An automatic non-invasive method for Parkinson's disease classification

https://doi.org/10.1016/j.cmpb.2017.04.007Get rights and content

Highlights

  • Gait Analysis based non-invasive identification of Parkinson subjects.

  • Wavelet based feature extraction using gait variables.

  • 100% classification accuracy using all spatio-temporal variables together.

Abstract

Background and objective

The automatic noninvasive identification of Parkinson's disease (PD) is attractive to clinicians and neuroscientist. Various analysis and classification approaches using spatiotemporal gait variables have been presented earlier in classifying Parkinson's gait. In this paper, we present a wavelet transform based representation of spatiotemporal gait variables to explore the potential of such representation in the identification of Parkinson's gait.

Methods

Here, we present wavelet analysis as an alternate method and show that wavelet analysis combined with support vector machine (SVM) can produce efficient classification accuracy. Computationally simplified features are extracted from the wavelet transformation and are fed to support vector machine for Parkinson's gait identification. We have assessed various gait parameters namely stride interval, swing interval, and stance interval (from both legs) to observe the best single parameter for such classification.

Results

By employing wavelet decomposition of the gait variables as an alternate method for the identification of Parkinson's subjects, the classification accuracy of 90.32% (Confidence Interval; 74.2%–97.9%) has been achieved, at par to recently reported accuracy, using only one gait parameter. Left stance interval performed equally good to Right swing interval showing classification accuracy of 90.32%. The classification accuracy improved to 100% when all the gait parameters from left leg were put together to form a larger feature vector. We have shown that Haar wavelet performed significantly better than db2 wavelet (p = 0.05) for certain gait variables e.g., right stride time series. The results show that wavelet analysis is a promising approach in reducing down the required number of gait variables, however at the cost of increased computations in wavelet analysis.

Conclusions

In this work a wavelet transform approach is explored to classify Parkinson's subjects and healthy subjects using their gait cycle variables. The results show that the proposed method can efficiently extract relevant features from the different levels of the wavelet towards the classification of Parkinson's and healthy subjects and thus, the present work is a potential candidate for the automatic noninvasive neurodegenerative disease classification.

Introduction

Approximately 60,000 Americans are diagnosed with Parkinson's disease (PD) each year in US and there is an estimate of seven to ten million people worldwide that are living with Parkinson's disease (statistics of Parkinson's disease foundation, [1]). Parkinson's is a neurodegenerative disease which is characterized by incapability of nerve cells to produce enough neurotransmitter in brain known as “dopamine”, affecting the gait pattern of an individual. The use of functional neuroimaging for diagnosis of neurodegenerative diseases including Parkinson's has been quite popular [2], [3] however still there is a need of an alternate diagnosis method which can offer quick, easy, and low cost measurement.

Initially walking pattern, known as gait cycle, was considered as an outcome of central pattern generators formed by spinal circuits [4], [5]. However, recent findings have demonstrated the significant role of cerebral cortex, basal ganglia, and the descending pathways to spinal cord in human locomotion [6], [7]. Thus, it becomes likely that any damage to corresponding area in cerebral cortex will severely affect the walking pattern. Such findings have offered an attractive alternate to diagnose the neurodegenerative disease like Parkinson's via gait cycle variables. Measurement of gait variables like swing interval, stance interval, stride intervals, etc., are measured by force sensitive resistors on the foot sole and hence may offer a quick, easy, and affordable measurement.

Many previous studies have analyzed the gait variables in neurodegenerative diseases, including Parkinson's disease, for various applications [8], [9], [10], [11]. It has been shown that the stride interval fluctuations are increased in PD and correlated to degree of severity [12]. The variability in Electromyogram (EMG) signal acquired from gastrocnemius muscle was found higher in PD patients [13]. A local intermittency measure from EMG of shoulder muscle has shown promise to distinguish the low-frequency behavior from healthy to Parkinson's subjects [14]. However, measurement of spatiotemporal gait variables are easy, quick and less prone to physiological parameter like sweating compared to EMG and hence spatial-temporal variables have been extensively used for the purpose of Parkinson gait. A multiple regression normalization approach in spatial-temporal gait variables significantly increased the PD classification accuracy [2]. In an another approach, Daliri [15] utilized basic features i.e. minimum, maximum, mean, and standard deviation from all possible gait variable time series of both legs i.e. left and right stride interval, left and right swing interval, left and right stance interval, and double support phase interval, for automatic identification of Parkinson's gait. In contrast to previous approaches of representing the time series by limited features, Khorasani and Daliri, [16] modeled the complete time series of gait variable using Hidden Markov Model (HMM). Similarly, Zeng and Wang, [17] used supercentral pattern generator (SCPG) model to represent swing and stance interval time series for Parkinson's gait identification. In a completely different approach, Dutta et al. [18] used features were derived from the cross-correlation the time series. Despite of the fact that these previous approaches have shown respectable classification accuracy, the potential of frequency domain features in a time series were not explored.

The studies with frequency domain representation of gait variable time series in Parkinson's disease classification have been very limited. Motivated from the promising results of signal turns count features of a time series in previous studies [19], [20], [21], Wu and Krishnan [22] used STC as one of the feature for time series representation of gait variable time series and reported a remarkable classification accuracy of 90.32%. STC is a deemed parameter for frequency representation [23] however suffers from two major limitations - (1) It loses the time instant information as it gives the information in a single number which represents the total variability in the time series, (2) The results of STC are subjective to chosen threshold value. Time-frequency representation of a time series overcomes the shortcomings of STC in a way that time-frequency representation provides time and frequency information simultaneously with certain resolution. Short-time-Fourier-transform (STFT) is one such transformation and has been used to represent the force data collected from 8 sensors placed underneath of each foot [24]. STFT suffers from the problem of fixed resolution and hence does not suit very well for analysis of non-stationary signals. The Choi-Williams distribution, Wigner-Ville distribution, makes interpretation nearly impossible for a complicated signal due to the cross-terms in its expression [25]. Wavelet transform, also known as multi resolution analysis (MRA), facilitates with multi-resolution analysis and provides an easy interpretation of wavelet coefficients. However, the continuous wavelet transform uses considerable resources and provides redundant information as compared to discrete wavelet transform (DWT). Hence we preferred DWT decomposition over other time-frequency analysis for its capabilities of multi-resolution, interpretable analysis, relatively easy computation, and non-redundant information. A DWT based approach using accelerometer data has shown potential for identification of Parkinson subjects [26]. We propose simplified computation features from wavelet analysis for healthy and Parkinson subjects classification. We present a wavelet transform based representation of spatiotemporal gait variables to explore the potential of such representation in identification of Parkinson's gait. Computationally simplified features are extracted from the wavelet transformation and are fed to support vector machine for Parkinson's gait identification. We selected each gait variable one at a time to observe the best single spatiotemporal gait variable and further compare left and right leg spatiotemporal gait variables. Present work will give an insight to assess various components of gait cycle in discriminating Parkinson's and Healthy controls with available instrumentation and computational resources. We finally discuss previous approaches in the literature and compare our findings to their outcomes.

Section snippets

Signal acquisition and processing

We used the database provided by Hausdorff et al. [27], [10], [12] on the web page of Physionet.1 The database supplies the records of spatiotemporal variables of gait cycle from the patients with Parkinson's disease (n = 15; 10 males and 5 females) and healthy control subjects (n = 16; 2 males and 14 females). According to the experiment [27], each subject was requested to walk at his or her normal pace along a straight

Results

Fig. 3(a) and (b) represent the Haar wavelet decomposition of the right swing interval of representative subject from each group and their respective feature values (maximum value, minimum value, mean value, variance and energy). Here XAi(n) represents the approximation (A) coefficients and XDi(n) represents the detail (D) coefficients of x(n) at the ith level (n is the number of coefficients).The results show that overall maximum classification accuracy of 90.32%(CI; 74.2%–97.9%) was produced

Discussion

The aim of this study was to use wavelet decomposition of the gait variables as an alternate method for the identification of Parkinson's subjects. The results obtained by using Haar wavelet correctly classified PD and healthy controls with an accuracy of 100% using all the left leg gait parameters. Pooling all the gait parameters could have generated a high dimensional space where SVM performed better than the feature space in one parameter. In a recent study on Parkinson's disease

Limitations

A number of limitations can be derived from this study: (1) the subject group that we have considered from the database does not match in their respective age, gender and height. These variations must be taken into account as they can have serious effects on the study. It has been shown that gait features become significantly different after using some multiple regression methods and anthropometric details [2], [41]. (2) In the present study gait rhythm is measured by stride time using a light

Conclusion

In this work a wavelet transform approach is explored to classify Parkinson's subjects and healthy subjects using their gait cycle variables. We found that maximum classification accuracy of 90.32% was obtained using Haar wavelet for right swing and left stance intervals features taken independently and 100% when all the left leg gait parameters were taken together creating a larger feature vector. In general, Haar wavelet performed better than the db2 wavelet. Among all gait variables, right

Acknowledgments

We thank Prof. J. M. Hausdroffand his research group for providing the clinical dataset freely available at Physionet. We thank anonymous reviewers for their valuable comments in improving the contents of manuscript.

References (42)

  • G. Bonora et al.

    Gait initiation is impaired in subjects with Parkinson's disease in the OFF state: evidence from the analysis of the anticipatory postural adjustments through wearable inertial sensors

    Gait Posture

    (2017)
  • Statistics on Parkinson's

  • F. Wahid et al.

    Classification of Parkinson's disease gait using spatial temporal gait features

    IEEE J. Biomed. Health Inform.

    (2015)
  • M. Niethammer et al.

    Functional neuroimaging in Parkinson's disease

    Cold Spring Harbor Perspect. Med.

    (2012)
  • F.P. Zehr

    Neural control of rhythmic human movement: the common core hypothesis

    Exerc.SportSci.

    (2005)
  • V. Dietz

    Proprioception and locomotion disorders

    Nat. Rev. Neurosci.

    (2002)
  • C. Camara et al.

    A fuzzy inference system for closed-loop deep brain stimulation in Parkinson's disease

    J. Med. Syst.

    (2015)
  • O. Henmi

    Spectral analysis of gait variability of stride interval time series: comparison of young, elderly and Parkinson's disease patients

    J. Phys. Ther. Sci.

    (2009)
  • J.M. Hausdroff et al.

    Gait variability and basal ganglia disorders: stride-to-stride variations of gait cycle timing in Parkinson's disease and Huntington's disease

    Mov. Disord.

    (1998)
  • L. Sugavaneswaran et al.

    Ambiguity domain-based identification of altered gait pattern in ALS disorder

    J. Neural Eng.

    (2012)
  • J.M. Hausdorff et al.

    Altered fractal dynamics of gait: reduced stride interval correlations with aging and Huntington's disease

    J. Appl. Physiol.

    (1997)
  • Cited by (95)

    • Age-gender specific prediction model for Parkinson's severity assessment using gait biomarkers

      2022, Engineering Science and Technology, an International Journal
      Citation Excerpt :

      The work articulated by Muniz et al. [27] studied the effect of deep brain stimulation of subthalamic nucleus (DBS-STN) and medication for advanced PD detection. Joshi et al. [28] achieved prediction rate of 90.32% by combined use of wavelet analysis with SVM to explore gait variability. The researcher in [29] studied the stride fluctuations of PD subjects using entropy parameters.

    View all citing articles on Scopus
    View full text