DC shifts, High Frequency Oscillations, Ripples and Fast Ripples in Relation to the Seizure Onset Zone

Abstract

Efforts to improve epilepsy surgery outcomes have led to increased interest in the study of electroencephalographic oscillations outside the conventional EEG bands. These include fast activity above the gamma band, known as high frequency oscillations (HFOs), and infraslow activity (ISA) below the delta band, sometimes referred to as direct current (DC) or ictal baseline shifts (IBS). HFOs in particular have been extensively studied as potential biomarkers for epileptogenic tissue in light of evidence showing that resection of brain tissue containing HFOs is associated with good surgical outcomes. Not all HFOs are conclusively pathological, however, as they can be recorded in nonepileptic tissue and induced by cognitive, visual, or motor tasks. Consequently, efforts to distinguish between pathological and physiological HFOs have identified several traits specific to pathological HFOs, such as coupling with interictal spikes, association with delta waves, and stereotypical morphologies. On the opposite end of the EEG spectrum, sub-delta oscillations have been shown to co-localize with the seizure onset zones (SOZ) and appear in a narrower spatial distribution than activity in the conventional EEG frequency bands. In this report, we review studies that implicate HFOs and ISA in ictogenesis and discuss current limitations such as inter-observer variability and poor standardization of recording techniques. Furthermore, we propose that HFOs and ISA should be analyzed in addition to activity in the conventional EEG band during intracranial presurgical EEG monitoring to identify the best possible surgical margin.

Introduction

The success of epilepsy surgery relies on the accurate localization and complete resection or disconnection of the epileptogenic zone (EZ) from surrounding brain tissue. The epileptogenic zone is defined as the minimum amount of cortex that must be resected to produce seizure freedom and is consequently an area that can only be identified retrospectively based on epilepsy surgery outcome [1]. In contrast, the seizure onset zone (SOZ) is defined as the area of cortex where the seizure is generated and can be detected using scalp or intracranial EEG. Because the EZ and SOZ overlap, recordings that capture the SOZ can be used to predict the location of the EZ [1]. In spite of advances in neuroimaging and technology, the overall long term seizure-freedom rate after epilepsy surgery is still less than 50% [2]. In the search for new methods to identify surgical margins, oscillations outside of the traditional EEG frequency bands, including activity below the delta band (i.e. infraslow activity, ISA) and above the gamma band (i.e. high frequency oscillations, HFOs) have attracted interest as possible biomarkers of the EZ. This review provides an update on recent progress on ISA and HFOs and their association with the SOZ. The term DC shift and ISA have been used interchangeably in many publications. In this review, we will use the term “ISA” rather than “ictal baseline shifts (IBS)” or “DC shifts” since most clinical EEG recording systems use AC (alternating current) rather than DC (direct current) amplifiers.

Physiological and pathological HFOs

HFOs are commonly divided into high gamma (80-150Hz), ripples (80-250Hz), fast ripples (FRs, 250-500Hz), and very high-frequency oscillations (VHFOs, 500Hz to 2kHz). These ranges are somewhat arbitrary, however, and the exact definition of these bands varies by author. In early research, ripples were considered to be “physiological” since they were observed in the hippocampus and parahippocampal structures of normal animals and humans [3]. Conversely, FRs were thought to be “pathological” and more specific to epileptic tissue [4]. Physiological HFOs were thought to be generated by the summation of synchronous inhibitory postsynaptic potentials of interneurons [5], whereas pathological HFOs were thought to represent the field potentials of excitatory neurons [4]. However, more recent studies have demonstrated that frequency alone is not enough to differentiate “normal” from “bad” HFOs. Ripples and FRs overlap significantly both within and outside of the SOZ [6, 7], and oscillations in the high gamma, ripple and FR range can all be induced by cognitive, visual and motor tasks in nonepileptic brain areas [8-10].

Despite the difficulties in prospectively differentiating pathological from physiological HFOs, several studies have characterized features thought to be specific to pathological HFOs. One study that looked at patient outcomes in neocortical epilepsy found that ripples associated with interictal epileptiform discharges were exclusively found in the SOZ or in the primary propagation area [11]. In Roehri et al. (2018), ripples, FR and spikes were all significantly more likely to be found inside the SOZ than outside of it. However, ripples and FR were not statistically better markers than spikes, and only the combination of both spikes and HFOs (either ripples or FR) was more specific to the epileptogenic tissue [12]. These mixed findings were thought to be due to the presence of physiological HFOs in healthy tissue and the absence of pathological HFOs in some epileptogenic areas. In Frauscher et al. (2015), pathological HFOs were frequently coupled with the transition from a depolarization state to hyperpolarization (“up” to “down”) state of the slow wave cycle in non-rapid eye movement sleep. Kerber et al. (2014) reported that surgical removal of areas with FRs (>200 Hz) and ripples on a flat background was correlated with good surgical outcomes. In comparison, removal of areas showing ripples on a continuously oscillatory background did not correlate with good outcomes [13]. Some studies have reported that the co-occurrence of 3-4 Hz slow waves may be a feature of pathological HFOs. In Nonoda et al. (2016), HFOs within the SOZ were more preferentially coupled with 3-4 Hz delta waves rather than 0.5 -1 Hz slow waves during slow wave sleep [14]. A study by limura et al. (2018) observed a high occurrence rate of FRs coupled with 3-4 Hz slow waves in children with epileptic spasms (ES). Moreover, resection of areas containing slow wave-coupled FRs resulted in seizure freedom [15]. In line with these results, Matoi et al. (2018) reported that class I outcomes were less likely when strong phase-amplitude coupling of HFOs and 3-4 Hz waves was observed in non-resected tissues [16].

Despite the complexities in the ongoing debate between physiological and pathological HFOs, several recent studies have provided evidence that supports the more traditional notion of FRs as markers of epileptogenic tissue. Cimbalnik et al. (2018) reported significant overlap in the frequency and duration of physiological HFOs and pathological HFOs, but also found that the amplitude of pathological HFOs was significantly higher than HFOs observed in patients without epilepsy. Additionally, failure to resect tissue generating pathological HFOs was associated with poor outcomes [8]. Liu et al. (2018) used automatic detection to demonstrate that HFOs within the SOZ have the highest degree of waveform similarity, whereas HFOs recorded in functional regions away from epileptogenic sites were embedded in random waveforms. This repetitive waveform pattern was more evident in FRs compared to ripples [17]. A multicenter study that examined HFOs in the nonepileptic human brain tissue during intracranial EEG monitoring reported high rates of ripples (0.7-11/min) in eloquent cortex (occipital cortex, medial and basal temporal region, transverse temporal gyrus and planum temporale, pre-and postcentral gyri, and medial parietal lobe). In comparison, rates of physiological FR were extremely low (mean 0.038/min) in both eloquent and noneloquent cortex, suggesting that FRs may be a good candidate for identifying epileptogenic tissue [6].

HFOs and seizure onset pattern

Seizures with different electrographic onset patterns likely have different mechanisms of generation [18], and differences in the associated HFOs may provide insight into such mechanisms. For example, FR power increases before hypersynchronous seizure onset and during hypersynchronous-to-low voltage fast (LVF) transitions, but this increase has not been noted during rhythmic spiking or in the polyspike seizure-onset pattern in patients with mesial temporal lobe epilepsy [19]. Ferrari-Marinho reported that the highest densities of both ripples and FRs occur during common seizure onset patterns such as high-amplitude polyspikes, delta brushes, and low amplitude fast activity. In contrast, <13 Hz sharp activity, the most common pattern for propagation, contains low rates of ripples and FRs, suggesting that HFOs may be more specific to seizure onset than propagation [18]. Moreover, studies have shown that the patterns of ictal HFOs can be useful in the determination of seizure semiology in epileptic spasms [20], and complete resection of areas with ictal HFOs is associated with a significantly higher rate of seizure-free outcome in pediatric patients [21].

HFO recording techniques

HFOs may be recorded using a variety of methods, and different techniques have offered a variety of insights into the role of HFOs in epilepsy. In earlier animal and human studies, HFOs were recorded with chronically implanted intracranial micro electrodes with sampling rates of 1000 samples/s or higher. Subsequently, Worrell et al. (2004) recorded HFOs using intracranial macro electrodes and demonstrated a significant increase of HFOs within the SOZ prior to seizure onset in adult patients with neocortical epilepsy [22]. More recently, some groups have advocated the use of HFOs during short-term intraoperative recordings. For example, Wu et al. (2010) and Hussain et al. (2016) have reported that removing the tissue generating interictal FRs on electrocorticography in the OR is linked to seizure freedom in children [23, 24]. These initial studies led to the “The HFO Trial”, a prospective randomized-control trial to determine whether intraoperative ECoG-tailored surgery using HFOs rather than interictal spikes leads to equal or better seizure outcomes [25]. The results of this trial were expected in 2018. Factors that may affect the intraoperative EEG recordings of HFOs, such as anesthesia or sampling rate, need to be better characterized. In addition to chronic intracranial and intraoperative recording methods, scalp EEG and MEG have been used to record HFOs [26, 27]. However, the utility of scalp HFOs without intracranial EEG correlation in the planning of epilepsy surgery is still unclear.

Accurate and meaningful analysis of fast frequency activity is contingent upon several features of the EEG recording system. According to the Nyquist theorem, sampling rates used to record HFOs should be at least two times the upper bound of the frequency band of interest, and even higher rates are required to determine wave morphology. This functionally translates to minimum sampling rates of 1 kHz for the identification of ripples, 2 kHz for FRs, and 5 kHz for VHFOs. In addition, a low noise amplifier (below ±2μV) should be used to record HFOs due to their low amplitude [28]. Once the signals are recorded, a low-frequency filter (LFF) is used to attenuate frequencies below the band of interest. For example a LFF of 80 Hz may be used to study ripples while 250 Hz may be used for FRs. This filtered signal can then be used to visualize and identify HFOs. HFOs are defined as oscillations of 80 Hz or higher with at least four oscillation cycles that clearly stand out from the background activity. Both automated detection systems and visual inspection have been used for analyzing HFOs [28]. Because visual analysis is time-consuming and prone to subjective errors, an automated system is preferred in many cases. However, visual inspection to confirm the findings of any automated system is necessary to prevent misidentifications. The most common cause of misidentification is contamination by non-cerebral artifacts such as EMG signals from temporal and ocular muscles [29]. Unwanted artifacts may also be present in intracranial recordings of the temporal lobe, particularly when depth electrodes are used. In such cases, using a bipolar montage may reduce contamination of EMG signals from distant sources [30]. The above covers only a few of the numerous considerations in HFO analysis, but a more in-depth discussion may be found in a review by Zijlmans et al. (2017) that systematically discusses the recording, evaluation and interpretation of HFOs and also provides practical guidelines for using HFOs in clinical research [28].

HFOs and epilepsy surgery outcomes

Numerous retrospective studies have demonstrated that resection of areas with high-rates of interictal FR and SOZ is correlated with better surgical outcomes [31-33]. Conversely, the presence of FRs in post-surgical electrocorticography (ECoG) after epilepsy surgery predicts poor seizure outcome [34]. These results have led to the general consensus that resecting more of the HFO-generating areas improves the likelihood of seizure freedom [35]. A recent prospective study highlighted the value of HFOs in guiding epilepsy surgery by demonstrating that all 13 patients achieved seizure freedom after full resection of the prospectively defined HFO area (i.e. contacts with the highest rate of ripples and FR) [36]. However, a multi-center prospective study contradicted this finding by pointing out that although removal of regions generating the highest counts of interictal ripples and FRs was correlated with seizure-free outcome overall, no correlation was found at 2 of 3 centers when a center-specific analysis was performed [37]. Several factors may contribute to this discrepancy, including differences in patient population and study methodology across the centers. For example, the center that showed a positive predictive value treated pediatric patients, while the other centers treated only adult patients. In addition, the pediatric patients included in the study all had lesional epilepsy, while one of the adult centers included patients with lesional or non-lesional epilepsy. In the center with only lesional pediatric patients, HFOs were identified based on acute intraoperative ECoG while HFOs at the adult centers were identified with chronically implanted subdural grids or stereotactic-depth electrodes. Another possibility is that surgical disruption of the networks involved in seizure generation is sufficient to achieve seizure freedom even if most of the HFO generating tissue remains unresected. In addition to ripples and FRs, VHFOs have also been shown to be useful in seizure localization. In a study of temporal lobe epilepsy patients by Brazdil et al. (2018), interictal VHFOs were recorded exclusively from mesiotemporal structures and significantly better seizure outcomes were observed when large areas containing FRs and VHFOs were resected [38].

Infraslow activity

EEG activity below the standard delta band (depending on the author, below 0.1-1Hz) is often referred to as infraslow activity (ISA), ictal baseline shift (IBS), or DC shift. The term DC refers to direct current, and its use originates from the requirement of a so-called DC amplifier to view the very low frequency signals. While HFOs have a duration in the ms-range, the subdelta oscillations have a duration >1s. Due to the long duration of these subdelta waves, they must be viewed on a compressed timescale that masks the more commonly assessed parameters in clinical EEG interpretation. Furthermore, it should be noted that DC amplifiers are not routinely used in clinical practice or in commercial EEG systems because they are notoriously unstable and saturate easily. Nonetheless, slow oscillations have been found to have value in localizing the seizure onset site [39-42].

Most research studying slower EEG fluctuations use an AC (alternating current) amplifier that includes a high pass filter to remove baseline shifts. Although some authors claim that baseline shifts can be recorded by these AC amplifiers (e.g., Ikeda et al., 1996), the term ‘DC shift’ strictly speaking is incorrect in this case, and the term “infraslow activity” is more appropriate. A true DC recording is difficult to obtain because at an electrode’s metal-substrate interface, low frequencies in the signal are attenuated - an effect that can be partially mitigated by the use of silver/silver-chloride electrodes. In addition, large infraslow artifacts (for example, sweating artifacts) quickly saturate the (pre)amplifier stage, so a high-pass filter is typically used prior to this step to remove low frequency components. Unfortunately, this process attenuates not only artefactual low frequency drifts, but also any ictal-associated low frequency activity. One way to mitigate this issue is to characterize the properties of these high-pass filters and use these properties to approximate the pre-filtered signal. The filters used in standard EEG recording systems attenuate target frequencies in a linear fashion, meaning that the effects of these filters on raw input signals can be mathematically described using an operation called convolution (Fig. 1A) [43]. Despite being attenuated, the EEG signals remain in the recorded data. The parameters of the convolution operation describe the changes the raw input signals undergo as they are processed by the EEG recording system. This transformation is a property of the EEG machine and its associated filters and can be probed using measurements of known calibration signals. Once the specifics of this convolution operation are known, they can be employed to design and perform the inverse operation (i.e. a deconvolution) on measured electrical activity to generate an approximate reconstruction of the original brain electrical activity (Fig. 1B). It should be noted that the deconvolution operation can only be applied for linear systems (such as the EEG filter), they can be unstable, and they are sensitive to system noise. Therefore, its validity must be checked rigorously with calibration signals in the frequency band of interest (the mathematical details will be published elsewhere). With the above notes of caution in mind, examples of ISA visualized by this deconvolution procedure are presented and discussed in the following section.

Figure 1. Convolution and deconvolution procedures involved in the EEG recording and infraslow activity (ISA) reconstruction.

A: The raw brain electrical activity (red trace) is recorded by an EEG machine with an AC amplifier that attenuates the extreme low (and high) frequencies. This procedure can be mathematically characterized as a convolution, a linear operation. B: When taking the correct precautions, the convolution effect can be undone by a deconvolution, thereby reconstructing the low frequency component that was present in the original brain electrical activity.

The association of infraslow activity, HFOs, and SOZ

Studies of epilepsy patients have shown that areas marked by ictal ISA are concordant with conventional presurgical evaluation results but considerably more spatially restricted than conventional ictal discharges [40, 44-46]. An example that illustrates the association of ISA, ictal HFOs and conventional seizure onset in a patient with right mesial temporal lobe epilepsy is shown in Figure 2. The ictal EEG was recorded with a stereotactic depth electrode placed through a burr hole in the occipital bone, sampling 12 locations along the right hippocampus. The frequency components in the signals are shown in Figure 2, and both ISA and HFOs can be seen at the channels within or closest to the SOZ. The patient underwent an amygdalo-hippocampotomy with MR-guided laser interstitial thermal therapy that removed the putative SOZ which coincided with the highest rates of ictal HFOs and highest amplitude of ictal ISA, as assessed retrospectively. The patient remained seizure free at the most recent post-surgical follow up (2.5 years). A study by Thompson et al. (2016) reported anatomically widespread ictal ISA, suggesting that ictal ISA may arise from a network process [47]. In a prospective study of six patients, full resection of areas with ISA and partial resection of the SOZ defined by conventional EEG led to Engel class I outcome in two patients and Engel class II in four patients [48]. Such results link both ISA and HFOs to seizure generation, but very few studies directly address the relationship of ISA, HFOs and the conventional EEG.

Figure 2: Co-localization of infraslow activity (ISA) and ictal HFOs at the seizure onset zone (SOZ).

The intracranial EEG was sampled at 1024 Hz per channel and the raw data was filtered by the amplifier hardware 0.16-344 Hz. In all panels we depict the signal (negative polarity up) and its spectrogram. The power per frequency is expressed in the decibel scale (dB; computed as 10×log10 (P), with P – raw computed power in the frequency band).

A: The low frequency component of the signal. High amplitude positive ISA is observed at hippocampal depth electrodes labeled RHD 5, 6, identified as the SOZ. A lower amplitude negative ISA is observed at the amygdala-hippocampal complex (RHD 4). The black traces are the ictal EEG displayed with filter settings of 0.16-15 Hz. The red traces are the ISA displayed with digital filter settings of 0-15 Hz after deconvolution. Note that the calibration for the ISA (red traces) is 15-fold larger than the AC recorded signal (black traces), and therefore high frequency signals are too compressed to be clearly visible in the red trace. The epoch used in B is indicated in the lower trace.

B: The conventional EEG band shows seizure onset with low amplitude fast activity at RHD 5, 6 (hippocampal body, SOZ). Conventional intracranial EEG is displayed after digital filtering between 1-100Hz. The epoch used in C is indicated in the lower trace.

C: Increased rate of ripples (150 Hz) at SOZ (RHD 5, 6) before EEG seizure onset and increased rates of ictal HFOs in the SOZ at and after the seizure onset. The black traces are the raw recorded signals and the red traces are the HFOs displayed after digital filtering between 80-344 Hz.

The above findings were representative for our findings in seven seizures recorded in this patient.

A better understanding of these relationships and their contributions to the SOZ could lead to less extensive yet more effective surgical resections. The spatial extents of ISA and ictal HFOs appear to be smaller than the area defined by conventional frequency activity in both neocortical and temporal lobe epilepsy patients [41, 49, 50]. Similarly, both ISA and HFOs appear to have a more restricted spatial distribution than conventional interictal spikes and therefore define a smaller SOZ [50]. A study by Modur et al. (2012) demonstrated that ISA correctly localized the region of seizure onset in 4/6 patients with neocortical epilepsy, and good surgical outcomes were seen in the patients whose resection areas were marked by both ISA and HFOs. Gnatkovsky et al. (2014) reported fast activity at 80-120 Hz, slow transient polarizing shift and voltage depression overlap within the epileptogenic zone in a retrospectively study [51]. The ability to home in on a more precise piece of epileptogenic tissue could be particularly useful in neocortical epilepsy, which tends to have a more widespread SOZ compared to mesial temporal lobe epilepsy.

The generators of spikes, HFOs, and ISA

While surface negative spikes have been attributed to excitatory post-synaptic currents of the pyramidal neurons, physiological HFOs are thought to arise from the summation of synchronous inhibitory postsynaptic potentials [5]. Pathological HFOs, in contrast, are thought to reflect summed action potentials from small groups of fast-firing neurons which are hypersynchronized through fast synaptic transmission or nonsynaptic mechanism such as gap junction coupling [4, 52, 53]. Some studies posit that ISA are due to massive glia cell depolarization secondary to neuronal activity, or from mixed generators of neurons and functionally related glial cells [54, 55]. Others suggest the HFOs and ISA are coupled. A study of epileptic spasms found that ictal HFOs were coupled with a phase of subdelta activitiy at <1 Hz, suggesting that HFOs and ISA may share a common cortical origin [56]. However, these mechanisms cannot explain the observation that ictal ISA precedes conventional frequency activity at seizure onset, leaving open the question of why these patterns colocalize with the seizure onset zone. Alternatively, the recently proposed distinction between core and penumbra territories of the network involved in seizure activity might explain how hyper-excitation in the core first leads to ISA followed by the oscillatory seizure activity of the territory surrounding the core [57, 58].

An HFO, just as any frequency in the EEG signal, can be attributed to a generator that produces a periodic signal at that particular frequency. For instance, an HFO of 135 Hz can be the result of a population of neurons that receive synaptic input at 135 activations per second in a synchronized fashion. A simple example of a sinusoidal driving frequency is depicted in Figure 3A. On the other hand, an oscillatory EEG signal in the HFO band may also be a component of a periodic signal of a lower frequency. This happens when a periodic signal produced by an active cell group is not purely sinusoidal in shape, as is the case with neuronal signals. Such a non-sinusoidal periodic signal can be decomposed into a series of sinusoidal waves, i.e. the Fourier series of the signal, and use of EEG frequency filters will show the sinusoidal waves that are located within those filters’ pass band [43]. For example, a periodic non-sinusoidal signal of 45 Hz can have sinusoidal components at 135 Hz, within the HFO frequency band. Thus, if a broadband EEG signal with such a 45 Hz non-sinusoidal signal is filtered with an 80-200 Hz band filter, the filtered signal will show oscillations at 135 Hz even though there are no generators firing at just 135 Hz. This example is depicted in Figure 3B. In the time domain, it can be seen that the filtered signal (red trace in Fig.3B1) of the raw non-sinusoidal signal (black trace in Fig.3B1) oscillates in the HFO band. A third possibility is that low frequency driving inputs are transformed non-linearly at the post-synaptic synapse; in this case higher frequencies would be present in the post-synaptic signal despite only low-frequency inputs. Such non-linear transformations are well described in mammalian neuronal systems (e.g., Rosenberg and Issa 2011) [59].

Figure 3: Possible mechanisms that create a signal in the HFO band.

A1: A straightforward example of a generator that produces a 135 Hz signal. The associated power spectrum is shown in panel A2. In the spectrum, relative power is expressed as a percentage of the total power in the spectrum.

B1: A 45 Hz non-sinusoidal signal, a square wave (black trace) in this example, can also include an oscillatory component in the HFO band (red trace). The associated relative power spectrum shown in panel B2 (the insert depicts a detail of the spectrum) confirms the presence of this HFO oscillation and shows a main peak at 45 Hz but also one at 135 Hz.

C: For the same reason as shown in B, HFO oscillations can be part of lower frequency non-sinusoidal signals; 11 Hz in this example. Interestingly the HFO bursts (red trace) intensify around the fast transitions in the raw signal (black trace). This example shows that the distinction between alternative scenarios in panels A (a high frequency generator) and B, C (HFOs as a component of a slower oscillation) cannot be made after the filter is applied. Only when the filtered signal is reviewed together with the raw data one can discriminate between these two cases. D: Ictal HFOs in an epoch in electrode RHD6 20 seconds after seizure onset of the example in Figure 2. This panel depicts the standard EEG (black trace), HFOs filtered at 80-250 Hz (red trace), and the associated spectrogram of the HFO band (similar to the spectrograms in Fig. 2; here the dB/Hz scale is 0-25). The HFO bursts that are marked by the arrows labeled by 1 and 2 show one example of a burst associated with a fast transition as in panel C (arrow 2) in the raw signal, and one that is not, as in panel A (arrow 1).

HFO component oscillations can be present in non-sinusoidal signals far below the HFO band such as non-sinusoidal oscillations in the conventional alpha band (Fig.3C). Square wave oscillations, while non-physiological, provide a useful demonstration of how HFOs can manifest with low-frequency non-sinusoidal signals. In any signal with sharp transitions between levels, like square waves, HFO bursts predominate around the fast transitions (square wave example in Fig.3C). Note that in the example of the clinical recording in Figure 2, some of the HFO bursts coincide with fast transitions in the raw signal (just as in Fig.3C), while some do not. This indicates that both scenarios depicted in Figure 3 may occur in recorded ictal activity. An in depth analysis of how this phenomenon plays a role in ictal EEG recordings is described by Eissa et al. [60]. Current techniques do not differentiate between HFOs directly generated by synaptic inputs and HFOs generated by the filtering of non-sinusoidal signals, but we believe both types will have clinical utility in identifying epileptogenic tissue.

Conclusions

Although evidence has been presented in the literature that HFOs and ISA may be used to localize the SOZ, there are also recent studies that contradict these findings. A large number of studies have demonstrated that HFOs are markers for epileptogenic tissue. Several of these studies provide compelling evidence that HFOs localize the SOZ more reliably than interictal spikes such as reports of good surgical outcomes associated with the resection of HFO-generating tissue. The value of ISA in epilepsy presurgical evaluation is not as clear as that of the HFOs. Currently these low frequency components are mainly used to complement conventional EEG findings because they may better determine the surgical margin in cases of a widely distributed SOZ. The overlap of ictal ISA, HFOs, and conventional activity at seizure onset suggest an association among the generators of the three different types EEG activity. Because there is no consensus in the field on the clinical value of HFOs and ISA, it seems a prudent strategy to analyze them along with the conventional EEG frequencies as part of the presurgical evaluation. Although technical pitfalls in the detection methods for the two patterns at both ends of the conventional EEG bands have been reported [61], methods are improving and there is sufficient evidence to motivate further investigation into the clinical value of HFO and ISA components. Such additional research will be necessary to standardize recording techniques, to discover the mechanism of the generators of these EEG patterns, and to better define the practical value and limitations of the different bands within the broadband EEG signal before they can be used as valid clinical tools.

Highlights.

• EEG activity in very high and low frequency bands are not typically utilized in the clinical settings during epilepsy monitoring.

• High frequency oscillations (HFOs) and infraslow activity (ISA) have been linked to the seizure onset zone by many studies.

• Utilization of high frequency and infraslow activity during epilepsy surgical evaluation may lead to better clinical outcomes.