Focal epileptic seizures anticipation based on patterns of heart rate variability parameters

https://doi.org/10.1016/j.cmpb.2019.05.032Get rights and content

Highlights

  • It investigates heart activity abnormalities manifested in the HRV parameters and provides a comprehensive analysis of their patterns during pre-ictal and ictal phases as well as the corresponding transition from one phase to the other.

  • It evaluates and ranks HRV parameters in terms of their relevance, significance and involvement in seizure anticipation and potentially prediction efficacy.

  • The proposed methodology provides a framework for selecting the most important HRV features, incorporating personalized reference information and developing a focal seizure anticipation model using only ECG signals.

Abstract

Background and Objective

Heart rate variability parameters are studied by the research community as potential valuable indices for seizure detection and anticipation. This paper investigates heart activity abnormalities during focal epileptic seizures in childhood.

Methods

Seizures affect both the sympathetic and parasympathetic system which is expressed as abnormal patterns of heart rate variability (HRV) parameters. In the present study, a clinical dataset containing 42 focal seizures in long-term electrocardiographic (ECG) recordings from drug-resistant pediatric epileptic patients (with age 8.2 ± 4.3 years) was analyzed.

Results

Results indicate that the time domain HRV parameters (heart rate, SDNN, standard deviation of heart rate, upper envelope) and spectral HRV parameters (LF/HF, normalized HF, normalized LF, total power) are significantly affected during ictal periods. The HRV features were ranked in terms of their relevance and efficacy to discriminate non-ictal/ictal periods and the top-ranked features were selected using the minimum Redundancy Maximum Relevance algorithm for further analysis. Then, a personalized anticipation algorithm based on multiple regression was introduced providing an “epileptic index” of imminent seizures. The performance of the system resulted in anticipation accuracy of 77.1% and an anticipation time of 21.8 s.

Conclusions

The results of this analysis could permit the anticipation of focal seizures only using electrocardiographic signals and the implementation of seizure anticipation strategies for a range of real-life clinical applications.

Introduction

Epilepsy is a chronic neurological disorder that affects about 50 million people worldwide [1]. In 2017, it is estimated that the prevalence and incidence of epilepsy are 6.38 and 0.67 per 1000 persons respectively [2]. Epilepsy is mainly treated with anti-epileptic drugs (AED). Approximately 65–70% of all cases can be controlled by medication, although only 15% of these achieve full seizure control without any side effects [3]. Moreover, 7–8% can be treated by surgical operation. For the remaining 25–30% of cases, the involved patients suffer from drug-resistant seizures, which cannot be controlled by any of the available treatments. For these cases, reliable seizure anticipation and prediction is of great significance for the patient's quality of life.

The terms anticipation and prediction are used interchangeably in the literature [4]. A recent review [5], extensively surveys seizure prediction methods with prediction time varying from 5 s to 90 min. In the present analysis, anticipation is considered as a short-term prediction with an uncertainty of the exact onset time [4].

EEG is by far the most widely adopted clinical technique for the diagnosis, detection, and anticipation and prediction of seizures in clinical practice [4], [5], [6], [7], [8], [9]. Although EEG is relative non-invasive and accurate, it is not always convenient for daily monitoring through wearable low-cost devices. Thus, alternative techniques for seizure anticipation and prediction are investigated. It is known that the anterior cingulate, insular, posterior orbitofrontal, and the prefrontal cortices play a significant role in the autonomic nervous system at the cortical level along with the amygdala and hypothalamus [10]. Epileptic seizures are caused due to dysregulations of the sympathetic (SNS) and parasympathetic nervous system (PNS), so heart activity which is also modulated by the autonomic nervous system (ANS) may provide an indicator of the ictal onset. Standardized HRV parameters (described in Table 1) are reliable markers for estimating various aspects of heart activity.

There are several studies investigating the long-term effects of epilepsy on heart activity and the influence of antiepileptic medication [11]. Heart activity is considered a significant factor in epilepsy as specific cardiac patterns relate to underlying mechanisms leading to Sudden Unexpected Death in EPilepsy (SUDEP), a major cause of death in epileptic patients, especially in the case of intractable epilepsy [12]. Epileptic patients present HRV changes such as increased sympathetic and decreased parasympathetic activity, e.g., increased SDNN [13], increased LF, LF/HF 24–36 h after seizure attack [13] and reduced HF [11], [13] in comparison to controls.

In addition, autonomic alterations expressed in ECG abnormalities, such as ST elevation, sinus arrhythmia, ictal tachycardias, atrioventricular (AV) block and asystole and others, occur during ictal periods [14], [15], [16]. ECG itself can also provide useful information in distinguishing episodes of asystolia from epileptic fits leading to clinical episodes originated from ictal arrhythmias [17].

Recently, there is increasing interest in investigating short-term heart activity patterns which are directly associated with seizures [18], just before (pre-ictal) or during seizures occurrence (ictal) serving for seizure anticipation and prediction. Heart rate (HR), the most straightforward and widely analyzed feature, has been reported to increase during or before seizures in the majority of studies [14], [15], [19], [20], [21], [22], [23], [24], [25]. In some cases, studies exploit heart rate increase (HRI) in the pre-ictal phase [26], [27] or nonlinear dynamics such as SD2/SD1 of Poincare plot [27]. The timing of these changes may provide useful clinical information and an additional clinical sign in determining seizures onset. In [21], Cooman proposes an automated online patient-independent seizure detection system based on differential measures related to HRI.

Most of the relevant studies either use ECG as complementary information to EEG [25], [28], or they investigate long-term HRV changes in epilepsy (i.e. without the presence of seizures) [11]. Other studies, use limited HRV parameters (e.g., only the HR or SDNN) [14], [20], [26] not assessing their significance/involvement or not in providing information about each HRV parameter pattern in pre-ictal/ictal phase [14] in order for someone to deploy this knowledge in a seizure anticipation model. There are few studies addressing the problem of seizure anticipation through only HRV analysis [20], [27], [29] and the mechanisms of HRV alterations remain limited [26].

This paper investigates heart activity abnormalities manifested in the HRV parameters and provides a comprehensive analysis of their patterns during pre-ictal and ictal phases as well as the corresponding transition from one phase to the other. In addition, it evaluates and ranks HRV parameters in terms of their relevance, significance and involvement in seizure anticipation and potentially prediction efficacy. The proposed methodology provides a framework for selecting the most important HRV features, incorporating personalized reference information and developing a focal seizure anticipation model only using ECG signals which is an important issue in epilepsy research.

Section snippets

ECG preprocessing

ECG signal was acquired by placing two Ag/AgCl electrodes in symmetric position of the chest. A typical preprocessed ECG signal acquired during this study is presented in Fig. 1. The signal was detrended and bandpass filtered in order to remove power line noise. Spikes and artifacts (due to the subject's activity like eye blinks, spikes, head movements, chewing, general discharges) were suppressed by applying dynamic threshold based on neighborhood envelope.

The R component was extracted using a

Inclusion criteria and ethics

Subjects participating in this study are patients diagnosed with non-idiopathic focal epilepsy. If a sustained seizure freedom is not ensured, then even the occurrence of at least one seizure event makes the subject eligible for inclusion in the study. There were some cases for which long-term video EEG and synchronous ECG were recorded, but when evaluated by the two neuropediatricians (see Section 3.2), they did not present any epileptic seizure. These cases were excluded from our analysis.

Results

The methodology for seizure anticipation was applied to the clinical dataset of this study. The HRV parameters were estimated from the ECG recordings using the methods described in Section 2.2. The sampling frequency of the ECG signal is fs = 256 Hz and a sliding temporal window of Δt = 30 s and step 2 s was used. The time window interval was chosen in order to be able to track HRV temporal dynamics, while it was checked that the increase of time window does not affect significantly system

Discussion

This study proposes a system for seizure anticipation based on HRV parameters on focal seizures in childhood. The results indicate significant cardiac autonomic alterations during seizures. Specifically, the HRV features HR, SDNN, HRstd, total power, LF/HF, LF, LFnorm, ECG upper envelope significantly increase and the features RR, HFnorm, pNN50 significantly decrease making them efficient features for seizure anticipation. The features were ranked in terms of their significance and relevance to

Declaration of Competing Interest

The authors declare no conflict of interest.

Acknowledgment

The authors would like to thank Spyridon Voutoufianakis, neuropediatrician, for the seizure annotation and Giorgos Livas for his valuable technical support.

Ethical approval statement

This study's protocol was approved by the appropriate scientific board of the University Hospital of Heraklion under the reference number 5631/15-5-14. Informed consent was obtained from all patients following a detailed explanation of the study objectives and protocol to each patient and/or caregiver. All caregivers/patients provided written informed consent prior to being monitored.

Funding and competing interests

All authors declare none.

References (47)

  • E. Miranda Dantas et al.

    Spectral analysis of heart rate variability with the autoregressive method: what model order to choose?

    Computers in Biology and Medicine

    (2012)
  • J. Żygierewicz et al.

    Computationally efficient approaches to calculating significant ERD/ERS changes in the time-frequency plane

    Journal of Neuroscience Methods

    (2005)
  • P. Geladi et al.

    Partial least-squares regression: a tutorial

    Analytica chimica acta

    (1986)
  • Z. Piotrowski et al.

    Classification of falling asleep states using HRV analysis

    Biocybernetics and Biomedical Engineering

    (2017)
  • N. Gosselin et al.

    Age difference in heart rate changes associated with micro-arousals in humans

    Clinical Neurophysiology

    (2002)
  • W.D. Dupont et al.

    Power and sample size calculations: a review and computer program

    Controlled Clinical Trials

    (1990)
  • World Health Organization (WHO)Report, fact sheet No 999, February...
  • K.M. Fiest et al.

    Prevalence and incidence of epilepsy A systematic review and meta-analysis of international studies

    Neurology

    (2017)
  • F. Mormann et al.

    Seizure prediction: the long and winding road

    Brain

    (2007)
  • G. Giannakakis et al.

    Methods for seizure detection and prediction: an overview, Modern Electroencephalographic Assessment Techniques

    (2015)
  • E.B. Assi et al.

    Towards accurate prediction of epileptic seizures: a review

    Biomedical Signal Processing and Control

    (2017)
  • G. Giannakakis et al.

    An approach to absence epileptic seizures detection using Approximate Entropy

  • V. Sakkalis et al.

    Absence seizure epilepsy detection using linear and nonlinear EEG analysis methods

  • Cited by (22)

    • Information fusion and multi-classifier system for miner fatigue recognition in plateau environments based on electrocardiography and electromyography signals

      2021, Computer Methods and Programs in Biomedicine
      Citation Excerpt :

      The threshold value is determined on an individual basis. To assess the expected amount of variation required for the fatigue state, a personalized threshold thr was determined using the formula thr = μ + kσ, as described in references [21,48]. Feature selection methods can be divided into three categories: filter methods, wrapper methods and embedding methods [49].

    • Heart rate variability analysis in horses for the diagnosis of arrhythmias

      2021, Veterinary Journal
      Citation Excerpt :

      Unfortunately, no such standardised guidelines exist for veterinarians, despite HRV being more widely used in recent times, both in the research setting and in clinical practice. In human medicine, HRV analysis has been utilised across a wide variety of fields, from detecting foetal distress, predicting heart failure following myocardial infarction, predicting the onset of seizures, assessing the effect of overtraining in athletes or to encourage lifestyle modification (Woo et al., 1992; Perini and Veicsteinas, 2003; Reed et al., 2005; Perkiomaki et al., 2014; Disertori et al., 2016; Singh et al., 2018; Giannakakis et al., 2019; Li et al., 2019). Traditional HRV analysis has been performed on longer-term ECG recordings, with arrhythmic and subsequent complexes being deliberately excluded during post processing (often by using software filters).

    • Seizure onset detection based on detection of changes in brain activity quantified by evolutionary game theory model

      2021, Computer Methods and Programs in Biomedicine
      Citation Excerpt :

      Of course, it is also important to keep the values for sensitivity and false alarm in an acceptable range. The use of various biological signals to detect the onset of epileptic seizures has been extensively investigated by researchers [11]. Affordability, high temporal resolution, and non-invasive nature of EEG signals have garnered great interest among epilepsy researchers in this multi-channel biological signal [12].

    • Linear and nonlinear analyses of normal and fatigue heart rate variability signals for miners in high-altitude and cold areas

      2020, Computer Methods and Programs in Biomedicine
      Citation Excerpt :

      The specific frequency bands and physiological effects of the relevant parameters are shown in Table 1. The specific derivation of SD1 and SD2 is shown in [58]. By observing the shape and parameters of the Poincaré plot, changes in the HRV of miners under normal and fatigue conditions are intuitively analyzed.

    View all citing articles on Scopus
    View full text