DOI 10.1515/bmt-2013-0139      Biomed Tech 2014; 59(4): 343–355

Diana Piper, Karin Schiecke, Lutz Leistritz, Britta Pester, Franz Benninger, Martha Feucht, Mihaela Ungureanu, Rodica Strungaru and Herbert Witte*

Synchronization analysis between heart rate variability and EEG activity before, during, and after epileptic seizure Abstract: An innovative concept for synchronization analysis between heart rate (HR) components and rhythms in EEG envelopes is represented; it applies timevariant analyses to heart rate variability (HRV) and EEG, and it was tested in children with temporal lobe epilepsy (TLE). After a removal of ocular and movement-related artifacts, EEG band activity was computed by means of the frequency-selective Hilbert transform providing envelopes of frequency bands. Synchronization between HRV and EEG envelopes was quantified by Morlet wavelet coherence. A surrogate data approach was adapted to test for statistical significance of time-variant coherences. Using this processing scheme, significant coherence values between a HRV low-frequency sub-band (0.08–0.12 Hz) and the EEG δ envelope (1.5–4 Hz) occurring both in the preictal and early postictal periods of a seizure can be shown. Investigations were performed for all electrodes at 20-s intervals and for selected electrode pairs (T3÷C3, T4÷C4) in a time-variant mode. Synchronization was more pronounced in the group of right hemispheric TLE patients than in the left hemispheric group. Such a group-specific augmentation of synchronization confirms the hypothesis of a right hemispheric lateralization of sympathetic cardiac control of the low-frequency HRV components. Keywords: EEG envelope; heart rate variability; synchronization; temporal lobe epilepsy; time-variant coherence. *Corresponding author: Prof. Dr. Herbert Witte, Institute of Medical Statistics, Computer Sciences and Documentation, Jena University Hospital, Friedrich Schiller University Jena, 07740 Jena, Phone: +49 3641933982, Fax: +49 3641933200, E-mail: [email protected] Diana Piper: Department of Applied Electronics and Information Engineering, Politehnica University of Bucharest, Romania; and Institute of Medical Statistics, Computer Sciences and Documentation, Jena University Hospital, Friedrich Schiller University Jena, Germany Karin Schiecke, Lutz Leistritz and Britta Pester: Institute of Medical Statistics, Computer Sciences and Documentation, Jena University Hospital, Friedrich Schiller University Jena, Germany

Franz Benninger and Martha Feucht: Epilepsy Monitoring Unit, Department of Child and Adolescent Neuropsychiatry, University Hospital Vienna, Austria Mihaela Ungureanu and Rodica Strungaru: Department of Applied Electronics and Information Engineering, Politehnica University of Bucharest, Romania

Introduction It is well-known that epileptic seizure activity influences the autonomic nervous system (ANS) in different ways. Accordingly, long-term (chronic) as well as short-term (acute) alterations of the ANS before, during, and after the seizure have been studied [34]. Changes in heart rate (HR) and HR variability (HRV) are the measures used most frequently to investigate both long-term and short-term alterations of the ANS in response to the type of epilepsy and to the evolution of the epileptic seizure [12, 20]. HRV can be considered as a mirror of neuronal influences on the cardiac pacemaker and as one of the important functions of the ANS [13]. It has been demonstrated by several studies that preictal HRV patterns alone can be beneficially used for seizure onset prediction (e.g., [16]). Results from basic research suggest that the dynamics of the HRV reactions, which are dependent on specific characteristics of the seizure, may provide more information on the organization of the ANS [13] and the mechanisms supporting ANS changes. This methodological study aims at the detection of synchronizations between HRV components and EEG activity before, during, and after a seizure in refractory temporal lobe epilepsy (TLE) patients (children and adolescents), in order to reveal functional relationships between the ANS and cortical processes. Time-variant EEG band activity is usually quantified by analyzing the envelope of the bandpass filtered EEG. The question arises: why might synchronization between HRV components and EEG envelopes be assumed? An impetus was provided by one of our previous studies, which investigated the time-evolution of HRV components in TLE patients before, during, and after the

Brought to you by | New York University Bobst Library Technical Services Authenticated Download Date | 6/7/15 7:26 PM

344      D. Piper et al.: Synchronization between heart rate variability and EEG seizure [32]. We demonstrated that a pronounced phaselocking between the low-frequency (HRV-LF) and the high-frequency (HRV-HF) component occurs in the preictal period. The HRV-LF component is closely associated with the Traube-Hering-Mayer waves (≈0.1 Hz; “Mayer waves”) of the systemic arterial blood pressure (ABP), and the HRV-HF component is known as respiratory sinus arrhythmia. Such an increased phase locking indicates that this component can be approximately described as an oscillation with a stable phase, in particular, 2 min before seizure onset and in the early postictal period (recuperation period). The increased phase locking is accompanied by a higher degree of predictability (value of the largest Lyapunov exponent decreases) and quadratic phase coupling between both components (bicoherence values increases). In a preliminary study regarding time-variant changes of the HRV components and EEG frequency bands, we found time epochs in which the EEG envelopes were characterized by a 0.1-Hz rhythmicity (e.g., of the δ and the α band [29]). Additionally, other studies have shown that such synchronizations between HRV-LF and the EEG frequency band activity exist (e.g., [28]). The following findings shaped our methodological approach and the analysis strategy: 1. Saleh et al. [31] argued that the time range before seizure onset is most favorable to investigate the mechanisms supporting ANS changes. This is because the spread of epileptic activity or seizure-related heart and circulatory alterations have not yet taken place. Therefore, the focus of our investigations is on preictal data analysis. 2. The epileptic focus of mesial temporal lobe epilepsy (mTLE) seizures is located in the limbic structures (“limbic seizure”), which are involved in the regulation of the ANS. The mTLE is the most common form of epilepsy, where the associated pathological substrate is usually hippocampal sclerosis [10]. The mTLE also appears to be one of the most medically refractory forms of human epilepsy [10]. Leutmezer et  al. [19] showed that the HR increases, which occur during the ictal period, are more pronounced in patients with mTLE compared with other TLE and epilepsy types. Therefore it would be interesting to investigate synchronization effects with a particular focus on mTLE. 3. It has also been shown that the HR is significantly increased in the preictal period (preictal tachycardia), when the focus of the TLE is in the right hemisphere [31]. In contrast, no statistically significant changes could be observed in a left-focus group. The authors stated that a right hemispheric lateralization of the sympathetic cardiac control can be assumed. These

results were confirmed for children with TLE. Mayer et al. [21] detected “significant differences in HR evolution depending on location and side of seizure onset”. They found that an early and high HR increase was primarily associated with right hemispheric mTLEs. Consequently, our strategy comprises a comparative analysis of two groups; one with left hemispheric mTLE (“left-focus group”) and the other with right hemispheric mTLE (“right-focus group”). 4. Time-frequency techniques are most appropriate to analyze acute changes in HRV and EEG [39]. Consequently, relationships between HRV and EEG should also be investigated by time-variant and frequencyselective approaches. The expected synchronization effect must be seen as a physiological epiphenomenon because HRV-generating and EEG-generating structures cannot interact with each other in a causal relationship. Therefore, we used time-variant coherence, as coherence is an established correlative measure for the detection and quantification of synchronization effects [40]. 5. As already mentioned, results from one of our recent studies [32] showed that premonitory information on imminent seizure onset can be derived from HRVLF, which is associated with the Mayer waves in ABP. Therefore, we focused our study on HRV-LF. The EEG of the interictal and preictal periods in TLE children is typically characterized by temporal spike or sharp-wave discharges and temporal intermittent rhythmic δ activity [24]. Consequently, the focus is on EEG δ activity. Our methodological study provides a new analysis strategy with the possibility of expanded application, taking into account all such previous findings from HRV and EEG analysis in TLE patients.

Subjects and methods Subjects The data were recorded during presurgical evaluation of the patients at the Vienna pediatric epilepsy center following a standard protocol as described by Mayer et  al. [21]. From the group of 20 patients, only those were selected who had at least one seizure with a recording time of 10 min (at least 5 min before and 5 min after seizure onset). Seizure onset and termination in the EEG was determined independently by two experienced neurologists. Four children were added to the patient group

Brought to you by | New York University Bobst Library Technical Services Authenticated Download Date | 6/7/15 7:26 PM

D. Piper et al.: Synchronization between heart rate variability and EEG      345

to achieve comparable numbers of male/female patients and right/left focused seizures as in [21]. The resulting group of 18 children (median age 9 years 4 months, range 6  years 6  months to 18  years 0 months; median seizure length 88 s, range 52–177 s) was subdivided into a group with left (n = 9) and a group with right hemispheric TLE (n = 9) (called left-focus and right-focus group). Patients’ demographic data and relevant information about the seizures of both subgroups are given in Table 1.

Data acquisition and preprocessing The extended 10–20 systems with additional temporal electrodes for the EEG (23 channels) and one channel for the ECG recording were used. All signals were recorded against the reference electrode CPz and filtered (1–70 Hz), before they were digitized by an analog-digital converter (sampling frequency 256 Hz, 12 bit) for further data analysis. A commercially available video-EEG monitoring system was used for data acquisition and off-line data processing (Phoenix EEG Monitoring System; EMS Co., Korneuburg, Austria). The first preprocessing step for the EEG was a downsampling to 64 Hz, using a low-pass IIR filter, Chebyshev type 1 order eight, that is applied in forward and reverse directions to eliminate phase delay. The filtering was followed by an artifact removal procedure, using independent component analysis (ICA) provided by the Field Trip toolbox [25]. Thereafter, a referencing of the EEG to an average reference montage was performed.

QRS detection was performed after digital bandpass filtering (10–50 Hz) of the ECG and interpolation by cubic splines (interpolated sampling frequency 1024 Hz) to detect the time of the maximum amplitude of each R-wave, and the resulting series of events was used for the HR computation, i.e., this series of events was low-pass filtered by means of a FFT-filter (cutoff frequency  ≤  half of the mean HR). The procedure is known as French-Holden algorithm [11], which leads to the low-pass filtered event series (LPFES), a standard HRV representation in the time domain. Theoretically, the LPFES is the result of an exact demodulation of the pulse-frequency modulated series of events (QRS) [6]. In contrast, the instantaneous heart rate (IHR) representation is an approximation of the demodulation. The superiority of the LPFES (vs. IHR) for the investigation of rhythmic HRV components was shown by Milde et al. [23]. The final HRV representation was obtained via multiplication of the LPFES with the sampling rate and with 60 beats per minute (bpm) and downsampled to 8 Hz. An artifact rejection was performed manually to minimize the influence of false QRS triggering.

Methods Frequency-selective Hilbert transform: The frequencyselective Hilbert transform of a signal x(t) can be calculated with the help of the Fourier transform [39]:

H [ x ]( t ) = F −1 [ −i⋅sign( f ) ⋅ BP( f ) ⋅ F [ x ]( f )]( t )



(1)

Table 1 Patients’ demographic data and relevant information on the seizures. Left-focus group Pat ID

 

Age  (y/m)

Gender   (m/f)

Local.   (M/L)

3 8 9 11 16 18 20 21 22

                 

6/11  17/7  9/4  10/0  6/6  9/5  8/2  9/5  11/7 

m m f f m f f m m

M L M M M L M M M

n = 9

  9/5    6/6    17/7 

                 

Median Min Max

Duration (in [s])

   

Right-focus group Pat ID

 

Age  (y/m)

Gender   (m/f)

Local   (M/L)

                 

8/7  12/1  13/4  12/4  8/7  7/8  9/4  6/10  18/0 

f m f f f f f m f

M M M L M M M M M

                 

74  52  89  74  72  177  70  111  100 

2 5 6 7 12 13 15 23 24

     

74  52  177 

n = 9

  9/4    6/10    18/0 

                 

Median Min Max

Duration (in [s])

                 

155 87 72 90 58 94 80 119 110

     

90 58 155

Relevant information divided according to left or right hemispheric seizure (18 children, n = 9 in each subgroup of patients). ID of patient, age (in years/months), gender (m, male; f, female), localization (M, mesial; L, Lateral) and duration of seizure (in [s]) are given for each patient.

Brought to you by | New York University Bobst Library Technical Services Authenticated Download Date | 6/7/15 7:26 PM

346      D. Piper et al.: Synchronization between heart rate variability and EEG where F–1 is the inverse of the Fourier transform, i is the imaginary unit, sign(f) is the signum function −1 for f 0 



(2) 

and BP(f) is a band-pass operator by which a frequency band in the frequency domain was selected before the inverse FFT was carried out. The Hilbert transform of x(t) is the imaginary part of a band-related analytic signal a x( t )BP with HT a HT



x( t )BP = x( t )BP + i⋅ H [ x ]( t )

(3)



where x(t)BP is the corresponding real part, which can be computed by an identical band-pass filtering (by using BP(f), FFT filter) of the signal x(t). The frequency bands 1.5–4 Hz (sub-δ), 4–8 Hz (θ), 8–13 Hz (α), and 13–18 Hz (β) were selected by using the frequency-selective HT implementation described by Witte et al. [41]. The envelope of a band-pass filtered signal is described by env( t ) = x 2 ( t )BP + H 2 [ x ] ( t ) .



(4)



Morlet wavelet transform: The wavelet transform uses a base waveform (window) called mother wavelet from which all other filters are obtained by scaling it. The preferred method for the time-scale (multifrequency) analysis is the continuous Morlet wavelet transform (CMWT). The CMWT can be formulated as in [38]. The mother wavelet of the CMWT is the complex-valued function ψ( t ) =

 ω2    t2  1  exp( iω0 t ) − exp  − 0    exp  −  .  4  2  2  π

(5) 

Note that the DC correction can be omitted for reasonably large ω0. The complex analytic signal for the scale s can be obtained by linear convolution



a  CMWT

x( t , s ) =

1

∞

 τ −t  d τ. s 

∫ x( τ ) ψ  s −∞

*



(6)

Here, * denotes the complex conjugate. Owing to the application in EEG analysis, a frequency-based notation is ω preferable. We obtain with  f = 0 2 πs    τ −t  ∞ 2 π f a  d τ. x( t , f ) = x ( τ ) ψ*  (7) CMWT ω0 ∫−∞  ω0   2 πf  

We adapted the tuning parameter ω0 = 2π to match the short-time characteristics of the signal, which ensures a minimization of artifacts due to inappropriate time-frequency resolution (signal-adapted CMWT). The time-variant power spectrum was computed by S( t , f ) =| CMWTa x( t , f )| 2 .



(8)



The time-frequency resolution of the CMWT is frequency-dependent. Higher frequencies lead to a better time resolution (TR) but also to worse frequency resolution (FR). The time-variant power spectra and coherences were calculated at first for the frequency interval 0–0.5 Hz. The frequency range 0.08–0.12 Hz was chosen for further computations of the time-variant coherence. We used the standard deviation of the Gauss envelope in the time domain and the standard deviation of the Gauss curve in the frequency domain as measure for the TR and FR [39]. The following TRs and FRs result for the frequencies we are most interested in: at 0.3 Hz (TR = 5 s, FR = 0.2 Hz) and at 0.1 Hz (TR = 15 s, FR = 0.06 Hz). Coherence: For the time-variant spectrum and coherence computation, the CMWT was applied. Coherence is calculated with the help of the time-variant CWMT cross spectrum SHRV/envelope (t, f) between HRV and the EEG envelope as well as of their time-variant spectra SHRV(t, f) and Senvelope (t, f): | CHRV / envelope |( t , f ) =



| SHRV / envelope ( t , f )| 2

SHRV ( t , f ) ⋅Senvelope ( t , f )

(9) 

with

SHRV / envelope ( t , f ) = CMWTa x HRV ( t , f ) * ⋅ CMWTa xenvelope ( t , f )



(10)

by using the CWMT-related analytic signals according to equation 7 (the superscript * is the complex conjugate). In order to compute the time-variant coherence, the envelopes were downsampled to 8  Hz to match the HRV’s sampling frequency, and a time smoothing of the cross-spectrum and of the two spectra was carried out to obtain an appropriate estimation. We used rectangular time windows with a length of 128 (16 s), 256 (32 s), and 512 (64 s) time points. The 128-point window provides a TR for HRV-LF, which is similar to that provided by CMWT. The best compromise between a sufficient smoothing (estimation properties) and a satisfactory TR was achieved by using the window with 512 points. The FR remains (≈0.06  Hz for 0.1 Hz, see above). The mean representing the frequency band 0.08–0.12  Hz was calculated for each time point of the time-variant coherence

Brought to you by | New York University Bobst Library Technical Services Authenticated Download Date | 6/7/15 7:26 PM

D. Piper et al.: Synchronization between heart rate variability and EEG      347

to obtain the time course of the band coherence. For the topographic analysis, the mean value of the time-variant coherence in a 20-s analysis interval was computed (mean over time). Statistics: The statistical hypothesis testing for coherence analysis was performed by using surrogate data. The null hypothesis is that there is no coupling (synchronization) between EEG envelopes and HRV. The surrogate data were obtained by destroying the phase information for both signals by means of phase randomization [36]. This was carried out for 1000 repetitions, and the 95th (90th) percentile of the surrogate time-variant coherence was computed. The mean over time of the 95th (90th) percentile was set as the “5% (10%) threshold” for statistically significant coherence values. For the topographic analysis, the mean value of the time-variant coherence in a 20-s analysis interval was compared with the 5% threshold to obtain the significant values for each electrode.

Processing concept The processing scheme used is represented in Figure 1. The first step of the EEG processing was the artifact removal by means of an ICA approach in order to reduce the influence of ocular and movement artifacts [18]. Thereafter, a re-referencing of the recordings to an average reference montage was performed. For analysis 20 channels (Figure 2) were used. Subsequently, the envelopes of

the frequency bands 1.5–4 Hz (sub-δ), 4–8 Hz (θ), 8–13 Hz (α), and 13–18  Hz (β) were computed by means of the frequency-selective Hilbert transform (equations 1–4). Each of these envelopes was used as one input signal for time-variant coherence analysis (equation 9). The HRV (LPFES representation) serves as the second input. The δ sub-band 1.5–4  Hz in combination with the restricted HRV-LF band (0.08–0.12 Hz) showed the best results and was chosen for this study. In Figure 2, examples of both EEG and HRV signals are shown for the whole analysis interval (600 s). Twenty EEG channels were used for the analysis (montage, Figure 2 right side). The EEG signal (A) of the channel T3 of one patient (ID 9) is shown. At 300 s, the seizure onset is localized. The average reference signal is depicted as overlay (gray) in Figure 2A and separately in Figure 2B. The average reference signal contains signal components, which simultaneously occur in all signals. In Figure 2C, the δ band activity (1.5–4 Hz) and in Figure 2D, the corresponding envelope are depicted. The HRV is represented below (Figure 2E). Both the envelope and the HRV rise immediately after seizure onset (300 s). The seizure lasts approximately 60 s (the median duration for all patients in both groups is 88 s). Approximately 50 s after onset, the HRV decreases toward the preictal mean HRV value. This is a typical ictal tachycardia, which can be observed in all patients of both groups. The postictal δ activity level remains higher than the level in the preictal period.

Figure 1 Processing scheme used for the synchronization analysis between HRV and EEG envelopes.

Brought to you by | New York University Bobst Library Technical Services Authenticated Download Date | 6/7/15 7:26 PM

348      D. Piper et al.: Synchronization between heart rate variability and EEG

Figure 2 Examples for recorded and processed signals. (A) The EEG at electrode T3 with the overlay (gray) of average reference activity, (B) average reference activity, (C) activity of the δ sub-band 1.5–4 Hz, (D) envelope of the δ sub-band and (E) HRV (linear trend was subtracted), all given for one child (ID 9). Additionally, overview of recorded EEG at all electrode positions is depicted for the same child.

Results Time-variant HRV analysis First, the HRV data of both groups were investigated by means of time-frequency spectral analysis (CMWT, equations 5–8). This initial analysis step aimed at the identification of HRV patterns, in particular, the timing and grouping of HRV-related Mayer waves in connection with corresponding characteristics of the respiratory sinus arrhythmia (HRV-HF). In a recent HRV study [32] (in which the patients were not divided according to left- or right-sided seizure) a clear separation of the Mayer waverelated HRV-LF (around 0.1 Hz) and the HRV-HF range (between 0.25 and 0.4 Hz) before seizure onset (300 s) was shown. The results of the time-variant HRV power spectrum analysis including all patients (n = 18) is represented in Figure 3A. The HRV-HF exists until and collapses with the onset of the seizure (when normal breathing changes [14]). Approximately 90 s after seizure onset, the HF range recur with strong power disturbances (390–480 s) and becomes less pronounced at the end of the analysis interval (500–600 s). Transient clusters of HRV-LF component can be observed in particular during the preictal period. For this study, the grand mean analysis for both subgroups (Figure 3B right-focus group, Figure 3C left-focus group) was repeated. It can be demonstrated that all the findings

Figure 3 Results of the time-variant HRV power spectrum analysis. (A) Grand mean over all 18 children, (B) mean of the right-focus group, and (C) mean of the left-focus group are displayed. The white horizontal rectangular frame designates the LF range (0.08–0.12 Hz), and the white elliptic frames indicate clusters of HRV-related Mayer waves. Time-frequency representations of power spectrum are given [color bar in (bpm2)). The marks 1–4 designate time intervals shown in Figure 6 in which significant coherence (HRV-LF vs. δ envelope) ranges occur.

Brought to you by | New York University Bobst Library Technical Services Authenticated Download Date | 6/7/15 7:26 PM

D. Piper et al.: Synchronization between heart rate variability and EEG      349

given above can be confirmed for the right-focus group. Moreover, the effects in the right-focus group are more enhanced (Figure 3B), but almost vanish in the left-focus group (Figure 3C). In Figure 3, the time-frequency ranges are indicated by white elliptic frames for which time intervals with significant coherences between HRV-LF and δ envelope (Figure 6) exist.

Results of the topographic analysis A time-variant topographic analysis was performed in order to obtain a rough overview of the time evolution, variability (stability), and laterality of topographic synchronization patterns. The mean coherence values between the HRV-LF and δ envelope were calculated for disjunct 20-s intervals for each electrode (mean over time and frequency). The mean coherence values are represented at the location of the corresponding electrode, thus, a coherence map sequence for each patient results. The group-related coherence map sequences result from an averaging of the patient-related sequences (i.e., mean over patients), which are represented in Figure 4. In addition, particular electrodes that can optimally describe these pattern characteristics were identified. For these electrodes, a more detailed analysis (higher TR) was carried out (Section Results for selected electrodes). It can be shown that the topographic coherence patterns change “continuously”, i.e., the transition from one to another pattern does not occur abruptly. Smearing effects caused by the necessary time smoothing of the single-trial time-variant coherence estimation contribute to the pattern evolution. However, the achieved TR is satisfactory and appropriate to the occurrence of HRV-related

Mayer wave clusters, which were analyzed with a fourfold higher TR. The synchronization patterns are most pronounced 1 min before seizure onset (interval 240–260 s). After onset, the topographic distributions tend toward the respective focus hemisphere, in particular, in the left-focus group. The right-focus group (Figure 4B) shows patterns with higher coherence values than the left-focus group (Figure 4A). This is particularly true for the preictal period. For the left-focus group, a stable pattern evolves after seizure onset in the left hemisphere. Starting from electrode C3, the pattern involves P3 and the neighboring central electrodes. It must be noted that C3 also shows high coherence values in the preictal period (maximum between 200 and 220 s), which decrease toward seizure onset. Each epileptic seizure is an extremely individual event (severity, focus localization, activity spreading, etc.). Therefore, on the one hand, the continuous evolution of the averaged topographic patterns indicates uniform (systematic) effects; on the other hand, blurring effects caused by individual variations cannot be excluded. It should be noted that our groups do not include only TLE patients with a mesial focus (Table 1, left-focus group: two patients with a lateral focus; right-focus group: one patient with a lateral focus). Therefore, two representative analysis results derived from one patient of each group should demonstrate that our processing concept enables an individual analysis. The results for each patient are depicted in Figure 5. These cases clearly show that (1) topographically extended areas of significant coherence occur 1 min before seizure onset in the patient with a righthemispheric mTLE, (2) immediately after seizure onset, no electrode shows any significant coherence values, and (3) a lateralization of locally circumscribed patterns (C3

Figure 4 Results of group-related mean coherence (HRV-LF vs. δ envelope) topography for subsequent 20-s intervals before and shortly after the seizure onset (300 s, red arrow). (A) The left-focus group (rectangular white frame = T3÷C3) and (B) the right-focus group (rectangular white frame = T4÷C4) are designated.

Brought to you by | New York University Bobst Library Technical Services Authenticated Download Date | 6/7/15 7:26 PM

350      D. Piper et al.: Synchronization between heart rate variability and EEG for the patient with the left-hemispheric and T4 for the patient with the right-hemispheric TLE) can be obtained at the end of the seizure and in the early postictal period (Figure 6, designations ). The electrode pairs T3÷C3 and T4÷C4 have been identified to describe the patterns’ characteristics best (temporal and topographic dynamics). In addition, it can be expected that T3 and T4 pick up the neuronal activity from the temporal lobe. The electrodes are designated by a frame (Figures 4 and 5).

Results for selected electrodes In Figure 6, the analysis results of both groups for the electrodes T3 and T4 are represented. For the right-focus group, the electrode T4 is on the focus side and T3 on the opposite side and, for the left-focus group, vice versa. In the right-focus group, approximately 100  s before the seizure onset, an increase of the coherence course at T4 occurs, which exceeds the 5% threshold for approximately 30  s (designation in Figure 6, covers the time segment 220–260 s). This duration agrees with the duration of a HRV-related Mayer wave cluster – approximately three Mayer waves – which can be seen in the time-variant HRV spectrum (designation in Figure 3B). In addition, the 10% threshold is exceeded in a time interval

before (140–170 s), i.e., between 140 s and 260 s, a strong coupling between the HRV-LF range and the δ envelope exists for the group data (n = 9). At the opposite electrode (T3), such an increase in the preictal period can also be observed, but the 5% threshold is only exceeded for some seconds. The corresponding points at which the thresholds exceeded the 10% threshold ( > 10 s, i.e., one period of a Mayer wave) are in the time range between 170 and 190 s. Additionally, a strong coupling between HRV-LF and the δ envelope can be observed in the postictal period at T3 (designation in Figure 6) at a time interval around 450 s. In this period, the HRV-LF amplitude is high (designation in Figure 3B). For the left-focus group, only a short over-crossing of the threshold in the preictal period can be observed (between 130 and 135 s). This result agrees with a reduced number of HRV-related Mayer waves (Figure 3C) in comparison with the preictal period of the right-focus group (Figure 3B). It was interesting to note that the left-focus group substantially exceeded the 5% threshold in the postictal period at both electrode sites (designations and ). High coherence values occur approximately 200 s after the seizure onset, and this event lasts approximately 100 s. These significant coherence values correlate to the occurrence of LF activity in the HRV (augmented occurrence of HRV-related Mayer waves in Figure 3C, designation ).

Figure 5 Results of coherence topography for subsequent 20-s intervals for two representative patients, one for each group. (A) Topographic coherence maps between HRV-LF and δ envelope and (B) significant coherence at electrodes (red designation) are given for one left-focus group member (ID 9, rectangular white frame = T3÷C3) and one right-focus group member (ID 2, rectangular white frame = T4÷C4). The time of the seizure onset is designated by a red arrow.

Brought to you by | New York University Bobst Library Technical Services Authenticated Download Date | 6/7/15 7:26 PM

D. Piper et al.: Synchronization between heart rate variability and EEG      351

Figure 6 Results of a group-related time-variant coherence analysis (HRV-LF vs. δ envelope) for the electrode sites T3 and T4 (abscissa in [s]). The dashed vertical line designates the seizure onset at 300 s. The red horizontal line shows the 5% threshold for the detection ­significant coherence values, and the green horizontal line shows the 10% threshold, accordingly. The gray rectangular frames represent the time ranges in which coherence exceeds the 10% threshold. The marks 1–4 designate time ranges in which coherence exceeds the 5% threshold longer than 10 s.

As mentioned above, such a postictal coherence “peak” (maximum 150 s after the onset, designation in Figure 6) can also be detected in the right-focus group, but only at the electrode on the non-focus side. HRV-related Mayer waves occur in the time range of this “peak” (Figure 3B). At the electrode C3, the highest coherence values (not illustrated) of all electrodes can be observed in the leftfocus group (C3 is at the focus side), where “peaks” before (maximum around 250 s) and immediately after the onset (between 350 and 400 s) exists. The minimum between both peaks is located at the seizure onset. Around the peaks, HRV-LF activity can also be observed, which is long-lasting before the onset.

Summary of the results The results of the coherence analysis can be summarized as follows:

– During the preictal and the postictal period, longer epochs exist, which are designated by significant coherence values between HRV-LF and the δ envelope. The HRV-related Mayer waves are pronounced during these epochs, i.e., the HRV-LF band shows clusters of high-amplitude events (waves). – The topographic analysis shows high coherence values at C3, T3 and C4, T4 in particular before and after the seizure. The coherence at T3 and T4 decreases during the seizure in both groups. The HRV analysis shows that during the seizure, the LF band is characterized by small amplitude values, and the HRV’s rhythmicity changes toward lower frequencies. – The averaged coherence curves as well as the topographic analysis of the preictal period show that the group with the right-hemispheric TLE is characterized by higher coherence values. The presence of HRVrelated Mayer waves (HRV-LF) is also massively pronounced for the right-focus group.

Brought to you by | New York University Bobst Library Technical Services Authenticated Download Date | 6/7/15 7:26 PM

352      D. Piper et al.: Synchronization between heart rate variability and EEG

Discussion Comparison with the results of other HRV-EEG coupling studies It is shown that our methodological approach allows the detection of synchronizations between HRV-LF rhythms and rhythmical EEG activity changes. To our knowledge, there has been no study, thus far, which has investigated the synchronization or the correlative coupling between HRV components and EEG activity in epileptic patients. The findings that preictal and postictal EEG δ activity is rhythmically modulated and that the modulation rhythm is correlated with the HRV-related Mayer waves (HRV-LF) are new. The time-variant coherence analysis is a linear time-frequency technique, which detects transient correlative relationships between signals (e.g., synchronizations). The coherence is amplitude independent; however, its values depend on the signal-to-noise-ratio (SNR) of both signals (low SNR causes low coherence) [4], i.e., for HRV and EEG envelopes. Therefore, a statistically defined threshold detecting significant coherence values is inevitable. It can be safely assumed that significant coherence values indicate synchronization between HVR and EEG activity (envelope). There are only a few studies investigating the coupling between HRV-LF and EEG activity modulations. During the quiet sleep of preterm neonates, the EEG alternates between a high-voltage burst discharge and a suppressed interburst activity (“tracé discontinue” EEG pattern), where a HR acceleration is coupled to the burst onset. We found that the higher the burst amplitude ( >  60 μV), the more pronounced is the HR change [33]. It should be noted here that preterm neonates have a deficit in ANS activity and a sympathetic-parasympathetic imbalance characterized by sympathetic predominance [17]. The synchronous changes of EEG and HR we have discussed are an indication for a coupling between cortical, thalamocortical, and central autonomic brain areas. Such a coupling between HR and EEG during the burst phases in anesthetized patients (burst-suppression patterns) has also been found [43]. Pfurtscheller et  al. [28] recently demonstrated a coupling over alternating epochs (duration approximately 100 s) between prefrontal oxyhemoglobin rhythms (0.07–0.13 Hz) and central EEG α and/or β envelopes in the resting brain. In two subjects, they found that oxyhemoglobin and EEG envelopes (β band) were approximately in-phase with ABP oscillations with an extremely high coupling between ABP and oxyhemoglobin rhythms. Roche-Labarbe et al.

[30] demonstrated that EEG bursts (quiet sleep period) in preterm neonates are accompanied by a transient stereotyped hemodynamic response involving a decrease in the oxyhemoglobin concentration followed by an increase. In sleep research, further studies exist, which show a temporal correlation between frequency band activity and HRV, e.g., during paradoxical sleep (between HRV and δ-θ bands [27]). Jurysta et al. [15] demonstrated that a closed connection between cardiac autonomic activity and spectral EEG bands exists. The δ band shows the highest variations in response to HRV-HF variability, and ANS activity precedes changes in the EEG during sleep in healthy young men. These results from the literature demonstrate that a correlation between HRV characteristics and EEG activity may occur in extreme physiological situations.

Physiological mechanisms Mayer waves in systemic ABP are strongly correlated with the oscillations of efferent sympathetic nervous activity, and the baroreflex plays a major role in the generation of Mayer waves. In contrast, the Mayer waveassociated HRV-LF component includes most probably both sympathetic and parasympathetic (vagal) influences [5]. A strong correlation between HR and pressure variations in the 0.1-Hz frequency range was shown [7]. In TLE patients, the baroreflex function is chronically impaired, e.g., the LF transfer function gain between ABP and HRV, which determines the baroreflex function [3, 9], is reduced. Other studies have shown that TLE patients are characterized by a dysfunction of the cardiovascular autonomic regulation (autonomic instability [13]), manifested as impaired HR responses to certain stimuli [1]. Acute HRV changes in TLE patients occur due to a chronic dysfunction in cardiovascular autonomic regulation, i.e., this dysfunction might enhance change in the organization of the Mayer waves in the preictal period. The cause of such a HRV-LF augmentation and, in particular, of the therewith associated synchronization between HRV-LF and the δ envelope must be associated with acute neuronal and non-neuronal brain processes, which evolve some minutes before the onset of the seizure and which cannot be detected by scalp EEG. It was recently shown that focal hemodynamic changes (cerebral blood flow (CBF) increases, and hemoglobin oxygenation decreases) precede seizure onset (humans and animals) by approximately 20 s [26, 44]. These changes can be measured (optical imaging)

Brought to you by | New York University Bobst Library Technical Services Authenticated Download Date | 6/7/15 7:26 PM

D. Piper et al.: Synchronization between heart rate variability and EEG      353

at the focus of the seizure. The etiology of the increase of CBF before seizure onset is unknown [26]. The authors hypothesize that these preictal CBF changes could be elicited by subtle neuronal or glial events, astrocyte- or pericyte-medicated signaling or local potassium, and local neurotransmitter/neuropeptide release. If such preseizure processes could influence the autonomic centers within the central nervous system, synchronization or resonance phenomena could result. CBF and its sympathetic regulation might play an important role. It is known that Mayer waves in systemic ABP create variations in CBF velocity in the intracranial arteries of the same frequency [35]. This establishes a link between ABP, HRV-LF, and CBF, which is connected with neuronal activity via neurovascular coupling [22]. Neurovascular coupling describes the relationship among neuronal activity, metabolism, tissue oxygenation, and CBF.

The influence of the focus side Jansen and Lagae [13] noted that “due to the hemispheric specific organization of the central autonomic nervous system, autonomic symptoms in epileptic seizures can provide lateralizing and localizing information”. Our results for the right-focus group confirm that the synchronization effects between HRV-LF and the δ envelope depends on the lateralization of the seizure. Additionally, we have found synchronizations between the HRV-LF and δ envelope in the early postictal period, i.e., immediately after the end of the epileptic seizure. This postictal synchronization was more pronounced in the left-focus group. Severe postictal disturbances (dysregulations) of the ANS over a time range of 5–6 h are described by Toth et al. [37] (HRV analysis). Therefore, it is not surprising that we found stronger synchronizations in both groups in comparison to those of the preictal period. The mechanisms discussed above might also contribute to postictal synchronization effects.

Outlook Several studies have suggested that the left hemisphere modulates the parasympathetic (vagal) tone. Accordingly, it would be interesting to investigate synchronization between EEG frequency band activity and HRV-HF (respiratory sinus arrhythmia). It would also be of interest to incorporate ABP and respiratory movements into the analysis. Subsequent analyses should include a cogent focus on the interictal period to investigate the “spontaneous” longterm organization of the HRV-related Mayer waves as well as their synchronization to EEG activity. However, such investigations require long-term recordings and monitoring of cardiovascular-cardiorespiratory parameters and the EEG. Our processing concept can be adapted to such requirements, e.g., the interval-based HT can be replaced by narrow-band Hilbert filters [2] and a filter bank-based CWMT implementation can be used for coherence computation. The threshold can be determined, for example, by a supervised classifier on the basis of representative training data [8]. Recordings from subdural and depth EEG electrodes in order to capture local cortical activity can be used. Most methods to determine seizure prediction use intracranial EEG recordings due to their higher fidelity in comparison to scalp EEG [42]. The reliability of our results, under conditions as described above and with a broadening of the methodological scope, will still require more intensive basic research before possible utilization in clinical settings to aid in the prediction of seizures. Importantly, our results confirm those of other studies and provide a deeper insight into the time-variant organization of interactions between the ANS and cortical processes. Acknowledgment: This work was supported by the DFG under Wi 1166/12-1 and by the Romanian Ministry of Labour, Family and Social Protection through the Financial Agreement POSDRU/107/1.5/S/76903 (D. Piper). Received December 17, 2013; accepted February 28, 2014; online first April 1, 2014

References [1] Ansakorpi H, Korpelainen JT, Suominen K, Tolonen U, Myllyla VV, Isojarvi JIT. Interictal cardiovascular autonomic responses in patients with temporal lobe epilepsy. Epilepsia 2000; 41: 42–47. [2] Arnold M, Doering A, Witte H, Dorschel J, Eisel M. Use of adaptive Hilbert transformation for eeg segmentation and calculation of instantaneous respiration rate in neonates. J Clin Monit 1996; 12: 43–60.

[3] Barbieri R, Bianchi AM, Triedman JK, Mainardi LT, Cerutti S, Saul JP. Model dependency of multivariate autoregressive spectral analyses. IEEE Eng Med Biol Mag 1997; 16: 74–85. [4] Bendat JS, Piersol AG. Random data: analysis and measurement procedures. 3rd ed. New York: John Wiley & Sons, Inc., 2000.

Brought to you by | New York University Bobst Library Technical Services Authenticated Download Date | 6/7/15 7:26 PM

354      D. Piper et al.: Synchronization between heart rate variability and EEG [5] Camm AJ, Malik M, Bigger JT, et al. Heart rate variability. Standards of measurement, physiological interpretation, and clinical use. Eur Heart J 1996; 17: 354–381. [6] Deboer RW, Karemaker JM, Strackee J. Comparing spectra of a series of point events particularly for heart-rate-variability data. IEEE Trans Biomed Eng 1984; 31: 384–387. [7] Deboer RW, Karemaker JM, Strackee J. Relationships between short-term blood-pressure fluctuations and heart-ratevariability in resting subjects. 1. A spectral-analysis approach. Med Biol Eng Comput 1985; 23: 352–358. [8] Doering A, Jager H, Witte H, et al. Adaptable preprocessing units and neural classification for the segmentation of EEG signals. Methods Inf Med 1999; 38: 214–224. [9] Dutsch M, Hilz MJ, Devinsky O. Impaired baroreflex function in temporal lobe epilepsy. J Neurol 2006; 253: 1300– 1308. [10] Engel J. Mesial temporal lobe epilepsy: what have we learned? Neuroscientist 2011; 7: 340–352. [11] French AS, Holden AV. Alias-free sampling of neuronal spike trains. Kybernetik 1971; 8: 165–171. [12] Harnod T, Yang CCH, Hsin YL, Wang PJ, Shieh KR, Kuo TBJ. Heart rate variability in patients with frontal lobe epilepsy. Seizure-Eur J Epilep 2009; 18: 21–25. [13] Jansen K, Lagae L. Cardiac changes in epilepsy. Seizure-Eur J Epilep 2010; 19: 455–460. [14] Jansen K, Varon C, Van Huffel S, Lagae L. Ictal and interictal respiratory changes in temporal lobe and absence epilepsy in childhood. Epilepsy Res 2013; 106: 410–416. [15] Jurysta F, van de Borne P, Migeotte PF, et al. A study of the dynamic interactions between sleep EEG and heart rate variability in healthy young men. Clin Neurophysiol 2003; 114: 2146–2155. [16] Kerem DH, Geva DH. Forecasting epilepsy from the heart rate signal. Med Biol Eng Comput 2005; 43: 230–239. [17] Landrot ID, Roche F, Pichot V, et al. Autonomic nervous system activity in premature and full-term infants from theoretical term to 7 years. Auton Neurosci 2007; 136: 105–109. [18] Leistritz L, Pester B, Doering A, et al. Time-variant partial directed coherence for analysing connectivity: a methodological study. Philos Trans R Soc A Math Phys Eng Sci 2013; 371: 20110616. [19] Leutmezer F, Schernthaner C, Lurger S, Potzelberger K, Baumgartner C. Electrocardiographic changes at the onset of epileptic seizures. Epilepsia 2003; 44: 348–354. [20] Lotufo PA, Valiengo L, Bensenor IM, Brunoni AR. A systematic review and meta-analysis of heart rate variability in epilepsy and antiepileptic drugs. Epilepsia 2012; 53: 272–282. [21] Mayer H, Benninger F, Urak L, Plattner B, Geldner J, Feucht M. EKG abnormalities in children and adolescents with symptomatic temporal lobe epilepsy. Neurology 2004; 63: 324–328. [22] Mesquita RC, Huppert TJ, Boas DA. Exploring neuro-vascular and neuro-metabolic coupling in rat somatosensory cortex. Phys Med Biol 2009; 54: 175–185. [23] Milde T, Schwab K, Walther M, et al. Time-variant partial directed coherence in analysis of the cardiovascular system. A methodological study. Physiol Meas 2011; 32: 1787–1805. [24] Nickels KC, Wong-Kisiel L, Moseley BD, Wirrell EC. Temporal lobe epilepsy in children. Epilepsy Res Treat 2012; 2012: 849540.

[25] Oostenveld PFR, Maris E, Schoffelen J-M. Fieldtrip: open source software for advanced analysis of MEG, EEG, and invasive electrophysiological data. Comput Intell Neurosci 2011; 2011: 156869. [26] Patel KS, Zhao MR, Ma HT, Schwartz TH. Imaging preictal hemodynamic changes in neocortical epilepsy. Neurosurg Focus 2013; 34: E10. [27] Pedemonte M, Rodriguez-Alvez A, Velluti RA. Electroencephalographic frequencies associated with heart changes in RR interval variability during paradoxical sleep. Auton Neurosci 2005; 123: 82–86. [28] Pfurtscheller G, Daly I, Bauernfeind G, Muller-Putz GR. Coupling between intrinsic prefrontal HbO2 and central EEG beta power oscillations in the resting brain. Plos One 2012; 7: e43640. [29] Piper D, Ungureanu M, Strungaru R, et al. Time-variant connectivity analysis between epileptic EEG signals and between EEG-envelopes and HRV. E-Health and Bioengineering Conference (EHB), 2013, Iasi, Romania, DOI: 10.1109/ EHB.2013.6707399 2013. [30] Roche-Labarbe N, Wallois F, Ponchel E, Kongolo G, Grebe R. Coupled oxygenation oscillation measured by NIRS and intermittent cerebral activation on EEG in premature infants. Neuroimage 2007; 36: 718–727. [31] Saleh Y, Kirchner A, Pauli E, Hilz MJ, Neundorfer B, Stefan H. Temporal lobe epilepsy: effect of focus side on the autonomic regulation of heart rate? Nervenarzt 2000; 71: 477–480. [32] Schiecke K, Wacker M, Piper D, Benninger F, Feucht M, Witte H. Time-variant, frequency-selective, linear and non-linear analysis of the heart rate variability in children with temporal lobe epilepsy. IEEE Trans Biomed Eng vol. DOI 10.1109/ TBME.2014.2307481, 2014. [33] Schwab K, Skupin H, Eiselt M, Walther M, Voss A, Witte H. Coordination of the EEG and the heart rate of preterm neonates during quiet sleep. Neurosci Lett 2009; 465: 252–256. [34] Sevcencu C, Struijk JJ. Autonomic alterations and cardiac changes in epilepsy. Epilepsia 2010; 51: 725–737. [35] Sliwka U, Harscher S, Diehl RR, van Schayck R, Niesen WD, Weiller C. Spontaneous oscillations in cerebral blood flow velocity give evidence of different autonomic dysfunctions in various types of headache. Headache 2001; 41: 157–163. [36] Theiler SEJ, Longtin A, Galdrikian B, Farmer JD. Testing for nonlinearity in time series: the method of surrogate data. Physica D 1992; 58: 77–94. [37] Toth V, Hejjel L, Fogarasi A, et al. Periictal heart rate variability analysis suggests long-term postictal autonomic disturbance in epilepsy. Eur J Neurol 2010; 17: 780–787. [38] Wacker M, Schiecke K, Putsche P, Eiselt M, Witte H. A processing scheme for time-variant phase analysis in EEG burst activity of premature and full-term newborns in quiet sleep: a methodological study. Biomed Eng-Biomed Tech 2012; 57: 491–505. [39] Wacker M, Witte H. Time-frequency techniques in biomedcal signal analysis. A tutorial review of similarities and differences. Methods Inf Med 2013; 52: 279–296. [40] Witte H, Schack B. Quantification of phase coupling and information transfer between electroencephalographic (EEG) signals: analysis strategies, models and simulations. Theory Biosci 2003; 122: 361–381.

Brought to you by | New York University Bobst Library Technical Services Authenticated Download Date | 6/7/15 7:26 PM

D. Piper et al.: Synchronization between heart rate variability and EEG      355 [41] Witte H, Eiselt M, Patakova I, et al. Use of discrete Hilbert transformation for automatic spike mapping – a methodological investigation. Med Biol Eng Comput 1991; 29: 242–248. [42] Witte H, Iasemidis LD, Litt B. Special issue on epileptic seizure prediction. IEEE Trans Biomed Eng 2003; 50: 537–539.

[43] Ylihankala A, Jantti V. EEG burst-suppression pattern correlates with the instantaneous heart-rate under isoflurane anesthesia. Acta Anaesthesiol Scand 1990; 34: 665–668. [44] Zhao MR, Suh MA, Ma HT, Perry C, Geneslaw A, Schwartz TH. Focal increases in perfusion and decreases in hemoglobin oxygenation precede seizure onset in spontaneous human epilepsy. Epilepsia 2007; 48: 2059–2067.

Brought to you by | New York University Bobst Library Technical Services Authenticated Download Date | 6/7/15 7:26 PM