Elsevier

Pattern Recognition Letters

Volume 94, 15 July 2017, Pages 180-188
Pattern Recognition Letters

Automated detection of focal EEG signals using features extracted from flexible analytic wavelet transform

https://doi.org/10.1016/j.patrec.2017.03.017Get rights and content

Highlights

  • A new method for detection of focal EEG signals has been proposed.

  • The proposed method is based on the flexible analytic wavelet transform.

  • The developed method has been compared with the other existing methods.

  • The obtained classification accuracy by proposed method is 94.41%.

Abstract

Epilepsy is a neurological disease which is difficult to diagnose accurately. An authentic detection of focal epilepsy will help the clinicians to provide proper treatment for the patients. Generally, focal electroencephalogram (EEG) signals are used to diagnose the epilepsy. In this paper, we have developed an automated system for the detection of focal EEG signals using differencing and flexible analytic wavelet transform (FAWT) methods. The differenced EEG signals are subjected to 15 levels of FAWT. Various entropies namely cross correntropy, Stein’s unbiased risk estimate (SURE) entropy, and log energy entropy are extracted from the reconstructed original signal and 16 sub-band signals. The statistically significant features are obtained from Kruskal–Wallis test based on (p < 0.05). K-nearest neighbor (KNN) and least squares support vector machine (LS-SVM) classifiers with different distances and kernels respectively are used for automated diagnosis. In the proposed methodology, we have achieved classification accuracy of 94.41% in detecting focal EEG signals using LS-SVM classifier with ten-fold cross validation strategy.

Introduction

The electroencephalogram (EEG) signal is utilized to examine the electrical activity of the brain function and comprises of many frequency components. These signals indicate the working of human brain and also possible neurological disorder.

According to the latest world health organization (WHO) report, nearly 50 million people suffer from epilepsy worldwide [43].

Epilepsy is a class of neurological disorder characterized by seizures which can lead to cortical dysplasia, cognitive disability, and also other health problems [29]. During focal epilepsy, seizures may not be managed with medications and hence it becomes important to localise the focal epileptic zone. However, scalp EEG may decline to gather ictal EEG changes during focal seizures which comes up from a small or deeply allocated focus [29]. Approximately, more than 20% patients have generalized epilepsy which affects the entire brain, whereas, more than 60% patients have focal or partial epilepsy starting from a limited part of the affected area [29]. Hence, it may be useful to develop signal processing methods to recognize and locate the surgically removable focal epileptic zones in the brain. In [2], the outcomes imply that EEG signals from an epileptogenic brain are barely random, stationary and more nonlinear-dependent compared to the signals from non-epileptogenic brain areas. In [28], a concept is introduced based on the functional connectivity to study the neurophysiological signals. In [20], the results helped in localizing the epileptogenic focus. The localization of the epileptogenic focus measured by comparing EEG signals shown asymmetry in delta activity with focal epilepsy patients [27].

Many computer based detection systems have been developed to aid the clinicians as it is difficult to detect the focal and non-focal EEG signals due to their low amplitudes (micro volts). Several nonlinear measures, such as Renyi, Shannon, Tsallis, fuzzy, sample, phase, permutation entropies and central tendency measures (CTM) have been presented to gather the dynamics of focal epileptic zones from EEG signals [7], [37], [38]. In [39], the features employed are average sample entropies and average variance of instantaneous frequency for the intrinsic mode functions (IMFs) obtained from empirical mode decomposition (EMD) of EEG signals. These features are fed into least squares support vector machine (LS-SVM) with radial basis function (RBF) as a kernel for classification of non-focal and focal EEG signals. This technique has been tested on 50 set of non-focal and 50 set of focal EEG signals and achieved an accuracy of 85%. In [38], entropy measures are applied on IMFs obtained from the EMD of EEG signals. This method achieved 87% accuracy with 50 set of non-focal and 50 set of focal EEG signals. In [37], an integrated index is formulated by using discrete wavelet transform (DWT) and entropy based features. The K-nearest neighbor (KNN) classifier gave 89.4% accuracy on 3750 set of non-focal and 3750 set of focal EEG signals [9]. In [7], the area parameters from CTM for various reconstructed phase space (RPS) plots are estimated to distinguish 50 focal and 50 non-focal EEG signals. Their method gave classification accuracy of 90% and 82.53% for 50 and 750 pairs of EEG signals, respectively. In [35], orthogonal wavelet filter banks and entropy features are used. This method achieved 94.25% accuracy using the entire Bern Barcelona database with LS-SVM classifier. In [6], the authors have developed multivariate sub-band fuzzy entropy which is derived from tunable-Q wavelet transform (TQWT). Using this multivariate sub-band fuzzy entropy together with the LS-SVM classifier, they obtained classification accuracy of 84.67% for classification of all the available focal and non-focal EEG signals of Bern Barcelona database.

Entropy based features are found to be very effective in detecting complexities in the EEG signals [9], [37], [38], [39]. The log energy entropy [12], [34], cross correntropy [19], and Stein’s unbiased risk estimate (SURE) entropy [10], [34] measures are useful to quantify the variation in the time series. The log energy entropy is also considered in [9] to differentiate focal and non-focal EEG signals. The motivation behind cross correntropy nonlinear measures comes from its previous success in non-Gaussian signal processing [19]. The motivation for studying SURE entropy is from the use of this entropy in [34].

In the present work, the main focus is to propose a new methodology for computer-aided detection of the focal EEG signals using flexible analytic wavelet transform (FAWT) [5] and entropy features. The features are classified using KNN and LS-SVM classifiers. The block diagram of the proposed methodology is shown in Fig. 1.

The paper is organized into following four sections: Section 1 describes the introduction, Section 2 provides brief description about data collection and explanation about the designed methodology. It also provides description on FAWT, entropy features, and classifiers. Results and discussions of the entire work is provided in Section 3. The Section 4 describes the future work and the conclusion of this paper.

Section snippets

Data collection

EEG signals are acquired from Bern Barcelona database [23]. In [2], focal epilepsy is monitored among 5 patients and all the patients went through long-range intracranial EEG recordings at the Neurology Department of the Bern University. We have used 3750 focal and 3750 non-focal EEG signals, which were sampled at a sampling rate of 512 Hz for 20 s with 10,240 samples. The data set comprised bivariate EEG signals and are depicted as “X” and “Y”. The plot of “X” time series of non-focal and

Results and discussion

In the present work, FAWT is employed to decompose the differenced EEG signals into fifteen sub-bands and one approximation band (15 level decomposition). The reconstruction of these sub-bands along with signal reconstructed with all bands are used to evaluate log energy entropy, cross correntropy, and SURE entropy. To improve the realization, statistically significant features with (p < 0.05) are decided using the Kruskal–Wallis statistical test [21], [36], [24], which suggest us to omit the

Conclusion

It is very strenuous and time-consuming to diagnose focal epilepsy manually. There is a need for an automated system to overcome these deficiencies and eliminate the possible errors due to manual readings. Early detection of epilepsy may save the patients from these serious brain disorders. In the present work, only focal epilepsy is considered. FAWT method is used to decompose the focal and non-focal EEG signals into sub-band signals. Various features namely cross correntropy, log energy

References (46)

  • R. Sharma et al.

    An integrated index for the identification of focal electroencephalogram signals using discrete wavelet transform and entropy measures

    Entropy

    (2015)
  • R. Sharma et al.

    Application of entropy measures on intrinsic mode functions for the automated identification of focal electroencephalogram signals

    Entropy

    (2015)
  • D.W. Aha et al.

    Instance-based learning algorithms

    Mach. Learn.

    (1991)
  • R.G. Andrzejak et al.

    Nonrandomness, nonlinear dependence, and nonstationarity of electroencephalographic recordings from epilepsy patients

    Phys. Rev. E

    (2012)
  • A.T. Azar et al.

    Performance analysis of support vector machines classifiers in breast cancer mammography recognition

    Neural Comput. Appl.

    (2014)
  • I. Bayram

    An analytic wavelet transform with a flexible time-frequency covering

    IEEE Trans. Signal Process.

    (2013)
  • A. Bhattacharyya et al.

    Tunable-Q wavelet transform based multivariate sub-band fuzzy entropy with application to focal EEG signal analysis

    Entropy

    (2017)
  • A. Bhattacharyya et al.

    A novel approach for automated detection of focal EEG signals using empirical wavelet transform

    Neural Comput. Appl.

    (2016)
  • S.H. Cha

    Comprehensive survey on distance/similarity measures between probability density functions

    International Journal of Mathematical Models in Applied Sciences

    (2007)
  • D.L. Donoho et al.

    Adapting to unknown smoothness via wavelet shrinkage

    J. Am. Stat. Assoc.

    (1995)
  • J. Han et al.

    Entropy feature extraction on flow pattern of gas/liquid two-phase flow based on cross-section measurement

    Journal of Physics: Conference Series

    (2009)
  • R.J. Hyndman, G. Athanasopoulos, Forecasting: principles and practice, OTexts: Melbourne, Australia. Section 8/1....
  • Istanbul Technical University

    An analytic wavelet transform with a flexible time-frequency covering

    (2013)
  • Cited by (114)

    • Sleep apnea detection using electrocardiogram signal input to FAWT and optimize ensemble classifier

      2022, Measurement: Journal of the International Measurement Confederation
    View all citing articles on Scopus
    View full text