Model for Testing Single-Leg Balance Performance of. Athletes after Knee .... provided by two physiotherapists/physiatrists for each trial (see section 2.4). Fig. 1.
An Interval Type-2 Fuzzy Logic Based Classification Model for Testing Single-Leg Balance Performance of Athletes after Knee Surgery Owais Ahmed Malik and S.M.N. Arosha Senanayake Faculty of Science, Universiti Brunei Darussalam, Gadong, Brunei Darussalam {11H1202, arosha.senanayake}@ubd.edu.bn
Abstract. Single-leg balance test is one of the most common assessment methods in order to evaluate the athletes’ ability to perform certain sports actions efficiently, quickly and safely. The balance and postural control of an athlete is usually affected after a lower limb injury. This study proposes an interval type2 fuzzy logic (FL) based automated classification model for single-leg balance assessment of subjects after knee surgery. The system uses the integrated kinematics and electromyography (EMG) data from the weight-bearing leg during the balance test in order to classify the performance of a subject. The data are recorded through wearable wireless motion and EMG sensors. The parameters for the membership functions of input and output features are determined using the data recorded from a group of athletes (healthy/having knee surgery) and the recommendations from physiotherapists and physiatrists, respectively. Four types of fuzzy logic systems namely type-1 non-singleton interval type-2 (NSFLS type-2), singleton type-2 (SFLS type-2), non-singleton type-1 (NSFLS type-1) and singleton type-1 (SFLS type-1) were designed and their performances were compared. The overall classification accuracy results show that the interval type-2 FL system outperforms the type-1 FL system in classifying the balance test performance of the subjects. This pilot study suggests that a fuzzy logic based automated model can be developed in order to facilitate the physiotherapists and physiatrists in determining the impairments in the balance control of the athletes after knee surgery. Keywords: single-leg balance, fuzzy logic, classification, knee injury, electromyography, kinematics
1
Introduction
Balance control and postural stability are very important for the athletes during sports activities requiring fast movements and changing directions quickly. A complex interaction of central nervous system and musculoskeletal system helps the athletes in maintaining balance during demanding sports actions. A lower limb injury (e.g. anterior cruciate ligament rupture) to an athlete may cause various complications including dynamic joint instability and neuromuscular/proprioception impairments which eventually affect his/her balance control during high-level sporting activities [1, 2]. A Ó Springer International Publishing Switzerland 2016 P. Chung et al. (eds.), Proceedings of the 10th International Symposium on Computer Science in Sports (ISCSS), Advances in Intelligent Systems and Computing 392, DOI 10.1007/978-3-319-24560-7_11
85
86
O.A. Malik and S.M.N. Arosha Senanayake
variety of tests are available to detect the balance/postural control changes after a knee injury or surgery. The single leg balance test measures postural stability (i.e. balance) of a subject and it is a simple, easy and effective method to screen for balance impairments in subjects after knee surgery. The single leg balance testing can be performed with opened or closed eyes and on flat or perturbed surface with different flexion positions of weight-bearing and contralateral knees. Anterior cruciate ligament (ACL) trauma, being one of the most common knee injuries in sports, alters the proprioceptive function of the joint and leads to changes in joint stability and thus affects the balance control. Proprioceptive structures within knee joint also influence the muscle activity around the joint. In single leg stance, the muscular activity is more intensive as compared to double leg stance due to less redundancy in exoskeleton system. For single leg eyes open and eyes closed balance tests of healthy subjects, a positive correlations between the extent of body sway and the magnitude of muscular activity for tibialis anterior, medial gastrocnemius, quadriceps femoris and gluteus medius muscles has been noted [3, 4]. For ACL injured/reconstructed subjects, the knee stabilizing muscles (vastus medialis, vastus lateralis, semitendinosus, biceps femoris and gastrocnemius) have greater role in single leg balancing task. The amount of activation of these muscles varies with the perturbing stimuli. In recent past, the physiotherapists and physiatrists have started using the computerized and sensors’ based methods for the objective assessment of balance control and postural sway of subjects [5]. Wearable motion sensors (e.g. accelerometers, gyroscopes etc.) have enabled objective measurement of balance control during various clinical tests [5]. However, the use of machine learning techniques with integration of different bio-signals has not been much explored in the previous literature for providing an automated classification of single-leg balance testing performance. In this study, an interval type-2 fuzzy logic (FL) has been used for assessment of single-leg balance performance of subjects after knee surgery (ACL reconstruction). Fuzzy logic is a form of multi-valued logic that helps in approximate reasoning by handling impreciseness and uncertainty present in both quantitative and qualitative information using a single model. FL can handle the impreciseness and uncertainties present in the actual measurements recorded through sensors due to noise and motion artifacts. Type-1 FL systems handle the imprecise data by choosing precise membership function value but these systems lack modeling of uncertainty involved in the measurements, definitions of input and output fuzzy sets, and rules. Thus, a type-2 FL based automated adaptive classification system has been investigated for distinguishing the dynamic balance control of healthy and knee injured subjects at different stages of recovery. The kinematics and EMG signals are recorded through wearable wireless sensors and relevant features are extracted from the data. The parameters for the membership functions (MFs) of antecedents are extracted from the data recorded for healthy and knee injured subjects after surgery during experiments and the parameters for MFs of consequent are determined based on the recommendations/observation of physiotherapists and physiatrists. Theses parameters are tuned by using steepest descent method. The classification performance of the proposed system was tested for singleton and non-singleton inputs and results were compared with type-1 FL system.
An Interval Type-2 Fuzzy Logic Based Classification Model …
87
The purpose of the system is to provide a complementary decision supporting tool to enable the trainers and physiotherapists to objectively monitor the balance control performance of athletes after knee surgery.
2
Methodology
2.1
General Framework
A fuzzy logic based general framework for classification of single-leg balance testing performance of subjects after knee surgery is shown in Fig. 1. The framework mainly consists of two modules: data collection and processing of different types of input signals, and an adaptive fuzzy logic based classification module. Although, there is variety of single-leg balance testing activities, but in this study kinematics and neuromuscular features were extracted for eyes-open single-leg balancing on BOSU balance trainer (shaded in Fig. 1) to develop the fuzzy logic based classification model. The output of the fuzzy logic system (FLS) was a defuzzified value transformed into classification (normal/healthy, average or poor) of balance testing performance as provided by two physiotherapists/physiatrists for each trial (see section 2.4).
Eyes Close
Eyes Open
Eyes Close
Kinematics EMG
Data Collection and Processing
Eyes Open
Different Types of Signals/Sensors
Single-Leg Balance on Flat Surface
Fuzzy Logic System (Type-1/Type-2)
Fuzzy set definitions
Balance Control Performance (Defuzzified Value/ Class)
Fuzzy Rules
Training Algorithm
Single-Leg Balance on BOSU Balance Trainer
Fig. 1. General framework for classification of single-leg balance control of subjects after knee surgery using adaptive fuzzy logic system
2.2
Data Collection and Processing
A total 10 subjects (three healthy and seven unilateral ACL reconstructed) were recruited for this study. The wireless micro-electro-mechanical system (MEMS) motion sensors (containing tri-axial accelerometer and gyroscope) from KinetiSense (ClevMed. Inc) were used in this study to collect the kinematics data. A BioCapture physiological monitoring system, consisting of BioRadio and USB receiver, was used to record the EMG signals from the relevant muscles around the knee joint. The motion sensors and EMG electrodes were setup on weight bearing leg of each subject (operated leg for the ACL reconstructed subjects and randomly selected leg for the
88
O.A. Malik and S.M.N. Arosha Senanayake
healthy subjects) as follows: two motion sensors attached to identified positions on lateral aspects of his/her thigh and shank using an adhesive medical tape to note the 3D kinematics of lower limb extremities, the EMG signals were recorded by placing disposable pre-gelled snap electrodes on vastus lateralis (VL), biceps femoris (BF) and gastrocnemius medialis (GM). The standard guidelines were followed for skin preparation, placements of sensors and electrodes on identified positions and filtering [6]. During each testing session, three trials were performed by every test subject for a duration of at-least 15-20 seconds while each subject was standing on single-leg with eyes open on BOSU balance trainer with knee fully extended for weight bearing leg. The data for multiple trials were collected in order to take care of variability in signals and generating a large dataset containing knee dynamics and neuromuscular signals. A custom software was developed in MATLAB 7.0 for processing of the recorded data. Five features (two kinematics and three EMG), namely mean knee abduction/adduction, range of knee abduction/adduction and normalized root mean squared (RMS) value for VL, BF and GM, were extracted from the processed data for each trial of the subjects as input to design the FLS.. 2.3
Adaptive Fuzzy Logic Classification System
An adaptive FLS for classification of balance testing performance was designed based on following steps [7]. Initialize the System. In order to initialize the type-2 FLS, definitions (types of MFs and their parameters) were determined for antecedents, consequents and inputs. The proposed FLS consisted of five antecedents (mean abduction/adduction, range of abduction/adduction and normalized RMS values of VL, BF and GM muscles) and one output (class for balance testing). The numerical data collected during experiments were used to obtain definitions of antecedents, while the consequent were determined based on the recommendations from physiotherapists and physiatrists for corresponding input. In this context the antecedents and consequent were considered to be type-2 Gaussian with uncertain mean and the input membership functions was type-1 Gaussian for non-singleton inputs. After defining the types of the membership functions, the antecedents’ intervals were divided into suitable number of fuzzy sets. In this study, three fuzzy sets (low, average and high) were assumed for all features. The initialization of the parameters (mean, standard deviation etc.) of antecedents' membership functions was done using the procedure described in [7]. The MFs for consequent were defined based on the judgments of physiotherapists and physiatrists about the balance testing for each trial of each subject. A scale of a range 1 through 10 was used to take their input and later mapped to three MFs as Poor, Average and Normal/Healthy to represent the current status of the balance control for each subject. Two types of input membership functions were initialized. In the case of singleton inputs the mean and standard deviation values of each input membership function was the corresponding mean of the 3 trials of the balance testing. For non-singleton inputs, the various measures were represented by type-1 Gaussian MFs. The mean and stand-