Elsevier

Expert Systems with Applications

Volume 104, 15 August 2018, Pages 153-167
Expert Systems with Applications

Robust detection of epileptic seizures based on L1-penalized robust regression of EEG signals

https://doi.org/10.1016/j.eswa.2018.03.022Get rights and content

Highlights

  • This study introduces a robust detection method of epileptic seizures.

  • The performance of the proposed method is examined under ideal and real-life conditions.

  • The proposed method achieves 100% classification accuracy under the ideal conditions.

  • The proposed method is also proven to be robust in real-life situations.

Abstract

Epilepsy is the second common brain disorder affecting 70 million people worldwide. Electroencephalogram (EEG) has been widely used for the diagnosis of epileptic seizures. However, most of the existing EEG-based seizure detection methods cannot maintain robust performance in real life conditions. This is where the EEG data are corrupted with different sources of noise and artifacts. This study presents a robust seizure detection system that works efficiently under the real life conditions as well as the ideal conditions. A feature learning method based on L1-penalized robust regression is developed and applied to the EEG spectra to recognize the most prominent features pertinent to epileptic seizures. The extracted features are then fed into the random forest classifier for seizure detection. Results on a public benchmark dataset show that the performance of this seizure detection system is superior to prior work. It first achieves seizure detection rates of 100.00% sensitivity, 100.00% specificity, and 100.00% classification accuracy under the ideal conditions. The proposed method is also proven to be robust in the presence of white noise and EEG artifacts mainly those arising from muscle activities and eyes-blinking. It is found to achieve seizure detection accuracies in the range of 90.00–100.00% when applied to noisy EEG data corrupted with high noise levels. To the best of our knowledge, there exists no work in the literature that detects seizures under these conditions.

Introduction

Epilepsy is a neurological disorder that affects around 70 million people worldwide (Fazel, Wolf, Långström, Newton, & Lichtenstein, 2013). The defining characteristic of epilepsy is recurrent seizures that strike without warning. Symptoms may range from brief suspension of awareness to loss of consciousness and sometimes violent convulsions (Iasemidis et al., 2003). Unlike other neurological disorders, such as stroke or Alzheimer’s disease, which tend to develop later in life, epilepsy often affects people in the prime of their lives, with the majority between 15 and 64 years of age (Hesdorffer, Hauser, Annegers, & Cascino, 2000). Early detection of epileptic seizures is of a high interest from the treatment aspect since it allows more tailored therapies and reduces the side effects of the anti-epileptic medicine (Myint, Staufenberg, & Sabanathan, 2006).

Studies of epilepsy often rely on the electroencephalogram (EEG) as the EEG can indicate the brain’s electrical activities associated with seizures (Smith, 2005). Much research and development in the area of automated seizure detection using EEG signals have been carried and several EEG-based seizure detection systems have been reported in the literature. Feature extraction and feature classification are the two essential modules that are necessary to build an automatic seizure detection system. Around ninety percent of the prior work has focused on developing effective feature extraction techniques that can discover the most discriminative EEG features for seizure diagnosis. Most of these techniques use hand-crafted features extracted from the time-domain (Meier, Dittrich, Schulze-Bonhage, Aertsen, 2008, Minasyan, Chatten, Chatten, Harner, 2010), frequency-domain (Aarabi, Fazel-Rezai, Aghakhani, 2009, Correa, Laciar, Patiño, Valentinuzzi, 2007, Polat, Güneş, 2007), wavelet-domain (Abibullaev, Seo, Kim, 2010, Güler, Übeyli, 2005, Tzallas, Tsipouras, Fotiadis, 2007), and sometimes from multiple domain representations of the EEG (Mitra et al., 2009). However, domain-based schemes experience two major difficulties. First, the non-stationarity nature of EEG makes it difficult to have a single seizure pattern, making the hand-crafted features less practical in clinical settings (Kaplan, Fingelkurts, Fingelkurts, Borisov, & Darkhovsky, 2005). Second, in practice, the EEG data is prone to different sources of artifacts such as muscle activities and eye-blinking as well as white noise. Artifacts and noise interfere with EEG signals, producing serious distortions that negatively affect the seizure detection performance (Abualsaud, Mahmuddin, Saleh, & Mohamed, 2015).

To overcome these challenges, we develop a robust seizure detection system that can accurately recognize epileptic seizures under real-life conditions as well as ideal conditions. A feature extraction method, based on L1-penalized robust regression (L1PRR), is proposed to provide the most informative seizure-associated features. The proposed scheme is applied to the EEG frequency spectra, and the extracted spectral features are used as an input to the random forest (RF) classifier for EEG training and classification. We used the benchmark clinical dataset provided by Bonn University (Andrzejak et al., 2001) to compare the seizure detection performance of our proposed method to those of the state-of-the-art techniques. We first examine how efficiently our proposed method performs under ideal conditions, i.e., when the EEG data is completely free of artifacts and noise. Results demonstrate that our approach achieves superior performance than those in literature; delivering the highest seizure detection rates of 100% sensitivity, 100% specificity, and 100% classification accuracy. More importantly, the performance of our approach is tested in the presence of muscle activities and eye-blinking artifacts as well as white noise. Our approach is proven to be robust against all of these interferences. It maintains high seizure detection performance even at high noise levels, making it more practical and workable in real-life and clinical settings.

Section snippets

Time-series EEG data

In this study, we conduct our seizure detection experiments on the benchmark clinical EEG dataset provided by Bonn University (Andrzejak et al., 2001). This is popular and well-known dataset for epileptic seizure detection. It comprises five different sets denoted A, B, C, D, and E; each set includes 100 single-channel EEG signals of 23.6 s duration each. Sets A and B contain surface EEG signals recorded from 5 healthy volunteers using the standard 10–20 system for EEG electrode placement (

Literature review

In this section, we briefly review the previous work in the context of epileptic seizure detection using EEG signals.

Methodology

This section demonstrates how the L1-penalized robust regression (L1PRR) model is applied to the EEG spectrum to extract the most discriminative EEG features pertinent to epileptic seizures. It also depicts how the L1PRR problem can be solved using a computationally-simple approach named block coordinate descent (BCD). Finally, it describes how the extracted features are used for training and testing random forest (RF) classifier.

Results and discussion

To evaluate the effectiveness of the proposed robust regression-based seizure detection approach, we compare its performance to those of the state-of-the-art detectors that use the same benchmark dataset. The detection performance was evaluated using the standard metrics of sensitivity, specificity, and classification accuracy (Hussein, Ward, Wand, & Mohamed, 2016).

Conclusion

This paper presents a robust method for automatic detection of epileptic seizures using EEG signals. This method relies on a robust feature extraction scheme that we developed based on L1-penalized robust regression (L1PRR). L1PRR can adaptively learn the most informative EEG features pertinent to seizures under ideal and real-life conditions. The performance of the proposed method is first examined under ideal conditions, i.e., when the EEG data is free of noise. Experimental results on a

References (75)

  • U. Herwig et al.

    Using the international 10–20 EEG system for positioning of transcranial magnetic stimulation

    Brain Topography

    (2003)
  • M. Kaleem et al.

    EEG seizure detection and epilepsy diagnosis using a novel variation of empirical mode decomposition

    Proceedings of the thirty-fifth IEEE annual international conference on engineering in medicine and biology society (EMBC)

    (2013)
  • Y.U. Khan et al.

    Automated seizure detection in scalp EEG using multiple wavelet scales

    Proceedings of the 2012 IEEE international conference on signal processing, computing and control (ISPCC)

    (2012)
  • A.M. Murugavel et al.

    Lyapunov features based EEG signal classification by multi-class SVM

    Proceedings of the 2011 world congress on information and communication technologies (WICT)

    (2011)
  • Y. Nesterov

    Efficiency of coordinate descent methods on huge-scale optimization problems

    SIAM Journal on Optimization

    (2012)
  • M. Niknazar et al.

    A new framework based on recurrence quantification analysis for epileptic seizure detection

    IEEE Journal of Biomedical and Health Informatics

    (2013)
  • M. Peker et al.

    A novel method for automated diagnosis of epilepsy using complex-valued classifiers

    IEEE Journal of Biomedical and Health Informatics

    (2016)
  • D. Rivero et al.

    A new signal classification technique by means of genetic algorithms and kNN

    Proceedings of the 2011 IEEE congress on evolutionary computation (CEC)

    (2011)
  • Siuly et al.

    A novel statistical algorithm for multiclass EEG signal classification

    Engineering Applications of Artificial Intelligence

    (2014)
  • S. Smith

    EEG in the diagnosis, classification, and management of patients with epilepsy

    Journal of Neurology, Neurosurgery & Psychiatry

    (2005)
  • A. Subasi

    EEG signal classification using wavelet feature extraction and a mixture of expert model

    Expert Systems with Applications

    (2007)
  • A.T. Tzallas et al.

    Automatic seizure detection based on time-frequency analysis and artificial neural networks

    Computational Intelligence and Neuroscience

    (2007)
  • E.D. Übeyli

    Combined neural network model employing wavelet coefficients for EEG signals classification

    Digital Signal Processing

    (2009)
  • E.D. Übeyli

    Decision support systems for time-varying biomedical signals: EEG signals classification

    Expert Systems with Applications

    (2009)
  • WangL. et al.

    Automatic epileptic seizure detection in EEG signals using multi-domain feature extraction and nonlinear analysis

    Entropy

    (2017)
  • YuanQ. et al.

    Epileptic EEG classification based on extreme learning machine and nonlinear features

    Epilepsy Research

    (2011)
  • ZhangT.

    Solving large scale linear prediction problems using stochastic gradient descent algorithms

    Proceedings of the twenty-first international conference on machine learning

    (2004)
  • B. Abibullaev et al.

    Epileptic spike detection using continuous wavelet transforms and artificial neural networks

    International Journal of Wavelets, Multiresolution and Information Processing

    (2010)
  • K. Abualsaud et al.

    Ensemble classifier for epileptic seizure detection for imperfect EEG data

    The Scientific World Journal

    (2015)
  • U.R. Acharya et al.

    Use of principal component analysis for automatic classification of epileptic EEG activities in wavelet framework

    Expert Systems with Applications

    (2012)
  • U.R. Acharya et al.

    Application of recurrence quantification analysis for the automated identification of epileptic EEG signals

    International Journal of Neural Systems

    (2011)
  • U.R. Acharya et al.

    Automated diagnosis of epilepsy using CWT, HOS and texture parameters

    International Journal of Neural Systems

    (2013)
  • S. Amin et al.

    A robust approach towards epileptic seizure detection

    Proceedings of the twenty-sixth IEEE international workshop on machine learning for signal processing (MLSP)

    (2016)
  • R. Andersen

    Modern methods for robust regression

    (2008)
  • R.G. Andrzejak et al.

    Indications of nonlinear deterministic and finite-dimensional structures in time series of brain electrical activity: dependence on recording region and brain state

    Physical Review E

    (2001)
  • D.S.T. Behara et al.

    Detection of epileptic seizure patterns in EEG through fragmented feature extraction

    Proceedings of the third international conference on computing for sustainable global development (IndiaCom)

    (2016)
  • A. Bhattacharyya et al.

    Tunable-Q wavelet transform based multiscale entropy measure for automated classification of epileptic EEG signals

    Applied Sciences

    (2017)
  • Cited by (44)

    • Automatic feature learning model combining functional connectivity network and graph regularization for depression detection

      2023, Biomedical Signal Processing and Control
      Citation Excerpt :

      Finally, the fused features were fed into the cascade forest model as the input for depression recognition, and 92.48% recognition accuracy could be achieved. Meantime, some researchers explored novel solutions from the perspective of constructing or improving optimization models, which are able to learn the EEG features flexibly and adaptively [30–33]. Compared with the handcrafted features, these features learnt adaptively perform better.

    • Epileptic-seizure classification using phase-space representation of FBSE-EWT based EEG sub-band signals and ensemble learners

      2022, Biomedical Signal Processing and Control
      Citation Excerpt :

      For analysing epileptic-seizure EEG signals, the authors have used a significant feature based on matrix determinant [20]. In this paper [21], the authors have computed a feature vector based on L1- penalized robust regression. Then it is passed to a random forest (RF) classifier as an input, achieves the highest classification accuracy of 100 % without artifacts, and while in the presence of white noise and artifacts, it achieves accuracy around 90 % for classifying seizure EEG signals.

    • Automatic seizure detection using orthogonal matching pursuit, discrete wavelet transform, and entropy based features of EEG signals

      2021, Computers in Biology and Medicine
      Citation Excerpt :

      Moreover, Zhang et al. analyzed EEG seizure activities using a novel feature calculation approach, i.e., combining the fuzzy distribution entropy and the wavelet packet decomposition [19]. Hussein et al. applied a new feature learning algorithm based on the L1-penalized robust regression for seizure recognition purposes [9]. Peachap et al. introduced a novel method based on wavelets by considering the differences and similarities between the Mexican hat and Gaussian wavelet families.

    • Epilepsy EEG signals classification based on sparse principal component logistic regression model

      2024, Computer Methods in Biomechanics and Biomedical Engineering
    View all citing articles on Scopus

    The first author is funded by Vanier Canada Graduate Scholarship from the Natural Sciences and Engineering Research Council of Canada (NSERC) http://www.vanier.gc.ca/en/nomination_processprocessus_de_mise_en_candidature_overview.html.

    View full text