Clinical Neurophysiology 126 (2015) 1505–1513

Contents lists available at ScienceDirect

Clinical Neurophysiology journal homepage: www.elsevier.com/locate/clinph

Identification of seizure onset zone and preictal state based on characteristics of high frequency oscillations Urszula Malinowska a,⇑, Gregory K. Bergey a, Jaroslaw Harezlak b, Christophe C. Jouny a a b

Department of Neurology, Johns Hopkins University School of Medicine, Baltimore, MD, USA Department of Biostatistics, Richard M. Fairbanks School of Public Health and School of Medicine, Indiana University, Indianapolis, USA

a r t i c l e

i n f o

Article history: Accepted 7 November 2014 Available online 18 November 2014 Keywords: High frequency oscillations Ripples Preictal state Seizure onset zone Automatic detection

h i g h l i g h t s  High frequency oscillations (80–200 Hz) (HFO) may occur in all channels of intracranial recordings.  Average HFO rate is higher within seizure onset zone and increase during the transition from interic-

tal to preictal and to ictal period.  Characteristics of HFO events within the seizure onset zone differ from those outside the seizure onset

zone, and change during the interictal to ictal transition.

a b s t r a c t Objective: We investigate the relevance of high frequency oscillations (HFO) for biomarkers of epileptogenic tissue and indicators of preictal state before complex partial seizures in humans. Methods: We introduce a novel automated HFO detection method based on the amplitude and features of the HFO events. We examined intracranial recordings from 33 patients and compared HFO rates and characteristics between channels within and outside the seizure onset zone (SOZ). We analyzed changes of HFO activity from interictal to preictal and to ictal periods. Results: The average HFO rate is higher for SOZ channels compared to non-SOZ channels during all periods. Amplitudes and durations of HFO are higher for events within the SOZ in all periods compared to non-SOZ events, while their frequency is lower. All analyzed HFO features increase for the ictal period. Conclusions: HFO may occur in all channels but their rate is significantly higher within SOZ and HFO characteristics differ from HFO outside the SOZ, but the effect size of difference is small. Significance: The present results show that based on accumulated dataset it is possible to distinguish HFO features different for SOZ and non-SOZ channels, and to show changes in HFO characteristics during the transition from interictal to preictal and to ictal periods. Ó 2014 International Federation of Clinical Neurophysiology. Published by Elsevier Ireland Ltd. All rights reserved.

1. Introduction High frequency oscillations (HFO) are events observed in EEG recordings whose descriptions and mechanisms are still under investigations. Commonly used definitions span a wide range of frequency, amplitude, and duration. Their occurrence often correlates with epileptogenicity, although it remains unclear if HFO reflect pathophysiology or are epiphenomena. However, HFO are also observed in primary visual and motor cortex and are considered ⇑ Corresponding author at: Johns Hopkins University School of Medicine, Department of Neurology, Epilepsy Research Laboratory, Meyer 2-147, 600 N Wolfe St., Baltimore, MD 21287, USA. Tel.: +1 410 614 8770; fax: +1 410 955 0751. E-mail address: [email protected] (U. Malinowska).

physiological, spontaneous or task-induced activity (Nagasawa et al., 2012; Matsumoto et al., 2013; Wang et al., 2013). High-frequency oscillations have been recorded during interictal (Staba et al., 2002; Urrestarazu et al., 2007), preictal (Jacobs et al., 2009) and ictal (Jirsch et al., 2006) periods. Bragin et al. (1999a,b), and Urrestarazu et al. (2007) reported higher rates of HFO in the seizure onset zone (SOZ) than in other areas during interictal periods and more frequently during slow wave sleep than during wakefulness (Staba et al., 2004; Bagshaw et al., 2009). HFO occur very frequently associated with EEG spikes, but have also been detected independently (Jacobs et al., 2008). Interictal and ictal HFO occur in similar regions (Zelmann et al., 2009), while spikes are more widely distributed, involving a wider area ictally than interictally (Zijlmans et al., 2011). Postsurgical studies

http://dx.doi.org/10.1016/j.clinph.2014.11.007 1388-2457/Ó 2014 International Federation of Clinical Neurophysiology. Published by Elsevier Ireland Ltd. All rights reserved.

1506

U. Malinowska et al. / Clinical Neurophysiology 126 (2015) 1505–1513

show a correlation between the removal of tissue under channels with high HFO rates and favorable surgical outcome (Jacobs et al., 2010). Therefore the systematic study of HFO has taken on a greater importance in clinical applications. Previous studies based mainly on visual marking of HFO have proven to be challenging and highly time consuming. HFO are not clinical events and require different approaches than usual EEG event detection which for the most part are correlated with clinical events and associated EEG features. Because of lack of a formal definition, detection of HFO remains subjective and the comparison of studies that use heterogeneous definitions is difficult. Automatic HFO detectors are a crucial step to get a more complete overview of the HFO characteristics and to further investigate their relationship with epilepsy, especially with the assessment of the seizure onset zone in the context of continuous monitoring in epilepsy monitoring units. A HFO pattern is usually defined as a finite EEG oscillation in the range of 80–500 Hz. This EEG activity was first recorded with microelectrodes (20–40 lm in diameter) implanted in temporal regions (Bragin et al., 1999a,b), and next with clinical macroelectrodes in temporal and neocortical regions (Jirsch et al., 2006; Urrestarazu et al., 2007; Worrell et al., 2008). When recorded with macro-electrodes, HFO are characterized by a typical duration of 30–100 ms, an inter-event interval of at least 25 ms, and amplitude of 10–1000 lV. The criteria selected by different investigators for HFO identification are varied, but commonly require at least four oscillations that can be clearly distinguished from background activity, and at least 25 ms apart from each other. Detection of such low-voltage events is technically challenging, and subject to false positives introduced by signal filtering. Therefore automated detection is usually combined with visual validation of the detected events by experts. Automatic HFO detectors are largely based on comparison of the signal energy of the EEG epoch that includes the event with a background period. A recent publication (Pail et al., 2013) compares automated detection of HFO based on line length method versus a visual assessment of SEEG traces and indicates that both contribute comparably to the identification of the SOZ in patients with focal epilepsy. Taking into account, how tedious and subjective the visual analysis can be, the use of fully automatic procedures which enables evaluation of HFO even in long duration recordings can bring objective quantification to the HFO analysis paradigm. Blanco et al. (2010) analyzed a large number of HFO (N = 290,273) detected by automatic analysis. After detection based on the energy threshold designed by Staba et al. (2002), the authors developed an algorithm for automated classification of HFO. Using an unsupervised clustering approach that did not specify the number of clusters, three distinct classes of transient oscillations within the 100–500 Hz frequency range were identified. Two of the classes were consistent with ripple and fast ripple oscillations, and a third consisted of mixed-frequency events. Blanco et al. (2011) present an analysis of these classified groups of events with respect to seizure onset zone channels and other regions. The same dataset and methodology was used (Pearce et al., 2013) to investigate temporal changes of different types of HFO, their rate and proportions during interictal, preictal, ictal and postictal periods. Using data from 5 patients (2 mesial; 3 neocortical), the authors did not show clear systematic trends in HFO behavior across patients but patient-specific changes in HFO morphology linked to fluctuation in the relative rate of ripples, fast ripples, and mixed frequency events were observed. Although many groups investigate HFO rate and their changes in time and relationship with spikes and seizure onset zone, so far only a few studies have tried to analyze features of HFO to find difference between pathological and physiological activity (Matsumoto et al., 2013; Nagasawa et al., 2012; Wang et al.,

2013). One other recent work (Kerber et al., 2014) shows that specific ripple patterns, ripples occurring during flat background activity, better help identify epileptogenic areas for surgical procedures. To investigate HFO as biomarkers of epileptogenic tissue we are proposing a novel automated HFO detection method based not on local energy but on the amplitude and shape of the HFO event. We examine changes of HFO activity before and during complex partial seizures in humans. We also assess the potential use HFO as a marker for preictal state testing if there are significant changes in the characteristics of HFO in the period leading to the seizure onset and if based on HFO rate and characteristic it is possible to identify the SOZ. 2. Methods 2.1. Data Forty-five consecutive patients with intractable partial epilepsy undergoing presurgical evaluations at the Johns Hopkins Epilepsy Center recorded between 2004 and 2006 were screened for this study. We selected records from patients diagnosed either with mesial temporal (N = 13) or neocortical (N = 20) onset seizures and with at least 2 h of interictal activity before the seizure during clinical monitoring. Our final pool of patients contains 33 patients (14 males, 19 females) age 27.4 ± 13.5 years. Intracranial recordings included combinations of subdural multicontact grids and strips placed over the area of interest and targeted multi-contact depth electrode arrays. Typical implantation includes 64–128 electrodes in 4 or 8 contact strips and grid arrays with 16–64 contacts. Seizures were recorded using a Stellate™ system with Schwarzer amplifiers. A 300 Hz low-pass, Butterworth anti-aliasing filter (order = 5; 20 dB/oct) was applied prior to 16-bit digitization over a ± 3196 lV range. Analyses were carried out using bipolar montages using neighboring electrodes for improved localization of ictal onset activity. For each patient between 20 and 102 channels were analyzed. Seizure events included complex partial seizures with or without secondary generalization. All seizures were spontaneous events occurring over the course of the evaluation (5–7 days), and then marked based on visual analysis of recorded EEG by an experienced neurologist. Only the first seizure for each patient was taken into consideration for this analysis. Seizure onset is defined as the onset of epileptiform activity leading to the ictal event without return to baseline in between seizures. The channel with the first changes of epileptiform activity and surroundings electrodes (up to 8 for grid only, 2 for strips only, and more if combinations of electrodes were localized around focus channel) were considered here as seizure onset zone channels (SOZ). The seizure onset patterns consisted of sustained low voltage fast activity (21 patients), sustained rhythmic spikes discharges (8 patients), or sustained rhythmic slow wave discharges (4 patients). The onsets with low voltage fast activity very often evolved into rhythmic spikes discharges. The research protocol was reviewed by the IRB and data were stored in compliance with HIPAA regulations. 2.2. HFO detection method Periods of 2 h before and up to 2 min after the seizure onset of each marked seizure (a total of 122 min) were divided into 2 min segments and filtered at 80–200 Hz (band-pass filter using two-way least-squares FIR filtering, EEGLab, Matlab™). For HFO detection we applied an automated method, which implements identification of HFO based on the following criteria. To be marked as an HFO, the event must consists of at least 4 consecutive oscillations on filtered EEG with amplitudes above 10 lV and be

1507

U. Malinowska et al. / Clinical Neurophysiology 126 (2015) 1505–1513

at least two times larger than the average amplitude of oscillations of the surrounding background (amplitude of 5 oscillations before and after considered event). To detect short, transient HFO rather than long continuous high frequency activity, a limitation of a maximum of 100 ms was applied for each event’s duration. Events of duration longer than 100 ms were rejected from the analysis. For each detected event, the following features were quantified (cf. Fig. 1): 1. duration (time difference between the time of the first oscillation (maximum) detected to the time of the last maximum), 2. frequency (average frequency calculated from the number of zero crossings of the filtered signal in the period between first maximum and last maximum point of the oscillation), 3. amplitude (mean amplitude of the oscillations), 4. energy (of filtered signal in the window between first maximum and last maximum point of the oscillation), 5. global/average HFO amplitude peak ratio (ratio of the maximal amplitude peak of filtered signal to average of other amplitude maxima, excluding the maximal peak). Other features characterized original signal in the window of detected HFO: 6. energy of the raw signal in the window between first maximum and last maximum point of the oscillation, 7. context of detected HFO including the maximum, minimum and the phase of the raw signal during the oscillation; from this maximum amplitude difference were calculated. Some of these features are highly correlated, so only subsets of those features were used in particular comparisons.

2.4.1. Comparison between interictal, preictal, ictal periods HFO rate during interictal periods was calculated as the average rate over 59 2-min segments before the preictal period. Given the lack of normality of the HFO rate in these periods, a Kruskal–Wallis test was used to compare rate of HFO events per channel between the three periods in each group of patients and for SOZ channels and non-SOZ channels separately. Post-hoc testing was done using a Wilcoxon rank sum test between these particular periods: interictal versus preictal, interictal versus ictal and preictal versus ictal period. 2.4.2. Comparison between SOZ and non-SOZ channels To assess the differences in HFO rate between channels within and outside SOZ in each of period (interictal, preictal, ictal) for all patients and to analyze differences between HFO from patients with mesial temporal lobe epilepsy (MTLE) and patients with neocortical onset seizures (NEO), the Wilcoxon rank sum test was used. All these procedures: data reading, filtering, detection of HFO, calculation and storage of parameters, and all statistical comparison were implemented and performed in Matlab™. 2.5. HFO feature analysis HFO rates are highly variable among channels and patients. To compare properties of HFO between analyzed periods and channels we considered detected HFO events as repeated for each patient observation obtained from the interictal, preictal and ictal periods as well as from multiple channels used for each subject. To properly adjust for the correlations between these observations, we used a hierarchical linear mixed model (Verbeke and Molenberghs, 2000): 

2.3. Classification of HFO events Events detected across all patients during a 2 h window before electrographic seizure onset (interictal period), 2 min prior to (considered preictal period) and 2 min following seizure onset (ictal period), were classified by K-means clustering analysis based on features characterizing the morphology of the HFO pattern: frequency, duration, energy, global/average peak ratio, and max difference of context amplitude. Events from each particular cluster were compared to the period of their occurrence and group of channels localized within the seizure onset zone (SOZ) and outside the seizure onset zone (non-SOZ). 2.4. HFO rate comparison Rates of HFO detection were compared between group of channels (SOZ and non-SOZ) and between the interictal, preictal and ictal periods.

Y ikl ðt j Þ ¼ b0 þ b1 period þ b2 electrode þ b3 period electrode þ b0k þ b0kðlÞ þ eijkl

ð1Þ

where bs are the fixed coefficients of the period, electrode group and their interaction respectively, and bs are the random coefficients inducing the correlation between channels (‘‘k’’ index), and across time within channels (‘‘i’’ index). The fixed coefficients in the model were tested for their equality to zero. The best model was chosen for each analysis for the 6 HFO-derived measures. These models were fitted using the R statistical software (R Foundation for Statistical Computing, http://www.R-project.org). The model can be used to estimate the period-by-channel group specific means as well as the variance explained by the random effects (b) and unexplained variation (e). All fitted values presented in the results section are based on the REML (restricted maximum likelihood) estimators of the random effects covariance matrix. We designed a two part analysis to investigate the differences of HFO characteristics surroundings ictal events. To that purpose, we first identify the differences between HFO relative to their occurrence in ictal periods or outside of the ictal periods (non-ictal

Fig. 1. Example of an HFO pattern and parameters describing its characteristics used in automatic detection: at least 4 consecutive oscillations (6 in this case) with amplitude above 10 lV and averaging at least 2 times larger than the average amplitude of surrounding oscillations. Mean amplitude is the average of the amplitude of the 6 peaks in the event window.

1508

U. Malinowska et al. / Clinical Neurophysiology 126 (2015) 1505–1513

events). Events detected during a 2 h window before the electrographic onset of a seizure and free of other ictal events were labeled as non-ictal events. HFO detected during a 2 min window following the electrographic onset of a seizure were labeled as ictal events. Although the two minute window can be greater than the seizure duration itself (Afra et al., 2008), the post-ictal period included reflects the propagation after the ictal event. In a second part of the analysis of HFO characteristics, we distinguish preictal period events as HFO events occurring within a two minute window prior to the onset of the seizure. Non-ictal events in the previous 1h 58 min are labelled as interictal. 3. Results 3.1. Detection and classifications results Over 4558 channel-hours (33 patients with on average 69 channels per patient, 2 h per channels) of recordings were analyzed. Automatic HFO detection identified over one million events (N = 1,045,989). These events were examined with clustering analysis, K-means algorithm to identify the optimal number of clusters which best discriminate HFO based on the subset of selected features. The optimum number of clusters was assessed by calculating the smallest number of cluster for which the summed distance of each data point (event) to its assigned centroid showed only a small change with the addition of an extra cluster. The threshold in the number of cluster for which extra cluster was not reducing the distance to centroids was for k = 4. Results of the clustering analysis with k = 4 are shown on Fig. 2. The first cluster contains events of long duration, fast frequency and high energy (cf. Fig. 3 – C1). The second cluster was for events of short duration and high frequency with average energy (Fig. 3 – C2). The third cluster contains events that were of long duration, mixed frequency content often indicating a superimposition of events, and average energy (cf. Fig. 3 – C3). The 4th cluster contains mostly artifacts (high amplitude shifts in signal cf. Fig. 3 – C4). The majority of ictal HFO (47.2% of all events detected in ictal period) belong to the first cluster, and are predominantly in channels within the seizure onset zone (79.2% vs 20.8% outside SOZ). Interictal and preictal HFO are spread within the second and third clusters. In interictal periods C1 represents 5.5% of all detected events, C2 55.5%, C3 32.4% and C4 6.6% of interictal HFO. In preictal period C1 is 3.8%, C2 is 58.4%, C3 35.4%, C4 2.4% of all detected

Fig. 2. Clustering of 1,045,989 events (across all patients) in the 3-dimensional space of their duration, frequency and energy characteristics. Each of the 4 clusters found by the K-mean algorithm is colored uniquely (C1-blue, C2-red, C3-green, C4black).

during preictal period events. In ictal periods: C1 is 47.2%, C2 28.2%, C3 19.3% and C4 5.3% of ictal detections. Results of clustering per periods are presented in details on Fig. 4. In all periods and for all clusters (except C4, which contains mostly artifacts, in interictal and preictal period) HFO is predominant in the SOZ (Fig. 4b). Although the number of events detected during interictal period is much higher (due to 2 h observation) than during only 2 min of preictal period, the distribution in each cluster is similar for these two periods. That distribution however changes for the ictal period. Because we apply an automatic detection method for HFO detection we performed the following steps for artifacts elimination: we selected only events from cluster 1–3, and events without sharp shifts in raw signal within only several data samples (indicating high likelihood of an electrical artefact). The remaining HFO were verified by a random sampling and visual inspection of 0.1% of all events by two reviewers. From the results of this inspection we estimate that still about 5.2% of HFO events could be resulting from recordings artifacts (amplitude shifts, suboptimal electrodes or signals with harmonics).

3.2. HFO rate The average rates of detected HFO (after artifact removal) per channel per 1 min, for each period are presented in Fig. 5. The average rate (and 95% confidence intervals, CI) of HFO detection in SOZ channels is 3.37/min (95% CI: 3.18–3.57), 4.01/min (95% CI: 2.38–5.65), and 19.34/min (95% CI: 15.15–23.53) per interictal, preictal and ictal period respectively. For non-SOZ channels rates are: 2.84/min (95% CI: 2.77–2.92), 3.15/min (95% CI: 2.59–3.72), 7.91/min (95% CI: 6.84–8.99) per each period. HFO rates for this pooled dataset are significantly higher in channels within the SOZ compared with channels outside the SOZ in all periods (all p < 0.001) and differ between periods. The average HFO rate per channel in the SOZ increase from interictal through preictal to ictal periods, all with a significant difference p < 0.01 (Fig. 5a). There are also significant differences in the average HFO rates between MTLE and NEO group in all analyzed periods and groups of channels (all p < 0.001), except non-SOZ channels during ictal periods (p = 0.34) and this HFO rate is consistently higher for MTLE (Fig. 5b).

3.3. HFO features: ictal versus non-ictal comparison The HFO features described earlier (Section 2.2) were considered: HFO amplitude, frequency, duration, energy (filtered signal), energy (raw signal) and global/average peak ratio. Before fitting the model (Section 2.5), we assessed the distributions of the six features and found that they were highly skewed towards higher values. To reduce the influence of the large observed values and stabilize the variance we log10-transformed all the measures and used them as outcomes in the model. The model was fit to each log-10 transformed measure and we selected the most parsimonious model which balances the variation explained and the number of explanatory variables for a given measure. For the three of the measures (amplitude, energy filtered, energy raw) the interaction model was chosen. For frequency the additive model: b3 = 0 was selected. For the global peak ratio the model with only the period: b2 = b3 = 0 was chosen. For duration the model with only the channel group: b1 = b3 = 0 was selected. The interaction model was selected with the p-values for the test of the coefficients less than 0.01 for all analyzed features, except frequency. Fitted values for the period-channel group combination are presented in Table 1 and in Fig. 6.

U. Malinowska et al. / Clinical Neurophysiology 126 (2015) 1505–1513

1509

Fig. 3. Typical HFO events for clusters C1 (blue on Fig. 2, events of long duration, fast frequency and high energy), C2 (red on Fig. 2, events of short duration and high frequency with average energy), C3 (green on Fig. 2, events that were of long duration, mixed frequency content often indicating a superimposition of events, and average energy), C4 (black on Fig. 2, mostly artifacts (high amplitude shifts in signal); for each example: on the top -raw signal, bipolar montage channel, bottom – the same signal filtered in the 80–200 Hz range with detected HFO.

Fig. 4. Results of clustering per periods. (a) Percentage of events in each cluster in each period. (b) Percentage of events in each cluster in each period with separation of events within and outside SOZ.

All considered HFO features differ between non-ictal and ictal period. Amplitude, energy (and energy of raw signal), duration and frequency are higher during ictal periods than non-ictal periods. All features during ictal periods were also higher within the SOZ (all p < 0.001) compared to outside the SOZ, except for the frequency. HFO from the SOZ channels have lower frequency than HFO outside the SOZ (Fig. 6). For non-ictal periods, significant differences were for energy (of raw signal) and duration. Higher values of energy of raw signal and duration indicated HFO from SOZ channels than outside. Also during non-ictal periods frequency

was lower for SOZ channels than outside in non-SOZ channels (p < 0.001).

3.4. HFO features: comparison of interictal to preictal periods In the second step of analysis, we compared only the interictal period to the preictal period.

1510

U. Malinowska et al. / Clinical Neurophysiology 126 (2015) 1505–1513

3.4.1. Interaction models (of channels group and period factor) The interaction model was selected when the p-value for the test of the interaction coefficient was less than 0.01. The estimated mean amplitude value for the period-electrode group combination is presented in Table 2 and on Fig. 7. Since the model indicates an interaction between the period and the SOZ factor, pairwise comparisons show that the amplitude of HFO is lower in non-SOZ than SOZ (p < 0.01), but during the preictal period the difference between the amplitude of HFO between SOZ and non-SOZ is larger. There was no significant difference (p = 0.44) in amplitudes between HFO of the SOZ in the interictal

and the preictal period, however the estimated amplitude of HFO in non-SOZ channels is smaller in the interictal period than in the preictal period (p < 0.001). The interaction model was selected for the energy of HFO events in the filtered signal. The estimated mean energy values for the period-electrode group combinations are presented in Table 2 and on Fig. 7. There was no significant difference (p = 0.66) in energy between HFO of the SOZ in the interictal period and the preictal period; however the estimated mean energy of HFO in non-SOZ channels is smaller in the interictal period than in the preictal period (p < 0.001). The energy of HFO (filtered signal) is larger in SOZ channels than outside in either interictal (p = 0.002) or preictal (p < 0.001) periods. The energy of the raw signal in the window of detected HFO is significantly different between SOZ and non-SOZ channels outside SOZ in both interictal (p < 0.001) and preictal (p < 0.001) periods. There was no significant difference (p = 0.07) in energy between HFO of the SOZ in the interictal period and preictal periods. The estimated mean values for the period-channel group combination are presented in Table 2 and in Fig. 7. 3.4.2. Non-interaction models (channels group + period) The additive model was selected for comparison of the frequency of HFO. The fitted values of frequency for the periodchannel group combinations are presented in Table 2 and in Fig. 7. All pairwise comparisons show significant differences between groups. The frequency of events in SOZ channels are lower than outside in interictal (p < 0.001) and preictal period (p < 0.001). Frequency of HFO in the SOZ is smaller in the interictal period than in the preictal period (p < 0.001) as well as outside of the SOZ (p < 0.001).

Fig. 5. (a) Average rate ± SE of HFO (from C1, C2 and C3) per channel per 1 min, for each period and for channels within and outside SOZ; (b) average rate of HFO (from C1, C2 and C3) per channel per 1 min for each period. SOZ: channels within the seizure onset zone, nonSOZ (nsoz on plot (b)): channels outside the onset zone, MTLE: mesial temporal onset seizure, NEO: neocortical onset seizures. Lines and stars indicate groups of significant differences in HFO rate.

3.4.3. Non-interaction models (channels group) For duration of HFO events the only significant predictor was the channel group which showed separation between channels (SOZ vs non-SOZ). The estimated mean durations for channel groups are presented in Table 2 and on Fig. 7. Duration of HFO in the SOZ is greater than duration of HFO outside of the SOZ (p < 0.001).

Table 1 Estimated mean values of selected HFO features and their standard errors (mean ± SE) for the period-electrode group combination. Model: comparison of non-ictal and ictal periods. Non-ictal

Amplitude [lV] Energy (filtered) [dB] Energy (raw) [dB] Frequency [Hz] Duration [sec] Glob/av. peak ratio

Ictal

SOZ

Non-SOZ

SOZ

Non-SOZ

20.46 ± 1.92 8.3 ± 0.2 13.29 ± 0.24 98.28 ± 1.32 0.059 ± 0.001 1.62 ± 0.08

20.42 ± 1.82 8.28 ± 0.19 12.69 ± 0.21 99.3 ± 1.29 0.058 ± 0.001 1.63 ± 0.08

43.2 ± 4.04 10.11 ± 0.24 16.11 ± 0.29 99.89 ± 1.34 0.064 ± 0.001 2.42 ± 0.13

33.36 ± 2.98 9.42 ± 0.22 14.74 ± 0.24 100.93 ± 1.32 0.061 ± 0.001 2.14 ± 0.11

Table 2 Estimated mean values of selected HFO features and their standard errors (mean ± SE) for the period-electrode group combination. Model: comparison of interictal and preictal periods. Interictal

Amplitude [lV] Energy (filtered) [dB] Energy (raw) [dB] Frequency [Hz] Duration [sec] Glob/av. peak ratio

Preictal

SOZ

Non-SOZ

SOZ

Non-SOZ

21.12 ± 0.98 8.37 ± 0.10 13.32 ± 0.13 98.44 ± 0.77 0.060 ± 0.001 1.633 ± 0.044

20.28 ± 0.90 8.26 ± 0.09 12.65 ± 0.11 99.44 ± 0.76 0.059 ± 0.001 1.633 ± 0.044

21.20 ± 0.98 8.38 ± 0.10 13.36 ± 0.13 98.18 ± 0.77 0.060 ± 0.001 1.602 ± 0.043

19.58 ± 0.86 8.18 ± 0.09 12.60 ± 0.11 99.18 ± 0.76 0.059 ± 0.001 1.602 ± 0.043

U. Malinowska et al. / Clinical Neurophysiology 126 (2015) 1505–1513

1511

Fig. 6. Estimated mean values of analyzed HFO features (mean ± SE) for the period-channel group combination. Model: comparison of non-ictal and ictal periods. Significantly different groups are marked by lines.

Fig. 7. Estimated mean values of analyzed HFO features (mean ± SE) for the period-channel group combination. Significant differences between groups are marked by black lines.

3.4.4. Non-interaction models (period) Only model of separation between periods (interictal vs preictal) was significant for global/average HFO amplitude’s peak ratio. This HFO feature is smaller in preictal period compared to the interictal period. The estimated mean values for these two periods are shown in Table 2 and on Fig. 7.

3.5. Summary of HFO features comparison Fig. 8 presents a schematic visualization of properties of HFO within SOZ and outside the SOZ, during the three time periods. In summary, for these visually distinguishable features: for all periods there are significant differences in fitted values of HFO

1512

U. Malinowska et al. / Clinical Neurophysiology 126 (2015) 1505–1513

Fig. 8. Schematic presentation of differences in properties of estimated HFO within SOZ (on the left) and outside the SOZ (on the right) of the three visually distinguishable features: A – amplitude, D – duration, F – frequency; and their significant changes during the transition from interictal to preictal and to ictal period.

amplitude, durations and frequency between SOZ channels and channels outside the SOZ. All features of ictal patterns are higher than during interictal and preictal periods. During the transition from interictal to preictal periods, the frequency of HFO and amplitude decrease but not amplitude of HFO within SOZ.

4. Discussion and conclusions Taking advantage of the automatic HFO detection method it was possible to investigate large numbers of HFO events and their characteristics. In this study we focus on detection of HFO patterns, not continuous fast oscillations but short transient oscillations clearly distinguished from background, as suggested recently (Kerber et al., 2014) to be better correlated with epileptogenic areas then other HFO activity. This distinction was possible through the used automated detection approach. SOZ considered in this study is not exactly as SOZ defined based on visual inspection practiced in clinical evaluations, but consist as a channel with the first changes of epileptiform activity and surroundings electrodes (which not always could be involved or show early ictal activity). However, based of accumulated dataset we were able to find significant differences of HFO rate and characteristics distinguish SOZ and the rest of the channels. The results of clustering analysis do not indicate separation of clusters or different distributions between clusters for interictal and preictal HFO. For the ictal period the distributions of clustertypes patterns indicates changes in the characteristics of HFO and the most prominent types of HFO belong to the first cluster: long, high amplitude and energy, average frequency, coming from SOZ channels. Most importantly, the average HFO rate is higher for SOZ channels compared to non-SOZ channels during ictal, preictal and also interictal periods. This result is consistent with the current approach in epilepsy research that HFO can be seen as an indicator of the epileptogenic zone. Our result also shows that HFO are more

frequent for MTLE patients than NEO during non-ictal and ictal periods but not for channels outside SOZ during ictal periods where their average rates do not differ significantly. Despite this finding, it is necessary to emphasize that HFO could be observed in all channels. It was also reported (Nagasawa et al., 2012; Wang et al., 2013) that primary visual cortex can normally generate HFO and resection of such primary visual HFO sites is not necessary to obtain long term seizures freedom. Based on the other analyzed HFO features the differentiation of HFO events from SOZ and non-SOZ channels indicates that energy and energy of the raw signal is higher for channels in the SOZ during interictal and preictal periods, the frequency is lower during interictal and preictal periods, and duration is higher for SOZ than non-SOZ channels. The same trend was found for ictal HFO when compared to the SOZ and non-SOZ HFO, although all analyzed features of ictal events were significantly higher compared to non-ictal period. Preictal HFO exhibit significant differences in some of the characteristics compared to interictal HFO. Comparison between interictal and preictal periods showed the following differences in HFO characteristics: decrease in amplitude, energy, and energy of raw signal of HFO outside SOZ, decreases of frequency of HFO from SOZ and non-SOZ, decrease in value of global/average peak ratio from interictal to preictal period. However, amplitude, energy, and energy of raw signal of HFO detected in SOZ channels did not change significantly. Differences in values of some of HFO features between interictal and preictal HFO could be related to EEG desynchronization during the preictal period or a higher number of artifacts during long interictal periods. Wang et al. (2013) reported that ripples superimposed on epileptiform discharges such as paroxysmal fast, spike or sharp waves better correlate with SOZ than ripples (or spikes) that occurred independently. In this study of an automated detection paradigm, we did not visually assess background activity of detected HFO. However, some of our features characterized the original signal in the window of detected HFO (energy of the raw signal, maximum and minimum amplitude in the window between first maximum and last maximum point of the HFO). The estimated energy of the raw signal (as presented on Figs. 6 and 7) and the maximum amplitude difference was always significantly higher for HFO from SOZ compared to rest of the channels. From this we can infer that HFO in SOZ channels occur most often in highly variable amplitude backgrounds, such as active spiking, and more frequently than HFO outside the SOZ channels. Using only three of the features that characterize the shape of the HFO events, we represented in Fig. 8 a schematic view of the HFO differences between the transition between interictal and preictal and between non-ictal and ictal events. During preictal transitions, HFO are slower both inside and outside the SOZ but also of lower amplitude outside the SOZ. During the non-ictal to ictal transitions, HFO are faster and of higher amplitude in both areas, but longer in the SOZ. Similar results were found by Matsumoto and colleagues (Matsumoto et al., 2013) when they studied task-induced physiological HFO and pathological HFO from epileptogenic brain. Pathological HFO had higher mean spectral amplitude, longer mean duration, and lower mean frequency than physiological HFO. However, the authors observed in one individual patient infrequent high-amplitude HFO in the motor cortex just before movement onset in the motor task and concluded that this raised the possibility that in epileptic brain, physiological-induced gamma can assume higher spectral amplitudes similar to those seen in pathologic HFO. Other studies which reported the occurrence of spontaneous and visually- induced HFO in occipital cortex (Nagasawa et al., 2012) did not find significant differences either in the spectral frequency band or amplitude of spontaneous occipital HFO from

U. Malinowska et al. / Clinical Neurophysiology 126 (2015) 1505–1513

those of epileptogenic areas (but very limited datasets were compared). In our dataset only 3 of the patients had any occipital electrodes implanted (67 channels, 2% of all channels analyzed), so the impact on the final results should be small if any. In conclusion, automated procedures are necessary to process the extremely large amount of HFO data, but caveats remain as to the meaning of the events detected without human supervision. Based on this analysis, taking into account large numbers of events, we were able to distinguish HFO features different for SOZ and non-SOZ channels, and changes in HFO characteristics during the transition from interictal to preictal and to ictal. The evidence that large numbers of HFO are required to identify the SOZ is a limitation to the utility of HFO as to be used as a tool to mark the seizure onset location in individual patients. However the mere existence of these differences suggests that the pathological nature of these HFO is true even if the difference between normal and pathological HFO is less than anticipated. Acknowledgment This work was supported by NIH grant NS75020. Conflict of interest: None of the authors have potential conflicts of interest to be disclosed. References Afra P, Jouny CC, Bergey GK. Duration of complex partial seizures: an intracranial EEG study. Epilepsia 2008;49:677–84. Bagshaw AP, Jacobs J, LeVan P, Dubeau F, Gotman J. Effect of sleep stage on interictal high-frequency oscillations recorded from depth macroelectrodes in patients with focal epilepsy. Epilepsia 2009;50:617–28. Blanco JA, Stead M, Krieger A, Stacey W, Maus D, Marsh E, et al. Data mining neocortical high-frequency oscillations in epilepsy and controls. Brain 2011;134:2948–59. Blanco JA, Stead M, Krieger A, Viventi J, Marsh WR, Lee KH, et al. Unsupervised classification of high-frequency oscillations in human neocortical epilepsy and control patients. J Neurophysiol 2010;104:2900–12. Bragin A, Engel J, Wilson CL, Fried I, Mathern GW. Hippocampal and entorhinal cortex high-frequency oscillations (100–500 Hz) in human epileptic brain and in kainic acid-treated rats with chronic seizures. Epilepsia 1999a;40:127–37. Bragin A, Engel J, Wilson CL, Vizentin E, Mathern GW. Electrophysiologic analysis of a chronic seizure model after unilateral hippocampal KA injection. Epilepsia 1999b;40:1210–21.

1513

Jacobs J, LeVan P, Chander R, Hall J, Dubeau F, Gotman J. Interictal high-frequency oscillations (80–500 Hz) are an indicator of seizure onset areas independent of spikes in the human epileptic brain. Epilepsia 2008;49:1893–907. Jacobs J, Zelmann R, Jirsch J, Chander R, Dubeau C-ECF, Gotman J. High frequency oscillations (80–500 Hz) in the preictal period in patients with focal seizures. Epilepsia 2009;50:1780–92. Jacobs J, Zijlmans M, Zelmann R, Chatillon C-E, Hall J, Olivier A, et al. High-frequency electroencephalographic oscillations correlate with outcome of epilepsy surgery. Ann Neurol 2010;67:209–20. Jirsch JD, Urrestarazu E, LeVan P, Olivier A, Dubeau F, Gotman J. High-frequency oscillations during human focal seizures. Brain 2006;129:1593–608. Kerber K, Dümpelmann M, Schelter B, Le Van P, Korinthenberg R, Schulze-Bonhage A, et al. Differentiation of specific ripple patterns helps to identify epileptogenic areas for surgical procedures. Clin Neurophysiol 2014;125: 1339–45. Matsumoto A, Brinkmann BH, Matthew Stead S, Matsumoto J, Kucewicz MT, Marsh WR, et al. Pathological and physiological high-frequency oscillations in focal human epilepsy. J Neurophysiol 2013;110:1958–64. Nagasawa T, Juhász C, Rothermel R, Hoechstetter K, Sood S, Asano E. Spontaneous and visually driven high-frequency oscillations in the occipital cortex: intracranial recording in epileptic patients. Hum Brain Mapp 2012;33: 569–83. Pail M, Halámek J, Daniel P, Kuba R, Tyrlíková I, Chrastina J, et al. Intracerebrally recorded high frequency oscillations: simple visual assessment versus automated detection. Clin Neurophysiol 2013;124:1935–42. Pearce A, Wulsin D, Blanco JA, Krieger A, Litt B, Stacey WC. Temporal changes of neocortical high frequency oscillations in epilepsy. J Neurophysiol 2013;110: 1167–79. Staba RJ, Wilson CL, Bragin A, Fried I, Engel J. Quantitative analysis of highfrequency oscillations (80–500 Hz) recorded in human epileptic hippocampus and entorhinal cortex. J Neurophysiol 2002;88:1743–52. Staba RJ, Wilson CL, Bragin A, Jhung D, Fried I, Engel J. High-frequency oscillations recorded in human medial temporal lobe during sleep. Ann Neurol 2004;56: 108–15. Urrestarazu E, Chander R, Dubeau F, Gotman J. Interictal high-frequency oscillations (100–500 Hz) in the intracerebral EEG of epileptic patients. Brain 2007;130: 2354–66. Verbeke G, Molenberghs G. Linear mixed models for longitudinal data. New York: Springer-Verlag; 2000. Wang S, Wang IZ, Bulacio JC, Mosher JC, Gonzalez-Martinez J, Alexopoulos AV, et al. Ripple classification helps to localize the seizure-onset zone in neocortical epilepsy. Epilepsia 2013;54:370–6. Worrell GA, Gardner AB, Stead SM, Hu S, Goerss S, Cascino GJ, et al. High-frequency oscillations in human temporal lobe: simultaneous microwire and clinical macroelectrode recordings. Brain 2008;131:928–37. Zelmann R, Zijlmans M, Jacobs J, Châtillon C-E, Gotman J. Improving the identification of high frequency oscillations. Clin Neurophysiol 2009;120: 1457–64. Zijlmans M, Jacobs J, Kahn YU, Zelmann R, Dubeau F, Gotman J. Ictal and interictal high frequency oscillations in patients with focal epilepsy. Clin Neurophysiol 2011;122:664–71.