Skip to content
BY 4.0 license Open Access Published by De Gruyter December 15, 2023

An overview of machine learning and deep learning techniques for predicting epileptic seizures

  • Marco Zurdo-Tabernero ORCID logo EMAIL logo , Ángel Canal-Alonso ORCID logo EMAIL logo , Fernando de la Prieta ORCID logo , Sara Rodríguez ORCID logo , Javier Prieto ORCID logo and Juan Manuel Corchado ORCID logo

Abstract

Epilepsy is a neurological disorder (the third most common, following stroke and migraines). A key aspect of its diagnosis is the presence of seizures that occur without a known cause and the potential for new seizures to occur. Machine learning has shown potential as a cost-effective alternative for rapid diagnosis. In this study, we review the current state of machine learning in the detection and prediction of epileptic seizures. The objective of this study is to portray the existing machine learning methods for seizure prediction. Internet bibliographical searches were conducted to identify relevant literature on the topic. Through cross-referencing from key articles, additional references were obtained to provide a comprehensive overview of the techniques. As the aim of this paper aims is not a pure bibliographical review of the subject, the publications here cited have been selected among many others based on their number of citations. To implement accurate diagnostic and treatment tools, it is necessary to achieve a balance between prediction time, sensitivity, and specificity. This balance can be achieved using deep learning algorithms. The best performance and results are often achieved by combining multiple techniques and features, but this approach can also increase computational requirements.

1 Introduction

Epilepsy is a neurological disorder that affects over 39 million people in the United States [1], making it the third most common disorder after stroke and migraines [2]. This chronic condition is known to impact the quality of life significantly, making it a focus of ongoing research and development of new therapies. According to the International League Against Epilepsy, an epileptic seizure is a “transient occurrence of signs and/or symptoms due to abnormal excessive or synchronous neuronal activity in the brain” [3].

Table 1:

List of the features that can be extracted for seizure prediction and detection.

Time domain Kurtosis
Mean
Skewness
Variance
Phase Correlation
Common Spatial Pattern
Nonlinear features Correlation dimension
Largest Lyapunov Exponent
Fractality Dimension
Lempel-Zic Complexity
Approximate Entropy
Spectral Entropy
Wavelet Entropy
Hjort Parameters Analysis
Wavelet Energy
Frequency Domain Features Short Time Fourier Transform
Fractional Fourier Transform
Spectral Power Analysis
Time-frequency domain features Discrete Wavelet Transform
Continuous Wavelet Transform
Higher-Order Spectra
Variational Model Decomposition
Table 2:

Feature extraction application.

Time domain Parvez and Paul [9]
Nonlinear features Aarabi et al. [12]
Shafique et al. [14]
D’Alessandro et al. [16]
Zhang et al. [19]
Direito et al. [21]
Gadhoumi et al. [23]
Frequency Domain Features Truong et al. [25]
Bandarabnadi et al. [28]
Time-frequency domain features Ouyang et al. [35]
Acharya et al. [32]
Kumat and Rao [34]
Table 3:

Feature selection methods.

Principal Component Analysis (PCA) Williamson et al. [36]
Partial Least Squares (PLS) Varatharajah et al. [45]
Minimum Redundancy Maximum relevance (MRMr) Bandarabadi et al. [39]
Gaussian Mixture Models Zandi et al. [41]
Deep Learning (Automatic Feature Selection) Wei et al. [43]
Yan et al. [44]
Tsiouris et al. [46]
Table 4:

Seizure prediction methods.

Author Features Method Pred. time Results
Animal models
Ouyang et al. [35] Wavelet transform Wavelet-based nonlinear similarity index 2.14 min Sensitivity > 90 %

Specificity = 6.25 %
Varatharajah et al. [45] Power in band (PIB) features, time domain correlations (TMCO), and spectral coherence (SPCO) between different pairs of channels Support Vector Machine, Artificial Neural Networks, and Random Forests Classifier 1 min Specificity = 70 %

Specificity = 75 %
Rajdev et al. [49] Filtered signal Adaptive Wiener algorithm 6.7 s Sensitivity = 94 %

False positive rate (FPR) = 0.009 min−1
Human subjects
Iasemidis et al. [50] Short-term maximum Lyapunov exponent Spatial dynamical entrainment of short-term maximum Lyapunov exponent 89 min Sensitivity = 91.3 %

FPR = 8.27 h−1
Aarabi et al. [12] Correlation dimension, correlation entropy, noise level, Lempel-Ziv complexity, largest Lyapunov exponent, and nonlinear inter-dependence Patient-specific rule-based algorithm 50 min Sensitivity >90 %

FPT <0.15 h−1
Zhang et al. [19] Higuchi Fractal Dimension Bayesian linear discriminant analysis as a classifier 39.39 min Sensitivity = 89.33 %

FPR = 0.20 h−1
Tsiouris et al. [46] Time domain, Frequency domain, Brain connectivity, and Graph-theoretic features LSTM 120 min Sensitivity > 99 % and FPR <0.06 h−1
Wei et al. [43] CNN automatic feature extraction Long-Term Convolutional Neural Network 21 min Sensitivity = 93.40 %

Specificity = 86.13 %
Yan et al. [44] STFT Transformers 30 min Sensitivity = 96.01 %

FPR = 0.047 h−1
Figure 1: 
The standard process for classifying or predicting epileptic seizures.
Figure 1:

The standard process for classifying or predicting epileptic seizures.

1.1 Issues

Electroencephalography (EEG) is the most commonly used method for diagnosing epilepsy. This technique involves recording the brain’s electrical activity through the measurement of voltage changes caused by the ionic currents of brain cells [4]. Electrodes placed on the scalp (sEEG) or inside the cranium (iEEG) are used to record the voltage changes produced by the brain’s electrical activity. These voltage changes result from the simultaneous activity of millions of neurons, which generate a voltage wave that the electrodes can detect.

The events observed in the EEG can be grouped into three categories [5]: ictal events, which occur during a seizure; preictal events, which occur before a seizure; and interictal events, which are all other events that are not part of the ictal or preictal phase. The duration of the preictal phase can vary widely, from a few minutes to several hours.

In animal and in vitro models, microelectrodes are sometimes used instead of EEG to perform electrophysiology recordings, due to the limitations of EEG in these contexts.

1.2 Objectives

This study aims to give an overview of the existing machine-learning techniques for predicting epileptic seizures based on features extracted from EEG signals. As it is not intended to portray a pure bibliographical review of the subject, the publications here cited have been selected among many others based on their relevance and number of citations.

The present document goes through the usual pipeline for seizure prediction (Figure 1). First of all, the processing of the EEG signal will be discussed, including artifact-removal techniques. Later, we will discuss the most relevant features that can be extracted from the preprocessed signal, and how we can manage to select those that contain the optimal amount of information needed to generate a predictive model. Finally, the most prevalent methods for epileptic seizure prediction will be exposed, separating studies using animal and human EEG.

1.3 Questions and hypothesis

The use of machine learning (ML) and deep learning (DL) methods for epileptic seizure detection and prediction has gained significant attention in recent years. However, several questions and hypotheses remain unanswered. In this paper, we aim to address the following questions and hypotheses:

  1. Which features are most relevant for epileptic seizure detection and prediction? Various features have been proposed for ML and DL models, but it is unclear which features are most relevant for accurate seizure detection and prediction.

  2. Which ML and DL models perform best for epileptic seizure detection and prediction? Several ML and DL models have been proposed for seizure detection and prediction, but it is unclear which models perform best and under what conditions.

By addressing these questions and hypotheses, we aim to provide a comprehensive understanding of the current state of the field of ML and DL methods for epileptic seizure detection and prediction and identify potential areas for future research.

2 Signal processing and feature extraction

EEG signals may be affected by various artifacts that can mask information related to epilepsy. The type and details of the EEG recording can influence the presence and nature of these artifacts.

A common step in processing EEG signals is removing background noise through filters to eliminate the 50–60 Hz powerline frequency. After this step, a large number of features can be extracted from the processed signal.

2.1 Methods

The present work is a narrative review [6], which is a traditional way of reviewing the existing literature on a particular topic. Narrative reviews attempt to summarize or synthesize what has been written on a particular topic but do not seek generalization or cumulative knowledge from what is reviewed. The primary goal of a narrative review is to determine the extent to which a body of literature provides evidence for a particular theory or hypothesis. Narrative reviews are helpful in obtaining a broad perspective on a topic. The present work aims to provide a comprehensive understanding of the current state of the field of epileptic seizure detection and prediction by summarizing and synthesizing the existing literature on the topic.

Internet bibliographical searches were conducted using the keywords “epileptic seizure prediction,” “artificial intelligence,” and “neural networks (computer)” to identify relevant literature on the topic. Through cross-referencing from key articles, additional references were obtained to provide a comprehensive overview of various artificial intelligence techniques for epileptic seizure prediction. The articles here referenced range from 1901 to 2022, and have been extracted from Web of Science and Scopus, along with the 142 databases available to the members of the Universidad de Salamanca [7].

This paper aims to present a detailed analysis of the current methodologies for the application of artificial intelligence techniques in epileptic seizure prediction, including classical ML algorithms, and DL algorithms such as dense neural networks (DNN), convolutional neural networks (CNNs), and recurrent neural networks (RNNs). As it is not intended to portray a pure bibliographical review of the subject, the publications here cited have been selected among many others based on their relevance and number of citations.

2.2 Feature extraction

2.2.1 Statistical features

EEG signals can be analyzed using statistical parameters such as kurtosis, mean, skewness, and variance [8] to track changes in distribution and amplitude. Phase correlation [9, 10] can be used to examine patterns in ictal and preictal events, and Common Spatial Pattern [11] can extract features through the decomposition of EEG signals.

2.2.2 Nonlinear features

Correlation Dimension [12, 13] can be used to measure the complexity of each event, and the Largest Lyapunov Exponent [14, 15] can calculate the level of chaos in EEG signals. Fractality Dimension [16, 17] can be used to compare rhythms between EEG events and identify self-similarity in the data. The Repeatability of the events can be quantified using Lempel-Zic Complexity [12, 18] and Approximate Entropy [19, 20]. Entropy can also be used as a measure of randomness with Spectral Entropy and level of disorganization using Wavelet Entropy.

The Hjorth parameters analysis [21, 22] includes activity (variance of the signal of a time function), mobility (proportion of the standard deviation), and complexity (change in frequency) as descriptors. Wavelet Energy [23, 24] can be used to examine the uniformity of different frequency bands in EEG data.

2.2.3 Frequency domain features

Short Time Fourier Transform [25], and Fractional Fourier Transform [26] are used to calculate phase and magnitude components. Spectral Power Analysis [27, 28] can be used to study different frequency bands in EEG.

2.2.4 Time-frequency domain features

The wavelet transform is a widely used time-frequency domain feature in the analysis of EEG. There are two types of wavelet transform: the Discrete Wavelet Transform [29] and the Continuous Wavelet Transform [30]. The CWT generates a scalogram through dilation and translation, while the DWT filters the signal and decomposes it into different levels.

Certain time-frequency domain features can be combined with nonlinear features to examine hidden data properties. Higher-Order Spectra [31, 32] and Variational Model Decomposition [33, 34] are among the most commonly used features in these combinations (Tables 1 and 2).

2.3 Feature selection

For most ML techniques to perform at their best, it is important to select an optimal number of features. Too many features that convey similar information about the target variable, or no information about it at all, can make the model unnecessarily complex and hinder its performance.

There are a variety of techniques that can be employed to identify the most suitable features for a particular study (Table 3). Statistical approaches such as Principal Components Analysis [36, 37] and Partial Least Squares [38] are relatively straightforward ways of reaching a conclusion. More complex techniques like Minimum Redundancy Maximum Relevance [39, 40] and Gaussian Mixture Models [41, 42] may be used in situations that require greater sensitivity, such as predicting human seizures.

However, some DL algorithms have the ability to automatically extract the most relevant information from the input variables using part of their architecture [43, 44]. Both feature extraction and selection can be bypassed in these cases and still yield competitive results.

3 Seizure detection and classification

Seizure detection using EEG has traditionally been performed manually by clinical professionals, who evaluate characteristics such as frequency, wavelength, voltage, amplitude, and waveforms. These features can be analyzed using automated learning algorithms. Automated detection of ictal events using EEG has become a crucial area of research in epilepsy since the first computer analysis of EEG records using wavelet transform in 2002 [47].

In recent years, numerous studies have been conducted to enhance the performance of automated seizure detection. A variety of ML and DL classifiers have been used [48]. Still, while detecting seizures may be helpful for research purposes, patients and clinical professionals require tools that can help prevent seizures rather than simply performing post-hoc analysis. This is where seizure prediction becomes necessary.

4 Seizure prediction

Having sufficient advance warning before a seizure occurs is a key aspect of clinical treatment. The process of seizure prediction is similar to seizure detection, but instead of classifying interictal and ictal events, it focuses on distinguishing interictal from preictal events, leaving out the actual seizures from the analysis.

Developing an appropriate stimulation protocol requires a sufficient advance warning before a seizure. However, early prediction often leads to reduced sensitivity and specificity, which can negatively impact overall performance (Table 4).

4.1 Animal models

The first experiments using wavelet functions as predictors achieved times of approximately 2.24 min and demonstrated their effectiveness [35]. The best time results in animal models were obtained by [45], which used canine EEG data and multiple ML algorithms to detect seizures 1 min in advance. While this work established a proof of concept, no diagnostic performance analyses were conducted. However, some studies have developed systems with sensitivities exceeding 90 %, although with somewhat longer prediction times. For example, Rajdev’s team (as described in [49]) developed a seizure prediction system for rat EEG recordings based on an adaptive Wiener filter, achieving a sensitivity of 92 %, which is the highest sensitivity achieved in works on animal EEG recordings.

4.2 Human subjects

The work of Iasemidis [50] established a maximum prediction time of 91 min with a precision of 91.3 % (and a sensitivity of 81.82 %). Other authors have sought to maintain a prediction time horizon within certain parameters. For example, in [12, 19], 30-min and 50-min horizons were fixed, resulting in sensitivities of 79.9 %–90.2 %.

Tsiouris [46] achieved a prediction with 15–120 min advance warning and a sensitivity of 99 %, using Long Short-term Memory networks (LSTM), which marked the first application of DL in this field.

Other DL approaches that automate feature extraction have achieved similar results. For example, Wei [43] used an image of the EEG as input to an architecture based on Convolutional Neural Networks (CNN) for feature extraction and Long Short-term Memory (LSTM) for sequence learning, achieving an average accuracy of 93.4 % with an average warning time of 21 min. Transformer models have also been used in a similar way [44], achieving a prediction sensitivity and False Positive Rate of 96.01 % and 0.047/h, respectively, with an average warning time ranging from 3 to 30 min.

5 Conclusions

Despite the significant progress made in automated seizure prediction since the first studies were conducted, there are still some challenges that need to be addressed.

To implement accurate diagnostic and treatment tools, it is necessary to achieve a balance between prediction time, sensitivity, and specificity. This balance can be achieved using DL techniques like CNNs, LSTMs, and Transformers. There have been some recent developments in this area that show promising results [43, 44, 51].

The best performance and results are often achieved by combining multiple techniques and features, but this approach can also increase computational requirements. Finding ways to effectively manage these requirements is a significant challenge in automated seizure prediction.

Finding a solution to the problem of automated seizure prediction has the potential to greatly improve the quality of life for epilepsy patients and reduce the burden on healthcare systems.


Corresponding authors: Marco Zurdo-Tabernero and Ángel Canal-Alonso, BISITE Research Group, University of Salamanca, Salamanca, Spain, E-mail: (M. Zurdo-Tabernero), (Á. Canal-Alonso) (M. Zurdo-Tabernero). (Á. Canal-Alonso).

Funding source: HERMES project

Award Identifier / Grant number: 824164

Funding source: XAI - XAI - Sistemas Inteligentes Auto Explicativos creados con Módulos de Mezcla de Expertos

Award Identifier / Grant number: SA082P20

  1. Research ethics: Not applicable.

  2. Author contributions: J.M.C. and J.P. have carried out this work's Conceptualization and Supervision. A.C. and M.H. did the Investigation and the Writing of the original draft. G.H., A.G., F.P., and S.R. took care of the Methodology. J.M.C. did the Project administration Funding acquisition. All the authors participated in the Reviews and Final Editing of the manuscript.

  3. Competing interests: The authors state no conflicts of interest.

  4. Research funding: This work was supported by the HERMES project, funded by the European Union under the Horizon 2020 FET-proactive program, Grant Agreement n. 824164., as well as the “XAI - XAI - Sistemas Inteligentes Auto Explicativos creados con Módulos de Mezcla de Expertos” project, ID SA082P20, financed by Junta Castilla y León, Consejería de Educación, and FEDER funds.

  5. Data availability: Not applicable.

References

1. GBD 2015 Disease and Injury Incidence and Prevalence Collaborators. Global, regional, and national incidence, prevalence, and years lived with disability for 310 diseases and injuries, 1990–2015: a systematic analysis for the global burden of disease study 2015. Lancet 2016;388:1545–602. https://doi.org/10.1016/S0140-6736(16)31678-6Search in Google Scholar PubMed PubMed Central

2. Hirtz, D, Thurman, DJ, Gwinn-Hardy, K, Mohamed, M, Chaudhuri, AR, Zalutsky, R. How common are the “common” neurologic disorders? Neurology 2007;68:326–37. https://doi.org/10.1212/01.wnl.0000252807.38124.a3.Search in Google Scholar PubMed

3. Robert, S, Fisher, CA, Arzimanoglou, A, Bogacz, A, Cross, JH, Elger, CE, et al.. ILAE official report: a practical clinical definition of epilepsy. Epilepsia 2014;55:475–82. https://doi.org/10.1111/epi.12550.Search in Google Scholar PubMed

4. Niedermeyer, E, da Silva, FL. Electroencephalography: basic principles, clinical applications, and related fields, 7th ed. Oxford, England: Oxford University Press; 2018.Search in Google Scholar

5. Cámpora, NE, Mininni, CJ, Kochen, S, Lew, SE. Seizure localization using pre ictal phase-amplitude coupling in intracranial electroencephalography. Sci Rep 2019;9:20022. https://doi.org/10.1038/s41598-019-56548-y.Search in Google Scholar PubMed PubMed Central

6. Trudel, MC, Jaana, M, Kitsiou, S, Paré, G. Synthesizing information systems knowledge: a typology of literature reviews. Inf Manag 2015;52:183–99. https://doi.org/10.1016/j.im.2014.08.008.Search in Google Scholar

7. University of Salamanca Library. BRUMARIO: catalog [Internet]. Salamanca: University of Salamanca; 2023. Available from: https://bibliotecas.usal.es/brumarioform.Search in Google Scholar

8. Press, WH, Flannery, BP, Teukolsky, SA, Vetterling, WT. Numerical recipes in C: the art of scientific computing. Cambridge, England: Cambridge University Press; 1992.Search in Google Scholar

9. Parvez, MZ, Paul, M. Epileptic seizure prediction by exploiting spatiotemporal relationship of EEG signals using phase correlation. IEEE Trans Neural Syst Rehabil Eng 2016;24:158–68. https://doi.org/10.1109/tnsre.2015.2458982.Search in Google Scholar PubMed

10. Reddy, BS, Chatteriji, BN. An fft-based technique for translation, rotation, and scaleinvariant image registration. IEEE Trans Image Process 1996;5:1266–71. https://doi.org/10.1109/83.506761.Search in Google Scholar PubMed

11. Ramoser, H, Mller-Gerking, J, Pfurtscheller, G. Optimal spatial filtering of single trial EEG during imagined hand movement. IEEE Trans Rehabil Eng 2000;8:441–6. https://doi.org/10.1109/86.895946.Search in Google Scholar PubMed

12. Aarabi, A, He, B. A rule-based seizure prediction method for focal neocortical epilepsy. Clin Neurophysiol 2012;123:1111–22. https://doi.org/10.1016/j.clinph.2012.01.014.Search in Google Scholar PubMed PubMed Central

13. Grassberger, P, Procaccia, I. Characterization of strange attractors. Phys Rev Lett 1983;50:346–9. https://doi.org/10.1103/physrevlett.50.346.Search in Google Scholar

14. Shafique, A, Sayeed, M, Tsakalis, K. Nonlinear dynamical systems with chaos and big data: a case study of epileptic seizure prediction and control. Guide to big data applications. New York, NY, US: Springer International Publishing; 2018.10.1007/978-3-319-53817-4_13Search in Google Scholar

15. Rosenstein, M, Colins, JJ, de Luca, CJ. A practical method for calculating largest lyapunov exponents from small data sets. Phys D 1993;65:117–34. https://doi.org/10.1016/0167-2789(93)90009-p.Search in Google Scholar

16. D’Alessandro, M, Esteller, R, Vachtsevanos, G, Hinson, A, Echauz, J, Litt, B. Epileptic seizure prediction using hybrid feature selection overmultiple intracranial EEG electrode contacts: a report of four patients. IEEE Trans Inf Theory 2003;50:603–15. https://doi.org/10.1109/tbme.2003.810706.Search in Google Scholar PubMed

17. Mandelbrot, BB. Geometry of nature. San Francisco, CA, US: Freeman; 1983.Search in Google Scholar

18. Ziv, J, Lempel, A. A universal algorithm for sequential data compression. IEEE Trans Inf Theory 1977;23:337–43. https://doi.org/10.1109/tit.1977.1055714.Search in Google Scholar

19. Zhang, Y, Zhou, W, Yuan, Q, Wu, Q. A low computation cost method for seizure prediction. Epilepsy Res 2014;108:1357–66. https://doi.org/10.1016/j.eplepsyres.2014.06.007.Search in Google Scholar PubMed

20. Richman, JS and Moorman, JR. Physiological time-series analysis using approximate entropy and sample entropy. Am J Physiol Heart Circ Physiol 2000;278:2039–49. https://doi.org/10.1152/ajpheart.2000.278.6.h2039.Search in Google Scholar

21. Direito, B, Teixeira, CA, Sales, F, Castelo-Branco, M, Dourado, A. A realistic seizure prediction study based on multiclass svm. Int J Neural Syst 2017;27:1750006–21. https://doi.org/10.1142/s012906571750006x.Search in Google Scholar

22. Hjorth, B, Elema-Schonander, AB. EEg analysis based on time domain properties. Electroencephalogr Clin Neurophysiol 1970;29:306–10. https://doi.org/10.1016/0013-4694(70)90143-4.Search in Google Scholar PubMed

23. Gadhoumi, K, Lina, J, Gotman, J. Seizure prediction in patients with mesial temporal lobe epilepsy using EEG measures of state similarity. Clin Neurophysiol 2013;124:1745–54. https://doi.org/10.1016/j.clinph.2013.04.006.Search in Google Scholar PubMed PubMed Central

24. Garg, G, Singh, V, Gupta, JRP, Mittal, AP. Relative wavelet energy as a new feature extractor for sleep classification using EEG signals. Int J Biomed Signal Process 2011;2:75–80.10.1007/s13534-012-0044-0Search in Google Scholar

25. Truong, ND, Nguyen, AD, Kuhlmann, L, Bonyadi, MR, Yang, J, Ippolito, S, et al.. Convolutional neural networks for seizure prediction using intracranial and scalp electroencephalogram. Neural Network 2018;105:104–11. https://doi.org/10.1016/j.neunet.2018.04.018.Search in Google Scholar PubMed

26. Ozaktas, HM, Zalevsky, Z, MA, K. The fractional Fourier transform. Hoboken, NJ, US: Wiley; 2001.10.23919/ECC.2001.7076127Search in Google Scholar

27. Dressler, O, Schneider, G, Stockmanns, G, Kochs, EF. Awareness and the EEG power spectrum: analysis of frequencies. Br J Anaesth 2004;93:806–9. https://doi.org/10.1093/bja/aeh270.Search in Google Scholar PubMed

28. Bandarabadi, M, Rasekhi, J, Teixeira, CA, Karami, MR, Dourado, A. On the proper selection of preictal period for seizure prediction. Epilepsy Behav 2015;45:158–66. https://doi.org/10.1016/j.yebeh.2015.03.010.Search in Google Scholar PubMed

29. Yves, M. Wavelets and operators. Cambridge, England: Cambridge University Press; 1992.Search in Google Scholar

30. Mallat, S, Hwang, WL. Singularity detection and processing with wavelets. IEEE Trans Inf Theory 1992;38:617–43. https://doi.org/10.1109/18.119727.Search in Google Scholar

31. Nikias, CL, Petropulu, AP. Higher order spectra analysis: a nonlinear signal processing framework. Englewood Cliffs, NJ, US: PTR Prentice Hall; 1993.Search in Google Scholar

32. UR Acharya, Yanti, R, Zheng, JW, Mookiah, MRK, Tan, JH, Martis, RJ, et al.. Automated diagnosis of epilepsy using CWT, HOS and texture parameters. Int J Neural Syst 2013;23:1350009. https://doi.org/10.1142/s0129065713500093.Search in Google Scholar PubMed

33. Dragomiretskiy, K, Zosso, D. Variational mode decomposition. IEEE Trans Signal Process 2014;62:531–44. https://doi.org/10.1109/tsp.2013.2288675.Search in Google Scholar

34. Kumar, MR, Rao, YS. Epileptic seizures classification in EEG signal based on semantic features and variational mode decomposition. Cluster Comput 2018;22:1–11.10.1007/s10586-018-1995-4Search in Google Scholar

35. Ouyang, G, Li, X, Guan, X. Application of wavelet-based similarity analysis to epileptic seizures prediction. Comput Biol Med 2007;37:430–7. https://doi.org/10.1016/j.compbiomed.2006.08.010.Search in Google Scholar PubMed

36. Williamson, JR, Bliss, DW, Browne, DW, Narayanan, JT. Seizure prediction using EEG spatiotemporal correlation structure. Epilepsy Behav 2012;25:230–8. https://doi.org/10.1016/j.yebeh.2012.07.007.Search in Google Scholar PubMed

37. Pearson, K. On lines and planes of closest fit to systems of points in space. Philos Mag A 1901;2:559–72. https://doi.org/10.1080/14786440109462720.Search in Google Scholar

38. AMKM, H, Kaplan, AM. A beginner’s guide to partial least squares analysis. Understand Stat 2004;3:283–97. https://doi.org/10.1207/s15328031us0304_4.Search in Google Scholar

39. Bandarabadi, M, Teixeira, CA, Rasekhi, J, Dourado, A. Epileptic seizure prediction using relative spectral power features. Clin Neurophysiol 2015;126:237–48. https://doi.org/10.1016/j.clinph.2014.05.022.Search in Google Scholar PubMed

40. Peng, HC, Long, F, Ding, C. Feature selection based on mutual information: criteria of max-dependency, max-relevance, and min-redundancy. IEEE Trans Pattern Anal Mach Intell 2005;27:1226–38. https://doi.org/10.1109/tpami.2005.159.Search in Google Scholar

41. Rasmussen, CE. The infinite Gaussian mixture model. In: Solla, SA, Leen, TK, Müller, K-R, editors. Advances in neural information processing systems 12. Cambridge, MA: MIT Press; 2000:554–60 pp.Search in Google Scholar

42. Zandi, AS, Tafreshi, R, Javidan, M, Dumont, G. Predicting epileptic seizures in scalp EEG based on a variational bayesian Gaussian mixture model of zero-crossing intervals. IEEE Trans Biomed Eng 2013;60:1401–13. https://doi.org/10.1109/tbme.2012.2237399.Search in Google Scholar PubMed

43. Wei, X, Zhou, L, Zhang, Z, Chen, Z, Zhou, Y. Early prediction of epileptic seizures using a long-term recurrent convolutional network. J Neurosci Methods 2019;327:108395. https://doi.org/10.1016/j.jneumeth.2019.108395.Search in Google Scholar PubMed

44. Yan, J, Li, J, Xu, H, Yu, Y, Xu, T. Seizure prediction based on transformer using scalp electroencephalogram. Appl Sci 2022;12:4158. https://doi.org/10.3390/app12094158.Search in Google Scholar

45. Varatharajah, Y, Iyer, RK, Berry, BM, Worrell, GA, Brinkmann, BH. Seizure forecasting and the preictal state in canine epilepsy. Nerv Syst 2017;27:1650046. https://doi.org/10.1142/s0129065716500465.Search in Google Scholar

46. Tiouris, KM, Pezoulas, V, Zervakis, M, Konitsiotis, S, Koutsouris, DD, Fotiadis, DI. A long short-term memory deep learning network for the prediction of epileptic seizures using EEG signals. Comput Biol Med 2018;99:24–37. https://doi.org/10.1016/j.compbiomed.2018.05.019.Search in Google Scholar PubMed

47. Adeli, H, Zhou, Z, Dadmehr, N. Analysis of EEG records in an epileptic patient using wavelet transform. J Neurosci Methods 2003;123:69–87. https://doi.org/10.1016/s0165-0270(02)00340-0.Search in Google Scholar PubMed

48. Lee, K, Jeong, H, Kim, S, Yang, D, Kang, H-C, Choi, E. Real-time seizure detection using eeg: a comprehensive comparison of recent approaches under a realistic setting; 2022. arXiv:2201.08780 [cs].Search in Google Scholar

49. Rajev, P, WArd, MP, Rickus, J, Worth, R, Irazoqui, PP. Real-time seizure prediction from local field potentials using an adaptive wiener algorithm. Comput Biol Med 2010;40:97–108. https://doi.org/10.1016/j.compbiomed.2009.11.006.Search in Google Scholar PubMed

50. Iasemidis, LD, Shiau, DS, Pardalos, PM, Chaovalitwongse, W, Narayanon, K, Prasad, A, et al.. Long-termprospective on-line real-time seizure prediction. Clin Neurophysiol 2005;116:532–44. https://doi.org/10.1016/j.clinph.2004.10.013.Search in Google Scholar PubMed

51. Acharya, UR, Oh, SL, Hagiwara, Y, Hojjat, A, Adeli, H. Deep convolutional neural network for the automated detection of seizure using eeg signals. Comput Biol Med 2017;100:270–8. https://doi.org/10.1016/j.compbiomed.2017.09.017.Search in Google Scholar PubMed

Received: 2023-01-17
Accepted: 2023-08-01
Published Online: 2023-12-15

© 2023 the author(s), published by De Gruyter, Berlin/Boston

This work is licensed under the Creative Commons Attribution 4.0 International License.

Downloaded on 2.3.2024 from https://www.degruyter.com/document/doi/10.1515/jib-2023-0002/html
Scroll to top button