference on Signals, Systems &
Computers, 1, 325329.
47. Kilby, J., and Hosseini, H.G., 2006, Extracting effective features
of SEMG using continuous wavelet transform. Conference
Proceedings IEEE Engineering in Medicine & Biology Society, 1,
17041707.
48. Jang, J.-S.R., 1993, ANFIS: Adaptive-network-based fuzzy inference system. IEEE Transactions on Systems, Man, and Cybernetics,
23, 665685.