The application of support vector machines for detecting recovery from knee replacement surgery using spatio-temporal gait parameters
Introduction
Knee osteoarthritis (OA) is one of the leading causes of disability among people aged 65 years and over in the worldwide population [1], [2], [3], [4], [5], [6]. Individuals with knee OA suffer from pain, stiffness and physical disability which consequently lead to a loss of functional independence [7]. Knee replacement is a common surgical procedure used worldwide for the management of knee OA with 171,335 procedures preformed in 2001 in the USA [8] and more than 25,000 procedures preformed annually in Australia [9]. Knee replacement provides pain relief and improves physical function and quality of life for OA patients [10].
Since walking patterns are adversely affected by lower extremity joint disease, quantitative gait measures such as spatio-temporal parameters (i.e. step length, step time, stride length, etc.) have clinical relevance in the assessment of motor pathologies. Spatio-temporal gait parameters have, therefore, been used to assess functional outcome after knee replacement surgery [11], [12], [13], [14].
Knee replacement surgery has been reported to improve the gait patterns of patients with OA [11], [12], [13], [15], [16]. However gait dysfunction, including altered spatio-temporal parameters and gait asymmetry may still persist after the surgery [17], [18], [19]. Since quantitative measurements of gait can reflect the integrity of a person's locomotor ability, it may be useful to monitor such parameters after surgical intervention as an indicator of functional outcome [11]. An understanding of how gait patterns of patients with knee OA and those who undergo knee replacement surgery differ from asymptomatic individuals may help in the development of physical therapy programs to improve the outcome of knee replacement surgery.
Computational intelligence techniques such as artificial neural networks, support vector machines (SVM) and fuzzy classifiers have recently been used for automated identification of gait pathologies [20], [21], [22], [23], [24], [25]. In this work, we have elected to use the SVM classifier which has been successfully used in various pattern recognition problems [25], [26]. The SVM classifier is a supervised learning formulation which learns nonlinear data relationships which are not evident when using classical statistical analysis based on linear techniques, e.g. linear discriminant analysis (LDA). The SVM has demonstrated better classification performance on test data, including spatio-temporal gait parameters [23], which indicates SVM's better generalization capability over linearity-based classical statistical analysis [27]. To the best of our knowledge, SVMs have neither been investigated in relation to gait parameters pertaining to knee OA nor used to detect post-operative gait changes.
In this paper, the SVM classifier used spatio-temporal gait parameters to distinguish between individuals with OA prior to undergoing knee replacement and healthy controls. The objective was to investigate if the classifier could detect changes at 2 and 12 months following knee replacement surgery using these differentiating features and to compare the predicted performance with a clinical assessment of the knee using American Knee Society knee score.
Section snippets
Knee osteoarthritis data
The study consisted of two groups from a research database within the Musculoskeletal Research Centre, La Trobe University: healthy controls and symptomatic individuals who had undergone knee replacement surgery. The control group (six females and six males) had a mean age, mass, height and body mass index (BMI) of 81.5 ± 4.6 years, 68.1 ± 15.6 kg, 166.4 ± 0.1 cm and 24.3 ± 2.5 kg m−2, respectively. The symptomatic group included 11 patients (2 females and 9 males) who had undergone unilateral knee
Statistical analysis of spatio-temporal features
We calculated the fundamental statistical values for all 12 features, computing the mean and standard deviation separately for two classes for each feature, followed by the individual feature ROC areas. As seen in Table 1, the means for cadence, step length and stride time do not differ significantly indicating poor discrimination between the two classes. For example, cadence for the OA group had mean (±S.D.) of 0.010 ± 0.935 while the control group was 0.208 ± 1.035. This is further confirmed by
Discussion
The linear SVM detected all spatio-temporal features as being sensitive to distinguish between OA and healthy gait. It was found that any reduction in the number of features affected the generalization performance (Table 3) even when using nonlinear models. Models with higher nonlinearity, such as the polynomial kernel required fewer features (high training accuracy) but were found to overfit the data (poor test accuracy). The SVM was successfully able to distinguish between OA and healthy gait
Conclusion
There is limited research related to the application of the SVM in pathological gaits. To the best of our knowledge, the present study is the first study that investigated the application of SVM in identifying gait pattern of patients with osteoarthritis and also predicted gait improvement or altered gait after knee replacement surgery using spatio-temporal measures. The SVM classifier was able to effectively recognize gait parameters that are altered due to knee OA before knee replacement
Acknowledgment
This project was funded by an Australian Research Council Linkage Grant (LP 0455460).
Conflict of interest statement
None.
References (41)
- et al.
Quantitative gait analysis after medial unicompartmental knee arthroplasty for osteoarthritis
J Arthroplasty
(2003) - et al.
Gait and clinical measurements in patients with knee osteoarthritis after surgery: a prospective 5-year follow-up study
Knee
(2005) - et al.
Altered loading during walking and sit-to-stand is affected by quadriceps weakness after total knee arthroplasty
J Orthop Res
(2005) - et al.
Gait analysis after total knee arthroplasty. Comparison of posterior cruciate retention and substitution
J Orthop Sci
(1998) - et al.
Quantitative evaluation of gait pattern in patients with osteoarthrosis of the knee before and after total knee arthroplasty. Gait analysis using a pressure measuring system
J Orthop Sci
(1999) A review of analytical techniques for gait data. Part 2. Neural network and wavelet methods
Gait Posture
(2001)- et al.
Validity of the GAITRite walkway system for the measurement of averaged and individual step parameters of gait
Gait Posture
(2005) - et al.
Evaluation of an instrumented walkway for measurement of the kinematic parameters of gait
Gait Posture
(2000) - et al.
Concurrent related validity of the GAITRite walkway system for quantification of the spatial and temporal parameters of gait
Gait Posture
(2003) - et al.
Gait analysis before and after unilateral total knee arthroplasty. Study using a linear regression model of normal controls—women without arthropathy
J Orthop Sci
(1999)