Abstract

With the modern development of electroencephalogram (EEG) recording technology (stereo-EEG, grid, strip, and microelectrodes), computational EEG analysis started to play an essential role in assisting traditional visual EEG evaluation. The number of recording electrodes and the temporal resolution of these modern recording technologies are far beyond the human capability to rapidly extract and integrate all necessary information for studies of epileptogenic networks. This chapter describes the state of the art of computational EEG analysis. It reviews and discusses different computational approaches and achievements in the approximation of the epileptogenic zone, in seizure pattern recognition, seizure classification, and in the understanding of seizure generation mechanisms.

Introduction

With the modern development of electroencephalogram (EEG) recording technology (stereo-EEG, grid, strip, and microelectrodes), computational EEG analysis started to play an essential role in assisting traditional visual EEG evaluation. The number of recording electrodes and the temporal resolution of these modern recording technologies are far beyond the human capability to rapidly extract and integrate all necessary information for studies of epileptogenic networks. Here we will describe the state of the art of computational EEG analysis. We will review and discuss different computational approaches and achievements in the approximation of the epileptogenic zone, in seizure pattern recognition, seizure classification, and in the understanding of seizure generation mechanisms.

When in 1996 Garry Kasparov as the acting chess world champion lost his legendary game against IBM’s deep blue supercomputer, he did not resign from chess (Kasparov, 2017). On the contrary, he embraced the new technology and initiated what then became known as “cyborg” or “advanced chess,” where players are allowed to use computers and chess software during games to reach unprecedented levels of strength. While neurologists have yet to be “defeated,” that is, surpassed by artificial intelligence in their ability to assess and interpret EEG signals, “cyborg epileptology,” that is, the use of computers and quantitative methods to support clinicians, has emerged long ago and has become an important tool of modern epileptology. Here, we set out to give an overview of these methods by first following an “outside to inside of the brain” approach. We describe scalp EEG, source imaging, and then intracranial “stereo” EEG and a set of linear and nonlinear quantitative signal characteristics. We then outline the more recent and (again) strongly technology-driven field of ultra-long-term EEG, which has provided crucial insights into larger-scale patterns of epileptic activity. We conclude by speculating about future developments, including the design of invasive brain machine interfaces with very large numbers of microelectrodes.

Scalp EEG

For decades, scalp EEG has remained the primary noninvasive tool to study electrical brain signals. Thanks to standardized electrode positions and montages, recorded data can nowadays be easily exchanged between different centers. Traditional visual evaluation can be applied to study physiological and pathological brain activity. Digitization of EEG recordings allows developing and applying different computational approaches to identify and analyze ictal and interictal epileptiform signals. In particular, identification and localization of epileptic spikes still are main goals of many computational tools developed for scalp EEG and MEG analysis. Primarily implemented in EEG software is the identification and representation of the negative voltage field maximum, that is, source localization. It is based on the assumption that the surface electrode that records the maximum field potential is closest to the spike generator. It provides a fast and straightforward method but is not always accurate. The assumption is only valid if the direction of the dipole is perpendicular to the head surface and matches the orientation of neurons. If the spike source is inside a sulcus, the pattern of voltage maximum on the surface is different, and the focus location’s conclusion may be wrong. More accurate methods developed to study voltage topography over the entire brain to localize epileptogenic brain areas are based on the principles of volume conductor theory (Gloor, 1985) and aim to resolve the forward (Huang et al., 1999; Mosher et al., 1999; Kybic et al., 2005; Gramfort et al., 2010) and inverse (Dale et al., 2000; Baillet et al., 2001; Pascual-Marqui, 2002) problem of epileptic spikes source localization. The forward problem relies on the fact that every dipolar source within a three-dimensional (3D) volume conductor can produce only one resultant distribution of its field on the two-dimensional (2D) surface of the conductor. In other terms, under perfect conditions, for any focal spike generator somewhere in the brain's 3D volume, only a single possible pattern of surface EEG signal exists. The inverse problem is more relevant in epileptology, where one needs to identify the spike source in the 3D brain from a scalp EEG recording. However, the inverse problem has an infinite number of solutions and in practice solutions are only found after applying restrictive assumptions, which often do not have a physiological basis (Michel and He, 2019). Common methods include the minimum norm estimation (MNE), dynamic statistical parametric maps (dsPMs), and standardized low-resolution brain electromagnetic tomography (sLORETA) or modifications of these. All three methods are considered distributed inverse methods, supposing the same initial assumptions to define active zones with different hypotheses. While MNE normalizes the current density map, dsPM uses the noise covariance for normalization, and sLORETA replaces the covariance of noise by the covariance of data. Obtained results may vary significantly between methods. Hence, it is necessary during presurgical evaluation to use several methods to address the inverse problem to improve the precision of approximating the generators (Beniczky et al., 2016).

Intracranial Stereo-EEG

Two-thirds of the cortex lies in in the depth of sulci and fissures and, together with mesial areas, can generate epileptiform signals that may be difficult or even impossible to evaluate with scalp EEG recordings alone. In particular, the focal and fast activity typically generated at seizure onsets is often challenging to detect from head surface electrodes. Direct recordings from the “deep” epileptogenic brain areas can be performed only with invasive intracranial stereo-EEG recording (SEEG), which allows the precise exploration of almost any brain region. During the last decade, stereotactic techniques have significantly evolved with increased accuracy, broad distribution in many epilepsy centers around the globe, high standardization, and low complication rates (Mullin et al., 2016; Willems et al., 2019). Boundaries of the epileptogenic zone (EZ), defined as the area of the cortex necessary and sufficient for seizure generation, are commonly delineated by visually reviewing EEG signals during seizures and the interictal state. Since intracerebral recordings are performed with many electrode contacts (15–20 electrodes, up to 200 contacts) positioned in preselected brain regions for several days, a time-consuming reviewing process by highly qualified and trained neurophysiologists is required. Computational algorithms and quantitative analysis can complement the traditional evaluation process. Compared to visual inspection, quantitative analysis of intracranial signals is considered to (1) expedite EZ delineation, (2) to provide precise and more objective results, and (3) to improve the understanding of EEG patterns in the EZ and surrounding areas.

Two principal approaches are available to analyze simultaneously recorded multichannel neuronal activity: linear and nonlinear. Frequency is one of the most used quantitative characteristics of linear EEG analysis.

Epileptogenicity Index

Dynamic changes in brain rhythms characterize seizure onset. Many different intracranial electroencephalographic seizure-onset patterns can be observed in different epileptogenic lesions (low-voltage fast activity, low-frequency high-amplitude periodic spikes, sharp activity, spike-and-wave activity, burst of high-amplitude polyspikes, burst suppression, and delta brush). Most commonly, low-voltage fast activity is detected (Perucca et al., 2014). Fabrice Wendling and coworkers were the first to propose a quantitative way to identify the epileptogenic zone based on low-voltage, fast-activity measurement. Using a sliding window Fourier analysis they defined a so-called Epileptogenicity Index (EI) (Bartolomei et al., 2008). This measure is based on both spectral (appearance of high-frequency oscillations replacing the slower background activity) and temporal (delay of appearance concerning seizure onset) properties of intracerebral EEG signals recorded during presurgical evaluation. High-frequency oscillations are frequently observed during the transition from interictal to ictal activity (Worrell et al., 2004; Jirsch et al., 2006). They are often referred to as “rapid discharges” as they constitute a typical electrophysiological pattern characterized by a noticeable increase of signal frequency. This pattern was initially described by Bancaud and coworkers (1965) and is now recognized as a characteristic marker of the onset of focal seizures, provided that electrodes are appropriately positioned in the target site. The spectral properties of rapid discharges (above 100 Hz) observed at seizure onset have been quantified in several studies (Allen et al., 1992; Alarcon et al., 1995; Wendling et al., 2003; Worrell et al., 2004; Jirsch et al., 2006). Classically, the rapid discharge is a transient phenomenon, which lasts for a few seconds, and which may or may not be associated with voltage reduction. To characterize both the propensity of a given brain structure to generate rapid discharge patterns and the delay of appearance of this discharge relative to seizure onset, the EI can be used (Fig. 19–1, “EI”). Thus, the EI goes beyond the description of the frequency content of signals. Instead, the EI makes use of (1) the signal energy in well-defined frequency bands (appearance of beta-gamma oscillations concomitantly with the attenuation of slower alpha-theta oscillations) and (2) the time at which rapid discharges occur. The combination of these two factors in a single quantity provides a good characterization of the ictogenicity of explored brain structures: the absence (resp. presence) of a rapid discharge and/or the late (resp. early) involvement in the seizure process contributes to low (resp. high) EI value. Several experimental (Traub et al., 2001) and computational modeling studies (Wendling et al., 2005) demonstrated the existence of a relationship between the epileptogenicity of the neuronal tissue and its propensity to generate fast oscillations. However, some parameters must be adjusted to control the sensitivity and the specificity of the method. Particularly, two parameters, that is, threshold and bias, must be set to detect the onset of the rapid discharge. These two parameters can be interactively adjusted in a semi-automated fashion. Once the operator agrees on the detection time provided by the respective algorithm, the EI value is computed. Future studies are needed to compare reported results with those obtained from a fully automated method.

Figure 19–1.

Comparison of different linear and nonlinear methods for EZ detection in four patients (from (Andrzejak et al., 2015)). Electrode positions on 3D brain schemes (top in each panel), IEEG recordings (left bottom in each panel), and mean values of indices (more...)

Statistical Parametric Mapping of Epileptogenicity Index

David and coworkers further developed the methodology of the EI adopting a neuroimaging approach to mapping of the seizure onset zone, and of networks of seizure propagation, based on the quantification of fast oscillations recorded in implanted epileptic patients (David et al., 2011). Assuming that many stereo-EEG recording sites can be obtained to generate statistical parametric maps of the EI, their method allows standard statistical testing, either at the patient or at group levels. Statistical parametric mapping is an established framework for comparing multidimensional image data and allows one to correct for inherent multiple comparisons. The methodology of Statistical Parametric Mapping of the Epileptogenicity Index (SPMEI) identifies brain areas whose high-frequency activity is significantly greater than baseline using standard techniques from imaging time series analysis (Fig. 19–1, “SPMEI”). This involves a simple categorical comparison (using t-tests) between mean activity at baseline and the mean activity over short windows (e.g., 4 s) at various times (e.g., 0–20 s) after seizure onset. The significance of these differences is evaluated with relation to the variability of fluctuations within each time segment. The main prior of the proposed approach is the identification of a frequency band of interest, for example, the high gamma range (60–100 Hz), which is known to be typically involved in temporal and insular-temporal seizures (Bartolomei et al., 2008, 2010). However, other frequency bands may also be relevant and will need to be systematically tested in future studies.

Quantified Frequency Analysis Index

Another computer-assisted method based on EEG signal analysis in frequency and space domains was proposed by Gnatkovsky et al. They developed a semi-automatic tool to perform real-time extraction and visualization of the principal SEEG biomarkers of the epileptogenic network such as individual contact spectral dynamics and slow-wave components (Gnatkovsky et al., 2011, 2014). The respective method combines three electrographic SEEG biomarkers associated with seizures: (1) specific fast activity (FA), (2) transient slow polarizing shift (SPS), and (3) electrographic trace depression or “flattening” (FLT) of background activity. This electrographic stereo-EEG biomarkers and time-frequency profiles can be quantified individually for a single patient and for each ictal event. After biomarker extraction a final combined index score can be calculated as Mean [Fast Activity + Slow polarizing shift + Flattening]/3, and contacts with threshold >60 have been suggested to be considered as EZ contacts (Gnatkovsky et al., 2014). Individual time-frequency profiles of the contacts with the highest index help to efficiently identify and categorize seizure patterns and evaluate their reproducibility in a single patient (Fig. 19–1, “QFAI”). As a major result of this approach, a prevalent neocortical focal seizure pattern was identified and described as a transient electrographic event with prompt low-voltage, fast-activity onset (around 120 Hz), superimposed on a slow potential shift, terminated with periodic bursting activity and prominent postictal depression (Gnatkovsky et al., 2019). These so-called P-type seizures were observed in almost 70% of patients, were of extratemporal origin, and had a mean duration of about 20 seconds. In contrast, so-called L-type seizures were observed in about 41% of patients and consistently involved mesial temporal structures. They lasted much longer (93 ± 48 seconds), started with 116 ± 21 Hz low-voltage, fast activity superimposed on a slow potential shift, and terminated with a large-amplitude, periodic bursting activity. In some patients, the L-type seizure was preceded by a P-seizure. A limitation of this methodology is that current analysis has relatively low spatial resolution given the limited number of electrodes (about 17 electrodes per implantation). Given the importance of the slow transient during seizures, unfiltered DC recording needs to be given special attention in the future.

Nonlinear Structure Index

Epileptic seizures are commonly characterized as “hypersynchronous states.” This habit is doubly misleading, because seizures are not necessarily synchronous and are not unchanging “states” but dynamic processes. Schindler and coworkers ( 2007) analyzed the temporal evolution of the correlation structure in the course of focal onset seizures of patients recorded by intracranial multichannel EEG. They applied a multivariate method by computing the eigenvalue spectrum of the zero-lag cross correlation matrix of a short sliding window. Thereby, they observed statistically significant changes of the correlation structure of focal onset seizures. They found that the zero-lag correlation of multichannel EEG either remained approximately unchanged or especially in the case of secondary generalization—decreased during the first half of the seizures. Then correlation gradually increased again before the seizures terminated. This development was qualitatively independent of the anatomical location of the seizure-onset zone and therefore seemed to be a generic property of focal onset seizures. It was suggested that the decorrelation of EEG activity was due to the different propagation times of locally synchronous ictal discharges from the seizure-onset zone to other brain areas. Moreover, the increase of correlation during the second half of the seizures was hypothesized to be causally related to seizure termination.

Later, Andrzejak et al. combined different univariate nonlinear measures and surrogates to analyze interictal EEG recordings from patients with epilepsy. They compared signals recorded from brain areas where the first ictal EEG signal changes were detected (“focal signals”) with signals recorded from brain areas that were not involved at seizure onset (“nonfocal signals”). Interictal EEG analysis showed that combinations of nonlinear measures with univariate surrogates allowed to localize the ictogenic brain regions where performances of both linear and nonlinear measures were weak. Based on combinations of nonlinear measures with surrogates, they found that interictal EEG signals from epileptogenic brain areas are less random, more nonlinear-dependent, and more stationary than signals recorded from nonepileptogenic brain areas (Andrzejak et al., 2012; Fig. 19–1, “NLSI”).

High-Frequency Oscillations

Research of recent years suggested that high-frequency oscillations (HFOs) are a promising biomarker of the EZ (Gardner et al., 2007; Jefferys et al., 2012; Gliske et al., 2016; Frauscher et al., 2017), and it will be addressed in detail in Chapter 13, this volume. As for any biomarkers, a prospective assessment of the use of HFOs for surgery planning using automatic detection of HFOs is needed in order to determine the true clinical value (Jacobs et al., 2018). Disentangling physiologic from pathologic HFOs also remains an important issue (Guragain et al., 2018). Considering the appearance and the topographic location of presumed physiologic HFOs could be immanent for the interpretation of HFO findings in a clinical context. Also, reliably recording HFOs noninvasively via scalp electroencephalography (EEG) and magnetoencephalography (MEG) is highly desired, as it provides us with the possibility to translate the use of HFOs to the scalp in a larger number of patients (Melani et al., 2013; Fahoum et al., 2014; Cai et al., 2021).

Ultra-Long-Term EEG

Ultra-long-term EEG recordings are another emerging subfield of epileptology, where computational methods are central. “Ultra-long-term” here refers to periods of months to years. There are at least two urgent needs for developing and improving methods and devices for ultra-long-term EEG. The first one is that seizure counts reported by patients often are inaccurate (Fisher et al., 2012; Elger and Hoppe, 2018). This is a fundamental but often neglected challenge because seizure occurrences reported by patients are one of the most important—if not the most important—measure to inform clinical treatment and to gauge the effectiveness of therapies, both pharmacologic and nonpharmacologic. The reasons for the inaccuracies of patients’ seizure counts are, for example, seizures that occur during sleep or that are very brief and impair consciousness so rapidly that the patients may not even realize that they just had a seizure. If nobody observed and then also informed the patient, this seizure will go unrecorded. Seizures may also involve neural networks crucial for memory—for example, in mesio-temporal lobe epilepsy—and thereby prevent the patient from remembering. And finally, even if the patient correctly realized the occurrence of a seizure, the event has then to be documented in an easy-to-use and reliable way, preferably with back-up and in a format that can make the information readily available to the treating epileptologist (Bruno et al., 2020; Chiang et al., 2020).

A second motivation for ultra-long-term EEG is the intriguing observation that interictal epileptiform activity (IEA) and the risk of seizure occurrence reveal cyclical patterns on large time scales (Cook et al., 2013; Baud et al., 2018; Karoly et al., 2021). These slow rhythms and their interactions with circadian and sleep-wake patterns (Leloup and Goldbeter, 2008) are impossible to detect with isolated short-term recordings as are currently the prevalent clinical practice (Karoly et al., 2021), leading to a severe under-sampling of the individual disease dynamics (Baud et al., 2021).

However, stigmatization of epilepsy patients is still prevalent—as impressively described in the recently published autobiography of author and journalist Kurt Eichenwald (Eichenwald, 2018). Therefore, devices and methods for ultra-long-term EEG recording have to be designed in a way that prevents or at least minimizes additional potential stigmatization of epilepsy patients. In practice, this implies that ultra-long-term EEG has to be as unobtrusive and as invisible to others as possible. There are several—not mutually exclusive—approaches to solving this crucial challenge. Miniaturized EEG electrodes may, for example, be integrated into personal objects of daily use such as glasses (Lee et al., 2020), discreet baseball caps, or individualized earpieces (Bleichner et al., 2015). They may also be inserted subcutaneously (Weisdorf et al., 2019; Duun-Henriksen et al., 2020) or implanted—together with the recording and analyzing device—into the skull (Sun and Morrell, 2014). Alternatively, one might also restrict EEG recordings to night-time by easy-to-use and robust headband systems (Arnal et al., 2020). Though these recordings would then not provide continuous information, they might still allow to track multi-dian or longer cycles of interictal epileptiform activity and thereby help to inform personalized treatments.

To minimize the obtrusiveness of all the above-mentioned systems, it is mandatory to take energy efficiency into account. Highly energy-efficient systems need, for example, smaller batteries and may thus meet much better the tight space constraints for intracranially implantable devices (Sun and Morrell, 2014) or allow to reduce weight and bulkiness of wearables. An important aspect of ultra-long-term EEG systems should also be their ability to not only record electrical brain signals but also to analyze them. Computing on the device would allow reducing the amount of data that has to be transmitted for further analyses, thereby reducing the energy used for communication and ideally also increasing data safety (Sun et al., 2018). However, processing EEG data on the device, for example, to detect epileptic seizures or assess interictal epileptiform activity, necessitates that both hardware and algorithms are energy-efficient. In recent years, several methods that meet this criterion have been developed and applied for the analysis of biosignals. One very elegant representative of this group of energy-efficient, brain-inspired approaches has been termed “hyperdimensional computing” (Kanerva, 2009) In hyperdimensional computing, information is typically represented by binary or bipolar vectors with a very high dimension of ~10,000. Importantly, this type of representation is extremely robust and allows compensating the variability often inherent in ultra-low-power hardware (Wu et al., 2018). Computation then consists of manipulating the high-dimensional vectors with clearly defined mathematical operations (Schindler and Rahimi, 2021) . Hyperdimensional computing has recently been demonstrated to enable highly energy-efficient detection of epileptic seizures from short-term and long-term intracranial EEG recordings with excellent sensitivity and specificity (Burrello et al., 2019; Burrello, Benatti, et al., 2020; Burrello, Schindler, et al., 2020). Recent intriguing developments aim at combining deep-learning methods with hyperdimensional computing (Karunaratne et al., 2021).

The Next Ten Years

Major limitations of currently available computational EEG analysis methods include that technical infrastructure varies between centers, and that there is no gold standard defined yet. Described methods are usually highly specific and complex, requiring expert mathematical and computational knowledge. They are developed and applied in a single center with no broad clinical applications.

Even more so they lack comparisons with other methods. There are several developments to address this issue. One is the creation of public databases for human EEG data, for example, www.ieeg.org (https://mni-open-ieegatlas.research.mcgill.ca/), www.epilepsy-database.eu, and http://ieeg-swez.ethz.ch/, which will allow for testing methods on larger cohorts. In addition, integrated databases, where different research domains are comprehensively covered, are urgently needed to identify correlations of EEG signals with other biomarker sources. Another approach is to organize open crowdsourcing seizure detection competitions via KAGGLE and other platforms (Baldassano et al., 2017), which not only help to attract bioinformaticians, data analysts, and engineers interested in big data from inside and outside the epilepsy field but also provide an objective performance measure on a common dataset that is used by all participants.

Though single units have been recorded in humans for decades, a very promising new development that will likely gain greater momentum over the next 10 years is the sampling of the neuronal network around single neurons or units with a very large number of microwires (Musk, 2019). Although they are not yet widely used in the clinic, electrodes with microwires are available for exploration. The limitations are for sure the data complexity and interpretation with regard to the whole epileptogenic network. From the positive side we have information of single-unit contribution in particular structures or lesions (e.g., hippocampus, FCD, tumor) for the ictal and interictal activity. Especially interesting is how this human single-unit data correlate with previously obtained experimental data on ictogenic mechanisms in animal models and organotypic slice preparations (Worrell et al., 2008; Truccolo et al., 2011; Niediek et al., 2016; Weiss et al., 2016).

Disclosure Statement

References

1. Alarcon, G. et al. (1995) ‘Power spectrum and intracranial EEG patterns at seizure onset in partial epilepsy’, Electroencephalogr Clin Neurophysiol, 94(5), pp. 326–337. [PubMed: 7774519]
2. Allen, P. J., Fish, D. R. and Smith, S. J. (1992) ‘Very high-frequency rhythmic activity during SEEG suppression in frontal lobe epilepsy’, Electroencephalogr Clin Neurophysiol, 82(2), pp. 155–159. [PubMed: 1370786]
3. Andrzejak, R. G. et al. (2015) ‘Localization of Epileptogenic Zone on Pre-surgical Intracranial EEG Recordings: Toward a Validation of Quantitative Signal Analysis Approaches.’, Brain topography, 28(6), pp. 832–7. doi: 10.1007/s10548-014-0380-8. [PubMed: 24929558]
4. Andrzejak, R. G., Schindler, K. and Rummel, C. (2012) ‘Nonrandomness, nonlinear dependence, and nonstationarity of electroencephalographic recordings from epilepsy patients’, Physical review. E, Statistical, nonlinear, and soft matter physics. 86(4 Pt 2), p. 046206. [PubMed: 23214662]
5. Arnal, P. J. et al. (2020) ‘The dreem headband compared to polysomnography for electroencephalographic signal acquisition and sleep staging’, Sleep, 43(11), pp. 1–13. doi: 10.1093/sleep/zsaa097. [PMC free article: PMC7751170] [PubMed: 32433768]
6. Baillet, S., Mosher, J. C. and Leahy, R. M. (2001) ‘Electromagnetic brain mapping’, IEEE Signal Processing Magazine, 18(6), pp. 14–30. doi: 10.1109/79.962275.
7. Baldassano, S. N. et al. (2017) ‘Crowdsourcing seizure detection: Algorithm development and validation on human implanted device recordings’, Brain, 140(6), pp. 1680–1691. doi: 10.1093/brain/awx098. [PMC free article: PMC6075622] [PubMed: 28459961]
8. Bancaud, J. et al. (1965) ‘La stéréoencéphalographie dans l’épilepsie. Informations neuro-physio-pathologiques apportées par l’investigation fonctionnelle stéréotaxique’, Paris : Masson, p. 321.
9. Bartolomei, F. et al. (2010) ‘From mesial temporal lobe to temporoperisylvian seizures: a quantified study of temporal lobe seizure networks’, Epilepsia. 2010/08/27, 51(10), pp. 2147–2158. doi: 10.1111/j.1528-1167.2010.02690.x. [PubMed: 20738379]
10. Bartolomei, F., Chauvel, P. and Wendling, F. (2008) ‘Epileptogenicity of brain structures in human temporal lobe epilepsy: a quantified study from intracerebral EEG’, Brain, 131(7), pp. 1818–1830. [PubMed: 18556663]
11. Baud, M. O. et al. (2018) ‘Multi-day rhythms modulate seizure risk in epilepsy’, Nature Communications, 9(1), 88. doi: 10.1038/s41467-017-02577-y. [PMC free article: PMC5758806] [PubMed: 29311566]
12. Baud, M. O., Schindler, K. and Rao, V. R. (2021) ‘Under-sampling in epilepsy: Limitations of conventional EEG’, Clinical Neurophysiology Practice. Elsevier B.V., pp. 41–49. doi: 10.1016/j.cnp.2020.12.002. [PMC free article: PMC7829106] [PubMed: 33532669]
13. Beniczky, S. et al. (2016) ‘Ictal EEG source imaging in presurgical evaluation: High agreement between analysis methods’, Seizure, 43, pp. 1–5. doi: 10.1016/j.seizure.2016.09.017. [PMC free article: PMC5176190] [PubMed: 27764709]
14. Bleichner, M. G. et al. (2015) ‘Exploring miniaturized EEG electrodes for brain-computer interfaces. An EEG you do not see?’, Physiological Reports, 3(4), e12362. doi: 10.14814/phy2.12362. [PMC free article: PMC4425967] [PubMed: 25847919]
15. Bruno, E. et al. (2020) ‘Remote assessment of disease and relapse in epilepsy: Protocol for a multicenter prospective cohort study’,  JMIR Research Protocols, 9(12), e21840. doi:10.2196/21840 [PMC free article: PMC7773514] [PubMed: 33325373]
16. Burrello, A. et al. (2019) ‘Laelaps: An Energy-Efficient Seizure Detection Algorithm from Long-term Human iEEG Recordings without False Alarms’, in Proceedings of the 2019 Design, Automation and Test in Europe Conference and Exhibition, DATE 2019. Institute of Electrical and Electronics Engineers Inc., pp. 752–757. doi: 10.23919/DATE.2019.8715186.
17. Burrello, A., Benatti, S., et al. (2020) ‘An Ensemble of Hyperdimensional Classifiers: Hardware-Friendly Short-Latency Seizure Detection With Automatic iEEG Electrode Selection’, IEEE Journal of Biomedical and Health Informatics, 25(4), pp. 935–946. doi: 10.1109/jbhi.2020.3022211. [PubMed: 32894725]
18. Burrello, A., Schindler, K., et al. (2020) ‘Hyperdimensional Computing with Local Binary Patterns: One-Shot Learning of Seizure Onset and Identification of Ictogenic Brain Regions Using Short-Time iEEG Recordings’, IEEE Transactions on Biomedical Engineering, 67(2), pp. 601–613. doi: 10.1109/TBME.2019.2919137. [PubMed: 31144620]
19. Cai, Z. et al. (2021) ‘Noninvasive high-frequency oscillations riding spikes delineates epileptogenic sources’, Proceedings of the National Academy of Sciences of the United States of America, 118(17). doi: 10.1073/pnas.2011130118. [PMC free article: PMC8092606] [PubMed: 33875582]
20. Chiang, S. et al. (2020) ‘Seizure detection devices and health-related quality of life: A patient- and caregiver-centered evaluation’, Epilepsy and Behavior, 105. doi: 10.1016/j.yebeh.2020.106963. [PubMed: 32092459]
21. Cook, M. J. et al. (2013) ‘Prediction of seizure likelihood with a long-term, implanted seizure advisory system in patients with drug-resistant epilepsy: A first-in-man study’, The Lancet Neurology, 12(6), pp. 563–571. doi: 10.1016/S1474-4422(13)70075-9. [PubMed: 23642342]
22. Dale, A. M. et al. (2000) ‘Dynamic statistical parametric mapping: Combining fMRI and MEG for high-resolution imaging of cortical activity’, Neuron, 26(1), pp. 55–67. doi: 10.1016/S0896-6273(00)81138-1. [PubMed: 10798392]
23. David, O. et al. (2011) ‘Imaging the seizure onset zone with stereo-electroencephalography’, Brain. 2011/10/07, 134(10), pp. 2898–2911. doi: awr238 [pii]10.1093/brain/awr238. [PubMed: 21975587]
24. Duun-Henriksen, J. et al. (2020) ‘A new era in electroencephalographic monitoring? Subscalp devices for ultra–long-term recordings’, Epilepsia, 61(9), pp. 1805–1817. doi: 10.1111/epi.16630. [PubMed: 32852091]
25. Eichenwald K (2018) A Mind Unraveled: A Memoir. New York: Ballantine Books.
26. Elger, C. E. and Hoppe, C. (2018) ‘Diagnostic challenges in epilepsy: seizure under-reporting and seizure detection’, The Lancet Neurology. Lancet Publishing Group, pp. 279–288. doi: 10.1016/S1474-4422(18)30038-3. [PubMed: 29452687]
27. Fahoum, F. et al. (2014) ‘Epileptic scalp ripples are associated with corticothalamic BOLD changes’, Epilepsia, 55(10), pp. 1611–1619. doi: 10.1111/EPI.12760. [PubMed: 25167766]
28. Fisher, R. S. et al. (2012) ‘Seizure diaries for clinical research and practice: Limitations and future prospects’,  Epilepsy and Behavior. Academic Press Inc., pp. 304–310. doi: 10.1016/j.yebeh.2012.04.128. [PubMed: 22652423]
29. Frauscher, B. et al. (2017) ‘High-frequency oscillations: The state of clinical research’,  Epilepsia. Blackwell Publishing Inc., pp. 1316–1329. doi: 10.1111/epi.13829. [PMC free article: PMC5806699] [PubMed: 28666056]
30. Gardner, A. B. et al. (2007) ‘Human and automated detection of high-frequency oscillations in clinical intracranial EEG recordings’, Clinical Neurophysiology, 118(5), pp. 1134–1143. doi: 10.1016/j.clinph.2006.12.019. [PMC free article: PMC2020804] [PubMed: 17382583]
31. Gliske, S. V. et al. (2016) ‘The intrinsic value of HFO features as a biomarker of epileptic activity’, in ICASSP, IEEE International Conference on Acoustics, Speech and Signal Processing—Proceedings. Institute of Electrical and Electronics Engineers Inc., pp. 6290–6294. doi: 10.1109/ICASSP.2016.7472887. [PMC free article: PMC4955749] [PubMed: 27453693]
32. Gloor, P. (1985) ‘Neuronal generators and the problem of localization in electroencephalography: Application of volume conductor theory to electroencephalography’, Journal of Clinical Neurophysiology, 2(4), pp. 327–354. doi: 10.1097/00004691-198510000-00002. [PubMed: 4056020]
33. Gnatkovsky, V. et al. (2011) ‘Identification of reproducible ictal patterns based on quantified frequency analysis of intracranial EEG signals’, Epilepsia. 2011/01/29, 52(3), pp. 477–488. doi: 10.1111/j.1528-1167.2010.02931.x. [PubMed: 21269289]
34. Gnatkovsky, V. et al. (2014) ‘Biomarkers of epileptogenic zone defined by quantified stereo-EEG analysis.’, Epilepsia, 55(2), pp. 296–305. doi: 10.1111/epi.12507. [PubMed: 24417731]
35. Gnatkovsky, V. et al. (2019) ‘Two main focal seizure patterns revealed by intracerebral electroencephalographic biomarker analysis’, Epilepsia, 60(1), pp. 96–106. doi: 10.1111/epi.14610. [PubMed: 30565671]
36. Gramfort, A. et al. (2010) ‘OpenMEEG: Opensource software for quasistatic bioelectromagnetics’, BioMedical Engineering Online, 9(1), pp. 1–20. doi: 10.1186/1475-925X-9-45. [PMC free article: PMC2949879] [PubMed: 20819204]
37. Guragain, H. et al. (2018) ‘Spatial variation in high-frequency oscillation rates and amplitudes in intracranial EEG’, Neurology, 90(8), pp. E639–E646. doi: 10.1212/WNL.0000000000004998. [PMC free article: PMC5818159] [PubMed: 29367441]
38. Huang, M. X., Mosher, J. C. and Leahy, R. M. (1999) ‘A sensor-weighted overlapping-sphere head model and exhaustive head model comparison for MEG’,  Physics in Medicine and Biology, 44(2), pp. 423–440. doi: 10.1088/0031-9155/44/2/010. [PubMed: 10070792]
39. Jacobs, J. et al. (2018) ‘Removing high-frequency oscillations: A prospective multicenter study on seizure outcome’, Neurology, 91(11), pp. e1040–e1052. doi: 10.1212/WNL.0000000000006158. [PMC free article: PMC6140372] [PubMed: 30120133]
40. Jefferys, J. G. R. et al. (2012) ‘Mechanisms of physiological and epileptic HFO generation’, Progress in Neurobiology, 98(3), pp. 250–264. doi: 10.1016/j.pneurobio.2012.02.005. [PMC free article: PMC4873284] [PubMed: 22420980]
41. Jirsch, J. D. et al. (2006) ‘High-frequency oscillations during human focal seizures’, Brain, 129(Pt 6), pp. 1593–1608. [PubMed: 16632553]
42. Kanerva, P. (2009) ‘Hyperdimensional computing: An introduction to computing in distributed representation with high-dimensional random vectors’, Cognitive Computation, 1(2), pp. 139–159. doi: 10.1007/s12559-009-9009-8.
43. Karoly, P. J. et al. (2021) ‘Cycles in epilepsy’, Nature Reviews Neurology. Nature Research. doi: 10.1038/s41582-021-00464-1. [PubMed: 33723459]
44. Karunaratne, G. et al. (2021) ‘Robust high-dimensional memory-augmented neural networks’, Nature Communications, 12(1), p. 2468. doi: 10.1038/s41467-021-22364-0. [PMC free article: PMC8084980] [PubMed: 33927202]
45. Kasparov, G. (2017) Deep Thinking: Where Machine Intelligence Ends and Human Creativity Begins. John Murray Press.
46. Kybic, J. et al. (2005) ‘A common formalism for the integral formulations of the forward EEG problem’, IEEE Transactions on Medical Imaging, 24(1), pp. 12–28. doi: 10.1109/TMI.2004.837363. [PubMed: 15638183]
47. Lee, J. H. et al. (2020) ‘3D Printed, Customizable, and Multifunctional Smart Electronic Eyeglasses for Wearable Healthcare Systems and Human-Machine Interfaces’, ACS Applied Materials and Interfaces, 12(19), pp. 21424–21432. doi: 10.1021/acsami.0c03110. [PubMed: 32319751]
48. Leloup, J. C. and Goldbeter, A. (2008) ‘Modeling the circadian clock: From molecular mechanism to physiological disorders’, BioEssays. Bioessays, pp. 590–600. doi: 10.1002/bies.20762. [PubMed: 18478538]
49. Melani, F. et al. (2013) ‘Occurrence of scalp-fast oscillations among patients with different spiking rate and their role as epileptogenicity marker’, Epilepsy Research, 106(3), pp. 345–356. doi: 10.1016/j.eplepsyres.2013.06.003. [PMC free article: PMC4490901] [PubMed: 23932079]
50. Michel, C. M. and He, B. (2019) ‘EEG source localization’, in Handbook of Clinical Neurology. Handb Clin Neurol, pp. 85–101. doi: 10.1016/B978-0-444-64032-1.00006-0. [PubMed: 31277878]
51. Mosher, J. C., Leahy, R. M. and Lewis, P. S. (1999) ‘EEG and MEG: Forward solutions for inverse methods’, IEEE Transactions on Biomedical Engineering, 46(3), pp. 245–259. doi: 10.1109/10.748978. [PubMed: 10097460]
52. Mullin, J. P. et al. (2016) ‘Is SEEG safe? A systematic review and meta-analysis of stereo-electroencephalography-related complications’, Epilepsia, 57(3), pp. 386–401. doi: 10.1111/epi.13298. [PubMed: 26899389]
53. Musk, E. (2019) ‘An integrated brain-machine interface platform with thousands of channels’, Journal of Medical Internet Research, 21(10). doi: 10.2196/16194. [PMC free article: PMC6914248] [PubMed: 31642810]
54. Niediek, J. et al. (2016) ‘Reliable analysis of single-unit recordings from the human brain under noisy conditions: Tracking neurons over hours’, PLoS ONE, 11(12), p. e0166598. doi: 10.1371/journal.pone.0166598. [PMC free article: PMC5145161] [PubMed: 27930664]
55. Pascual-Marqui, R. D. (2002) ‘Standardized low-resolution brain electromagnetic tomography (sLORETA): Technical details’, in Methods and Findings in Experimental and Clinical Pharmacology, pp. 5–12. [PubMed: 12575463]
56. Perucca, P., Dubeau, F. and Gotman, J. (2014) ‘Intracranial electroencephalographic seizure-onset patterns: Effect of underlying pathology’, Brain, 137(1), pp. 183–196. doi: 10.1093/brain/awt299. [PubMed: 24176980]
57. Schindler, K. et al. (2007) ‘Assessing seizure dynamics by analysing the correlation structure of multichannel intracranial EEG’, Brain, 130(Pt 1), pp. 65–77. [PubMed: 17082199]
58. Schindler, K. A. and Rahimi, A. (2021) ‘A Primer on Hyperdimensional Computing for iEEG Seizure Detection’, Frontiers in Neurology, 12:701791. doi: 10.3389/fneur.2021.701791. [PMC free article: PMC8329339] [PubMed: 34354666]
59. Sun, F. T. and Morrell, M. J. (2014) ‘The RNS System: Responsive cortical stimulation for the treatment of refractory partial epilepsy’, Expert Review of Medical Devices, 11(6), pp. 563–572. doi: 10.1586/17434440.2014.947274. [PubMed: 25141960]
60. Sun, H. et al. (2018) ‘Wearable communications in 5g: Challenges and enabling technologies’, IEEE Vehicular Technology Magazine, 13(3), pp. 100–109. doi: 10.1109/MVT.2018.2810317.
61. Traub, R. D., Miles, M. A. and Whittington, A. (2001) ‘A possible role for gap junctions in generation of very fast EEG oscillations preceding the onset of, and perhaps initiating, seizures’, Epilepsia, 42(2), pp. 153–170. doi: 10.1046/j.1528-1157.2001.26900.x. [PubMed: 11240585]
62. Truccolo, W. et al. (2011) ‘Single-neuron dynamics in human focal epilepsy’, Nat Neurosci. 2011/03/29, 14(5), pp. 635–641. doi: 10.1038/nn.2782. [PMC free article: PMC3134302] [PubMed: 21441925]
63. Weisdorf, S. et al. (2019) ‘Ultra-long-term subcutaneous home monitoring of epilepsy—490 days of EEG from nine patients’,  Epilepsia, 60(11), pp. 2204–2214. doi: 10.1111/epi.16360. [PMC free article: PMC6899579] [PubMed: 31608435]
64. Weiss, S. A. et al. (2016) ‘Ictal onset patterns of local field potentials, high frequency oscillations, and unit activity in human mesial temporal lobe epilepsy’, Epilepsia, 57(1), pp. 111–121. doi: 10.1111/epi.13251. [PMC free article: PMC4862409] [PubMed: 26611159]
65. Wendling, F. et al. (2003) ‘Epileptic fast intracerebral EEG activity: evidence for spatial decorrelation at seizure onset’, Brain, 126(6), pp. 1449–1459. [PMC free article: PMC2040489] [PubMed: 12764064]
66. Wendling, F. et al. (2005) ‘Interictal to ictal transition in human temporal lobe epilepsy: insights from a computational model of intracerebral EEG’,  Journal of clinical neurophysiology : official publication of the American Electroencephalographic Society, 22(5), pp. 343–356. [PMC free article: PMC2443706] [PubMed: 16357638]
67. Willems, L. M. et al. (2019) ‘Invasive EEG-electrodes in presurgical evaluation of epilepsies: Systematic analysis of implantation-, video-EEG-monitoring- and explantation-related complications, and review of literature’, Epilepsy and Behavior, 91, pp. 30–37. doi: 10.1016/j.yebeh.2018.05.012. [PubMed: 29907526]
68. Worrell, G. A. et al. (2004) ‘High-frequency oscillations and seizure generation in neocortical epilepsy’, Brain, 127(7), pp. 1496–1506. doi: 10.1093/brain/awh149. [PubMed: 15155522]
69. Worrell, G. A. et al. (2008) ‘High-frequency oscillations in human temporal lobe: Simultaneous microwire and clinical macroelectrode recordings’, Brain, 131(4), pp. 928–937. doi: 10.1093/brain/awn006. [PMC free article: PMC2760070] [PubMed: 18263625]
70. Wu, T. F. et al. (2018) ‘Hyperdimensional computing exploiting carbon nanotube FETs, resistive RAM, and their monolithic 3d integration’, IEEE Journal of Solid-State Circuits, 53(11), pp. 3183–3196. doi: 10.1109/JSSC.2018.2870560.