Short CommunicationUsing Blind Source Separation on accelerometry data to analyze and distinguish the toe walking gait from normal gait in ITW children
Introduction
Idiopathic Toe Walkers (ITW) are children who normally tend to walk on their toes with the ankle plantar flexed. These children are called idiopathic because they are considered to walk on their toes for no apparent reason [1]. The severity of toe walking varies as some children walk on the tip of the toes while some have their heels just off the ground [2]. Consequences of toe walking often lead to an abnormal gait associated with back pain or a bouncy gait if they are not treated and monitored. The treatment options for these children vary as per the severity of toe walking. However, even after the treatment there is a tendency for these children to revert back to toe walking due to habit. Hence monitoring their gait after treatment is very important.
Currently, the ITW children are assessed by measuring the ankle flexibility at the clinics. Parental feedback is another form of gait assessment method in ITW children. However, both these methods are subjective and confined to a gait laboratory and clinics [1], [2]. There are other methods for monitoring the gait using the wireless sensor technology [2]. Sensors such as accelerometers, gyroscopes, pressure sensors are very commonly used for gait analysis. However each of these sensors has certain limitations in terms of measurements. Gyroscopes can be used for gait analysis but are preferred for applications with angle or tilt measurement and hence useful to determine the angle of the foot with respect to ground [3], [4], [5]. Pressure sensors are also used for gait analysis in clinical settings [6], [7], [8], [9]. The drawback of using the pressure sensors is that we need multiple pressure sensors on the foot (one on heel, one on balls of the foot, etc.) as they provide output (pressure in analog value) only when that part of the foot is in contact with the ground [8], [9]. An accelerometer provides output acceleration (analog or digital) of the foot even when the foot is not in contact with the ground. Therefore it is possible to determine the acceleration of the foot during the full gait cycle and also determine the different phases (stance & swing) in the gait cycle using accelerometers [10], [11], [12], [13], [14]. Hence, for this study we have chosen a dual axis accelerometer.
The new activity monitoring devices uses wireless technology integrated with multiple sensors such as accelerometers, gyroscopes and pressure sensors in order to determine the gait activity [10], [11], [12], [13], [14]. However, these sensors are highly sensitive to vibrations or movement. Even a very small change in one sensor can affect the other sensors. Hence, the signal associated with one sensor could record signals originating from adjacent sensors and lead to cross-talk (noise). This cross talk could lead to an overlapping spectrum of signals. The cross talk or the noise often goes unobserved by the researcher or the clinician as the level (amplitude) of the cross talk can vary depending on the gait activity. If the cross talk is substantial, it can also lead to improper assessment while monitoring the gait activity in subjects.
To reliably identify the gait activities from multiple sensor data or multiple axes from a single sensor, there is a need to decompose and separate each sensor's signals accurately. Spectral and temporal filtering is not suitable for this because of overlapping signals from the multiple sensors used due to simultaneous movements of the foot in the horizontal and vertical planes. BSS techniques provide a solution for such a situation. Independent component analysis (ICA) is a widely used BSS technique that is suitable for estimating Independent Components (ICs) from a mixture. It has been used very successfully for audio and bio-signal applications [15], [16], [17]. BSS is also widely used in source separation and identification of other biomedical signals such as Electroencephalography (EEG), Electrocardiography (ECG) and Electromyography (EMG), which are weak and overlapping.
In this research, we propose an automated classification of heel accelerometer data (ITW and normal gait) using ICA and KNN classifiers. Initially, the ICA algorithm is used to extract features; a feature vector by a coefficient matrix is formed. The feature vectors are then classified by a K-means clustering algorithm. The rest of the paper is organized as follows; the basic principle of heel accelerometer data and ICA are described in Section 2. In Section 3, experimental sensor setup and sensor calibration is introduced. Section 4 explains the data analysis and the classification results of normal and toe walking gait. Section 5 discusses the outcome of the proposed research technique. Finally, in Section 6, the proposed method is summarized and suggestions for future work are mentioned.
Section snippets
Theory
The following section explains the concept of heel accelerometer data used to assess the gait and ICA in brief.
Experimental sensor setup and sensor calibration
University ethics committee granted approval to conduct experiments on human subjects and acquire accelerometer data. For the data acquisition, the ADXL 202EB (MEMS Technology) was selected as it is a miniature accelerometer board, with dual axis accelerometer with range ±2 g and has both analog and digital output interfaces. ADXL 202 can also be used for motion sensing and tilt sensing. A tilt in the accelerometer results in an output voltage whose amplitude is proportional to the gravitational
Data analysis and results
Data analyses were performed in windows platform, using the Matlab based FastICA algorithm [22]. Initially, normal walking data from two channels (ax and ay) were fed to FastICA algorithm and the sources were estimated. The same process was repeated for the toe walking data. One concern of the use of ICA is that it suffers from the amplitude and order ambiguity. Hence, the ICA separated sources (estimated accelerometer data) are normalized according to the amplitude of the original signals. The
Discussion
In previous work [1], detection of ITW gait from accelerometry data was achieved by comparing the threshold (constant) value of the acceleration signal to distinguish toe-walking stance from normal stance in a gait cycle. This solution proved tedious when several children required quick diagnosis. In this work, we demonstrated that BSS separated statistical features of the heel accelerometry data together with the K-means clustering classifier provided an automatic recognition method for
Conclusion
Typically, biomedical sensor data are affected by large measurement errors, mainly due to the non-invasive nature of the measurement process or the severe constraints to keep the input signal as low as possible for safety and bio-ethical reasons. To extract the relevant information for diagnosis and therapy, BSS techniques can be widely used. Hence, the proposed BSS method along with KNN classifier is a model based technique which could be used as a significant software tool by the clinicians
References (26)
- et al.
Linking clinical measurements and kinematic gait patterns of toe-walking using fuzzy decision trees
Gait Posture
(2007) - et al.
Classification of idiopathic toe walking based on gait analysis: development and application of the ITW severity classification
Gait Posture
(2007) - et al.
A practical gait analysis system using gyroscopes
Med. Eng. Phys.
(1999) - et al.
Spatio-temporal parameters of gait measured by an ambulatory system using miniature gyroscopes
J. Biomech.
(2002) - et al.
Methods for gait event detection and analysis in ambulatory systems
Med. Eng. Phys.
(2010) - et al.
In-shoe plantar pressure measurement: a review
The Foot
(1992) - et al.
Evaluation of instrumented shoes for ambulatory assessment of ground reaction forces
Gait Posture
(2007) - et al.
In-shoe plantar pressure measurements for patients with knee osteoarthritis: reliability and effects of lateral heel wedges
Gait Posture
(2011) - et al.
Real-time gait assessment utilizing a new way of accelerometry
J. Biomech.
(1990) A new method for evaluating motor control in gait under real-life environmental conditions. Part 2: Gait analysis
Clin. Biomech.
(1998)
Estimation of gait cycle characteristics by trunk accelerometry
J. Biomech.
Independent component analysis: algorithms and applications
Neural Networks
Automated method to distinguish toe walking strides from normal strides in the gait of idiopathic toe walking children from heel accelerometry data
Gait Posture
Cited by (40)
Faster and better sparse blind source separation through mini-batch optimization
2020, Digital Signal Processing: A Review JournalUnderdetermined blind separation of source using l<inf>p</inf>-norm diversity measures
2020, NeurocomputingCitation Excerpt :Most traditional BSS algorithms assume that the number of sources is known and less than the number of sensors, i.e., in the overdetermined case. Independent component analysis (ICA) is a statistical method for decomposing multivariate data into components that are statistically as independent as possible, which has wide applications in various areas such as biomedical signal processing [2,3], audio signal processing [4–6], and mechanical engineering [7–9]. However, when the number of sources is larger than the number of sensors, i.e., in the underdetermined case.
The utility of normative foot floor angle data in assessing toe-walking
2018, FootCitation Excerpt :Despite its importance, the detection of the nature of initial contact, i.e., heel or toe strike, is often elusive with conventional kinematic measures. Various methods have been proposed for measuring toe-walking: identifying sagittal kinematic parameters of ankle dorsiflexion angle at initial contact (ADA IC) or peak ankle dorsiflexion in stance and swing [12,13]; quantifying improvements in the range of passive ankle dorsiflexion [12]; and using dynamic foot pressure [14] or calf muscle-tendon length measurements [15], pedobarometric evaluations [16], or heel accelerometry [17,18]. However, none of these tools directly assess the angle of the sole relative to the floor to infer heel strike, foot-flat, or toe strike during initial contact.
Output-only modal estimation using sparse component analysis and density-based clustering algorithm
2018, Measurement: Journal of the International Measurement ConfederationCitation Excerpt :The blind source separation (BSS) aims to separate the sources from mixed signals without prior knowledge of the mixing procedure or sources. The BSS technique has been widely applied in various fields, such as image processing [11], biomedical analysis [12], speech signal processing [13], communications [14]. Many BSS methods have been already applied for operational modal analysis (OMA) successfully.
Double-layer ensemble monitoring of non-gaussian processes using modified independent component analysis
2017, ISA TransactionsCitation Excerpt :An examination of the existing literature on ICA shows that the FastICA iterative algorithm proposed by Hyvärinen and Oja [16] is used as a “default” method for model construction because it can greatly reduce the computation time. The FastICA algorithm has been intensively adopted for a functional data processing tool, especially in the area of blind source separation [17–20]. However, the original FastICA algorithm has some drawbacks from practical viewpoint.
Least-Squares Algorithms for Complex-Valued Blind Source Separation
2024, Circuits, Systems, and Signal Processing