Seizure detection in EGG signal with novel optimization algorithm for selecting optimal thresholded offset Gaussian feature

https://doi.org/10.1016/j.bspc.2019.101708Get rights and content

Highlights

  • In detecting the epileptic seizure disorder, the EEG signals of the patient are diagnosed. In this analysis, the occurrence of epileptiform discharges in between the happening of seizure disorder should be identified.

  • For obtaining the accurate classification result, the features should be carefully extracted. Though it is better to take more features in order to obtain clear and detail information of the signal, the classifier performance will be affected when more features are applied.

  • In this work the Divergence State Estimation based Bio-geography based Optimization (DSEBBO) is presented for feature selection process whereas the for feature extraction method, Thresholded Offset Gaussian (TOG) method is employed.

  • The selected optimal features are applied to the SVM classifier and the results are validated for the proposed method by comparing with the other optimization algorithms as well as other feature extraction methods.

  • From the classification results, it is clearly elucidated that the TOG feature extraction and the utilization of DSEBBO based optimal feature selection strategy gives the better accuracy which 4.64% more than the method that does not employ feature selection technique.

Abstract

EEG analysis is a responsible device in assessing the neurophysiological disorders associated to the postsynaptic movement happened in the neocortex. In detecting the epileptic seizure disorder, the EEG signals of the patient are diagnosed. In this analysis, the occurrence of epileptiform discharges in between the happening of seizure disorder should be identified. The introduction of automatic computerized process is required to analyze the EEG signals for detecting the seizure disorder. For obtaining the accurate classification result, the features should be carefully extracted. Though it is better to take many features in order to obtain clear and detail information of the signal, the classifier performance will be affected when large number of features is applied. Hence the optimal set of features must be obtained using the suitable optimization algorithm. In this work the Divergence State Estimation based Bio-geography based Optimization (DSEBBO) is presented for feature selection process whereas the for feature extraction method, Thresholded Offset Gaussian (TOG) method is employed. The selected optimal features are applied to the SVM classifier and the results are validated for the proposed method by comparing with the other optimization algorithms as well as other feature extraction methods. From the classification results, it is clearly elucidated that the TOG feature extraction and the utilization of DSEBBO based optimal feature selection strategy gives the better accuracy which is 4.64% more than the method that does not employ feature selection technique.

Introduction

Epileptic seizure is the disease caused by the transitory electrical disturbance occurred in the brain. The event of seizures often confused with the occurrence of some other events like stroke or even become undetected in some occasions. Normally, one person of the population of 100 people may undergo seizure in their lifetime [1]. It is always unpropitious since the happening of the event of seizures is unstable and less understandable. In order to diagnose the brain disorders to understand the cause for the event exhaustively, it is recommended to examine the electroencephalograph (EEG) records of the patient. For diagnosing epileptic disorders, it is significant to detect the occurrence of epileptiform discharges in the EEG signal in between the happening of seizure disorder [2]. The World Health Organization (WHO) says that roughly 50 million persons are affected by epileptic seizure [3]. It has also reported that about 4–10 person of population of 1000 people suffer by the epileptic seizure. In a year, approximately 2.4 million persons are diagnosed with epilepsy. The persons are diagnosed for epilepsy when a minimum of two malevolent seizures happened. It is essential to perform the diagnosis on time for providing the medical aids to cure the disease and to decrease the chance of occurrence of seizure in future [4]. The current practice of diagnosing epileptic disorder is to perform the analysis of patient history, neurological tests and some other tests like neuro-imaging and EEG analysis. There are two different epileptiform abnormalities which are termed as inter-ictal that occurs between seizures and ictal that occurs during seizure. These two events can be detected in EEG signals. The epileptiform waves are detected by the presence of spikes, sharp waves and polyspike complexes in the EEG signals [5].

EEG analysis is a responsible device in assessing the neurophysiological disorders associated to the postsynaptic movement happened in the neocortex [6]. The physicians used EEG signals in examining various functions of the brain and also to identify the anomalies like seizure disorder, stroke, Alzheimer disease, sleep apnea, trauma and other mental disorders. Normally, the physicians analysed the EEG signals by performing the visual analysis for detecting the epileptic disorder. However, it is a time consuming process and requires many number of electrodes for recording the signal. Hence, it is essential to introduce an automatic computerized algorithm for the effective classification of EEG signals in detecting the epileptic disorder. In the classification process, the features take vital role in obtaining the better results. The feature extraction step should be done meticulously. For arriving at a better classification results, the application of sensitive features is judiciously imposed. Thus the extraction of perceptive features from the EEG signal is of greater concern in automatic classification of EEG data. The features that show discrete effects in EEG signal of a person having epileptic disorder should be considered. In general, the EEG signal consists of many undue features which effect nothing on the result since certain segment of waveform repeats again at different time period. Hence by applying the large amount of features, the accuracy and reliability of the classification result may be disrupted. In addition, the time consumption for this process is substantially high.

There are many works for extracting the features from the EEG signals for applying to the classification algorithms. The standard methods for extracting features from the EEG signals are Short Time Fourier transform (STFT) [7], Fast Fourier transform (FFT) [8] and the wavelet transform (WT) [9]. The application of suitable feature extraction method is analytically indispensable since it affects the classifier results and size of the data processed in the classifier algorithm. Different algorithms are in practice for so many years, such as support vector machine (SVM), artificial neural network (ANN), k-nearest neighbor (KNN), Adaptive Neuro-Fuzzy Inference System (ANFIS) classifier and extreme machine learning (ELM) algorithm. After extracting the features using a suitable feature extraction technique, they can be applied to the feature selection process in order to make the classifier performance more effective. The problem selecting the subset form the higher volume of sets of features is of greater concern. For this purpose, the meta-heuristics algorithms can be applicable. Many meta-heuristics algorithms like Genetic algorithm (GA) [10], Particle Swarm Optimization (PSO) [11], Ant colony optimization (ACO) [12] and Social Spider optimization (SSO) [13] are employed for the feature selection process. They are random feature selection methods and results in a local optimum result and it is proved by Simon in [14] that the Biogeography Based Optimization (BBO) algorithms yields better comparing with these methods.

There are many bio-inspired optimization algorithms available in the literatures. Recently the new algorithms like Krill Herd [15], grey wolf optimizer [16], elephant herding optimization [17], lion optimization [18], whale optimization [19], dragon fly optimization [20] and coral reef optimization [21] were presented. In this work BBO algorithm is presented. The standard BBO algorithm is first presented by Simon in [14] for optimizing the metaheurisitc problem. For the recent years, many authors have been worked on BBO algorithm to introduce several modifications for improving the performance of the standard BBO algorithm in order to increase the accuracy of the global best result arrived and the convergence time. The modified BBO algorithms are introduced in literatures from [[22], [23], [24], [25], [26], [27], [28], [29], [30]]. In works [[31], [32], [33], [34], [35], [36], [37]], the hybrid algorithms of BBO with other meta-heuristic algorithms are presented. The merits and demerits of the papers presented on different BBO works are discussed in the next Section. The crossover operation appearing in many evolutionary-based algorithms may disturb the fitness of many good solutions, which may be higher quality during the initial stage and consequently drops their quality in succeeding stages. In BBO, since there is no crossover operation, the solutions are steadily modified using migration and mutation operations. In [23], Markov model is developed for representing population distribution in BBO using a small proportion which requires more computation time and it is also a complex process. In [24], Gaussian mutation, Cauchy mutation and Levy mutation operators are employed to obtain improved mutation process. This method shows better exploration ability comparing with the standard BBO method. This method suits for high-dimensional problems. In [26], improved BBO algorithm is presented by applying nonlinear variations of immigration rate and emigration rate. However this work does not reduce the computational time significantly. Divergence State Estimation (DSE) based BBO (DSEBBO) method presented in this work used probabilistic immigration and emigration rate which is computed using less complex expressions and requires less time for processing.

In this work, DSEBBO is presented for feature selection approach. The preliminary steps such as the pre-processing of the given is done to remove noises and other artifacts by using the Finite Impulse Response (FIR) filter. Then the preprocessed EEG signal is applied for the feature extraction process. The given EEG signal is applied for Thresholded Offset Gaussian (TOG) feature extraction. The features are obtained for different offset distances. Hence, the resultant number of features increases. If all of these features are applied to the classifier, then the classifier yields poor result. This is for the reason that the high volume of features can instigate over fitting of the classifier algorithm [37]. Hence, in this work DSEBBO is presented for obtaining the optimum features set only. The selected feature set is applied to the SVM classifier and the seizure disorder is detected by classifying the EEG signal based on the features.

The contributions of the proposed work are illustrated as follows. Comparing with the conventional BBO algorithm, the proposed DSEBBO algorithm converges faster since the immigration process are done using the divergence state estimation process, which consumes less time and reduces the complexity of the problem significantly. The implementation of TOG for feature extraction process helps to obtain extensive details regarding the spikes occurred in the EEG signals and the time duration of the spikes which helps in classifying the signal. This eliminates the presence of noise or enhancement of noise in the signal while extracting the features. It also reduces the complexity when comparing with the computation of Gaussian derivative model. The performance of the proposed methodology in detecting the seizure disorder is justified by comparing with the previous works on seizure detection.

The paper is organized with five major Sections. The works done on detection of seizure, works presented for feature selection and works for the BBO algorithm are illustrated in Section 2. Section 3 provides the details of the dataset used in the work. Section 4 illustrates the brief theory about the different methodologies involved in the work. The validation of results of the proposed work and the comparison of performance with the previous works are described in Section 5.

Section snippets

Literature survey

For the past two decades, there have been several research works on the task of detecting the epileptic disorder by analyzing EEG signals. The time domain and frequency domain features take crucial role in analyzing the EEG signals. In work [38], the independent component analysis (ICA) technique was presented to remove artifacts in EEG signal and to separate the signal into individual sources. However the drawback is that the ICA methods works when the signal is completely noise free and there

Dataset details

Though there are some databases available for the EEG signals, the database collected by Andrzejak et al. [64], which is available at Bonn University, Germany, is used in many works. The dataset consists of five different classes of signals representing normal EEG signals (set is denoted by A and B), preictal EEG signal (set is denoted by C and D) and seizure EEG signal (set is denoted by E). The signals are obtained using continuous multichannel EEG measurements. In each dataset comprises 100

Proposed methodology

This Section describes the different methods involved in this work. The preprocessed EEG signals are initially considered for feature extraction process. In obtaining effective classification results, the extracted features take significant role. Hence, feature extraction should provide essential information. Concurrently, the extracted features should not be higher in volume since the high dimensional features may over fit the classifier performance. Therefore, the feature selection, which

Classification

The selected optimal feature set is applied to the SVM classifier [67]. The sample signals in the given dataset including the normal patients and seizure affected patients are distributed for training and testing. The datasets A and B are jointly considered as the class having normal patients’ dataset. On the other hand, the datasets C, D and E are jointly considered as the class with seizure affected patients’ data. After taking 30,000 samples from each dataset, the samples are distributed for

Result and discussion

To evaluate the proposed technique in automatic detection of seizure disorder from EEG signals, the dataset described in Section 3 are taken and tested in Matlab software. The original EEG signal of a normal person is depicted in Fig. 4.

The EEG signals are applied to threshold offset Gaussian function and TOG signal is obtained. In Gaussian, the σ values can be changed to change the trend of the signal. Thus by using 10 σ values, ten TOG signals are obtained. For each of them 3 statistical

Conclusion

The proposed methodology presented an automatic process of EEG signal classification for detecting the seizure disorder. In order to detect the seizure events from the EEG signal in an accurate manner, we have presented TOG feature extraction method. For obtaining accurate results many TOG features are extracted. But it affects the classifier performance when they are applied directly. Hence the optimal set of features are extracted which contain the set of features that provide better accuracy

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

V. Sutha Jebakumari completed her B.E in Computer Science & Engineering in the year 1992 from Thiagarajar College of Engineering, Madurai and M.E. Computer science and Engineering in from Manonmaniam Sundaranar University, Tamil Nadu. She is currently working as an Assistant Professor in Department of Computer Science & Engineering, Kamaraj College of Engineering and Technology, Virudhunagar. She is a member of professional societies like ISTE and CSI. Her research interests include Soft

References (72)

  • Wenyin Gong

    A real-coded biogeography-based optimization with mutation

    Appl. Math. Comput.

    (2010)
  • Dan Simon

    A dynamic system model of biogeography-based optimization

    Appl. Soft Comput.

    (2011)
  • Haiping Ma et al.

    Blended biogeography-based optimization for constrained optimization

    Eng. Appl. Artif. Intell.

    (2011)
  • Amitava Chatterjee

    An improved biogeography based optimization approach for segmentation of human head CT-scan images employing fuzzy entropy

    Eng. Appl. Artif. Intell.

    (2012)
  • M.R. Lohokare

    Accelerated biogeography-based optimization with neighborhood search for optimization

    Appl. Soft Comput.

    (2013)
  • Poonam Savsani et al.

    Effect of hybridizing biogeography-based optimization (BBO) technique with artificial immune algorithm (AIA) and ant colony optimization (ACO)

    Appl. Soft Comp.

    (2014)
  • Yu-Jun Zheng et al.

    Ecogeography-based optimization: enhancing biogeography-based optimization with ecogeographic barriers and differentiations

    Comput. Oper. Res.

    (2014)
  • N. Kannathal

    Entropies for detection of epilepsy in EEG

    Comp. Methods Programs Biomed.

    (2005)
  • A.H. Jahidin

    Classification of intelligence quotient via brainwave sub-band power ratio features and artificial neural network

    Comp. Methods Programs Biomed.

    (2014)
  • Manish Sharma et al.

    A new approach to characterize epileptic seizures using analytic time-frequency flexible wavelet transform and fractal dimension

    Pattern Recognit. Lett.

    (2017)
  • Kaveh Samiee et al.

    Epileptic seizure detection in long-term EEG records using sparse rational decomposition and local Gabor binary patterns feature extraction

    Knowledge Based Syst.

    (2017)
  • Abeg Kumar Jaiswal et al.

    Local pattern transformation based feature extraction techniques for classification of epileptic EEG signals

    Biomed. Signal Proc. Control

    (2017)
  • Y.ılmaz Kaya

    1D-local binary pattern based feature extraction for classification of epileptic EEG signals

    Appl. Math. Comp.

    (2014)
  • Ling Guo

    Automatic feature extraction using genetic programming: an application to epileptic EEG classification

    Expert Syst. Appl.

    (2011)
  • Abdulhamit Subasi et al.

    EEG signal classification using PCA, ICA, LDA and support vector machines

    Expert Syst. Appl.

    (2010)
  • U. Rajendra Acharya

    Deep convolutional neural network for the automated detection and diagnosis of seizure using EEG signals

    Comp. Biol. Med.

    (2018)
  • ...
  • A. Krumholz et al.

    Evidence-based guideline: management of an unprovoked first seizure in adults: report of the Guideline Development Subcommittee of the American Academy of Neurology and the American Epilepsy Society

    Neurology

    (2015)
  • Laura Anna Van de Pol

    Strong relation between an EEG functional connectivity measure and postmenstrual age: a new potential tool for measuring neonatal brain maturation

    Front. Hum. Neurosci.

    (2018)
  • Hamid R. Mohseni

    Automatc detection of epileptic seizure using time-frequency distributions." Advances in medical

  • Asrul Adam

    Feature selection and classifier parameters estimation for EEG signals peak detection using particle swarm optimization

    Sci. World J.

    (2014)
  • Morteza Behnam et al.

    Periodogram pattern feature-based seizure detection algorithm using optimized hybrid model of MLP and ant colony

  • Ahmed M. Anter

    Feature selection approach based on social spider algorithm: case study on abdominal CT liver tumor

  • Dan Simon

    Biogeography-based optimization

    IEEE Trans. Evol. Comp.

    (2008)
  • Gai-Ge Wang et al.

    Elephant herding optimization

  • Seyedali Mirjalili

    Dragonfly algorithm: a new meta-heuristic optimization technique for solving single-objective, discrete, and multi-objective problems

    Neural Comput. Appl.

    (2016)
  • Cited by (8)

    • Epileptic seizure prediction in intracranial EEG using critical nucleus based on phase transition

      2022, Computer Methods and Programs in Biomedicine
      Citation Excerpt :

      Therefore, seizure prediction can effectively avoid patients suffering from sudden injuries, which improves the patients’ quality of life. Electroencephalogram (EEG) is the recording of the electrical activity of the brain, which has been extensively used in diagnosing and detecting epileptic seizures [3–5] due to its high temporal resolution, low cost, and continuous monitoring. Considerable pieces of evidence [1,6–11] have shown that the underlying mechanism of seizure based on the epileptic EEG shares the general principles of other complex non-linear systems.

    • Patient-specific method of sleep electroencephalography using wavelet packet transform and Bi-LSTM for epileptic seizure prediction

      2021, Biomedical Signal Processing and Control
      Citation Excerpt :

      The filtered EEG datasets of each case were segmented using a sliding window. Subsequently, these segments are decomposed by WPT; then, the signal is divided into 16 sub-bands, and the energy values are obtained by calculating the corresponding sub-band coefficients to form a feature vector [52]. The feature vector is used as input to Bi-LSTM to classify the EEG into preictal/interictal.

    View all citing articles on Scopus

    V. Sutha Jebakumari completed her B.E in Computer Science & Engineering in the year 1992 from Thiagarajar College of Engineering, Madurai and M.E. Computer science and Engineering in from Manonmaniam Sundaranar University, Tamil Nadu. She is currently working as an Assistant Professor in Department of Computer Science & Engineering, Kamaraj College of Engineering and Technology, Virudhunagar. She is a member of professional societies like ISTE and CSI. Her research interests include Soft Computing and Machine Learning. She has published more than 10 papers in International Conferences and Journals.

    Dr. D. Shanthi Saravanan received her B.E. Computer Science and Engineering in 1992 from Thiagarajar College of Engineering, Madurai, Tamil Nadu and M.E. Computer science and Engineering in from Manonmaniam Sundaranar University, Tamil Nadu and PhD in Soft Computing from Birla Institute of Technology, Ranchi. She is currently working as a Professor in Department of Computer Science & Engineering, PSNA College of Engineering and Technology. She has more than 21 years of Teaching and Programming Experience. She is member of various professional societies like IEEE, CSTA, IAENG and IACSIT. Her research interests include Genetic Algorithms, Neural Networks, Intelligent Systems, Image Processing, Embedded System, M-Learning and Green Computing. She has published more than 25 papers in International Conferences and Journals and also published 5 books in computing and applications. She is the reviewer of various international journals.

    D. Devaraj completed his B.E and M.E in Electrical & Electronics Engineering and Power System Engineering in the year 1992 and 1994, respectively, from Thiagarajar College of Engineering, Madurai. He obtained his Ph.D degree from IIT Madras in the year 2001. He has organized 8 Conferences and 10 Seminars. He has Supervised 20 Ph.D, 2 M.S and 25 M.E thesis. Presently, he is guiding 8 Ph.D scholars. He has undertaken 3 sponsored research projects. His research interest includes Power system security, Voltage stability and Evolutionary Algorithm. At present, he is the Dean, School of Electronics and Electrical Technology at Kalasalingam Academy of Research and Education, Krishnankoil, India.

    View full text