tmp357A.tmp

215

International Journal of Hybrid Intelligent Systems 10 (2013) 215–235
DOI 10.3233/HIS-130178
IOS Press

CO
PY

Anterior cruciate ligament recovery
monitoring system using hybrid
computational intelligent techniques
S.M.N. Arosha Senanaykea,∗, Owais Ahmed Malika , Pg. Mohammad Iskandara and Dansih Zaheerb
a

b

Universiti Brunei Darussalam, Gadong, BE, Brunei Darussalam
Sports Medicine and Research Center, Hassan Bolkiah National Stadium, Berakas, Brunei Darussalam

TH
OR

Abstract. A recovery monitoring system, based on hybrid computational intelligent techniques, is presented for post anterior
cruciate ligament (ACL) injured/reconstructed subjects. The case based reasoning methodology has been combined with fuzzy
and neuro-fuzzy techniques in order to develop a knowledge base and a learning model for classification of recovery stages and
monitoring the progress of ACL-reconstructed subjects during the convalescence regimen. The system records kinematics and
neuromuscular parameters from lower limbs of healthy and ACL-reconstructed subjects using body-mounted wireless sensors
and a combined feature set is generated by performing data transformation and feature reduction techniques. In order to classify
the recovery stage of subjects, fuzzy k-nearest neighbor technique and adaptive neuro-fuzzy inference system have been applied
and results have been compared. The system was successfully tested on a group of healthy and post-operated athletes for analyzing their performance during ambulation and single leg balance testing activities. A semi-automatic process has been employed
for case adaptation and retention, requiring input from the physiotherapists and physiatrists. The system can be utilized by physiatrists, physiotherapists, sports trainers and clinicians for multiple purposes including maintaining athletes’ profile, monitoring
progress of recovery, classifying recovery status, adapting recovery protocols and predicting athletes’ sports performance
Keywords: Case based reasoning (CBR), fuzzy/neuro-fuzzy system, anterior cruciate ligament (ACL), wireless sensors, knee
injury, recovery monitoring

AU

1. Introduction

A successful and timely return to previous healthy
activities, without any residual deficits, following an
injury is fundamental to effective treatment and efficient convalescence protocol. The lower limb injuries
in individuals may lead to have certain short and long
term disabilities such as altered gait patterns seen due
to defective cure and inadequate rehabilitation. Joints
and ligaments together with bony framework act in collaboration with intact neuromuscular system for normal locomotion. The normal locomotry patterns can be
∗ Corresponding author: S.M.N. Arosha Senanayke, Universiti
Brunei Darussalam (UBD), Gadong, BE 1410, Brunei Darussalam.
E-mail: [email protected].

c 2013 – IOS Press. All rights reserved
1448-5869/13/$27.50 

affected by either fracture of bony structure, tear and
rupture of ligaments or defective/impaired neuromuscular control. Coordination among bones, ligaments,
joints and neuromuscular control can be reverted back
to near normal level by skilled repair and well monitored rehabilitation process [34]. This can also avoid or
minimize various long term complications.
Recently, different machine learning and Artificial
Intelligence (AI) techniques have been utilized for rehabilitation of lower limb movements after neurological disorders and implantation of artificial limbs [37,
61]. These intelligent techniques combined with bodymounted sensors have been used in the fields of medical diagnosis and pathology classification [20,44,45].
The common applications of this blend are in the fields
of stroke diagnosis, hand gesture recognition, stress
monitoring, identification of hand motion commands,

216 S.M.N.A. Senanayke et al. / Anterior cruciate ligament recovery monitoring system using hybrid computational intelligent techniques

CO
PY

functional and static knee stability, range of motion)
are suggested to safe return to sports rather than relying on some specific time period for rehabilitation [16,
42,51]. For measuring ACL reconstruction outcome
and assessing functional ability, combination of subjective and objective evaluations have been reported
as no single score or instrument is comprehensive in
measuring all parameters [57]. Commonly used objective outcome measures include range of motion (ROM)
measurement using goniometer [21,41], circumference
measurement [14,19,22,40,59], functional hop tests [6,
7,51] and testing knee stability/laxity using arthrometer [18,25,27]. On the other hand, a variety of subjective outcome measures exist which are based on assessing different scores including knee rating scores
(e.g. IKDC, Lyshom) and analog scores (e.g. NRS,
VAS) [57]. Additionally, there are few measures (e.g.
SIRAS) available to indicate the subject’s rehabilitation adherence in the clinical environment following
ACL reconstruction [11].
A comprehensive system which can assist in preventing ACL injury, intelligently monitoring the rehabilitation progress of athletes and predicting their
sports performance after ACL reconstruction is useful
for both physiatrists as well as athletes. Such an exhaustive system can be built by developing a knowledge base of athletes’ profiles which would store current and past information (i.e. before and after ACL injury/surgery) about athletes’ knee dynamics and other
relevant parameters. This information can be transformed for developing a learning model from past experiences and strategies followed for improving the recovery after ACL injury/reconstruction. A suitable approach to build such system is to use Case Based Reasoning (CBR) methodology which maintains a knowledge base of past experiences (old problem-solution
pairs) and solves the new problems by using or adapting the existing solutions [1]. CBR has been used in
variety of medical applications for diagnosis and classification [8,13]. CBR approach has been proved successful in providing diagnosis and prognosis for stroke
patients based on the data collected through robotic
tool [5]. A potential solution for stroke diagnosis is
provided by finding similar cases from the repository
of stroke patients with explicit diagnosis and prognosis. A gait disorder analysis system has been proposed
in [35] for general practitioners using CBR approach. It
uses accelerometer data for lateral and forward movements of center of mass to retrieve similar cases based
on experts’ knowledge. CBR has also been combined
with other techniques including fuzzy logic and neu-

AU

TH
OR

post-stroke rehabilitation, electromyography (EMG)
classification for neuromuscular diseases and in general for human motion/gait analysis [5,9,31,38,53,63].
The data from subjects are acquired through lightweight wearable sensors and generally, the kinematics,
kinetics, neuromuscular or other relevant signals are
recorded and stored. Further processing and classification of the data are performed by using pattern matching and/or soft computing techniques. The main advantage of these computational intelligent techniques
is to provide efficient assistive tools to clinicians, in
addition to existing clinical methods, for assessing the
health conditions of subjects. In recent studies, the
data from multiple sensors are integrated before applying the intelligent mechanisms. This multi-modal approach proved to be successful in providing a broad
analysis of patients by using a more detailed feature set
rather than relying only on few parameters [55].
Although, the usage of different sensors has been investigated in recent studies for observing the changes
in knee joint dynamics and gait pattern during rehabilitation process after knee surgery [4,23,52] but few
efforts have been made for developing an intelligent
assistive tool for monitoring recovery of athletes after
Anterior Cruciate Ligament (ACL) injury and reconstruction [36,56]. ACL injury causes deterioration in
the sports performance or premature end to the career
for athletes [42]. ACL is one of the crucial ligaments
for knee joint stability, maintaining normal gait patterns and preventing anterior tibial translation, controlling knee axial rotation and varus movements. It can
be completely or partially ruptured as a result of deceleration due to sudden changes of direction, twisting and/or pivoting during sports, like soccer and basketball. The tearing/rupture of ACL affects gait patterns and variability, and causes changes in kinematics, kinetics and neuromuscular activities of athlete.
The absence of ACL results in the loss of mechanical control of the knee, loss of proprioception due to
unavailability of mechanoreceptors present in the ligament and neurophysiologic dysfunction, hindering the
athletes to participate in sports activities. ACL rupture
not only requires in-time reconstruction but also needs
a supervised objectively assessable rehabilitation program. Efficient and effective rehabilitation programs
are essential for athletes following ACL reconstruction
as well as for those having ACL deficiency. The rehabilitation programs are designed to rebuild muscle
strength, re-establish joint and neuromuscular control
and to enable the athletes to return to pre-injury activity level. The goal based criteria (muscle strength,

S.M.N.A. Senanayke et al. / Anterior cruciate ligament recovery monitoring system using hybrid computational intelligent techniques 217

CO
PY

search Center at the Ministry of Sports, Brunei Darussalam. The healthy subjects were having an age (mean
± std) of 31.0 ± 8.29 years, height (mean ± std)
164.5 ± 13.03 cm, and weight (mean ± std) 65.25
± 20.17 kg. For ACL reconstructed subjects the age
(mean ± std), height (mean ± std) and weight (mean ±
std) were 30.8 ± 3.73 years, 166.9 ± 7.18 cm and 68.2
± 14.46 kg, respectively. The ACL-reconstructed subjects were at different stages of recovery ranging from
3 months to more than a year after surgery. 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.
2.2. Activities monitored

During rehabilitation period of the subjects after
ACL surgery, different activities are assessed by the
physiotherapists/physiatrists as per the rehabilitation
protocol. Due to variations in the post-surgery time periods of the participants and based on the guidelines
from the physiatrist, two activities (walking at different
speeds and single leg flat surface open eye balance testing) were selected from the rehabilitation protocol for
the data acquisition. All of the subjects walked at three
different speeds: 4 km/h, 5 km/h and 6 km/h for 30–
35 seconds time on a motorized treadmill. The balance
testing data were collected from the subjects standing
on single leg on a flat surface for a time period of 15–
20 seconds. The balance test was performed for both
legs of the ACL reconstructed subjects. The data were
collected for two sets for each activity.

AU

TH
OR

ral network to develop more robust classification systems [49]. A fuzzy CBR has been used to design a decision support system, with accuracy around 88%, for
stress diagnosis in different subjects [9]. The imprecision and uncertainty in sensors’ measurements was
dealt by using a fuzzy similarity metric for case retrieval. Alahakone [2] proposed a hybrid system combining Self Organizing Maps (SOM) and CBR for
evaluating postural control based on trunk sway obtained during a tandem Romberg stability test. The prediction accuracy of the system was claimed to be more
than 90%.
This paper presents a novel approach for monitoring and classification of recovery status and evaluating athletes’ sports performance after ACL reconstruction by combining machine learning techniques with
CBR system. The kinematics and neuromuscular data
are collected from ACL reconstructed and healthy subjects using body-mounted wireless sensors. The extraction of important features from electromyography
data has been performed by using wavelet decomposition technique and a feature reduction method has
been applied to reduce the large number of features. A
knowledge base has been designed to store subjects’
profiles, recovery sessions’ data and problem/solution
pairs. In order to retrieve the most similar cases, two
techniques namely fuzzy k-nearest neighbor algorithm
and adaptive neuro-fuzzy inference system, have been
compared and case selection is done by using case density function. Once relevant cases are selected, adaptation is performed by adjusting the recovery protocols
for individuals and the performance evaluation can be
done. This system will facilitate the clinicians, physiotherapists, physiatrists and sports trainers to identify the subjects at various stages of recovery process
and those lacking behind the desired level. In addition, it will also assist them to do the needful intervention or accelerating the ongoing activity level of subjects. Moreover, objectively assessed recovery status
provides subjects’ self-satisfaction for the current stage
and ensures their positive feedback in future.

3. Hybrid intelligent framework for ACL recovery
monitoring
The hybrid intelligent framework for ACL recovery
monitoring is shown in Fig. 1. The components and
sequential operations of the system are elaborated in
the following sub-sections.

2. Experimental setup

3.1. System hardware

2.1. Subjects

The system hardware is composed of two subsystems operating simultaneously: 1) wireless microelectro-mechanical systems (MEMS) motion sensors
units with command module and a radio for wireless
data transmission (KinetiSense from ClevMed Inc.)
and 2) wireless electromyography sensors unit (Bio-

Fourteen subjects (4 healthy with no ACL or other
knee injury and 10 unilateral ACL reconstructed) were
recruited from Performance Optimization Centre at
Ministry of Defense and Sports Medicine and Re-

Formulate
Problem

EMG Sensors

Motion Sensors

Find matching case(s) from
clusters using intelligent
retrieval mechanism

Data Acquisition
No
Generate Case

Found?
Yes

Initial Processing

Reuse and Revise
Solution
Data Storage
Case & Knowledge
Base
Feature Extraction
Re-Grouping &
Re-Training
Feature Projection
Initial Grouping
and Case
Generation

Identify the treatment and
changes in recovery protocol

CO
PY

Data Collection and Processing

Retrieve Case

Retain Learned
Case

Performance Evaluation and
Prognosis

Data Analysis and Classification

System
Hardware

218 S.M.N.A. Senanayke et al. / Anterior cruciate ligament recovery monitoring system using hybrid computational intelligent techniques

Fig. 1. Hybrid intelligent framework for monitoring post ACL injury recovery and sports performance.

angular rates and linear acceleration from all four sensors were sent simultaneously in serial format to the
Command Module which transferred the digital data to
the computer through the USB Receiver connected to
the computer’s USB port. The raw data were viewed
and saved using KinetiSense software package for processing.
The surface EMG sensors record the action potential
of skeletal muscles which indicate the force of muscles. The amplitude of the EMG measurements (usually in millivolts) is related to the amount of force generated from the muscle’s contraction while performing
the motion. The EMG signals were recorded by placing foam snap electrodes on four different knee extensor and flexor muscles including vastus medialis (VM),
vastus lateralis (VL), semitendinosus (ST) and biceps
femoris (BF) on both legs of the subjects during walking. For balance testing activity, EMG data were also
collected from gastrocnemius medialis (GM) muscle
in addtition to above four muscles. SENIAM guidelines were followed for skin preparation, placing the
electrodes and EMG recording [26]. The sampling rate
to collect EMG signals was set to 960 Hz (maximum
available) at 12 bit Analog/Digital conversion and 2-D
linear acceleration was also recorded from BioRadio
unit to synchronize both KinetiSense and BioCapture
devices. The EMG data recorded using BioRadio were
transferred to the computer wirelessly through USB receiver connected to the computer’s USB port. These
signals were viewed and stored using BioCapture software package for processing. The data from both sys-

AU

TH
OR

Radio), electrodes and a USB receiver for wireless
data transmission (BioCapture System from ClevMed.
Inc.). The body-mounted MEMS motion sensors measure the 3-D motion using three orthogonal gyroscopes
and three orthogonal accelerometers. The BioRadio,
worn by the subjects, records the EMG signals through
surface electrodes attached to the target muscles and
then wirelessly transmits them to the computer using
USB receiver. These small, light-weight and untethered sensors can provide required parameters during
ambulation, one leg jumping or balance testing activities without obstructing the lower limb movements.
Additionally, a video camera provides the visualization
of human motion by capturing the video of the experiments.
3.2. Data collection and processing

In order to collect data and perform initial processing, following steps were performed.
3.2.1. Data acquisition
The sensing units were setup for recording signals
from the motion and EMG sensors. Two motion sensors were attached to each leg (one on thigh and one
on shank) of a subject using flexible bulk and Velcro
straps to note the knee joint movements. Each motion
sensor recorded the 3-D angular rates and 3-D linear
acceleration during the motion performed by the subjects. Motion sensor data were sampled at a frequency
of 128 Hz and digitized in the sensor unit. The recorded

CO
PY

S.M.N.A. Senanayke et al. / Anterior cruciate ligament recovery monitoring system using hybrid computational intelligent techniques 219

Fig. 2. Variations in knee flexion/extension during different gait phases.

tems were recorded in parallel and transmitted to the
same computer. These stored signals (angular rate, acceleration, and raw EMG data) were then exported to
the files for further processing.

AU

TH
OR

3.2.2. Initial processing
In order to prepare the data for feature set generation and extraction, the raw signals from motion
and EMG sensors were filtered and transformed into
the required format. For each motion sensor, measurements for zero-referencing were obtained prior to starting the experiment (actual motion) when the subjects
were standing in upright position and these measurements were subtracted from angular rate measurements
of corresponding sensor during the experiment. The
measurements obtained from the MEMS gyroscopes
were low pass filtered using 6th order Butterworth filter before computing the orientations. Similarly, the
raw EMG data from all muscles were band-pass filtered using 4th order Butterworth filter and the mean
of each signal was subtracted from it.

namics was generated from combined kinematics and
EMG signals. Before feature set generation, both signals were synchronized (due to different sampling rates
and recording delays) and then based on the monitored
activity (normal/brisk walking or balance testing), the
features set generation and selection steps were performed. For normal/brisk walking, the gait cycles were
identified for each subject for all sessions and then features were computed for each phase of a gait cycle
(Fig. 2). The knee kinematics and relevant muscles’
strength vary in each phase of a gait cycle and these
changes reflect the progress of the recovery after the
ACL injury/reconstruction. The marking of gait phases
and, segmentation of data from multiple EMG channels and motion sensors were based on the percentages
defined in [46].
For balance testing activity, a window of four seconds was chosen (based on experiments) as a data
segment for kinematics and EMG signals. The details
about kinematics and EMG features are described in
the subsequent sections.

3.2.3. Data storage
The processed data are stored in the knowledge base
and a preliminary profile of the subject is generated.
This data can also be used for visualizing the superimposed kinematics and neuromuscular signals for each
subject [11].
3.2.4. Feature set generation
Feature set generation is one of the most important steps for data analysis and classification tasks. The
selection of right features is crucial for minimizing
the assessment error and achieving high classification
accuracy. In this research, a feature set for knee dy-

Kinematics features The kinematics of the knee changes after ACL injury/reconstruction. In order to monitor these changes, the flexion/extension measurements
were computed for each gait cycle using angular rates
recorded through motion sensor units placed on the
thigh and shank segments of both legs. The sensors
were aligned to provide knee angle about the sagittal
plane. The orientations of lower extremities (θRThigh ,
θRShank , θLThigh and θLShank ) were estimated by applying trapezoidal integration method on respective angular rates (ωRThigh , ωRShank , ωLThigh and ωLShank ) of both
lower limbs. If ω(t) represents the angular rate of either
thigh or shank at time t and Δt is the sampling time,

CO
PY

220 S.M.N.A. Senanayke et al. / Anterior cruciate ligament recovery monitoring system using hybrid computational intelligent techniques

(a)

TH
OR

(b)

(c)

Fig. 3. (a) Knee flexion/extension (degrees), (b) Angular rate for thigh (rad/sec) (c) Angular rate for shank (rad/sec) for multiple gait cycles of an
ACL intact leg.

θ(t) =

AU

then the estimated orientation (θ(t) ) of thigh and shank
at time t is computed using Eq. (1).
ω(t) + ω(t+1)
× Δt
2

(1)

The knee angle was computed by subtracting θShank
from θThigh for both legs. Knee flexion/extension and
corresponding angular rates for thigh and shank are
shown for a healthy (ACL intact) leg in Fig. 3. After
marking the seven gait phases, three statistics values
(root mean square value, standard deviation and maximum knee angle) were computed for a gait cycle. Thus,
a total of twenty one kinematics features were generated for each gait cycle for one leg.
EMG features The EMG features represent the electrophysiological properties of the muscle fibers dur-

ing contraction. In different applications, various timedomain, frequency-domain and time-frequency-domain features extracted from EMG signals have been
used [32,60,63]. Due to random and non-stationary nature of EMG signals, the wavelet transformation has
been found more appropriate for analyzing such biosignals as compared to only time/amplitude or frequency analysis. The wavelet analysis of EMG signals
can help in visualization of variations in energy levels
of muscles during different segments of the activities
being monitored and localizing the singularities in the
EMG data. The original EMG signal for vastus lateralis muscles of a healthy subject is shown in Fig. 4(a)
for multiple gait cycles. The filtered and rectified version of the same signal is shown in Fig. 4(b) and its
corresponding CWT coefficients representing the en-

CO
PY

S.M.N.A. Senanayke et al. / Anterior cruciate ligament recovery monitoring system using hybrid computational intelligent techniques 221

(a)

TH
OR

(b)

(c)

Fig. 4. 3-D representation of coefficients energy for multiple gait cycles for an ACL intact subject (a) EMG signals for vastus lateralis muscle of
a healthy subject for multiple gait cycles, (b) filtered and rectified signal and (c) corresponding 3-D plot for CWT coefficients.

AU

ergy distribution of signal are shown in Fig. 4(c). The
onset of each muscle and its duration of activation can
easily be identified by plotting the wavelet coefficients
for different muscles during each gait cycle for walking activity or for relevant segments for balance testing
activity. In Fig. 5, the muscle strength of vastus lateralis is shown for an ACL intact leg during a gait cycle.
The corresponding wavelet coefficient plot represents
the bright region where the muscle activity/strength is
higher and the darker region shows the low activity in
the muscle. Such representation can be useful to detect
the muscles which are less activated during each gait
cycle or during balance training.
Although such representations are useful to display the behavior of muscles during different segments
of activities but for recovery and performance evaluation using intelligent mechanisms, a feature vector
is required. In [56], an EMG feature set consisting
of features from time-domain, frequency-domain and

wavelet coefficients from continuous wavelet transform (CWT) has been proved effective in classifying
subjects at different recovery stages after ACL reconstruction for slow walking speeds. In this study, an
EMG feature set has been generated based on multilevel discrete wavelet decomposition analysis. By employing discrete wavelet transform (DWT), EMG signal can be iteratively transformed into multi-resolution
subsets of coefficients using suitable wavelet basis
function. The time-domain EMG signal is passed
through various high pass and low pass filters to obtain the approximation coefficient subsets (e.g. cA1
. . . cA3 as shown in Fig. 6) and the detail coefficient subsets (e.g. cD1 . . . cD3 as shown in Fig. 6)
where the level of decomposition can be pre-defined
(e.g. 3 in Fig. 6). The choices of level and mother
wavelet depend on the domain and applications, but
for EMG analysis Daubechies 04/05 mother wavelet
with four/five levels of decomposition has shown bet-

CO
PY

222 S.M.N.A. Senanayke et al. / Anterior cruciate ligament recovery monitoring system using hybrid computational intelligent techniques

AU

TH
OR

Fig. 5. Muscle activation during a gait cycle for ALC intact leg (above) and corresponding wavelet energy levels (below) – bright regions depict
high activation of muscles.

Fig. 6. EMG signal for vastus lateralis of a healthy (ACL intact) subject walking at 5 km/h analyzed by DWT with Daubechies 05 mother wavelet
at 3-levels of decomposition.

ter performance results [47,48,60]. In this research,
Daubechies 05 wavelet basis function with five levels
of decomposition has been used to compute EMG features. An example of EMG signal analyzed by DWT
with mother wavelet Daubechies 05 (db05) with 3levels of decomposition is shown in Fig. 6.

For both walking and balance testing activities,
wavelet coefficients (cD1–cD5 and cA5) were computed which represent the energy distribution of the
EMG signals from four/five identified muscles. From
these coefficients, following five statistical features
were calculated for each phase of gait cycle and bal-

S.M.N.A. Senanayke et al. / Anterior cruciate ligament recovery monitoring system using hybrid computational intelligent techniques 223

M AV =

N
1 
|emgi |
N i=1

(2)

Kij = {knp }
Eij

=

TH
OR

where N is the phase/segment length.
– Standard deviation of the coefficients in each
phase/segment.
– Average power of the coefficients in each phase/
segment.
These features were chosen based upon previous
studies for bio-signal classification [24,60] and the experiments performed in this study for recovery and performance assessment. After computing the features, a
feature set for each activity was generated as follows:
Let Kij and Eij represent the kinematics and EMG features’ set for ith subject and j th gait cycle (for walking activity for each speed) or j th session (for balance
testing activity), respectively then
(3)

{mp1l , mp2l , mp3l , mp4l }

(4)

F =

{fij }

AU

where knp represents the nth kinematics feature (n =
1 . . . 3) for pth phase, mpl represents the lth EMG feature (l = 1 . . . 30, six values for 5 features) for pth
phase for muscles. The value for p depends on the activity being monitored (p = 1 . . . 7 for walking activity. Let fij represents the feature vector for ith subject
for j th gait cycle/session and total feature set for an
activity is F then
fij = {Kij } ∪ {Eij }

length for balance testing activity for each segment was
153 (6 × 5 EMG features × 5 muscles + 3 kinematics feature). In order to reduce the length of these feature vectors, different feature selection and reduction
algorithms were investigated. The feature selection algorithms select only a subset of the features to represent the model and this subset is searched to minimize
the classification error subject to certain constraints.
However, the selection algorithms such as sequential
forward selection (SFS) or sequential backward selection (SBS) have a disadvantage due to their inability
to re-assess the feature’s importance as once a feature
is added to the subset it cannot be removed. Moreover, due to large number of features, these algorithms
found to be time consuming and slow to test each subset for the prediction error. In contrast to feature selection algorithms, the feature projection/transformation
algorithms try to determine the best combination of
the original features to form a new and smaller feature
set. The features selection techniques discard some of
the features completely and so the information provided by those features is completely lost. In order
to avoid this problem, Two popular feature reduction
techniques, principal component analysis (PCA) and
independent component analysis (ICA) have been used
to extract features from bio-signals including EMG,
electrocardiograph (ECG) and electroencephalograph
(EEG) signals [17,32,54,62,64]. These techniques reduce the dimension of data set by removing the redundancy in the data and replacing the group of variables
with a single variable while still not rejecting some
of the features completely from the data set. ICA performs well when applied as a preprocessing step in different domains including EMG data [3,12], but in order
to reduce the features the PCA was found to be a better choice based on the experiments results performed
in this study for combine kinematics and EMG features. PCA transforms the original feature set of variables f ∈ F ⊆ RN into a new feature set of variables
v ∈ V ⊆ RM of reduced dimension by minimizing
the mean-square error (MSE) between the original set
F and projected set V [29]. For a given set of input vectors, fi (i = 1 . . . n), where each fi is of dimension N ,
PCA linearly transforms each vector fi into new vector
vi by using Eq. (7).

CO
PY

ance testing segment.
– Maximum of the coefficient values in each phase/
segment.
– Minimum of the coefficient values in each phase/
segment.
– Mean absolute value (MAV) of the coefficients in
each phase/segment.

(5)
(6)

3.2.5. Dimensionality reduction
The proposed set of features for recovery classification is based on kinematics and neuromuscular signals. In this study all of the above features were extracted for walking activity at different speeds and balance testing for open/closed eyes. The feature vector
length for walking activity for each speed was 841 (6 ×
5 EMG features × 4 muscles × 7 phases + 3 kinematics feature × 7 phases). Similarly, the feature vector

vi = AT fi

(7)

where A is the N ×N orthogonal matrix whose ith column ai is the ith eigenvector of the sample covariance
matrix in Eq. (8).

224 S.M.N.A. Senanayke et al. / Anterior cruciate ligament recovery monitoring system using hybrid computational intelligent techniques
n

1
fi fiT
n i=1

(8)

These new variables, called principal components,
are the linear combination of original variables, which
form an orthogonal basis for the space of data. It is often the first few principle components, explaining the
maximum variance of the original data, which are chosen for the classification task.

Fig. 7. Overall structure of knowledge base containing different
types of data.

speeds on the treadmill and the balance testing activity.
The fcm function in Fuzzy Logic Toolbox from MATLAB starts with an initial guess for the cluster centers
for marking the mean location of each cluster. Each
feature vector as a data point is then assigned a membership grade for each cluster. fcm follows an iterative process to minimize the objective function Eq. (9)
and then decides the right cluster centers Eq. (10) and
membership grade Eq. (11) for each feature vector.

TH
OR

3.2.6. Initial grouping and case generation
Initially when the case repository was empty, the
data collected from the ACL reconstructed and healthy
subjects were clustered using fuzzy clustering technique (elaborated in next section) to form the groups
based on their current recovery stage. This initial
grouping was done due to the variations in the rehabilitation stages of these subjects. The labels for these
groups were identified by finding the distances of their
centers from the center of healthy subjects’ cluster and
were manually verified. The case generation was a
semi-automatic process in this research. The extracted
features and recommendation from the clinicians were
used to generate a case, based on the proposed case
representation (explained in Section 3.4) and stored in
the case repository.

CO
PY

Cov =

3.3. Groups formation using fuzzy clustering

AU

In order to retrieve the most similar cases efficiently,
the clustered case base organization has been used.
These clusters represent the groups to which ACLreconstructed/ healthy subjects belong. Due to the imprecise nature of motion and neuromuscular parameters, the fuzzy clustering has been adopted as opposed
to the crisp/classical clustering algorithms which assign an object to only one group. This is generally difficult in domains like recovery classification or gait analysis where variations in data are more common and
one object may belong to different groups with different degree of memberships. Fuzzy clustering partitions
the sample space and organizes the data into approximate clusters [10,15]. O’Malley et al. [43] has applied
fuzzy clustering for classifying gaits of children with
cerebral palsy into different groups. Fuzzy clustering
approach has also been used to identify the effect of
temporal patterns on the walking speed based on foot
switches [39].
The fuzzy C-means (FCM) has been applied to the
transformed feature set ‘V ’ of kinematics and neuromuscular data collected during walking at different

C
N 


2
um
ij ||vi − cj ||

(9)

i=1 j=1

cj =

N


um
ij vi

i=1
N


i=1

uij =

C

k=1

(10)
um
ij


1
||vi − cj ||
||vi − ck ||

2
 m−1

(11)

where vi is the ith of M  -dimensional data from PCA,
cj is the M  -dimension center of the cluster, uij is the
degree of membership of vi in the cluster j, || || represents the similarity (Euclidian distance) between any
measured data and the center, and m is the number of
clusters.
In order to avoid any pre-assigned number of clusters, the cluster validity measure proposed in [50] has
been modified to identify the number of initial clusters.

S.M.N.A. Senanayke et al. / Anterior cruciate ligament recovery monitoring system using hybrid computational intelligent techniques 225

k

i=1

Validity =


∀v⊆V

μvci × v − ci 2/KN
2

min(ci − cj  )
(12)

Subjects and session The Subject table contains the
biographical and some injury/surgery related information for each subject based on the form submitted before the start of experiments. This Session table contains the information about each session of the experiments i.e. whenever the data are acquired from
a subject (pre-injury, post-injury or post surgery), a
new session is created and the data are stored for
each activity monitored during that session. Pre-injury
data are utilized in comparing the post-injury or postoperated performance differences when in case an athlete gets an ACL or other knee injury. Post-injury
data are utilized to compare the pre- and post-operated
changes in kinematics and muscles attributes. Postoperated/Recovery data are recorded for the visits
made by the subjects during different recovery stages
and this data acquisition is based on the milestones
set in the rehabilitation protocol and the advice of the
physiotherapist/physiatrist.

TH
OR

where ci and cj are the cluster centers, N is the number
of records in the data set, k is the maximum number of
clusters, v is a subset (selected features) of V and ||v −
ci ||2 and ||ci − cj ||2 represent the intra and inter cluster
distances, respectively. During the initial stage of CBR
system, the number of identified clusters varies as there
is less number of subjects in each group. As the case
repository is populated with more cases and it contains
enough data from subjects at different recovery stages,
the number of clusters becomes almost fixed.

data during convalescence. In order to manage the
knowledge base repository, a relational database has
been used (see Fig. 8) to reduce the storage redundancy
and flexibility.
The knowledge base evolves with the time-period
when new problems are presented and new cases are
added to the system. This evolution process makes it
more useful for domains where subject’s specific monitoring and prognosis mechanisms are required. The
important components of the KB’s implementation are
briefly explained as below.

CO
PY

In [50], a simple validity measure based upon the intra vs. inter cluster distance has been designed for Kmeans clustering. This measure has been adapted, as in
(9) by including fuzzy membership value μ assigned
to each of the clusters and minimum validity index has
been selected.

3.4. Data analysis using case based reasoning system

The data analysis and classification of recovery was
done by using case based reasoning approach. CBR
paradigm is based on the concept of solving new problems by using/modifying the similar previous experiences (problem-solution pairs). The problem-solving is
a four step process in CBR system [1]:

AU

– Retrieve: Finding similar case(s) from the knowledge base whose problem description best matches with the given problem.
– Reuse: Reusing the solution of most similar case
to solve the new problem.
– Revise: Adapting/Modifying the chosen solution
according to the differences in new problem.
– Retain: Storing the new problem-solution pair as
a case once it has been solved.
The details about these processes and other components of the CBR system designed for this study are
discussed in next subsections.

3.4.1. Knowledge base (KB)
The overall structure of knowledge base is depicted
in Fig. 7. The knowledge base contains different types
of information including case library, raw and processed data, domain knowledge, historical data available for subjects (pre-injury, post-injury) and session

Activity and sets During each session, data for different activities (walking, running, one leg jumping
and balance testing etc.) can be recorded. In the postoperated data collection, the selection of activity depends on the current stage of recovery for an individual. For each activity, multiple datasets are stored to
identify outliers and avoid any measurements error.
Kinematics and EMG data The data recorded during
each set are stored in these tables after data processing
step. The link to the corresponding raw data file is also
stored along with each record.
Cluster and case The clustering is performed in order to group the subjects according to their similarities. This table contains the cluster indices and cases
under each cluster. In the initial stage of populating the
case library, the variation in number of clusters indices
is possible which is reduced as more cases are added
to the repository. The Case table contains all informa-

CO
PY

226 S.M.N.A. Senanayke et al. / Anterior cruciate ligament recovery monitoring system using hybrid computational intelligent techniques

Fig. 8. Knowledge base implementation using relational database.

the identified cluster by using case density function
given in Eq. (13) [58].

SMee

TH
OR

tion related to each solved case stored in the repository including problem description, transformed attribute/value pairs, recovery stage, class, recommendations and protocols followed during rehabilitation etc.,
as described in the next section.

AU

3.4.2. Case representation
Each case in the knowledge base is composed of
two components: activity set and overall set. The activity set consists of subsets of activity-based (walking at
different speeds and balance testing) problem-solution
pair. The problem part of each subset is represented
as attribute/value pairs for the selected kinematics and
EMG features and the corresponding solution part is
made up of the recovery classification and the treatments suggested for the next stage. The overall set contains the performance evaluation for the athletes, treatments given and followed at current stage of recovery,
link to the previous sessions’ outcomes (if any) and
the case description. The case description contains the
subject’s biographical, sports and other relevant clinical information. This information may provide assistance in refining the retrieval and adaption of similar
cases.
3.4.3. Retrieval of similar cases
In order to retrieve the most similar cases for the
given scenario, a two-step process is performed. In the
first step, the best matching class is identified for the
new problem (after transforming it using PCA coefficients) by using machine learning techniques. In the
second step, the most similar cases are retrieved from

Case_Density(e, C) =

e ∈C−{e}

|C| − 1

(13)

where SMee represents the similarity between case
e and e (computed by cosine/Euclidian measure) and
|C| is the number of cases in class C. The retrieved
cases are arranged in the descending order of case densities and first k cases are selected where k is the user’s
choice.
In order to perform the first step i.e. finding the best
matching class, following two techniques were investigated.

Fuzzy k-nearest neighbor classifier The principal of
fuzzy k-nearest neighbor (f-knn) is similar to the crisp
k-nn algorithm where the properties of an input vector v, with unknown class, are expected to be matching
with the vectors in its neighborhood. In f-knn, instead
of assigning a single class to v, represented by majority
of its k nearest neighbors, a membership value of the
class is assigned to it [30]. Thus a vector can belong to
multiple classes with different degrees of membership.
The membership value for vector v is computed as a
function of distance from its k-nearest neighbors and
the degree of membership of these neighbors to available classes [30]. Such technique is quite useful for domains like recovery classification where a subject can
belong to multiple classes of recovery stage based on
their performance in each activity.

S.M.N.A. Senanayke et al. / Anterior cruciate ligament recovery monitoring system using hybrid computational intelligent techniques 227

y=

n


n


w i fi =

i=1

wi fi

i=1
n


i=1

j=1

μij (xj )

3.4.5. Retaining the learned case
The new case may be retained in the library after formulating a solution based on the adjustments of parameters for individuals. The attribute/value pairs for the
selected features are generated by taking an average of
their values for all sets. Whenever a new case is added,
the fuzzy clustering algorithm generates new clusters
(if required) and assigns different membership grades
to the related cases. The cluster table is also updated in
the KB.

(15)

TH
OR

wi =

n


(14)
wi

3.4.4. Reuse and repair of cases
After retrieving the most similar case(s), the next
step is to use and adapt the solution of this case to
improve the recovery process of the athlete for the
next stage. This semi-automatic process requires the
involvement of the clinicians/physiotherapist to decide
any changes in the rehabilitation protocol based on the
recommendations and indication of performance level
from the retrieved case(s). Additionally, modifications
may also be done in the recommendation section of the
previous stages of the new problem.

CO
PY

Adaptive neuro-fuzzy inference system (ANFIS) The
ANFIS is a fuzzy Sugeno model that adapts the membership function parameters using neural network and
learns from the given data set [28]. The variations in
kinematics signals and non-stationary nature of EMG
data lead the recovery classification task challenging.
The ANFIS can be more useful for building models
for such inputs. It can effectively identify the stochastic changes in bio-signals, and can also deal with the
impreciseness in measurements and variations due to
subjects’ physiological conditions [31,33]. The system
adjusts the membership function (μ) parameters based
on the given data, and the number of rules and output
of fuzzy rules is minimized. The overall output of an
n-input system is given in Eq. (14) where fi is the previous layer’s output and wi is called the firing strengths
of the rules Eq. (15).

AU

The subtractive-clustering method was used to partition the data due to large number of inputs, no requirement of setting the number of clusters in advance and noise robustness. It is one-pass algorithm
for estimating number of clusters by finding the high
density data point regions in feature space. The cluster center is the point with the highest number of
neighbors. The learning parameters of membership
functions (premise parameters) and outputs (consequent parameters) were tuned using hybrid learning
algorithm. This algorithm combines the least square
method and gradient descent method which makes the
convergence faster in the large search space. The forward pass (least square method) and backward passes
(gradient descent method) are used to optimize the
consequent and premise parameters, respectively. After determining the consequent parameters, the output of ANFIS is calculated and the premise parameters are adjusted based on output error by using backpropagation algorithm [28].
The classifiers (f-knn and ANFIS) were trained separately for each activity (three walking speeds and balance testing). The results of the retrieval step are: 1)
the recovery classification of the given problem i.e. the
recovery classification of an athlete for each activity
based on the values of currently measured parameters
and 2) retrieval of cases for adaptation process.

4. Results and discussion
In order to provide an objective assessment of the recovery status after ACL reconstruction and a long term
analysis of subjects, the developed system was tested
for the selected activities. The feature set reduction was
an important step before applying the CBR technique
for case generation and selection. Due to large feature vector size it was impractical to consider all features directly for classification so PCA was used to reduce the feature sets for walking and balance testing
activities. PCA was initially applied on original feature
set consisting of 841 and 153 features for all walking
speeds and balance testing activities, respectively. The
maximum and minimum coefficient values for each
phase/segment were further removed from the feature
set for all activities due to their low effect on class separation and classification accuracy. The PCA was then
applied on new feature set consisting of 525 and 93
features for all walking speeds and balance testing activities, respectively.
The variations explained by principal components
(PCs) for these activities are shown in Figs 9 to 12. A
careful analysis of these PCs showed that the number
of PCs required increasing the cumulative distribution
of variance in data to 90% or more vary among activities: 42 PCs for walking at 4 km/h, 39 PCs for walking

CO
PY

228 S.M.N.A. Senanayke et al. / Anterior cruciate ligament recovery monitoring system using hybrid computational intelligent techniques

TH
OR

Fig. 9. Percentage of variance explained by first 50 principal components for walking at 4 km/h.

AU

Fig. 10. Percentage of variance explained by first 50 principal components for walking at 5 km/h.

Fig. 11. Percentage of variance explained by first 50 principal components for walking at 6 km/h.

Fig. 12. Percentage of variance explained by first 50 principal components for single leg eyes open flat surface balance testing.

CO
PY

S.M.N.A. Senanayke et al. / Anterior cruciate ligament recovery monitoring system using hybrid computational intelligent techniques 229

Fig. 13. Original data projected on first two principal components.

TH
OR

Fig. 14. Clusters’ centers identified by FCM for walking at 5 km/h – cluster 1 (×) cluster 2 (o) cluster 3 (+). (Colours are visible in the online
version of the article; http://dx.doi.org/10.3233/HIS-130178)

AU

Fig. 15. Clusters’ centers identified by FCM for walking at 6 km/h – cluster 1 (×) cluster 2 (o) cluster 3 (+). (Colours are visible in the online
version of the article; http://dx.doi.org/10.3233/HIS-130178)

at 5 km/h, 37 PCs for walking at 6 km/h and 6 PCs for
balance testing. After finding the PCS, the coefficients
matrix and these transformed features were stored for
data analysis. Figure 13 shows the original data projected on first three PCs for walking at 4 km/h. It is
visible from Fig. 13 that the PCA is also useful in identifying the subjects who have very different values of
the input parameters (marked inside circles).
The transformed features were clustered using FCM
to form the groups of subjects who were healthy or
at similar stage of recovery after ACL reconstruction.
This step is part of the initial grouping and case generation process. Three clusters were initially identified
for each activity using proposed validity index Eq. (11)
which were manually verified and found to be appro-

priate. Figures 14 and 15 show 3-D scatter plot for
the first three PCs and the cluster centers identified by
FCM for walking at 5 km/h and 6 km/h respectively,
where clusters 1, 2 and 3 represent subjects less than
6 months after surgery, subjects after 1 year of surgery
and subjects without ACL injury, respectively. Some
of the data points lie on the boundary of second and
third clusters which depicts that some of the subjects
belong to both clusters with more or less equal membership grades and cannot be completely categorized
into one group. These are the subjects who have recovered to a level where they are similar to the subjects in
healthy group. This is natural as even after following
the same rehabilitation protocol, the recovery may depend on individuals’ other physical parameters. After

230 S.M.N.A. Senanayke et al. / Anterior cruciate ligament recovery monitoring system using hybrid computational intelligent techniques

Technique Activity

f-knn

ANFIS

Classification
Classification
accuracy
accuracy
mean ± SD (%) maximum (%)
Walking at 4 km/h 47.00 ± 11.67
64.00
Walking at 5 km/h 52.39 ± 8.58
80.00
Walking at 6 km/h 44.57 ± 16.56
74.28
Balance testing
43.33 ± 26.17
83.33

Walking at 4 km/h
Walking at 5 km/h
Walking at 6 km/h
Balance testing

85.93 ± 4.47
90.48 ± 4.39
99.34 ± 0.54
98.89 ± 0.62

91.29
99.89
99.94
99.66

Activity
Walking 4 km/h

Walking 5 km/h

AU

Class Sensitivity Specificity Precision
(%)
(%)
(%)
1
100.00
93.75
62.50
2
82.61
96.67
95.00
3
96.00
96.43
96.00
1
2
3

100.00
92.00
100.00

94.29
100.00
100.00

71.43
100.00
100.00

Walking 6 km/h

1
2
3

100.00
98.04
100.00

98.82
100.00
100.00

96.15
100.00
100.00

Single leg balance test

1
2
3

100.00
96.15
100.00

98.08
100.00
100.00

93.75
100.00
100.00

This study demonstrates the successful use of CBR
methodology with fuzzy/neuro-fuzzy techniques for
developing a framework for objectively monitoring
and assessing the recovery progress for ACL reconstructed subjects. The kinematics and electromyography data from walking and balance testing activities
provide useful statistical features for differentiating
subjects at various levels of recuperation. The principal
components extracted by using PCA offer more effective information about the human locomotion in lower
dimensions and higher classification accuracy can be
achieved by utilizing fewer PCs.
Anterior cruciate ligament injury has long term effects on the human locomotion including cartilage
degeneration and early onset of osteoarthritis. The
changes in the kinematics and neuromuscular parameters persist even after 1–2 years of ACL surgery. In
order to handle these challenges and minimizing the
risk of re-injury, a comprehensive system based on
CBR methodology has been developed. The pre-injury,
post-injury and during recovery data storage can provide detailed analysis about the recovery of subjects
and a comparison of current and past performance in
any specific activity. The collection of relevant existing
experiences for rehabilitation can help in intervening
the current rehabilitation protocol of new subjects and
making adjustments as per the requirements based on
the objective monitoring. This will help in designing a
rehabilitation protocol specific to the individual needs
as well as improving the generalized rehabilitation process.
The integration of kinematics and electromyography
data has proved to be quite effective in identification of
stage of recovery of ACL reconstructed subjects. The
electromyography signals are non-stationary in nature
so wavelet analysis was performed in order to extract

TH
OR

forming the clusters, the case generation step was performed by storing problem-solution pairs, corresponding treatments given/followed by the subjects and the
recommendations from the physiatrists.
In order to classify a new subject as part of the case
retrieval, the input parameters were first transformed
using the coefficient matrix and then f-knn and ANFIS were used to classify the subject based on trained
clustered data. The cross validations of both methods
were done by partitioning the data into two groups:
training data (90% of the total data) and test data (10%
of the total data). The partitions were made by using
cvpartition function from MATLAB which randomly
partitions observations into a training set and a test set
with stratification such that both training and test sets
have roughly the same class proportions as in the clusters’ groups (original groups). The tenfold cross validation was performed for all activities using f-knn and
ANFIS techniques. The ANFIS networks were trained
for 100 epochs and the step size was set to an initial value of 0.01. The experiments showed that only
first few PCs were enough to achieve high classification accuracy and inclusion of further PCS deteriorated
the classification performance. The effect of number
of PCs on classification accuracy of f-knn and ANFIS
are presented in Figs 16 to 23. In general, it was found
that first 2 to 5 PCs produced that maximum classification accuracy for both techniques for all activities. The
maximum and average accuracy of each technique are
shown in Table 1.
The ANFIS technique proved to be more accurate in
classifying the recovery status of the subjects as compared to f-knn for all activities. Table 2 presents the
classification performance of ANFIS networks for all
activities. After classifying the recovery stage, the relevant cases are selected using case density function
Eq. (13). The performance of each subject is compared
with the most similar retrieved cases by using their recommendations and next stage results.

Table 2
ANFIS classification performance for all activities

CO
PY

Table 1
Overall mean and maximum classification accuracy

CO
PY

S.M.N.A. Senanayke et al. / Anterior cruciate ligament recovery monitoring system using hybrid computational intelligent techniques 231

TH
OR

Fig. 16. Classification accuracy for walking at 4 km/h using fuzzy k-nearest neighbor technique.

Fig. 17. Classification accuracy for walking at 5 km/h using fuzzy k-nearest neighbor technique.

AU

Fig. 18. Classification accuracy for walking at 6 km/h using fuzzy k-nearest neighbor technique.

Fig. 19. Classification accuracy for single leg eyes open balance testing using fuzzy k-nearest neighbor technique.

CO
PY

232 S.M.N.A. Senanayke et al. / Anterior cruciate ligament recovery monitoring system using hybrid computational intelligent techniques

Fig. 20. Classification accuracy for walking at 4 km/h using adaptive neuro-fuzzy inference system technique.

TH
OR

Fig. 21. Classification accuracy for walking at 5 km/h using adaptive neuro-fuzzy inference system technique.

AU

Fig. 22. Classification accuracy for walking at 6 km/h using adaptive neuro-fuzzy inference system technique.

Fig. 23. Classification accuracy for single leg eyes open balance testing using adaptive neuro-fuzzy inference system technique.

S.M.N.A. Senanayke et al. / Anterior cruciate ligament recovery monitoring system using hybrid computational intelligent techniques 233

CO
PY

monitoring and performance evaluation. The system
has been successfully tested for normal/brisk walking
and single leg eyes open balance testing activities for a
group of ACL reconstructed and healthy subjects and
high classification accuracy has been demonstrated by
using ANFIS for the monitored activities. The system
will be enhanced in future by including data for further activities from the rehabilitation protocol. The preand post-injury data from subjects are being collected
to build the data repositories and to compare the performance differences after ACL reconstruction or ACL
injury. Additionally, the effect of other parameters (e.g.
gender, age, type of sports) will also be considered during case retrieval task.
Acknowledgments

The authors would like to acknowledge the support provided by Sport Medicine and Research Center
(SMRC) and Performance Optimization Center (POC),
Brunei Darussalam for providing test subjects for this
study.

AU

TH
OR

useful information from EMG data. CWT analysis of
EMG signals provides visualization of the activation
and strength of different muscles during walking and
balance activities (Fig. 4). The bright/dark energy coefficients from CWT analysis can help in differentiating active and non-active muscles during each phase
of an activity (Fig. 5). In order to prepare a feature set
for classification, discrete wavelet transform was used
on EMG data from relevant muscles to perform timefrequency analysis and initially five different statistical
features were generated from the wavelet coefficients
(Fig. 6). The final selection of EMG features was based
on the performance of feature reduction and classification techniques. Three (MAV, standard deviation and
power of wavelet coefficients in each phase/segment)
out of five EMG features in combination with three
kinematics features were found useful in getting higher
classification accuracy.
The data were collected from subjects who were at
their different stages of recovery so instead of manually labeling the subjects’ data, fuzzy clustering was
used to generate initial groups based on kinematics
and EMG parameters. These labels were later used for
classification task. Fuzzy k-nearest neighbor and adaptive neuro-fuzzy inference system were applied and
compared for retrieving the similar cases by identifying the class of the given problem. The performance
demonstrated by ANFIS shows that the combination
of fuzzy logic with the learning capabilities of neural network is well suited for classifying the recovery
progress of ACL reconstructed subjects for different
activities based on the given parameters. With the addition of more cases, the system will become more robust due to its learning capability.

5. Conclusion and future work

The results of this study demonstrate that the use of
hybrid computational intelligent techniques can help
in developing a system for clinicians and physiatrist
to assess the rehabilitation status of the subjects after
ACL injury/surgery. This assistive tool can determine
an athlete’s ACL injury profile and, potential knee joint
and neuromuscular problems during rehabilitation and
provide valuable feedback in clinical environment as
decision support system. A knowledge base consisting of athletes’ pre-injury, post-injury and post-surgery
profiles, and possible set of solutions for assistance
during rehabilitation period provides a comprehensive
platform for ACL injury prevention, recovery progress

References
[1]

A. Aamodt and E. Plaza, Case-based reasoning: Foundational
issues, methodological variations, and system approaches, AI
Commun 7 (1994), 39–59.
[2] A.U. Alahakone and S.M.N.A. Senanayake, A hybrid intelligent system combining self organizing maps and case based
reasoning for evaluating postural control, in: IEEE Conference on Multisensor Fusion and Integration for Intelligent
Systems (MFI) (2010), 38–43.
[3] G.B. Ali, H. Al-Timemy, N. Outram and J. Escudero, Reduction in classification errors for myoelectric control of hand
movements with independent component analysis, in: The 5th
International Conference on Information Technology (ICIT)
(2011).
[4] L. Atallah, G.G. Jones, R. Ali, J.J.H. Leong, B. Lo and G.-Z.
Yang, Observing recovery from knee-replacement surgery by
using wearable sensors, in: Proceedings of the 2011 International Conference on Body Sensor Networks IEEE Computer
Society (2011), 29–34.
[5] M. Baig, Case-based reasoning: An effective paradigm to facilitate diagnostic support for stroke patients, VDM Verlag
2009.
[6] W.D. Bandy, K.R. Rusche and F.Y. Tekulve, Reliability and
limb symmetry for five unilateral functional tests of the lower
extremities, Isokinetics and Exercise Science 4 (1994), 108–
111.
[7] S.D. Barber, F.R. Noyes, R. Mangine and M. DeMaio, Rehabilitation after acl reconstruction: Function testing, Orthopedics 15 (1992), 969–974.
[8] S. Begum, M.U. Ahmed, P. Funk, N. Xiong and M. Folke,
Case-based reasoning systems in the health sciences: A survey
of recent trends and developments, Trans Sys Man Cyber Part
C 41 (2011), 421–434.

234 S.M.N.A. Senanayke et al. / Anterior cruciate ligament recovery monitoring system using hybrid computational intelligent techniques
[25]

W.P. Hanten and M.B. Pace, Reliability of measuring anterior laxity of the knee joint using a knee ligament arthrometer,
Phys Ther 67 (1987), 357–359.
[26] H.J. Hermens, B. Freriks, C. Disselhorst-Klug and G. Rau,
Development of recommendations for SEMG sensors and
sensor placement procedures, J Electromyogr Kinesiol 10
(2000), 361–374.
[27] C.L. Highgenboten, A. Jackson and N.B. Meske, Genucom,
KT-1000, and Stryker knee laxity measuring device comparisons. Device reproducibility and interdevice comparison in
asymptomatic subjects, Am J Sports Med 17 (1989), 743–746.
[28] J.S.R. Jang, ANFIS: Adaptive-network-based fuzzy inference
system, IEEE Transactions on Systems Man, and Cybernetics
23 (1993), 665–685.
[29] I.T. Jolliffe, Principal component analysis, Springer-Verlag
New York, Inc, New York, 2002.
[30] J.M. Keller, M.R. Gray and J.A.G. Jr, A fuzzy k-nearest neighbor algorithm, IEEE Transactions on Systems Man and Cybernetics 15 (1985).
[31] M. Khezri and M. Jahed, A neuro–fuzzy inference system for
sEMG-based identification of hand motion commands, IEEE
Transactions on Industrial Electronics 58 (2011), 1952–1960.
[32] R.N.A.J. Khushaba and Adel, Fuzzy wavelet packet based
feature extraction method for multifunction myoelectric control, International Journal of Biomedical Sciences 2 (2007),
186–194.
[33] S. Kocer, Classification of EMG signals using neuro-fuzzy
system and diagnosis of neuromuscular diseases, J Med Syst
34 (2010), 321–329.
[34] L.M. Kruse, B. Gray and R.W. Wright, Rehabilitation after
anterior cruciate ligament reconstruction: A systematic review, J Bone Joint Surg Am 94 (2012), 1737–1748.
[35] O.A. Kuchar, Y.L. Lazukova, D. Riordan and J.L. Leahey, A
case-based reasoning system involving a quantized method
for gait disorder diagnosis, (1998).
[36] G. Leporace, L.A. Batista, A.M. Muniz, G. Zeitoune, T. Luciano, L. Metsavaht and J. Nadal, Classification of gait kinematics of anterior cruciate ligament reconstructed subjects using principal component analysis and regressions modelling,
Conf Proc IEEE Eng Med Biol Soc (2012), 6514–6517.
[37] Y. Li, Y. Tian and W. Chen, sEMG signal classification for
the motion pattern of intelligent bionic artificial limb, in: Advances in neural networks – ISNN 2011 SE – 58, D. Liu, H.
Zhang, M. Polycarpou, C. Alippi and H.D. He, eds, Springer
Berlin Heidelberg DA – 2011/01/01, 2011, pp. 505–513.
[38] A. Mannini and A.M. Sabatini, Machine learning methods for
classifying human physical activity from on-body accelerometers, Sensors 10 (2010), 1154–1175.
[39] A. Manurung and J. Yoon, Fuzzy clustering of temporal parameters of gait during stance phase for walking speed estimation, in: International Conference on Control Automation
and Systems (ICCAS) Gyeonggi-do, 2010.
[40] E. Maylia, J.A. Fairclough, L.D.M. Nokes and M.D. Jones,
Can thigh girth be measured accurately? A preliminary investigation, Journal of Sport Rehabilitation 8 (1999), 43–49.
[41] P.J. Miller, Assessment of joint motion, in: Measurement in
Physical Therapy Churchill Livingstone, J.M. Rothstein, ed.,
New York, 1985, pp. 103–136.
[42] G.D. Myer, M.V. Paterno, K.R. Ford, C.E. Quatman and T.E.
Hewett, Rehabilitation after anterior cruciate ligament reconstruction: criteria-based progression through the return-tosport phase, J Orthop Sports Phys Ther 36 (2006), 385–402.
[43] M.J. O’Malley, M.F. Abel, D.L. Damiano and C.L. Vaughan,
Fuzzy clustering of children with cerebral palsy based on

CO
PY

S. Begum, M.U. Ahmed, P. Funk, N. Xiong and B.
Scheele, Classify and diagnose individual stress using calibration and fuzzy case-based reasoning, in: Proceedings of
the 7th International Conference on Case-based reasoning:
Case-based Reasoning Research and Development SpringerVerlag, Belfast, Northern Ireland, UK, 2007, pp. 478–491.
[10] J.C. Bezdek, Pattern recognition with fuzzy objective function
algorithms, Kluwer Academic Publishers, 1981.
[11] B.W. Brewer, J.L. Van Raalte, A.J. Petitpas, J.H. Sklar, M.H.
Pohlman, R.J. Krushell, T.D. Ditmar, J.M. Daly and J. Weinstock, Preliminary psychometric evaluation of a measure of
adherence to clinic-based sport injury rehabilitation, Physical
Therapy in Sport 1 (2000), 68–74.
[12] L.J. Cao, K.S. Chua, W.K. Chong, H.P. Lee and Q.M. Gu, A
comparison of PCA, KPCA and ICA for dimensionality reduction in support vector machine, Neurocomputing Support
Vector Machines 55 (2003), 321–336.
[13] J.F. De Paz, S. RodrÃ-guez, J. Bajo and J.M. Corchado, Casebased reasoning as a decision support system for cancer diagnosis: A case study, International Journal of Hybrid Intelligent Systems 6 (2009), 97–110.
[14] J. Dietrichson and T.O. Souryal, Preoperative and postoperative rehabilitation after anterior cruciate ligament tears,
Orthopaedic Physical Therapy Clinics of North America 3
(1994), 539–554.
[15] J. Dunn, A fuzzy relative of the ISODATA process and its
use in detecting compact well-separated clusters, Journal of
Cybernetics 3 (1973), 32–57.
[16] J. Ekstrand, Six versus eight months of rehabilitation after reconstruction of the ACL: A prospective randomized study on
soccer players, Science & Football 3 (1991), 31–36.
[17] K. Englehart, B. Hudgins, P.A. Parker and M. Stevenson,
Classification of the myoelectric signal using time-frequency
based representations, Med Eng Phys 21 (1999), 431–438.
[18] I. Fiebert, J. Gresley, S. Hoffman and K. Kunkel, Comparative
measurements of anterior tibial translation using the KT-1000
knee arthrometer with the leg in neutral, internal rotation, and
external rotation, J Orthop Sports Phys Ther 19 (1994), 331–
334.
[19] P.J. Fowler and W.D. Regan, The patient with symptomatic
chronic anterior cruciate ligament insufficiency, results of
minimal arthroscopic surgery and rehabilitation, Am J Sports
Med 15 (1987), 321–325.
[20] M. Frize, N. Yu and S. Weyand, Effectiveness of a hybrid pattern classifier for medical applications, International Journal
of Hybrid Intelligent Systems 8 (2011), 71–79.
[21] P.P. Gogia, J.H. Braatz, S.J. Rose and B.J. Norton, Reliability
and validity of goniometric measurements at the knee, Phys
Ther 67 (1987), 192–195.
[22] R.S. Gotlin, A.L. Sherman, N. Sierra, M.A. Kelly, Z. Pappas and W.N. Scott, Measurement of brake response time after right anterior cruciate ligament reconstruction, Arch Phys
Med Rehabil 81 (2000), 201–204.
[23] D. Gouwanda and S.M.N.A. Senanayake, Emerging trends of
body-mounted sensors in sports and human gait analysis in:
4th Kuala Lumpur International Conference on Biomedical
Engineering 2008, (Vol. 21), N.A. Abu Osman, F. Ibrahim,
W.A.B. Wan Abas, H.S. Abdul Rahman, H.-N. Ting and R.
Magjarevic, eds, Springer Berlin Heidelberg, 2008, pp. 715–
718.
[24] I. Güler and E.D.E.D.Ü. Übeyli, Adaptive neuro-fuzzy inference system for classification of EEG signals using wavelet
coefficients, Journal of Neuroscience Methods 148 (2005),
113–121.

AU

TH
OR

[9]

S.M.N.A. Senanayke et al. / Anterior cruciate ligament recovery monitoring system using hybrid computational intelligent techniques 235
S.F.M.J. Sahin, Ugur pattern recognition with surface emg
signal based wavelet transformation, in: IEEE International
Conference onSystems, Man, and Cybernetics (SMC) (2012),
295–300.
[55] S.M.N.A. Senanayake, O.A. Malik and P.M. Iskandar, Wireless multi-sensor integration for acl rehabilitation using
biofeedback mechanism, in: Proceedings of ASME 2012 International Mechanical Engineering Congress & Exposition
IMECE12, Houston, Texas, USA, (2012).
[56] S.M.N.A. Senanayake, O.A. Malik, P.M. Iskandar and D. Zaheer, A hybrid intelligent system for recovery and performance evaluation after anterior cruciate ligament injury, in:
12th International Conference on Intelligent Systems Design
and Applications (ISDA) (2012), 986–991.
[57] T. Shaw, L.S. Chipchase and M.T. Williams, A users guide to
outcome measurement following ACL reconstruction, Physical Therapy in Sport 5 (2004), 57–67.
[58] S. Shiu and S. Pal, Foundations of Soft Case-Based Reasoning (Wiley Series on Intelligent Systems), Wiley-Interscience
2004.
[59] G.L. Soderberg, B.T. Ballantyne and L.L. Kestel, Reliability
of lower extremity girth measurements after anterior cruciate
ligament reconstruction, Physiother Res Int 1 (1996), 7–16.
[60] A. Subasi, Classification of EMG signals using combined features and soft computing techniques, Applied Soft Computing
12 (2012), 2188–2198.
[61] B.L. Zhang and Y. Zhang, Classification of cerebral palsy gait
by kernel fisher discriminant analysis, International Journal
of Hybrid Intelligent Systems 5 (2008), 209–218.
[62] L.Z.L. Zhang, H.P.H. Peng and C.Y.C. Yu, An approach
for ECG classification based on wavelet feature extraction and decision tree, (2010), 1–4, DO – 10.1109/WCSP.
2010.5633782.
[63] X. Zhang, X. Chen, Y. Li, V. Lantz, K. Wang and J. Yang, A
framework for hand gesture recognition based on accelerometer and emg sensors, IEEE Transactions on Systems, Man and
Cybernetics, Part A: Systems and Humans 41 (2011), 1064–
1076.
[64] Z. Zhou and B. Wan, Wavelet packet-based independent component analysis for feature extraction from motor imagery
EEG of complex movements, Clin Neurophysiol 123 (2012),
1779–1788.

CO
PY

[54]

AU

TH
OR

temporal-distance gait parameters, IEEE Trans Rehabil Eng 5
(1997), 300–309.
[44] A. Pantelopoulos and N.G. Bourbakis, Prognosis–a wearable
health-monitoring system for people at risk: Methodology and
modeling, IEEE Transactions on Information Technology in
Biomedicine 14 (2010), 613–621.
[45] A. Pantelopoulos and N.G. Bourbakis, A survey on wearable
sensor based systems for health monitoring and prognosis,
IEEE Transactions on Systems, Man, and Cybernetics Part C:
Applications and Reviews 40 (2010), 1–12.
[46] J. Perry, Gait analysis: Normal and pathological function, Delmar Learning (1992).
[47] A. Phinyomark, C. Limsakul and P. Phukpattaranont, Application of wavelet analysis in emg feature extraction for pattern classification, Measurement Science Review 11 (2011),
45–52.
[48] A. Phinyomark, A. Nuidod, P. Phukpattaranont and C. Limsakul, Feature extraction and reduction of wavelet transform
coefficients for EMG pattern classification, Elektronika ir
Elektrotechnika 122 (2012), 27–32.
[49] J. Prentzas and I. Hatzilygeroudis, Combinations of casebased reasoning with other intelligent methods, International
Journal of Hybrid Intelligent Systems 6 (2009), 189–209.
[50] S. Ray and R.H. Turi, Determination of number of clusters
in k-means clustering and application in color image segmentation, in: Proceedings of the 4th International Conference
on Advances in Pattern Recognition and Digital Techniques
(1999), 137–143.
[51] G.S. Roi, D. Creta, G. Nanni, M. Marcacci, S. Zaffagnini and
L. Snyder-Mackler, Return to official Italian first division soccer games within 90 days after anterior cruciate ligament reconstruction: A case report, J Orthop Sports Phys Ther 35
(2005), 52–61, discussion 61–56.
[52] B. Rosenhahn, U.G. Kersting, A.W. Smith, J.K. Gurney, T.
Brox and R. Klette, A system for marker-less human motion
estimation, in: Proceedings of the 27th DAGM Conference on
Pattern Recognition Springer-Verlag, Vienna, Austria, (2005),
230–237.
[53] S.H. Roy, M.S. Cheng, S.S. Chang, J. Moore, G. De Luca,
S.H. Nawab and C.J. De Luca, A combined sEMG and
accelerometer system for monitoring functional activity in
stroke, IEEE Trans Neural Syst Rehabil Eng 17 (2009), 585–
594.