Spatiotemporal Profiles of Electrical Rhythmic Activity in Human Extratemporal Lobe Epilepsy

Transcription

1 Spatiotemporal Profiles of Electrical Rhythmic Activity in Human Extratemporal Lobe Epilepsy by Marija Cotic A thesis submitted in conformity with the requirements for the degree of Doctor of Philosophy Institute of Biomaterials and Biomedical Engineering University of Toronto Copyright by Marija Cotic 2014

2 ii Spatiotemporal Profiles of Electrical Rhythmic Activity in Human Extratemporal Lobe Epilepsy Abstract Marija Cotic Doctor of Philosophy Institute of Biomaterials and Biomedical Engineering University of Toronto 2014 With the advent of new recording techniques, very fast brain activity (>80Hz) has become a new research focus in the area of seizure genesis. Initially identified in animals, high frequency oscillations (HFOs) have recently been recorded in epilepsy patients and proposed as possible novel biomarkers of epileptogenicity. Investigating the characteristics of HFOs-those that correlate with the clinical manifestation of seizures-may yield additional insights for delineating seizure onset zones (SOZs) and epileptogenic regions of the brain. To that end, this dissertation explores the spatial and temporal coherence patterns of HFOs ( Hz) in the human brain under epileptic conditions, as the pathological activity observed during seizures is indicative of abnormal dynamic interactions between neuronal networks. The coherence of neuronal rhythms, although not a new idea, has not yet been explored in relation to HFOs, recently captured in the electroencephalogram (EEG) of epileptics. Similarly, the relationship between the spatiotemporal coherence patterns of slow and fast neuronal oscillations has yet to be explored. We retrospectively analyzed the intracranial EEG (ieeg) of patients undergoing pre-surgical evaluation of drug resistant extratemporal lobe epilepsy (ETLE). Wavelet phase coherence (WPC) analysis was applied to ieeg signals-sampled at 2000 Hz-to characterize the spatial and

3 iii temporal relationships of low-frequency oscillations (LFOs; < 80Hz) and HFOs ( Hz), during seizure (ictal) and non-seizure (interictal) activity. WPC analysis permitted the differentiation of ictal and interictal activity. Elevated HFO ( Hz) coherence was observed in consistent and spatially focused electrode clusters during seizures. Furthermore, cortical regions possessing elevated ictal HFO coherence coincided with regions exhibiting 1) high ictal HFO intensity (related to the square-root of the power) and 2) elevated interictal LFO coherence (in the 5-12 Hz frequency range), relative to all other electrodes on the implanted patient grids. As HFOs have been shown to localize to the epileptogenic zone, and we have demonstrated a correlation between ictal HFO intensity with both ictal HFO and interictal LFO coherence, we propose that LFO/HFO coherence can act as an epilepsy biomarker. Similarly, in comparing the more active HFO and LFO associated regions of the cortex with the clinical data, a positive surgical outcome was observed for patients in whom the clinically marked SOZs and/or surgically excised tissue were in close proximity to coherence highlighted regions of interest (ROIs). Recent research has highlighted the involvement of both fast and slow brain rhythms in pathology. Studies have demonstrated that the surgical removal of cortical areas generating HFOs correlated with positive seizure-free surgical outcomes. Similarly, the presence of strong theta activity (4-8 Hz) in awake adults has been linked to abnormal EEG activity. The results presented here suggest that elevated ictal HFO coherence and interictal LFO coherence are preferentially localized to brain regions generating seizures. The identification of epilepsy biomarkers is particularly valuable for ETLEs, which present a significant challenge for SOZ localization. In contrast to temporal epilepsies, which initiate in the temporal lobe, extratemporal seizures can originate in any of the other three lobes of the brain-a substantial area of the

4 iv remaining cortex-and are thus associated with lower seizure-free outcomes following surgery. As such, we propose that targeting the coherence of ictal HFO activity in the ( Hz) frequency range and/or interictal LFO activity in the (5-12 Hz) frequency range, may enhance the detection and delineation of seizure-related ROIs in the ieegs of patients with ETLE.

5 v Acknowledgments To my supervisors Dr. Berj Bardakjian and Dr. Peter Carlen for their scientific guidance and mentorship, and their ongoing support, encouragement and friendship. To Dr. Shokrollah Jahromi, for providing me with a research foundation and instilling a belief in my own abilities. To my supervisory committee, Dr. Konstantinos Plataniotis, Dr Milos Popovic and Dr. Liang Zhang, for your invaluable feedback on my work and for providing me with fresh outlooks on research problems. To my research collaborators, Dr. Hui Ye, Dr. Yotin Chinvarun, Dr. Martin del Campo and Dr. Damian Shin, thank you for all your help, time and knowledge. A special thanks to Dr. Hui Ye for your mentorship and friendship. To my fellow labmates-past and present-for their support and help, for all the fun times, in and out of the lab, and above all, their sincere and generous friendships. That means you Eunji Kang, Osbert Zalay, Sinisa Colic, Joshua Dian, Mirna Guirgis, Vasily Grigorovsky, Alan Chiu, Demitre Serletis, Sam Talasila, Daniel Jacobs, Vanessa Breton, David Stanely and Angela Lee. Special thanks to Eunji Kang, not only a much admired colleague, but invaluable friend. Thank you for all the sanity-restoring talks, memorable travels and shared companionship on this journey. To my friends, for their encouragement and love, and providing me with a refuge for my brain to relax and reboot. To my parents, siblings and family, for their unconditional love and support, for never doubting my capabilities and always encouraging me to perform at my very best.

6 vi Contents Acknowledgments...v Contents... vi List of Tables... ix List of Figures... xi List of Abbreviations...xv Chapter Introduction Motivation and Background The EEG Brain Oscillations Epilepsy Focal Epilepsy Presurgical Evaluation of Epilepsy Surgical Outcomes High Frequency Oscillations (HFOs) Physiological HFOs Epilepsy and Pathological HFOs (phfos) HFOs and the ieeg Objectives and Hypothesis Objectives and Contributions...25

7 vii Hypothesis Outline of Chapters...28 Chapter Methodology Human ieeg in the Epileptic Brain Patient Clinical Data, Electrode Placement and Electrode Description Intracranial EEG Acquisition and Referencing Classification of SOZs and Surgical Outcome Intracranial EEG analysis Time-Frequency Analysis Wavelet Transform Frequency Normalized Wavelet Transform Coherence Analysis Wavelet-based Phase Coherence Surrogate Test Estimation Statistical Analysis One-way Analysis of Variance (ANOVA) Tukey Honestly Significant Difference Test...49 Chapter Mapping the Coherence and Spectral Profiles of Ictal HFOs in Human Extratemporal Lobe Epilepsy Characterization of HFOs via wavelet intensity Characterization of HFO interactions within subdural grids via wavelet phase coherence Extracting HFO features from WPC matrices Clinical significance of HFO intensity and coherence mappings...77

8 viii 3.5 Discussion and Concluding Remarks...82 Chapter Extratemporal Lobe Seizure Localization: Comparison of Clinical SOZs, Ictal HFOs and Interictal LFO Activity Characterization of ictal LFOs via wavelet phase coherence Characterization of interictal LFOs via wavelet phase coherence Comparison of HFO-defined and LFO-defined ROIs Clinical significance of interictal LFO and Ictal HFO mappings Discussion and Concluding Remarks Chapter Results and Future Work Coherence in the Epileptic Brain Hypothesis Revisited Future Work References...128

9 ix List of Tables 1.1 Seizure outcomes according to surgery and age. From Hindi-Ling et. al., Engel outcome groups. From Valentin and Alarcon, Studies investigating physiological high frequency activity in humans Studies investigating pathological high frequency activity in human epilepsies Quantitative analysis techniques applied to the study of HFOs in epilepsy Quantitative characterization studies of HFO properties in epilepsy Patient clinical information SOZ identification and Engel outcome Intracranial EEG analysis techniques ieeg acquisition and analysis parameters Patient interictal ieeg, Sampling Rate = 2000 Hz Patient ictal ieeg, Sampling Rate = 2000 Hz Electrodes inside SOZs, resection areas and HFO-defined ROIs...81

10 x 4.1 Patient ictal ieeg, Sampling Rate = 2000 Hz Patient interictal ieeg, Sampling Rate = 200 Hz and 2000 Hz Suprathreshold electrodes for all patients (threshold = +4σ) Suprathreshold electrodes for all patients (threshold = +5σ)...103

11 xi List of Figures 1-1 Distribution of epilepsy etiologies by age. From (Bhalla et al., 2011) Typical scalp EEG and ieeg recording in a patient. Modified from Panayiotopoulos, Neural oscillations in the human ieeg ILAE Organization of Seizures and Epilepsies. Modified from The four sections of the brain: frontal, parietal, temporal and occipital lobes The epileptogenic zone. Modified from (Lozano et al., 2009) HFOs in clinical epilepsy The spatial placement, coverage and number of implanted ieeg electrodes are dictated by the size and location of the SOZ Example of subdural grid placement Subdural ieeg recorded frontal lobe seizure onset (patient 1) Subdural ieeg recorded frontal lobe seizure offset (patient1) Seizure and non-seizure segments analyzed for patient

12 xii 2-5 Frequency normalized wavelet transform Sequence of Processing Module Illustrative high frequency activity during interictal and ictal periods HFO intensity increases during seizures in select electrodes Normalized wavelet frequency distributions across electrodes and seizures Surrogate test estimation of significant WPC between all electrodes with electrode Surrogate test estimation of significant WPC between all electrodes with electrode Surrogate test estimation of significant WPC for one pairing from all patients High HFO WPC values were observed in select electrode clusters, during seizures for six of seven patients Strong HFO WPC values emerge in select electrode clusters during seizures HFO coherence increases during seizures in select electrodes Histograms of mean seizure HFO-WPC values for all electrode pairings Max WPC highlights regions of interest Average WPC before, during and following seizure onset Spatial locations of elevated HFO coherence and intensity during seizures...78

13 xiii 3-14 Mapping of clinically marked SOZs and HFO-defined ROIs Mapping of tissue resection areas and HFO-defined ROIs Increased coherence emerged in low and high frequency bands during seizures Average WPC before, during and following seizure onset for patient Average WPC before, during and following seizure onset for patient Strong WPC is present in slow frequencies during interictal activity Strong LFO WPC values emerge in select electrode clusters during interictal activity Mean LFO interictal activity for each patient Mean interictal LFO (5-12 Hz) WPC matrices and histograms Spatiotemporal patterns of average LFO and HFO coherence during interictal and seizure activity Spatial locations of cohered suprathreshold electrodes during seizure and non-seizure ieeg activity Mapping of tissue resection areas and LFO-defined ROIs Mapping of clinically marked SOZs of neurologist A with LFO/HFO defined ROIs...109

14 xiv 4-12 Mapping of clinically marked SOZs of neurologist B with LFO/HFO defined ROIs Mapping of tissue resection areas with LFO/HFO defined ROIs Spatial overlap of cohered interictal LFOs with cohered ictal HFOs highlight seizure related ROIs Only patients with refractory focal epilepsies benefit from the identification of HFOs in ieeg. From (Jacobs et al., 2012)

15 xv List of Abbreviations AED anti-epileptic drug ANOVA analysis of variance CWT continuous wavelet transform DF dorsolateral frontal ECoG electrocorticogram EEG electroencephalogram; electroencephalography ETLE extratemporal lobe epilepsy EZ epileptogenic zone F frontal FFT Fast Fourier transform FR fast ripple FT frontotemporal HFO high frequency oscillation

16 xvi HSD honestly significance difference IAFFT iterated amplitude adjusted Fourier transform ieeg intracranial electroencephalogram ILAE International League Against Epilepsy LFO low frequency oscillation MEG magnetoencephalography MRI magnetic resonance imaging MSE mean square error nhfo normal high frequency oscillation NT neocortical temporal O occipital P parietal PET positron emission tomography phfo pathologic high frequency oscillation ROI region of interest SOZ seizure onset zone SR sampling rate

17 xvii SUDEP sudden unexpected death due to epilepsy T temporal TF time-frequency TLE temporal lobe epilepsy WPC wavelet phase coherence WT wavelet transform

18 1 Chapter 1 Introduction "If epilepsy falls once upon a person or falls many times, it is as the result of possession by a demon or a departed spirit." Sakikku Translation (Wilson and Reynolds, 1990) The first mention of epilepsy-in written text-is attributed to a series of Babylonian clay tablets, encompassing a medical diagnostic series known as Sakikku, dating back to the first millennium BC. The description of epilepsy begins on tablet 26, and lists various symptoms and seizure types recognized today, yet lacks an understanding of disease pathology, as each seizure type is associated with a particular spirit or god, who is usually evil in origin (Wilson and Reynolds, 1990). The Babylonians believed an epileptic episode was the result of possession of the body by demons and ghosts. This idea reflected the widespread ancient belief, that seizures were of supernatural origin, perhaps due to the fact that seizures did not present with any external symptoms, other than sudden attacks, fits, convulsions and a loss of consciousness (Engel, 2013). This longstanding hypothesis was only briefly challenged by a physician of the school of Hippocrates in the fifth century BC, with the hypothesis that "the most acute, most powerful, and most deadly diseases, and those which are most difficult to be understood by the inexperienced,

19 2 fall upon the brain" (Jones et al., 1948). Five hundred years would pass before Galen put forward the idea that epileptic symptoms arose due to influences on the brain from external sources. However, this concept-that epilepsy resulted from some form of dysfunction to the brain-was almost generally ignored and did not re-emerge again until the eighteenth century (Engel, 2013). Bromide, the first successful anti-epileptic drug (AED), was introduced in 1857 by Charles Locock, who was attempting to treat hysterical seizures in women. Until then, epilepsy was treated through various means, including religious, occult and magical cures. It was common to observe Romans afflicted with epilepsy drinking human blood, believed to possess curative powers. During the Middle Ages it was of widespread practice to spit if one came in contact with an epileptic, in order to repel any chance of infection. Many Christians turned to saints to cure them of their epilepsy, with St. Valentine, the recognized patron saint of epilepsy, one of the most popular. Castration was largely in practice until the 19th century, in an attempt to prevent genetic transmission, as epilepsy was generally regarded to be a mental disorder (Engel, 2013). By the 19th century, several key individuals would offer significant contributions that altered epilepsy research and clinical practices. In 1849, Robert Todd, who was largely influenced by his contemporary Michael Faraday, was the first to propose an electrical basis for seizure discharges in the brain (Reynolds, 2007). Some two decades later, Hughlings Jackson hypothesized that seizures resulted from exaggerated chemical discharges in brain areas functionally related to the clinical symptoms observed during an epileptic episode (Reynolds, 2007; Engel, 2013). This led to the introduction of surgical interventions for the treatment of focal epilepsies by the early 1880's. Following closely was the invention of electroencephalography in 1924 by Hans Berger, which still remains one of the most central diagnostic tools in clinical neurophysiology. By 1929,

20 3 Berger published the first successful recording of a human electroencephalogram (EEG) and went on to describe various EEG features associated with seizures (Engel, 2013). Figure 1-1 Distribution of epilepsy etiologies by age. Epilepsy is a heterogeneous neurological disease of which multiple etiologies have been identified. For clinical purposes seizure etiologies are classified into one of three broadly defined categories: genetic, structural-metabolic and unknown. From (Bhalla et al., 2011). Following many research and technological developments of the 20th century, epilepsy is now considered a heterogeneous neurological disease that encompasses a broad category of seizure types and clinical features. In addition, multiple etiologies have been identified, highlighting a

21 4 large spectrum of risk factors associated with seizure incidence (Fig. 1.1). The underlying causes have been organized into three subgroups consisting of genetic, structural-metabolic and unknown etiologies (Berg et al., 2010). Age appears as a risk factor as well, with recent studies reporting the greatest incidence in the young and elderly (Bhalla et al., 2011; Engel, 2013). Individuals living with epilepsy experience seizures which greatly affect quality of life. As a seizure may come on suddenly and with no warning, epileptics who cannot achieve complete seizure control typically can't drive a car, or use dangerous machinery. Seizures may interfere or limit work or education opportunities. Epileptics can also injure themselves, drown, or suffer sudden unexpected death due to epilepsy (SUDEP). Current treatment therapies do not result in seizure control or freedom for all patients. Medication and surgery are the most common forms of treatment at present. However, antiepileptic drugs, when they work, are only effective in suppressing seizures-the symptoms of epilepsy-and not the underlying disease processes (Kwan and Sander, 2004), while surgical interventions are only currently available to a select subset of patients and does not guarantee long-term seizure freedom (Engel, 2013). The standard clinical practice for the diagnosis, identification and treatment of suspected seizures continues to rely on a visual analysis of the EEG by a trained specialist. These visual EEG inspections are subjective and qualitative estimates of seizure activity which may at times lead to incorrect or imprecise findings. Recent technological advancements now allow for the sampling of EEG activity that contain components which are too fast for the human eye to see, without first applying filters to the EEG data and/or increasing time and amplitude scales. These new EEG recording techniques, along with various algorithms and signal processing tools, present

22 5 neuroscientists and neural engineers the opportunity to study additional characteristics of EEG activity which may positively impact current therapeutic and diagnostic treatments of epilepsy. 1.1 Motivation and Background This thesis is motivated by the objective of quantitatively characterizing electrical activity in the EEG of the human epileptic brain, in order to identify potential biomarkers of epileptic tissue. More specifically, this characterization involved the study of the presence, as well as the interactions of various EEG signal components in the brain, both during and in between seizure episodes. A successful and reliable biomarker should be able to identify the epileptogenic zone in the brain and thus be used to support clinical diagnoses and treatment phases The EEG The EEG provides a measure of the mean electrical activity generated by different sites in the brain. More specifically, the EEG measures the electrical current that flows when a group of neurons in the brain fire together (Engel, 2013). This ability to provide a measurement of neuronal electrical activity makes the EEG an important clinical tool when evaluating brain conditions, as people with neurological disorders and diseases can display atypical patterns of brain activity. EEG recordings can be obtained from outside the head (scalp EEG) or from within the head, via the surgical implantation of recording electrodes (intracranial EEG; ieeg). Scalp EEG is a noninvasive method for patient evaluation, and provides a broad spatial coverage of brain activity, as the typical electrode placement, the system, involves the uniform distribution

23 6 Figure 1-2 EEG recordings are obtained from outside the head along the scalp (left) while ieeg recordings are acquired from within the head (right), either from the surface of the brain (ECoG) or from within deeper brain structures. Modified from (Panayiotopoulos, 2005). of recording electrodes across the entire scalp. Intracranial EEGs can be recorded from the actual surface of the brain (electrocorticogram; ECoG), as well as from electrodes inserted into brain tissue (depth electrodes) (Fig. 1-2), which greatly limits the effects of muscle artifacts compared to surface EEGs. While ieeg is recorded over smaller spatial scales, these recordings provide a more detailed representation of the underlying electrical activity, and allow access to deeper brain structures whose activity may not be detected by scalp EEG (Werz and Pita-Garcia, 2010) Brain Oscillations Beginning with the prominent alpha waves first identified by Berger in normal awake adults, several other neuronal electrical oscillations have since been identified in EEG recordings (Fig.

24 7 1.3) (Buzsaki and Draguhn, 2004). These rhythms have been clinically categorized in humans, with frequencies ranging at the lower extreme from 0.5 Hz - 4 Hz (delta) and at the higher from 250 Hz Hz (fast ripples). It was only once digital EEG became prevalent in the 1980's, that it became possible to record electrical activities > 70 Hz. With the advent of such new recording techniques, very fast activities have become a new focus in brain research. Initially identified in animals (Buzsaki et al., 1992), very fast frequencies (>80 Hz) were soon successfully recorded in humans (Bragin et al., 1999a). These fast rhythms are commonly referred to in the literature as Figure 1-3 Neural oscillations recorded in human ieeg. Rhythmic brain activity has been clinically classified into several frequency bands, ranging from slow delta oscillations (0.5-4 Hz) to fast ripples ( Hz).

25 8 high frequency oscillations (HFO) and are typically classified into two frequency bands: ripples ( Hz) and fast ripples ( Hz) (Zijlmans et al., 2012). Neuronal oscillations have been linked to and proposed to play a role in various physiological processes. Several studies have begun to correlate the oscillatory activity of the brain with such cognitive processes as memory, attention, sleep and consciousness (Ward, 2003; Buzsaki and Draguhn, 2004; Engel, 2013). In the pathological brain, neural rhythms similarly remain a strong focus as abnormal oscillatory patterns have been observed for several disorders, including epilepsy (Schnitzler and Gross, 2005). As oscillatory processes involve the large scale integration of distributed neuronal populations, the coupling and binding of various informationprocessing areas is required. As such, there is a general interest in studying both the presence and integration of neural activity, by examining the frequency spectra and interactions of neuronal oscillations recorded simultaneously at different sites in the brain. Rhythmic activity is being closely examined in relation to epilepsy, as it appears abnormally dominate during seizures (Buzsaki and Draguhn, 2004; Uhlhaas and Singer, 2006). Many studies have investigated the properties of slower rhythms (< 80 Hz) in epilepsy (Uhlhaas and Singer, 2006). The recent identification of HFOs in neuronal tissues generating seizures has highlighted their potential involvement in epileptic activity, necessitating a need for further characterization of these fast frequencies, including their interactions with other slower oscillations present in the brain.

26 Epilepsy Epilepsy is a common neurological disease, affecting people of all ages. Approximately 50 million people worldwide are epileptic. 1 Epileptics possess a pathological tendency for recurrent seizures, "transient occurrence[s] of signs and/or symptoms due to abnormal excessive or synchronous neuronal activity in the brain" (Fisher et al., 2014). Patient symptoms are very diverse, as seizure activity manifests due to different underlying causes, which have been recently categorized (by the International League Against Epilepsy; ILAE) into genetic, structural-metabolic or unknown causes (Berg et al., 2010). Seizure characteristics also vary, depending on where in the brain disturbances arise and/or spread to. The ILAE has newly classified the epilepsies according to their clinical phenomena, into three different groups: generalized, focal and unknown epilepsies (Fig. 1.4) (Berg et al., 2010). In generalized epilepsy, seizure activity is observed in both brain hemispheres at seizure onset, while seizures in localization-related focal epilepsies arise in a specific region of the brain. Regardless of etiology, seizures are generally believed to result from abnormalities in brain cell networks, resulting in unchecked electrical disturbances (Uhlhaas and Singer, 2006). 1

27 Figure 1-4 ILAE organization of seizures and epilepsies. Modified from 10

28 Focal Epilepsy Focal epilepsy has an incidence of approximately 60% (Rosenow and Luders, 2001; Panayiotopoulos, 2005) and is characterized by seizures that arise in a location-specific area of the brain. Seizure types are typically described according to their anatomical origin in the brain. Temporal lobe epilepsy (TLE) is the most common adult focal epilepsy, affecting approximately 50% of patients (Panayiotopoulos, 2005). Focal epilepsy is otherwise extratemporal-outside of the temporal lobe. This includes seizures originating from the frontal, parietal and occipital lobes (Fig 1-5). The quality of life of epileptics is dependent on the achievement of complete seizure control. AEDs are the primary means for controlling seizures. Up to 70% of newly diagnosed children and adults respond to AEDs (Engel, 2013). Focal epilepsies which do not respond to medication may be treated via secondary methods such as surgery. Figure 1-5 The four sections of the brain: frontal, parietal, temporal and occipital lobes

29 12 The most common type of surgery is resective surgery, where brain regions involved in seizure activity are removed. Resective surgery is an option for patients whose seizure onset-where the seizure starts-can be clearly identified, and whether or not the resection of said brain region will not result in the loss of critical cognitive functions. In TLEs, seizures consistently originate within the core structures of this bounded brain area, while extratemporal lobe epilepsies (ETLEs) are harder to localize, as seizures can originate in any of the other three lobes which cover a large area of the cortex (Fig 1-5). As a result, ETLEs are associated with lower seizurefree outcomes after surgery compared to TLEs (Table 1.1) (Hindi-Ling et al., 2011). Table 1.1 Seizure outcomes according to surgery and age. From (Hindi-Ling et al., 2011) Presurgical Evaluation of Epilepsy In order to achieve a seizure free state, with no side effects, a patient's epilepsy must be thoroughly characterized prior to surgery. This presurgical evaluation involves identifying the epileptogenic zone (EZ), the area of cortex that can generate epileptic seizures, and that if removed, would stop seizure activity (Rosenow and Luders, 2001). However, it is not possible

30 13 to definitively identify the EZ in advance of surgery. It is only possible to estimate the tissue boundaries of this area using a variety of diagnostic tools, amongst which include the EEG and neuroimaging techniques. The EEG aids in locating the seizure onset zone (SOZ), the cortical region from which seizure generation can objectively be measured. Seizure onset and offset times are identified by a neurologist using standard visual analysis of the raw EEG, which may include both scalp and intracranial recordings. SOZs are defined electrographically as the electrode(s) with the earliest seizure activity. The SOZ and EZ do not necessarily overlap, as clinical results have demonstrated both positive and negative results, in relation to seizure freedom, following the removal of SOZs. It is hypothesized that the EZ is a combination of the SOZ and a potential seizure onset zone (figure 1-6), as tissue areas in unresected potential SOZs have been shown to trigger seizures post epilepsy surgery (Luders et al., 2006). Figure 1-6 The epileptogenic zone. Modified from (Lozano et al., 2009).

31 14 Neuroimaging techniques are used to detect abnormalities of the brain, both in structure and function. Magnetic resonance imaging (MRI) is the preoperative imaging of choice for the discovery of morphological brain abnormalities or lesions. Such abnormalities, which include tissue scaring (sclerosis), vascular and developmental malformations and tumors, are investigated to determine their involvement in seizure activity. Similarly, positron emission tomography (PET) is a technique that measures the cellular activity in the brain. Areas of the brain with abnormal levels, high or low, may also point to the EZ (Werz and Pita-Garcia, 2010). While brain malformations may aid in the identification of the EZ, a larger number of patients present with unrelated brain irregularities or none at all. It is critical to identify the exact cortical region responsible for seizure generation, not only to allow for a positive surgical outcome, but also to prevent postoperative neurological deficits. Eloquent cortex describes brain tissue that, if removed, will result in a functional deficit, ranging from paralysis, losses in sensory processing to cognitive deficits. As a result, there is a fine balance in maximizing the excision of the EZ while minimizing that of the eloquent cortex during epilepsy surgery Surgical Outcomes A categorization for epilepsy surgery outcome was proposed by Engel in 1985, which groups the success of surgical treatments into four classes (I-IV) ranging from complete seizure freedom (group I) to no worthwhile improvement (group IV) (Table 1.2) (Valentin and Alarcon, 2012).

32 15 This scheme is still commonly used to rank post-operative patients. This thesis will employ the Engel outcome groups to classify the results of epilepsy surgery. Table 1.2 Engel outcome groups. From (Valentin and Alarcon, 2012) High Frequency Oscillations (HFOs) Both animal and human studies have revealed the presence of HFOs in the brain (Staba and Bragin, 2011; Buzsaki and Silva, 2012). HFOs are believed to possess broad-band or bandlimited spectral characteristics, and occur as transiently bursting or continuous events (Jefferys et al., 2012). Recent studies indicate that HFOs seem to be characteristic of both regular and pathological activity in the brain; these fast oscillations appear linked to various physiological functions of the normal brain (Buzsaki and Draguhn, 2004), yet have also been identified as probable epileptogenic biomarkers (Worrell and Gotman, 2011).

33 Physiological HFOs Physiological HFOs, hereafter referred to as normal high frequency oscillations (nhfos), have been identified in various brain structures during numerous task-related activities, including the motor cortex during finger movements (Huo et al., 2010), the temporal lobe during working memory tasks (Axmacher et al., 2008) and the somatosensory cortex during electrical stimulation of the upper limb (Curio et al., 1994). As nhfos have been observed in several cortical regions, and typically during sensory or motor events, it is hypothesized that this fast activity is an index of cortical processing itself, rather than a marker for any given brain region (Lachaux et al., 2012). Normal high frequency oscillations have been observed in relation to several complex cognitive tasks, including memory, language, attention and motor processes, as well as visual, auditory and somatosensory perception (Table 1.3). In general, studies have reported task-induced broadband spectral energy changes for various ranges of high frequencies (Table 1.3). These spectral power changes-increases and decreases-as well as the frequency band under analysis, have appeared time-locked to specific cognitive events (Jerbi et al., 2009; Lachaux et al., 2012). Also, the spatial organization of nhfos, which has been shown to similarly depend on the cognitive process under investigation (Jerbi et al., 2009), is characterized by a focal spatial distribution, in contrast to that of lower frequency activity (Buzsaki and Draguhn, 2004; Jerbi et al., 2009). As ieeg signals allow for the recording of brain activity with high temporal and spatial resolution, and at a high signal to noise ratio, recent studies investigating nhfos have largely focused on ECoG and depth recordings to study regular brain function. Due to the invasive nature of ieeg recordings, study subjects are usually derived from a pool of patients undergoing

34 17 Table 1.3 Studies investigating physiological high frequency activity in humans Study Site Task Electrodes Frequency (Hz) Number of Patients (Curio et al., 1994) somatosensory cortex median nerve stimulation scalp (Hashimoto et al., 1996) somatosensory cortex median nerve stimulation MEG (Crone et al., 1998) sensorimotor cortex visual-motor ECoG (Tanji et al., 2005) temporal lobe reading, naming ECoG (Lachaux et al., 2005) occipital, temporal and parietal visual depth lobes (Edwards et al., 2005) frontal, temporal lobes auditory ECoG (Brovelli et al., 2005) premotor cortex attention/memory depth (Vidal et al., 2006) occipital lobe visual MEG (Mainy et al., 2008) frontal and temporal lobes word recognition depth (Mainy et al., 2008) Broca's area and reading depth prefrontal cortex (Axmacher et al., 2008) hippocampus, entorhinal cortex memory consolidation depth (Wyart and Tallon-Baudry, 2008) visual cortex spatial attention MEG (Miller et al., 2009a) motor and visual cortex rest/active ECoG (Le Van Quyen et al., 2010) parahippocampal gyri sleep depth (Huo et al., 2010) primary motor cortex motor MEG (Miller et al., 2011) mouth motor area, Broca s area language ECoG (Ossandon et al., 2011) frontal and temporal lobes visual depth (Chang et al., 2011) left hemisphere auditory ECoG (Gaona et al., 2011) sensorimotor cortex and Broca's auditory, visual ECoG area (Bastin et al., 2013) parahippocampal gyri visual depth (Matsumoto et al., 2013) motor cortex, occipital lobe visual, motor depth, ECoG

35 18 invasive monitoring for epilepsy surgery. These patients may be implanted with ieeg electrodes, from 1-3 weeks, during the surgical preplanning phase, and are generally capable of performing complex cognitive tasks between clinical procedures. The number of subjects involved in a typical ieeg study is usually low compared to non-invasive recording methods such as scalp EEG or magnetoencephalography (MEG), as the number of patients undergoing ieeg monitoring for clinical purposes is usually small. In table 1.3 the studies using intracranial EEG recordings range in patient numbers from 1-20, with an average of 7.8 patients per study. Due to the high spatial selectivity of ieeg recordings, it is generally accepted that statistically significant results can be obtained from such small patient numbers Epilepsy and Pathological HFOs (phfos) HFOs were initially described in the hippocampus and entorhinal cortex of patients with TLE, using microelectrodes, (Bragin et al., 1999a; Staba et al., 2002) and have since been identified in patients with TLE and neocortical epilepsies, using depth and grid clinical electrodes (Jirsch et al., 2006; Urrestarazu et al., 2007). In studying the spatial distribution of spontaneously recorded phfos, it appears as though their presence is not widespread throughout the brain; rather, they have been shown to occur reliably in cortical areas generating seizures (Jacobs et al., 2008; Crepon et al., 2010). Since their detection, the relationship of HFOs in epilepsy has become a new area of focus in research (Table 1.4). Several subsequent findings have recently emerged (figure 1-7). HFOs have been found to associate well spatially with the SOZ(s), localized by neurologists during the

36 19 Table 1.4 Studies investigating pathological high frequency activity in human epilepsies Study Site Electrodes Frequency (Hz) Epilepsy Type No. Patients with HFOs (Bragin et al., 1999a) Hip, EC depth TLE 6/9 (Bragin et al., 1999b) Hip, EC depth TLE 6/9 (Staba et al., 2002) Hip, EC depth TLE 23/25 (Worrell et al., 2004) temporal, frontal depth, ECoG neocortical 6 (Staba et al., 2004) Hip, EC depth TLE 25 (Jirsch et al., 2006) temporal, frontal lobes depth TLE, neocortical 7/10 (Ochi et al., 2007) frontal, parietal, temporal lobes ECoG neocortical 9 (Staba et al., 2007) Hip, EC depth TLE 13 (Urrestarazu et al., 2007) frontal, parietal, temporal lobes depth, ECoG TLE 7 (Jacobs et al., 2008) frontal, parietal, temporal lobes depth, ECoG TLE, neocortical 10 (Worrell et al., 2008) Hip depth TLE 7 (Schevon et al., 2009) frontal, temporal lobes microwire array TLE, ETLE 4 (Jacobs et al., 2009b) all brain lobes depth, ECoG TLE, ETLE 12 (Zijlmans et al., 2009) frontal, parietal, temporal lobes depth TLE, ETLE 12 (Khosravani et al., 2009) all brain lobes depth, ECoG TLE 7 (Wu et al., 2010) all brain lobes ECoG focal, spasms 24/30 (Crepon et al., 2010) temporal, occipital, parietal lobes depth, ECoG TLE, neocortical 16 (Jacobs et al., 2010) all brain lobes depth TLE, ETLE 20 (Usui et al., 2010) frontal and parietal lobes ECoG neocortical 5 (Zijlmans et al., 2011) all brain lobes depth TLE, TLE 25 (Nariai et al., 2011) all brain lobes ECoG spasms 11 (Akiyama et al., 2011) all brain lobes depth, ECoG focal, spasms 28 (Modur et al., 2011) all brain lobes depth, ECoG neocortical 6 (Fujiwara et al., 2012) all brain lobes ECoG ETLE 41/44 (Haegelen et al., 2013) (Kerber et al., 2014) frontal, temporal, occipital lobes all brain lobes depth, ECoG ECoG TLE, ETLE TLE, ETLE 30 22

37 20 Figure 1-7. HFO characteristics in clinical epilepsy. pre-surgical planning phase. A focal increase of HFO rates has been observed in SOZs during seizures, compared to their presence in surrounding tissue (Jirsch et al., 2006; Modur et al., 2011; Zijlmans et al., 2011) and their presence has also been correlated with atrophic tissue, which is characteristic of some epilepsies (Ogren et al., 2009). Similarly, increased spectral power has been shown to characterize HFOs residing in the SOZ (Akiyama et al., 2006; Fujiwara et al., 2012). Of particular clinical interest are recent studies examining the relationship between HFOs and therapeutic treatments, both AEDs and surgery. Firstly, HFOs have been shown to increase

38 21 in number when medication was withdrawn (Zijlmans et al., 2009). Secondly, a positive correlation exists between the removal of tissue areas exhibiting increased ictal and interictal HFO rates with a favourable post-surgical outcome (Ochi et al., 2007; Jacobs et al., 2010; Wu et al., 2010; Akiyama et al., 2011; Modur et al., 2011; Fujiwara et al., 2012; Haegelen et al., 2013). While several groups have strived to automate the detection of HFOs from long recordings of ieeg activity (Worrell and Gotman, 2011; Kalitzin et al., 2012), HFO-based human studies (Table 1.4) can be divided into two subgroups focused on a) the quantitative analysis of HFOs and b) the qualitative characterization of HFOs. Table 1.5 lists the various analysis techniques which have been applied to the study of HFOs, and which include spectral analysis and crossfrequency coupling investigations, applied to both spontaneous and evoked HFOs. Table 1.6 lists the various studies examining the relationship between HFOs and various features of epilepsy, such as SOZs, brain lesions and AEDs. Table 1.5. Quantitative analysis techniques applied to the study of HFOs in epilepsy Analysis of HFOs Study Amplitude, Duration (Bragin et al., 1999b), (Usui et al., 2010) Power spectral analysis FFT based (Bragin et al., 1999a), (Staba et al., 2002), (Worrell et al., 2004), (Staba et al., 2004), (Jirsch et al., 2006), (Staba et al., 2007), (Fujiwara et al., 2012) Multiple band (Ochi et al., 2007), (Akiyama et al., 2006) frequency analysis Wavelet (Urrestarazu et al., 2007), (Crepon et al., 2010), (van 't Klooster et al., 2011) Line-length measure (Worrell et al., 2008) Cross-frequency coupling (Nariai et al, 2011), (Ibrahim et al., 2014) Relative phase clustering index (Kalitzin et al., 2012), (da Silva et al., 2005)

39 22 Table 1.6 Quantitative characterization studies of HFO properties in epilepsy Characterization of HFOs Spatial extent, classification and distribution (with respect to SOZs, resected tissue areas) Relationship between HFOs and ieeg spikes Relationship between HFOs and brain lesions Study (Schevon et al., 2009), (Wu et al., 2010), (Jacobs et al., 2010), (Zijlmans et al., 2010), (Akiyama et al., 2011), (Modur et al., 2011), (Haegelen et al., 2013) (Jacobs et al., 2008) (Jacobs et al., 2009b) Relationship between HFOs and AEDs (Zijlmans et al., 2009) Relationship between HFOs and tissue atrophy (Ogren et al., 2009) While many of the previously discussed studies strongly support the idea that HFOs are biomarkers of epileptogenic tissue areas, there is still much to learn about these fast brain waves and their potential role in epilepsy. Despite their presence in SOZs, it has been observed that in most patients the presence of HFOs is not limited to the ROIs but extends beyond (Jacobs et al., 2012). Furthermore, many studies have used HFO threshold rates to distinguish epileptogenic regions, but, it is not possible to define an absolute threshold rate, as they can be changed by such things as sleep (Staba et al., 2004; Bagshaw et al., 2009) or medication (Zijlmans et al., 2009), and appear to be brain region and/or patient specific (Jacobs et al., 2009a). Also, while Kerber et al. (Kerber et al., 2014) have hypothesized that transient, amplitude modulated HFOs represent pathogenic physiology, as compared to steady oscillations which are possibly related to normal activity, the ability to differentiate physiological HFOs from pathological HFOs is not straightforward, as their oscillations both reside in similar and/or overlapping, frequency bands (Engel et al., 2009). Thus, it is not completely clear which characteristics of this fast activity are clinically relevant and advocates the need for additional analysis techniques.

40 HFOs and the ieeg While the ieeg provides a very appealing avenue for the study of both nhfos and phfos, several challenges exist which presently limit and/or dictate the study of HFOs in relation to the brain: 1. As ieeg data is only available from a low number of patients undergoing surgery for the treatment of epilepsy, most of our information about human epilepsy and phfos is limited to a subset of focal epilepsies, which are both drug resistant and operable. Thus, the specificity of HFOs as a biomarker for all epilepsies still remains an open question. 2. Along the same lines, the acquisition of ieeg recordings from healthy humans, as a comparative measure, is not feasible due to the invasiveness of the measurement technique. Several groups have attempted to study nhfos in epileptic patients by recording in brain areas removed from the epileptic zone or by collecting data that is free of epileptiform activity. However, the potential effects of epilepsy on normal cortical networks may exist. 3. In clinical settings, ieeg is typically acquired at low sampling frequencies (i.e. as low as 250 Hz), which are not conductive to the analysis of fast EEG signal components. These low acquisition rates are a reflection of current standard clinical practices: the visual analysis of the EEG, for the evaluation of patients with suspected seizures, typically involves slower activities (< 70 Hz). This further limits the availability of human data sets for research purposes.

41 24 Figure 1-8 The spatial placement, coverage and number of implanted ieeg electrodes are dictated by the size and location of the SOZ, as identified during the pre-surgical planning phase. a: 2x8 electrode grid within the interhemispheric space; b: 2x8 electrode grid on the temporal lobe; c: 4x4 grid over the frontal and temporal lobes; d: 8x8 grid over the frontal lobe. From (Voorhies and Cohen-Gadol, 2013). 4. The spatial placement, coverage and number of implanted ieeg electrodes, as well as the duration of recording, are all clinically determined and dictated by the size and location of the SOZ, as identified by neurologists during the pre-surgical planning phase (figure 1-

42 25 8). As a result, the location (on/in the brain) and the amount of data collected differ from patient to patient. 1.2 Objectives and Hypothesis Objectives and Contributions Seizures are generally believed to result from abnormalities in brain cell networks, resulting in unchecked electrical activity that becomes hyper-excitable (Buzsaki and Draguhn, 2004). This thesis will study the characteristics of these electrical rhythms, particularly those of HFOs, under epileptogenic conditions. Using human ieeg signals recorded from implanted subdural grids, the spatiotemporal characteristics of various oscillations will be examined, in relation to their 1) spectral frequency properties and 2) phase interactions, during both non-seizure and seizure events. The phase interactions of neuronal activity, although not a new idea, has not yet been explored in relation to fast interactions in the human epileptic brain. SOZs that are visually identified by neurologists during the presurgical planning phase are only estimates of seizure foci and can be imprecise. The resulting spectral profiles and coupling patterns, of the rhythmic activity studied here, will be used to identify possible seizure-related ROIs in the epileptic brain. These results will be compared with the neurologist identified SOZs, as well as the results of tissue resection surgery, to assess the clinical relevance of the seizurerelated ROIs identified in this thesis and their potential to support the presurgical planning phase, when defining SOZ(s).

43 26 The contributions of this thesis are the following: 1. Identification of relevant frequency bands in relation to high frequency oscillations (HFOs) and extratemporal lobe epilepsy. While HFO activity has been typically classified into two frequency bands: ripples ( Hz) and fast ripples ( Hz) (Zijlmans et al., 2012), the definitions of ripple and FR frequency bands are not universal throughout the literature and overlap in frequency and across patients. More importantly, due to the power spectrum scaling properties of ieeg activity (Miller et al., 2009b), it is difficult to visually identify physiological HFO bandwidths from time-frequency distributions. As a result, to define relevant HFO frequency bands for patients with ETLE, a frequency normalized wavelet transform was applied and evaluated, defining relevant HFO frequency bands on a patient by patient basis. While HFO coherence and intensity changes were broadly observed in the (80-270) Hz frequency band, the bandwidth of frequencies possessing the strongest increases were selective to each patient. Thus, a more tailored (i.e. frequency specific) patient approach may prove beneficial for ETLEs, in contrast to other HFO studies which typically maintain rigid frequency bandwidths (Jacobs et al., 2012). 2. Investigation of high frequency coherence as potential biomarker in ETLE. The coherence metric has been closely examined in relation to epilepsy, as rhythmic activity appears abnormally dominate during seizures (Penfield and Jasper, 1954; Buzsaki and Draguhn, 2004; Uhlhaas and Singer, 2006). While the coherence of LFOs has been extensively studied, the coherence of high-frequency activities has yet to be explored. An efficient coherence measure must account for the low amplitude characteristic of faster brain activities (i.e. >80 Hz), in contrast to high-amplitude low-frequency activity.

44 27 Furthermore, HFO couplings are transient, dynamic and frequency-specific (Worrell et al., 2012). As classical coherence techniques do not allow for the separation of phase components from amplitude, we evaluated the performance of wavelet phase coherence (WPC) analysis (i.e. surrogate testing), a time-frequency sensitive tool for analyzing the phase-locking of HFOs. Furthermore, we investigated the correlation of the computed WPC profiles as a potential biomarker of epileptogenic tissue. WPC analysis permitted the differentiation of ictal and interictal activity and elevated HFO coherence was observed in consistent and spatially focused electrode clusters during seizures. In comparing these more active HFO associated regions of the cortex, with the clinical data, a positive surgical outcome was observed for patients in whom the clinically marked SOZs and/or surgically excised tissue were in close proximity. 3. Investigation of spatiotemporal coherence patterns of HFOs and LFOs. While the coherence of LFOs has been extensively studied in relation to epilepsy, the relationship between the spatiotemporal coherence patterns of LFOs and HFOs has yet to be explored. Here, we demonstrated a positive correlation between HFO and LFO defined ROIs. As HFOs have been shown to localize to the epileptogenic zone, and we demonstrated a correlation between ictal HFO intensity with both ictal HFO and interictal LFO coherence, this suggests that HFO/LFO coherence is preferentially localized to brain regions generating seizures. This result has broad clinical applications, as LFOs offer a more practical avenue as an epilepsy biomarker. LFOs satisfy the present clinical acquisition settings, are more accessible in scalp EEG and would remove the necessity for potentially harmful ictal recordings during clinical treatment.

45 Hypothesis The hypothesis of this thesis is the following: H 1 : The spatiotemporal interactions of low and high frequency electrocorticographic oscillations delineate seizure onset zones in epileptic patients. Specifically, phase coherence between subdural grid electrodes captures frequency-dependent pathological neuronal interactions in the epileptic brain. H 0 : Phase coherence between subdural grid electrodes show no differences in frequency specific neuronal interactions in seizure onset zones of the epileptic brain. 1.3 Outline of Chapters Chapter 1: provided the motivation and background for the study of electrical rhythms as possible markers of epileptogenic brain tissue. Chapter 2: will review the experimental protocols and analysis techniques used in this thesis. Chapter 3: will focus on the spatiotemporal characterization, intensity and coherence, of fast electrical rhythmic activity during interictal and ictal activity. Chapter 4: will present the spatiotemporal characterization of slow electrical oscillations during seizure and non-seizure activity.

46 29 Chapter 5: will discuss the clinical relevance and results of this thesis as well as the future directions for this work.

47 30 Chapter 2 Methodology 2.1 Human ieeg in the Epileptic Brain While the pre-surgical evaluation of epilepsy performed in most epilepsy centers is comprised of an extensive medical history and various imaging tests (ex. MRI, PET, MEG)-which search for structural pathologies-the standard clinical practice for the diagnosis, identification and treatment of suspected seizures continues to heavily rely on the conventional visual analysis of the ieeg by a trained specialist Patient Clinical Data, Electrode Placement and Electrode Description Due to the invasive nature of ieeg recordings, study subjects are usually derived from a small pool of patients undergoing invasive monitoring for epilepsy surgery. Thus, the availability and size of patient ieeg data sets are dictated by the clinical practices (ex. number of surgeries) implemented at epilepsy treatment centers. Accordingly, the number of subjects involved in a typical ieeg study (table 1.3) is usually low compared to non-invasive recording methods such as scalp EEG or MEG.

48 31 In clinical settings, the ieeg is typically acquired at low sampling rates (i.e. as low as 200 Hz), which are not conductive to the analysis of fast ieeg signal components. These low acquisition rates reflect current standard clinical practices, which typically focus on slower ieeg activity (<70 Hz). As a result, at present, the availability of ieeg data acquired at high sampling rates (>1 khz) is in low supply. In this thesis, we retrospectively analyzed the ieeg data of patients who underwent intracranial epilepsy monitoring conducted by the Thailand Comprehensive Epilepsy Program, at the Phramongkutklao Hospital (Bangkok, Thailand). Intracranial EEG data was collected from seven patients, (3 females and 4 males; Table 2.1) with intractable extratemporal lobe epilepsy. The institutional review board of the Phramongkutklao Hospital approved the study protocol and all patients gave informed consent. All patients underwent surgery for the placement of intracranial subdural grids (ECoG) to map their respective seizure foci. As it is neither safe nor practical to cover the entire surface of the brain with recording electrodes, the spatial placement, coverage and number of implanted ieeg electrodes are dictated by the size and location of the SOZs, which have been previously identified by the treating neurologist. As a result, the locations of implanted electrodes differ from patient to patient. This spatial constraint-regarding electrode coverage-exists for all human ieeg studies (table 1.3). Electrodes consisted of platinum disks embedded in a sheath. Subdural electrodes grids were configured from 8x8 electrodes for a total of 64 contacts (PMT, Chanhassen, MN, U.S.A.), and

49 32 Figure 2-1. Example of subdural grid placement for patient 2. Lateral and frontal view x-rays of one implanted grid containing 64 contacts. placed directly on the surface of the cortex (Note: 48 contacts were used for patient 7). Electrode contacts possessed a diameter of 3 mm and an interelectrode distance of 10 mm (center to center). For the patients (n=7) studied here, electrode coverage encompassed either the frontal, temporal, parietal and/or occipital lobes. The placement of a subdural grid electrode (patient 2) is depicted in figure 2.1. Patients exhibited both lesional and non-lesional epilepsies, as determined via MRI imaging. Patient clinical information is summarized in Table Intracranial EEG Acquisition and Referencing Digitized ieeg signals were sampled at 200 Hz and 2000 Hz (Stellate Systems, Montreal, QC, Canada). All recordings were referenced to an electrode located behind the ears, and a ground electrode placed on the forehead. The choice of reference electrode, when measuring ieeg

50 33 signals in the brain, is always an important issue, as the electrical potentials measured at different sites in the brain will be sensitive to fluctuations generated in the vicinity of the reference electrode. Many intracranial studies have used external reference electrodes (ex. linked ears) as described above, as they are located in a region with attenuated electric fields; however, external references may become contaminated by muscular artifacts. To address this issue, clinicians often employ an offline bipolar montage, which consists of taking the difference between two nearby electrodes, so as to cancel out any influence from the reference electrode. The 64-contact ieeg recordings studied here were reformatted offline in a bipolar arrangement in order to diminish noise and artifacts. The bipolar reformatting consisted of taking the difference between pairs of neighboring electrodes, thereby reducing the number of channels for analysis to 32. Electrodes possessing persistent artifacts were excluded from study. Electrical noise, mains interference and harmonics, were removed using finite impulse response notch filtering. Intracranial EEG signals were acquired with simultaneous video recordings of patient behavior to provide information on clinical seizure onset and offset times, as well as clinical manifestations during seizures. Intracranial EEG was recorded continuously, for long periods, during the presurgical planning period. The ieeg data that was analyzed in this thesis was reviewed and segmented from the hours of ieeg data mentioned above, by the treating neurologist (figures 2.2 and 2.3). Analyzed ieeg data consisted of selected segments of seizure and non-seizure recordings for each patient (figure 2.4). Seizure segments were comprised of a seizure episode, as well as (on average) one minute of ieeg leading up to and following the seizure, to allow for the study of ieeg activity immediately preceding and following seizures. Interictal activity was recorded during periods when patients (a) did not experience clinical seizures and (b) they were at rest and/or undergoing minimal movement during the analyzed interictal ieeg activity.

51 34 Table 2.1 Patient clinical information Patient Age/Sex Epilepsy duration (years) MRI findings Sites of intracranial recording Intracranial seizure onset (electrode N ) 1 36/F 20 Abnormal intensity lesion, perisylvian Left: FT 22, , /M 4 Cortical dysplasia Right: DF 17-19, 34, 35, 41, /M 16 Hippocampal atrophy; dilated perisylvian Right: NT, F /M 16 Non-lesional Left: FT 34-37, 43, 50, 52-54, /F 8 Non-lesional Left: P 15, 16, 23, /F 18 Non-lesional Left: O, P, T 1-4, 49-51, 57, /M 10 Non-lesional Left: F 19, 26-32, 33-40, DF: dorsolateral frontal, F: frontal, FT: frontotemporal, NT: neocortical temporal, O: occipital, P: parietal, T: Temporal

52 35 Figure 2-2 Subdural intracranial EEG recorded frontal lobe seizure (patient 1). Bipolar montage ieeg from 8 x 8 (64 contact) subdural grid (contacts 1-32 are displayed). Beta frequency oscillations of low amplitude at seizure onset (red arrow) involving electrode contacts 1 and 2 (red box).

53 36 Figure 2-3 Ictal offset of front lobe seizure displayed in Fig Seizure offset is indicated by a (red) dashed line. Background suppression was observed during post-ictal activity (red box).

54 37 Figure 2-4. Seizure and non-seizure segments analyzed for patient 1 (Note: time is not to scale). Mean duration of each segment: four minutes Classification of SOZs and Surgical Outcome The recorded ieeg data were independently reviewed off-line by two neurologists (neurologists A and B, Table 2.2) to clinically delineate the SOZs for all seven patients. SOZ identification (performed by both neurologists) was completed separately from the algorithms used in this thesis. The SOZs identified by both neurologists were defined electrographically as the electrode(s) with the earliest seizure activity. In addition, neurologist B was blinded to all clinical information available from the pre-surgical planning phase.

55 38 Six of the seven patients studied undertook epilepsy surgery. Brain tissue resection was limited to the areas subjacent to electrodes located in the electrographically defined SOZs (according to the SOZs defined by neurologist A). The algorithms applied in this study were not used for surgical planning. Patient 1 did not undergo surgery due to the close proximity of the SOZ to eloquent cortex. Patient 4 underwent a limited resection as a portion of the SOZ was also in close proximity to eloquent cortex (see SOZs and resected regions, Table 2.2). Each patient's surgical outcome was categorized according to Engel's classification (Engel and Rasmussen, 1993) as described in Table 1.2: a) class 1: free, b) class 2: rare disabling seizures, c) class 3: worthwhile improvement and d) class 4: no worthwhile improvement. Table 2.2 SOZ identification and Engel outcome Patient Intracranial SOZ (electrode N ) Pathology Engel Neurologist A Neurologist B Outcome 1 22, 23, 30, 31 1, 2 No resection No resection , 34, 35, 41, 42 32, 55-56, Normal Class IV , 47-48, Atypical neuron with reactive gliosis Class II , 43, 50, 52-54, , 40, 43-44, 58 Cortical microdysgenesis Class III 5 15, 16, 23, 24 undefined Cortical microdysgenesis Class II 6 1-4, 49-51, 57, , 49-51, 57, 58 Cortical microdysgenesis, atypical neuron with reactive gliosis Class I 7 19, 26-32, 33-40, , 20, 30, 31, 39, 40 Abnormal Class I

56 Intracranial EEG analysis The analysis techniques used in this thesis are listed in table 2.3 and expanded upon in the following sections. Table 2.3 Intracranial EEG analysis techniques Methodology Novelty Contribution Frequency Normalized Wavelet Transform (WT) Wavelet Phase Coherence (WPC) Iterated Amplitude Adjusted Fourier Transform (IAAFT) Previous studies have applied standard WT technique to study HFOs in human ieeg (table 1.5). Have implemented a frequency normalized WT. Never applied to ieeg activity ( >100 Hz) in humans None Identify relevant (patientspecific) HFO frequency bands, rather than using the 'static' HFO bands defined in literature (i.e Hz, Hz) Fast brain waves are typically of low amplitude; phase based measures can provide a measure of phase relations between signals independent of amplitude. Surrogate data technique to evaluate the significance level of WPC analysis Time-Frequency Analysis Wavelet Transform Time-frequency (TF) measures, such as the wavelet transform (WT) have been developed for the analysis of time varying signals, such as those recorded in the brain's EEG. The WT is able to

57 40 decompose a signal into a time and frequency representation. The continuous wavelet transform (CWT) is a measure of the correlation between a signal and a wavelet basis, ψ(t), termed the mother wavelet, for different scales, s, and time shifts, τ (Daubechies, 1988) and is defined as and where W ) ( s, ) x( t) s, ( t dt R, (2.1) * 1 t s ( t) o s s, (2.2) is the mother wavelet function and * denotes the complex conjugate. The mother wavelet taken here is the complex Morlet wavelet of the following form (Goupillaud et al., 1984): 2 1 t o ( t) exp( ict ) (2.3) 2 2 with angular frequency ω c = 5.1 rad/s corresponding to the maximum modulus of the Fourier transform of equation (2.3). In practice, the CWT is computed by convolution of the timesampled signal (which is finite and bandlimited) with scaled and translated wavelet functions, yielding a complex-valued coefficient matrix W( s, ) w( s, ) iw~ ( s, ), (2.4) with M rows corresponding to the number of chosen scales, and N columns equal to the length of x(t). The frequency localization of the wavelet transform scales inversely with s (Abry, 1997):

58 41 f c ~. (2.5) 2s Frequency Normalized Wavelet Transform Wavelet transforms were computed using the complex Morlet WT (section ). Timefrequency (t-f) spectrograms were constructed from the magnitude of the coefficients of the complex Morlet WT (related to the square-root of the wavelet power). Due to the power spectrum scaling properties of ieeg activity (Miller et al., 2009b), the wavelet coefficient magnitudes were standard normalized with respect to frequency, prior to further analysis: W W ( f, t) ( f ) norm( f, t) t 2 ( f ) t1 t2 t1, (2.6) where the variables and σ represent the mean and standard deviation of a baseline segment of wavelet coefficient magnitudes for each corresponding frequency scale f (figure 2-5) Coherence Analysis There is an interest in studying cohered brain activity by means of phase analyses, as in contrast to classical coherence techniques, phase coherence allows for the separation of phase components from amplitude for a given frequency or frequency range. Standard coherence techniques provide a measure of spectral covariance, but do not separate the effects of amplitude and phase, when measuring the relations between two signals. As a result, the measured coherence values cannot be directly ascribed to changes in either amplitude or phase

59 42 Figure 2-5. Frequency normalized wavelet transform (section ). Due to the power spectrum scaling properties of ieeg activity, the wavelet coefficient magnitudes were standard normalized with respect to frequency, prior to further analysis

60 43 Table 2.4 ieeg acquisition and analysis parameters Patient Sampling Rate (Hz) # of subdural electrodes Electrodes Electrode placement Left: FT Reference and ground electrodes Signal Processing Algorithms 2 200, Right: DF 3 200, Right: NT, F Platinum d = 3mm ied = 10mm c.c Left: FT Left: P Left: O, P, T Left: F Linked ears, forehead CWT: Complex Morlet wavelet 1:1:400 Hz Surrogate Test Estimation: 200 surrogates 1:1: 400 Hz WPC: Complex Morlet wavelet 1:1:400 Hz d: diameter; ied: inter electrode distance; c.c: centre to centre; DF: dorsolateral frontal, F: frontal, FT: frontotemporal, NT: neocortical temporal, O: occipital, P: parietal, T: Temporal

61 44 (Le Van Quyen and Bragin, 2007). The contribution of phase and amplitude to signal synchrony is an important issue when studying EEG signals, as two signals simultaneously recorded in the brain can possess a high degree of phase locking, yet display a parallel disparity in amplitude (Lachaux et al., 1999). Furthermore, as fast brain activities are typically associated with low amplitudes, phase-based measures are effective tools for the study of spatiotemporal relationships spanning the frequency domain in general and higher frequency activities in particular Wavelet-based Phase Coherence The method for obtaining a phase coherence measure from the continuous wavelet transform follows from the work of Hoke et al. (Hoke et al., 1989), Tass et al. (Tass et al., 1998), Mormann et al. (Mormann et al., 2000), Lachaux et al.(lachaux et al., 2002) and Li et al. (Li et al., 2007). The original signals are transformed into complex-valued signals by convolution with a complex wavelet (Zalay, 2012). For an arbitrary complex-valued signal s ( t ) y ( t ) i y ( ), the instantaneous phase angle is computed over the range [-π,π] as ~ t ~ y ( t) ( t) arctan y( t) (2.7) and the relative phase difference between two signals over the same range is expressed by the relationship (Rosenblum and Kurths, 1998; Mormann et al., 2000): ~ y ~ 1( t) y2 ( t) y1( t) y ( ) ( ) arctan ( ) ( ) ~ ( ) ~ 2 t t (2.8) y t y t y t y ( t)

62 45 Therefore, the relative phase difference between the complex wavelet coefficients of two signals, ), ( 1 s W and ), ( 2 s W, for different scales, s, and time shifts, τ, can be written as follows for s = s m : ), ( ) ~, ( ~ ), ( ), ( ), ( ) ~, ( ), ( ), ( ~ arctan ), ( m m m m m m m m m s w s w s w s w s w s w s w s w s (2.9) As the WT provides a measure of the correlation between the original signal and the wavelet functions, over time and frequency, the phase relationship between the wavelet coefficients of two separate signals, for a given scale, corresponds to the phase relationship between the two signals for the frequency represented by that given scale. Accordingly, the relative wavelet phase coherence (WPC) between two signals for a given scale (frequency), s, and time segment centered at time t=t k, and for sampling period Δt, is obtained as follows: )), ( exp( ), ( n m n m s i s (2.10) 2 / 2 / )), ( exp( 1) ( 1 N k N k j m t j s i N (2.11) The relative phase coherence varies between 0 (independent signals) and 1 (constant phase-lag between two signals). This enables the construction of an N M ( ) N N WPC matrix covering the time-frequency plane of the signals being compared: ), ( ), ( ), ( ), ( ), ( ), ( ), ( ), ( ), ( ), ( N M M M N N s s s s s s s s s s (2.12)

63 46 Fig 2-6. Sequence of Processing Module. Intracranial EEG was analyzed via a frequency normalized WT and the WPC algorithm. Due to the power spectrum scaling properties of ieeg activity, the wavelet coefficient magnitudes were standard normalized with respect to frequency (section ). The frequency distribution of the normalized WT and WPC distributions were further investigated to identify HFO-relevant frequency bands (chapter 3). Average intensity and WPC estimates yielded LFO/HFO defined ROIs (chapters 3 and 4) on patient subdural grids.

64 Surrogate Test Estimation When analyzing neurophysiological data, a statistical analysis can be performed to infer the significance level or confidence limits of the analysis, so as to distinguish physiological interactions from random coincidence. A common approach involves the generation of a chance distribution, derived from surrogate time series sharing the statistical properties of the original data under examination. The basic procedure of the surrogate technique is as follows: 1. Given two neural signals X and Y, a statistic of interest is calculated (i.e. wavelet phase coherence; WPC(X,Y)). 2. Resample signal Y, via a statistical method, to obtain a surrogate dataset Y'. Repeat N times to generate N shuffled versions of the original signal Y (i.e. Y i ', i = 1,..., N). 3. Compute the statistic of interest for each surrogate set (i.e. WPC(X, Y i ')) In the above procedure, the primary aim is to choose an appropriate re-sampling method for the statistic of interest. When investigating the phase relations between signals, the goal is to preserve the power spectrum of the original series while randomizing the phase. Non-linear resampling methods, such as the iterated amplitude adjusted Fourier transform (IAAFT) are suitable for estimating the confidence levels of estimators dependent on phase relationships. In order to evaluate the significance of the WPC profiles generated we performed surrogate data tests as discussed above. Phase-randomized surrogate signals were generated using the IAAFT method (Schreiber and Schmitz, 1996). The IAAFT algorithm conserves the amplitude distribution and reproduces the power spectrum of the original data fairly accurately. For a given

65 48 pairing of two ieeg signals, Ea and Eb, we generated 200 new surrogate signals for Eb. For each surrogate signal, Ebi' (i = 1:200), we measured the WPC between Ea and Ebi' for several time windows and across frequency scales. These 200 surrogate WPC profiles were used to estimate the significance of the WPC between the original signals Ea and Eb, by calculating the proportion of surrogate WPC values higher than the original WPC values between Ea and Eb (across time and frequency). Here we have used a criterion of 5% to characterize significant synchrony between pairs of ieeg signals. Thus, if the percentage of surrogate WPC values higher than the original WPC (between Ea and Eb) was less than 5% (i.e. 10/200), said WPC value was deemed significant Statistical Analysis One-way Analysis of Variance (ANOVA) One-way analysis of variance (ANOVA) is a statistical technique for determining if differences exist between the means of two or more groups. ANOVA compares the means between the given groups by testing the null hypothesis (Walpole et al., 1998): Ho : k (2.13) where represents the group mean and k the number of groups. If the one-way ANOVA rejects the null hypothesis H o, the alternative hypothesis is accepted, that is, that at least 2 group means are significantly different from each other. To determine which specific groups are statistically different, a post-hoc multiple comparison test is necessary.

66 Tukey Honestly Significant Difference Test Post-hoc testing is performed in conjuncture with ANOVA, and is used to determine which group means are statistically different from each other. To begin, each group mean is compared with every other group mean, such that all pairwise group mean combinations are determined. Next, the value of the difference between each pair of means is determined. Tukey's critical value, the honestly significant difference (HSD), is calculated from the following (Lane, 2010): MSE HSD q, where (2.14) N MSE k n i i1 j1 N k x x i 2 (2.15) where the variable q is dependent on the alpha level (the significance level for rejecting H o ), and the number of means in H o. MSE represents the mean square error, N the total number of observations in each group and k the number of groups. Each group mean is represented by x i, and the calculated mean for all observations from all groups is equal to x. To be significantly different, the differences in means being tested must exceed this critical value. The ANOVA and Tukey multiple comparison test were used to test for statistical significance in this thesis unless otherwise stated.

67 50 Chapter 3 Mapping the Coherence and Spectral Profiles of Ictal HFOs in Human Extratemporal Lobe Epilepsy With the advent of new recording techniques, very fast brain activities have become a new research focus in the area of seizure genesis. High frequency oscillations, in the lower frequency end (i.e Hz), were initially described in the intracranial EEG recordings of epilepsy patients in very localized tissue areas, assumed to represent the seizure onset zone, at the onset of focal seizures (Allen et al., 1992; Fisher et al., 1992; Worrell et al., 2004). The presence of HFOs in the ( Hz) frequency range was soon after described in the hippocampus and entorhinal cortex of patients with TLE (Bragin et al., 1999a; Bragin et al., 1999b; Fried et al., 1999; Bragin et al., 2002; Staba et al., 2002; Staba et al., 2004), and have since been identified in patients with ETLE and neocortical epilepsies (Jirsch et al., 2006; Schevon et al., 2009). Recent studies have proposed high frequency activity as a possible biomarker of epileptogenicity (Jacobs et al., 2012; Zijlmans et al., 2012). More importantly, several groups have demonstrated a correlation between the removal of HFO areas and a positive outcome following epilepsy

68 51 surgery (Ochi et al., 2007; Jacobs et al., 2010; Wu et al., 2010; Akiyama et al., 2011; Fujiwara et al., 2012). Despite these promising results, several challenges have also emerged when exploiting these fast activities as epilepsy markers. Firstly, HFOs have been recorded in the normal brain, where seizures are not observed, and it appears that spectral frequency alone cannot resolve physiological from pathological HFOs (Engel et al., 2009). Secondly, HFOs are not limited to SOZs, but can extend beyond as well (Jacobs et al., 2009a; Jacobs et al., 2012). Furthermore, studies have shown that HFO rates differ amongst both anatomical regions and patients (Jacobs et al., 2009a), creating thresholding problems for standardizing the identification of epileptogenic regions. Exploring additional HFO characteristics, particularly those that correlate with the clinical manifestation of seizures may yield additional insights for delineating SOZs and/or epileptogenic areas in the brain. In this chapter, we first explored HFO intensity profiles (as related to the square-root of the wavelet power), to characterize the span of relevant frequency bands during extratemporal lobe seizures and the manner in which these spectral profiles varied temporally and spatially, as well as across patients. Several groups have proposed spectral power changes as possible markers of the ictal state, and similarly, as identifiers of SOZs and epileptogenic tissue (Akiyama et al., 2006; Jirsch et al., 2006; Jacobs et al., 2008). Here, we examined frequency-normalized wavelet intensity profiles, as the frequency bandwidth of relevant HFO intensity changes are more easily visible after normalizing the ieeg time-frequency distribution. Along the same lines, the frequency bandwidths of different groups of HFOs (i.e. ripple and FR) may be more readily observed. Next we applied a wavelet phase coherence analysis to examine the coordinated interactions of high frequency oscillations. As seizures are characterized by pathologically entrained neuronal

69 52 activity (Uhlhaas and Singer, 2006), we examined the phase relations between HFOs, across implanted patient grids, during interictal and ictal activity. We also examined the correlation between HFO coherence changes with the SOZ and/or resected cortical tissue. The coherence of neuronal activity, although not a new idea, has not yet been explored in relation to fast interactions in the human epileptic brain. It is important to note that the phenomenon of phaselocking studied here, loosely referred to at times as coherence and synchrony in this thesis, differ from the classical measures of spectral covariance or coherence that have been addressed in section Characterization of HFOs via wavelet intensity Normalized time-frequency (TF) spectrograms, as outlined in section , were calculated for interictal and ictal ieeg segments for all patients. The seizure and non-seizure ieeg patient data were as described in section 2.1 and only the ieeg data sampled at 2000 Hz were analyzed in this chapter. Seizure segments were comprised of a seizure episode, as well as (on average) one minute of ieeg leading up to and following the seizure, to allow for the study of ieeg activity immediately preceding and following seizures (table 3.2). Interictal activity was recorded during periods when patients (a) did not experience clinical seizures and (b) they were at rest and/or undergoing minimal movement. The results presented in this chapter in regards to interictal observations refer to these ieeg segments (table 3.1).

70 53 Table 3.1: Patient interictal ieeg, Sampling Rate = 2000 Hz Table 3.2: Patient ictal ieeg, Sampling Rate = 2000 Hz Patient/interictal segment No. of channels analyzed Duration of recording Time from seizure onset/offset Patient/Seizure No. of channels analyzed Duration of recording 1/1 1/2 2/1 3/1 3/2 4/1 4/2 4/3 4/4 4/5 4/6 5/1 6/1 6/2 6/3 7/1 7/2 7/ seconds 240 seconds 240 seconds 60 seconds 60 seconds 24 seconds 20 seconds 31 seconds 21 seconds 39 seconds 42 seconds 240 seconds 326 seconds 221 seconds 128 seconds 120 seconds 120 seconds 30 seconds 6.4 minutes 14.8 minutes 5.8 hours 1.6 minutes 1.9 minutes 10.5 hours 6.4 hours 20.6 minutes 4.8 minutes 8.3 hours 8.8 hours 10.6 minutes sec. to min. sec. to min. sec. to min. 1.2 minute 0.33 minutes 0.7 minutes 1/1 1/2 1/3 2/1 2/2 3/1 3/2 4/1 4/2 4/3 5/1 5/2 5/3 6/1 6/2 6/3 6/4 6/5 6/ seconds 240 seconds 240 seconds 240 seconds 240 seconds 240 seconds 240 seconds 401 seconds 237 seconds 238 seconds 240 seconds 240 seconds 360 seconds 86 seconds 139 seconds 152 seconds 221 seconds 88 seconds 326 seconds 7/1 7/2 7/ seconds 504 seconds 178 seconds

71 54 Figure 3-1. Illustrative high frequency activity during interictal and ictal periods. A: ieeg from patient 1, bandpass filtered from Hz with the corresponding regular and normalized wavelet below (electrode 1). B: 2-second plot of ieeg activity (indicated by gray rectangle in A). C: 2-second plot of ictal activity from same electrode. High frequency events increased in number during seizures. D: 2-second plot of interictal activity from six electrodes on the same grid. Normalized wavelet intensity activity (in gray rectangle) at right. E: 2-second plot of ictal activity from six electrodes on the same grid. Normalized wavelet intensity activity (in gray rectangle) at right.

72 55 The wavelet coefficient magnitudes for each analyzed ieeg segment were standard normalized with respect to frequency, prior to further analysis. This involved calculating the mean and standard deviation of a baseline ieeg segment, for each corresponding frequency scale f. Values for each scale were calculated from a 20-s period of baseline ieeg activity. The baseline was chosen before the seizure, ending at least 30 seconds prior to the observation of the earliest seizure activity. The 20-s baseline was selected long enough to minimize the effect of transient events and to obtain a representative mean value of electrical activity at various frequencies. These normalized t-f distributions were also averaged across frequency to observe average frequency activity across time, as well as across frequency and time to observe average ieeg intensity activities during seizure and non-seizure activity. In general, the interictal ieeg was characterized by transient high frequency activity, which appeared as burst-like events, in relation to the lower-amplitude background activity (figure 3.1). During seizures, the incidence of high frequency bursts increased and persisted during the course of the seizure episode (Figure 3-1C). A representative trace of this typical interictal and ictal activity is depicted in figure 3-1. The frequency spread of high frequency activity was more clearly observed in the normalized wavelet distribution (Figures 3-1A, 3-2 B,C). These normalized intensity profiles highlighted HFO events, in relation to mean HFO coefficient magnitude, by which each frequency scale was self-normalized. It was observed that HFOs were not present in all electrodes during interictal and ictal activity (figure 3-1D, E). Figure 3-2 illustrates a normalized wavelet transform (WT) for patient 1, ranging from Hz, over 170 seconds of ieeg data. A single seizure episode is indicated by a dashed box (seizure onset and offset times were identified by a neurologist using standard visual analysis of the raw ieeg). Increase were evident in both low and high frequency bands during the ictal state.

73 Figure 3-2. HFO intensity increased during seizures in select electrodes. (Figure caption continued on next page). 56

74 57 Figure 3-2 (Continued) A: The ieeg activity recorded from electrode 1 (patient 1) along with the bandpass filtered HFO activity ( Hz) below. Corresponding wavelet (B; no normalization) and normalized (C) wavelet distributions of seizure episode from A. The maximum wavelet magnitude, Wmax, for electrode 1, was observed at seconds and 224 Hz (fmax). The HFO bandwidth was bounded at frequency values equal to 0.37Wmax, yielding an HFO frequency band of Hz for patient 1 (D1). The HFO ranges for patients 2-7 were similarly estimated and ranged from: Hz (patient 2), Hz (patient 3), Hz (patient 4), Hz (patient 5), Hz (patient 6) and Hz (patient 7). Wavelet coefficients were averaged across frequency for the respective patient HFO bands, and over 1- second windows, yielding time plots of average HFO intensity (D). Average HFO plots are shown for electrodes 1-4. A comprehensive exploration of all electrodes on the implanted subdural grids yielded regions of high-intensity ictal HFOs for all 7 patients (see rectangle at right). Units of normalized wavelets and average intensity are in standard deviations (SD) from the mean.

75 58 In general, during seizures, electrodes displayed the following spectral patterns: increases in: a) slow activity only, b) fast activity only and c) both slow and fast activity. Electrodes possessing high-intensity HFOs during seizures (i.e. channels with statistically significant HFO intensity), were further explored to elucidate the frequency spread of fast ieeg ictal activities. To characterize the relative HFO bandwidth or spread, we examined the normalized wavelet frequency profile to define frequency bounds located at 1/e or 37% of the peak value, on either side. While frequency bandwidths and peak frequencies varied in space and time and across seizures and patients, the defined HFO bandwidth for each patient was based upon the widest frequency range of HFO activity identified across all electrodes in the implanted grids and across all recorded time intervals (figure 3-3). In figure 3-2 (for patient 1), the maximum wavelet magnitude, Wmax, was observed at s and 224Hz (f max ). The HFO bandwidth bounded at frequency values located at 0.37Wmax (where Wmax is the peak normalized wavelet profile pertaining to f max ), yielded an HFO frequency band of Hz for patient 1. The HFO ranges for patients 2-7 were similarly estimated and yielded the following ranges: Hz (patient 2), Hz (patient 3), Hz (patient 4), Hz (patient 5), Hz (patient 6) and Hz (patient 7). The analysis of HFO frequencies (at maximum wavelet intensity) across all patients illustrated one continuous spectral band or distribution, suggesting the presence of one HFO frequency band for all patients (figure 3-3). Wavelet coefficient magnitudes were averaged across frequency, for the respective patient-defined HFO bands, and over 1-second windows, yielding time series related to electrode specific HFO activity. Electrode-specific averaged HFO activity plots are shown for electrodes 1-4 (patient 1, Figure 3-2). Select electrodes indicated strong increases in HFO activity during seizures. A comprehensive exploration of all electrodes on the implanted grids yielded regions of high-intensity ictal HFOs for all 7 patients (Figure 3-2), although the intensity of HFO activity

76 59 Figure 3-3. The normalized wavelet magnitude distributions for electrodes 1, 5, 6, 26 and 32 for seizure 1, patient 1. The normalized wavelet frequency distribution was plotted at the time corresponding to the incidence of each electrode's max HFO coherence (W max ). While frequency bandwidths and peak frequencies varied in space and time, across seizures, the widest HFO frequency band identified was chosen to capture the HFO spectral features of all electrodes on the implanted grids. The spectral bands plotted above suggest the presence of one HFO frequency band.

77 60 was observed to vary considerably across both electrodes and patients. The largest values were observed for patient 7, which ranged from and the lowest for patient 5, with a range of Characterization of HFO interactions within subdural grids via wavelet phase coherence The ieeg segments from the same seven patients were evaluated by wavelet phase coherence as outlined in section WPC was applied to every possible combination of electrode contacts from the implanted subdural grids. We calculated WPC, using the complex Morlet wavelet, for frequency scales f m, where f m ranged from 1 to 400 Hz in steps of 1 Hz. A moving window of (1/f m )*10 second duration was applied to each ieeg pairing, at each frequency scale f m. This yielded a WPC matrix (for each pairing) with 400 rows, equal to the number of frequency scales, and N columns, equal to the number of time windows. The window size was a function of the given frequency scale f m, and chosen large enough to contain several signal oscillations (i.e. 10), yet brief enough to reduce smoothing. To obtain average WPC profiles, these matrices were averaged across time or both time and frequency for the indicated frequency ranges and time windows. To examine the coupling between electrodes on the implanted grids, we calculated WPC between electrode pairings. When analyzing neurophysiological data, a statistical analysis can be performed to infer the significance level or confidence limits of the analysis, so as to distinguish physiological interactions from random coincidence. A common approach involves the

78 61 generation of a chance distribution, derived from surrogate time series, sharing the statistical properties of the original data under examination. When investigating the phase relations between signals, the goal is to preserve the power spectrum of the original series while randomizing the phase. Non-linear re-sampling methods, such as the iterated amplitude adjusted Fourier transform (IAAFT) are suitable for estimating the confidence levels of estimators dependent on phase relationships. In order to evaluate the significance of the WPC profiles generated, we applied a statistical method based on surrogate techniques, or resampling. Phase-randomized surrogate signals were generated using the IAAFT method described in section We used 200 surrogates for each WPC pairing, to generate a distribution for chance signal coherence. From this, we were able to estimate whether the coherence between electrode pairings resulted from random coincidence or from significant interactions. We used a criterion of 5% to characterize significant coherence between pairs of ieeg signals. Thus, if the percentage of surrogate WPC values higher than the original WPC (between the two signals under analysis) was less than 5% (i.e. 10/200), said WPC value was deemed significant. Only the phase coherence values (across time and frequency) that satisfied the indicated significance level were plotted (figures 3-4, 3-5, 3-6). Figure 3-4 and 3-5 illustrate the results of surrogate testing for one seizure from patient 1, where electrodes 5 and 15 were chosen as seed electrodes. That is, significant WPC is shown between all electrodes on the grid with two different seed electrodes: electrode 5 (figure 3-4) and electrode 15 (figure 3-5). Surrogate data tests were performed for all patients. One pairing from patients 1-4 are shown in figure 3-6. All four electrode pairings shown here demonstrated significant coherence at fast frequencies during the seizure episode, indicating that wavelet phase coherence is a reliable measure for identifying physiological interactions (A: patient 1; B: patient 2; C: patient 3; D: patient 4).

79 62 Figure 3-4. Surrogate test estimation of significant WPC between all electrodes on the grid with electrode 5. Only the WPC values that satisfied a 5% significance level were plotted. Illustrative ieeg activity is shown at top (seizure activity is indicated by shaded gray area). Electrode pairings in the lower left corner of the grid possessed the highest WPC values (colorbar shown at right).

80 63 Figure 3-5. Surrogate test estimation of significant WPC between all electrodes on the grid with electrode 15. Only the WPC values that satisfied a 5% significance level were plotted. Illustrative ieeg activity is shown at top (seizure activity is indicated by shaded gray area).

81 64 Figure 3-6. In order to evaluate the significance of patient WPC profiles, we applied a statistical method based on surrogate techniques, or re-sampling. Surrogate test estimations of significant WPC distributions from several patients is shown above. Two hundred surrogates, for each WPC pairing, were used to estimate chance signal coherence. That is, whether the coherence between electrode pairings resulted from random coincidence or from significant interactions. All four electrode pairings depicted demonstrated significant WPC at fast frequencies during the seizure episode, indicating that wavelet phase coherence is a reliable measure for identifying physiological interactions. A: patient 1 (electrodes 5 and 9), B: patient 2 (electrodes 27 and 28), C: patient 3 (electrodes 26 and 30), D: patient 4 (electrodes 4 and 16).

82 65 Wavelet phase coherence profiles were calculated for interictal and ictal ieeg segments, for all possible electrode combinations. The WPC profiles of fast activities did not reveal any spatial selectivity during interictal activity, for all patients. HFO ( >80 Hz) coherence was consistently transient and of weak to moderate strength during interictal activity, for all electrode pairs across the entire grids. In contrast, high HFO WPC values were observed in select electrode clusters, during ictal activity, in 6/7 patients (figure 3-7). No discernible differences were observed between the ictal and interictal HFO WPC profiles of patient 5. Closer electrode pairings typically exhibited stronger coherence. However, it was observed that the pairings which displayed the strongest coherence showed a preference that was not always indicative of the physical distance separating them on the electrode grid. That is, it became apparent that the strength of WPC was not dependent on just the distance between the electrode pairing, but whether the electrodes fell within the section of the grid later determined to possess the strongest HFO coherence (figure 3-8 for patient 1). Electrode pairs possessing high ictal HFO coherence (i.e. electrode pairings with mean HFO- WPC values greater than the indicated thresholds in figures 3-9, 3-10) during seizures were further explored, to elucidate the frequency spread of cohered ictal HFO activity. As observed with HFO intensity, electrode clusters possessing strong HFO coherence varied in bandwidth across patients. In figure 3-9 (for patient 1, electrode pairing 1 and 5), the maximum WPC was observed at s for 113Hz (f max ). The HFO bandwidth was bounded at frequency values located at 0.37WPC max, yielding an HFO frequency band of Hz for patient 1. The HFO ranges for patients 2-4 and 6-7 were similarly estimated and yielded the following ranges: Hz (patient 2), Hz (patient 3), Hz (patient 4), Hz (patient 6) and Hz (patient 7).

83 66 Figure 3-7. WPC profiles of ictal HFOs. High HFO WPC values were observed in select electrode clusters, during seizures, for 6/7 patients. An illustrative pairing is shown above for each patient. No discernible differences were observed between the ictal and interictal HFO WPC profiles of patient 5. Average WPC (over the entire segment) is displayed to the right for all patients, except patient 6, which was averaged over the ictal discharge (at ~3 seconds).

84 67 Figure 3-8. Strong HFO WPC values emerged in select electrode clusters during seizures. Five WPC profiles, for the indicated pairings, are shown above (paring indicated on grid at right) for patient 1. Close distances between electrode pairs do not always result in stronger WPC. While pairing E1-E5 exhibited strong WPC, pairing E7-E8 showed minimal HFO coherence.

85 Figure 3-9. Average wavelet phase coherence (WPC) matrices of HFO activity. (Figure caption continued on next page). 68

86 69 Figure 3-9 (Continued) A: The ieeg activity recorded from electrode 1 (patient 1) along with the bandpass filtered HFO activity ( Hz) below. B: Corresponding WPC distribution of seizure episode from A (electrode pairing: E1 and E5). The maximum WPC (WPCmax) for pair E1-E5 was observed at seconds and 113 Hz (fmax). The HFO bandwidth was bounded at frequency values equal to 0.37WPCmax, yielding an HFO frequency band of Hz for patient 1 (C1). The HFO ranges for patients 2-4, 6 and 7 were similarly estimated and ranged from: Hz (patient 2), Hz (patient 3), and Hz (patient 4), Hz (patient 6) and Hz (patient 7). C: WPC values were averaged across frequency, for the respective patient HFO bands, and over 1-second windows, yielding time plots of average WPC. Average WPC plots are shown for electrode pairings E1-E5, E2-E5, E6-E5, E9-E5 and E10-E5. A comprehensive exploration of all electrode pairings on the implanted subdural grids yielded average interictal and ictal matrices. For patient one, the strongest coherence was present in electrodes 5 and 9. Electrode pairings with average WPC higher than the indicated threshold are shaded in green. Electrode grids for patients 1-4, 6 and 7 are shown at right for all analyzed seizures.

87 70 Figure Histograms of mean seizure HFO-WPC values for all electrode pairings for one seizure (patients 1-4, 6, 7). An expanded portion of each histogram (x-axis: ) is shown at right to more clearly illustrate these electrode pairings. Vertical dotted lines indicate thresholds differentiating pairings situated more than three or four standard deviations from the mean. A comprehensive exploration of all electrode pairings on the implanted subdural grids, during ictal activity, yielded the spatial locations of strongly cohered electrode clusters. Similar to the method by which electrode-specific HFO activity was depicted, WPC values were averaged across frequency and time to generate an average matrix consisting of mean WPC strength for each electrode pairing. To isolate HFO activity, the averaging was completed using the patient-

88 71 defined HFO frequency bands. A 10-second window was used to quantify mean seizure and nonseizure HFO-WPC values for all patients except patients 2 and 6, where a shorter time window (i.e. 6 seconds for patient 2 and <1 second for patient 6) were chosen to accommodate the shorter seizure durations and corresponding briefer intervals of HFO coherence. An example of an average WPC matrix is shown in figure 3-9D1, for patient 1. (Note, as each matrix is symmetric, only half of each matrix is displayed for clarity). Electrode pairings with mean WPC values greater than the indicated threshold are highlighted. Thresholds were obtained from histograms of averaged ictal HFO-WPC values across all electrode pairings (figure 3-10), whereby the threshold chosen was greater than three or four standard deviations from the mean (figure 3-9 D2, D3). Electrode grids for all patients are shown at right with respective suprathreshold electrodes in green. For all seizures and patients, a threshold of +3σ or +4σ highlighted clusters of electrodes with elevated HFO coherence. 3.3 Extracting HFO features from WPC matrices Next, maximum and global HFO wavelet coherence were investigated. To obtain the maximum HFO coherence, WPC values were averaged across a defined HFO frequency band, for all electrode pairings on the grid, providing a single HFO coherence value for each pairing at each time interval. The maximum HFO coherence value, across all pairings, was identified for each time window and the spatial location of the corresponding electrodes were plotted. In figure 3-11 A1, for patient 1, the band ranged from Hz with a window size of t=0.033s. During ictal activity, the electrodes corresponding to the maximally cohered pairing belonged to a consistent electrode cluster (figure 3-11 A1, A2, C2). A similar trend was observed across all patients (Figure 3-11D). In contrast, max HFO WPC was weaker and spatially distributed during

89 Figure Max WPC highlights ROIs. (Figure Caption continued on next page). 72

90 73 Figure 3-11 (Continued) Maximum HFO WPC highlights ROIs. WPC values were averaged across (HFO) frequency, for all electrode pairings on the grid, providing a single HFO coherence value for each pairing at each time interval. The max HFO coherence value across all pairings was identified for each time segment and the spatial location of the corresponding electrodes were plotted. Spatial location of max WPC pairing across time for patient 1 are shown in A1. The 3D plot illustrates the spatial distribution of the max pairs and corresponds to the grid at right. During ictal activity, the majority of max WPC pairs appeared in a limited area of the grid (corresponding to the bottom left corner of the grid). A2: Electrodes identified in the maximally cohered HFO pairings for the indicated time windows. B: The bandpass filtered HFO activity ( Hz) for electrode 5. C1, C2, C3: The three grids illustrate the distribution of electrodes involved in the maximally cohered pairing across time at three separate time intervals of 20-s duration. The two interictal intervals illustrate the absence of any spatial selectivity for strongly cohered HFO activity during interictal activity, in contrast to the ictal grid in C2. D: Grid distribution of electrodes (involved in the maximally cohered pairing across time) for patients 2-4 during interictal and ictal segments (time interval = 5 seconds).

91 74 interictal activity. The 3D plot in figure 3-11A illustrates the spatial distribution of the max pairs for patient 1 and corresponds to the grid at right. During seizure activity, the majority of max WPC pairs appeared in a limited area of the grid (corresponding to the bottom left corner of the grid). In figure 3-11 A1 and A2, the distributions of electrodes identified in the maximum pairs, for the indicated time windows, illustrates the temporal changes in strongly cohered electrode pairings during the seizure. High and spatially focused HFO coherence was consistently observed to occur only after seizure onset, where seizure onset was defined electrographically from the electrode(s) with the earliest seizure activity. For example, in figure 3-11, ictal onset occurred at 101 seconds, while the earliest signs of spatial selectivity appeared at 115 seconds (figure 3-11, A2). In figure 3-11 (C1-C3), the coloured grids illustrate the distribution (i.e. counts) of electrodes involved in the maximally cohered pairing across time, at three separate time intervals of 20-s duration. The two interictal intervals illustrate the absence of any spatial selectivity for strongly cohered HFO coherence during interictal activity, as the entire grid is active. In contrast, the ictal grid showed maximal activity in a small cluster of electrodes. Global HFO WPC was computed to qualitatively characterize the spatiotemporal coherence patterns of HFO activity. WPC values-in the patient-defined HFO frequency bands-were averaged in space (across all possible electrode pairs), and in time (within 1-second windows), yielding a global WPC value for each electrode. These global WPC values were arranged in the same spatial layout as the subdural grid electrodes. Global WPC values for electrode contacts from patients 1and 3 were computed over the range of Hz and Hz respectively (figure 3-12). Fifteen successive windows are shown for various segments of the plotted ieeg

92 75 Figure Average WPC before, during and following seizure onset. (Figure Caption continued on next page).

93 76 Figure 3-12 (Continued) Average HFO WPC highlights ROIs. Time and frequency averaged HFO WPC matrices were computed for each electrode using all possible pairings (for patients 1 and 3). WPC was averaged for the Hz and Hz frequency band, for patients 1 and 3 respectively, across 1-second time windows. Fifteen successive 1-second windows are shown for various segments of the plotted ieeg activity. Strong average ictal HFO coherence increased in a select area of the grid, briefly expanded to other electrodes, before returning to highlight a small number of electrodes. These mappings highlight HFO regions of interest in time and space during seizures.

94 77 activity (before, during and following seizure onset). It was observed that global HFO coherence increased and remained highest in a limited area of the grid during seizure activity. As averaging was performed over all possible electrode combinations, weaker or transient connections were smoothed out while stronger, consistent connections remained. 3.4 Clinical significance of HFO intensity and coherence mappings The spatial locations of electrodes exhibiting strong high-frequency coherence (i.e. WPC > threshold; figure 3-9) and statistically significant high-frequency activity (figure 3-2) are mapped as HFO-defined ROIs in figures 3-13 and Comparisons of mean HFO intensity (between electrodes) were done using a one-way ANOVA, followed by Tukey's multiple comparison test (alpha = 0.01). In general, regions exhibiting strong HFO coherence were in agreement with regions of high intensity HFOs. Only the clinical SOZ electrodes, identified by neurologist A, pertained to the cortical areas that were surgically excised in each patient. Post-surgical patient outcomes were as follows: class I for patients 6 and 7, class II for patients 3 and 5, class III for patient 4 and class IV for patient 2, where class I is the best and class IV the worst surgical result (Table 3.3). The best surgical outcome was observed when the patient SOZs and HFO-defined ROIs were in close proximity. Analogously, in cases where SOZs and ROIs were completely incongruent, a poorer surgical outcome was obtained.

95 78 Figure The spatial locations of electrodes exhibiting strong high-frequency coherence (i.e. WPC > threshold; figure 3-9) and statistically significant high-frequency activity (figure 3-2) are mapped to patient grids above. Electrodes on the patient grids possessing elevated HFO coherence during seizures demonstrated a topographical overlap with areas possessing elevated HFO intensity during ictal activity.

96 79 Figure Mapping of clinically marked SOZs and HFO-defined ROIs. It was observed that the best surgical outcome resulted when the neurologist defined seizure onset zones (SOZs) and HFO defined regions of interest (ROIs) were in proximity.

97 80 Figure Mapping of tissue resection areas and HFO defined ROIs. Only the clinical SOZ electrodes, identified by neurologist A, pertained to the cortical areas that were surgically excised in each patient. The best surgical outcome was observed when the resected areas and HFO defined ROIs were in close proximity. For patients with Engel class I and II outcomes, several HFO-defined electrodes were found to lie within and/or proximal to excised tissue, while the electrodes of patients with Engel class III or IV outcomes were observed to lie farther from HFO-defined electrodes. The areas of surgical resection covered additional electrodes (compared to the SOZs in Fig 3-15), unless the SOZ was found to be in proximity to functional cortex.

98 81 Table 3.3 Electrodes inside SOZs, resection areas and HFO-defined ROIs Patient Electrode Location SOZs of neurologist A ( N ) Electrodes with high ictal WPC (N ) Electrodes with high ictal intensity (N ) Resected electrodes (N ) Type of Surgery Surgical Outcome Pathology 1 L F-T 22, 23, 30, 31 1, 2, 9, 10, 17, 18, 25, , 9, 10, 17, No resection (proximity to functional cortex) R D-F 17-19, 34, 35,41, , 63, 64 61, , 34, 35,41, 42 R F lesionectomy Class IV Normal 3 R NT-F , , R NT-F corticectomy Class II Atypical neuron with reactive gliosis 4 L F-T 34-37, 43, 50, 52-54, , 8, 13, 14, , , 43, 52, 53 L F-T corticectomy (incomplete resectionproximity to functional cortex) Class III Cortical microdysgenesis 5 L P 15, 16, 23, 24 ALL 21, 22, 29, 30, 37, 38 15, 16, 22-24, 30-32, 37, 38 L P corticectomy Class II Cortical microdysgenesis 6 L O-P-T 1-4, 49-51, 57, , , 50, 55, 56, 59, , 49-51, corticectomy Class I Cortical microdysgenesis, atypical neuron with reactive gliosis 7 L F 19, 26-32, 33-40, , 24, 31, 32, 39, 40, 47, 48 21, 22, 31, 32, 39, 40, 47, , 25-30, 33-38, Anterior lobectomy Class I Cortical dysplasia L: left, R: right, DF: dorsolateral frontal, F: frontal, FT: frontotemporal, NT: neocortical temporal, O: occipital, P: parietal, T: temporal

99 82 We compared our HFO mapping with the SOZs observed by the treating neurologist (A) and a second neurologist (B), who was blinded to all clinical and surgical outcome information. In patients 1 and 2, HFO-defined ROIs were in agreement with the SOZs of neurologist B, while the SOZs of patients 3 and 4 were in close proximity. Neurologist B determined that the brain region responsible for seizure onset was not discernible in the ieeg of patient 5 and concluded that the SOZ originated in an area not covered by the implanted grid. Visually demarcated SOZs are subjective and coarse estimates of the true SOZ or epileptogenic zone. Seizure freedom, post surgery, is what provides evidence that the entire epileptogenic zone was included in the resected areas. In comparing the HFO-defined ROIs and resected tissue areas, the best surgical outcome was similarly observed when the resected zones and HFOdefined ROIs were in close proximity. More importantly, for patients with Engel class I and II outcomes, several HFO-defined electrodes were found to lie within and/or proximal to excised tissue, while patients with an Engel class III or IV were observed to lie farther from HFOdefined electrodes. The areas of surgically excised cortex were generally larger than the SOZs identified by neurologists, unless the SOZ was found to be in proximity to functional/eloquent cortex (figure 3-14). 3.5 Discussion and Concluding Remarks In the human epileptic brain, several groups have studied the presence of HFOs and their spectral changes as possible classifiers of the ictal state, as well as possible electrophysiological identifiers of seizure-related ROIs in epileptic tissue (Akiyama et al., 2006; Jirsch et al., 2006; Jacobs et al., 2008; Jacobs et al., 2012). Moreover, several groups have recently demonstrated

100 83 that the surgical removal of regions generating ictal HFO increases, correlate positively with a seizure-free post-surgical outcome (Ochi et al., 2007; Jacobs et al., 2010; Wu et al., 2010; Akiyama et al., 2011; Modur et al., 2011; Nariai et al., 2011; Fujiwara et al., 2012). In Wu et al., a two stage epilepsy surgery was performed, whereby seizure freedom was only achieved following the complete removal of HFO generating tissue in the second surgery (Wu et al., 2010). While a great potential for HFOs may be inferred from such results, problems yet linger in the concrete identification and characterization of pathological HFOs. For example, HFOs do not appear limited to ROIs but extend beyond. Furthermore, many studies have implemented HFO rates to distinguish epileptogenic regions. However, it is not possible to define an absolute threshold rate, as high-frequency activities can change during sleep (Staba et al., 2004; Bagshaw et al., 2009) or from the use of medication (Zijlmans et al., 2009), and appear to be brain region and/or patient specific (Jacobs et al., 2009a). As such, additional characterization of HFO activity may complement these existing techniques, while yielding further insights into the workings of fast oscillations in the epileptic brain. In this chapter we applied wavelet phase coherence analysis and studied frequency normalized wavelet coefficient changes to investigate whether certain electrodes demonstrated strong HFO coherence and/or high-intensity HFOs, relative to all other electrodes on the implanted grids, during extratemporal lobe seizures. We speculated that regions of strong HFO coherence identified epileptogenic networks, which are thought to possess a pathological locking nature, in relation to regular brain activity. While phase coherence analyses have been applied to slower rhythms (further described in Chapter 4), the phase-locking properties of HFOs in the epileptic brain have yet to be explored.

101 84 In this study we observed clusters of electrodes possessing both high intensity HFOs and strongly cohered HFOs during seizures, relative to all other electrodes on the implanted grids. Several groups have identified increased HFO rates and spectral power changes during seizures (Akiyama et al., 2006; Jirsch et al., 2006; Jacobs et al., 2008; Jacobs et al., 2012), and we observed spatially coinciding areas possessing strongly cohered electrodes with simultaneous increases in HFO intensity. This result suggests that strong HFO coherence may be used to distinguish epileptogenic regions. Furthermore, although both HFO intensity and HFO coherence identified similar regions on the implanted grids, the coherence measure demonstrated less variability across patients in the choice of threshold, which was used to highlight electrodes displaying significant HFO activity. It should be noted that HFO intensity changes were more readily visible after normalizing the ieeg time-frequency distributions. As a result, it was observed that even though both HFO coherence and intensity changes were broadly observed in the Hz frequency band, the bandwidth of frequencies possessing the strongest WPC and intensity increases were selective to each patient. A more tailored (i.e. frequency specific) patient approach may prove beneficial for ETLEs, as in general, extratemporal seizures are associated with lower seizure-free outcomes after surgery compared to temporal lobe resections (Spencer and Huh, 2008). ETLEs are harder to localize, when compared to TLEs, as extratemporal seizures can originate in any of the other three lobes of the brain, which cover a large area of the cortex. In a recent study by Haegelen et al. (Haegelen et al., 2013), the authors concluded that while the removal of HFO-generating regions appeared to lead to improved surgical outcomes in TLE, less consistent findings emerged for ETLE. Normalizing the time-frequency profiles of patients may allow investigators to more precisely isolate HFO bands on a patient-by-patient basis, to more accurately identify seizurerelated ROIs.

102 85 In comparing the more active HFO-associated regions of the cortex with the clinical data, a good surgical outcome generally resulted for patients in whom the resected tissue (including SOZs identified by neurologist A) was in closer proximity to HFO-defined regions of interest (i.e. patients 5 and 6 in figure 3-15). For patient 7, the SOZs of three seizures were included in the resected area. However, as only one seizure showed a SOZ on the posterior part of the grid (i.e. contacts 31, 32, 38-40), neurologists included only contact 38 in the resection zone. As with patient 6, patient 7 also possessed an Engel class 1 outcome. However, it is important to note that the patient had status epilepticus (i.e. state of persistent seizure), which may have affected the analysis (ex. irritative zone from status epilepticus) and that patient 7 was the only patient to undergo a lobectomy, compared to the corticectomies performed for all other patients. It is also of interest to note that the SOZs of neurologist B-who was blinded to all patient data-colocalized with the HFO ROIs for patient 7. Tellingly, the patients who had poorer outcomes had either clinically-defined SOZs whereby there was disagreement between neurologist A and B (figure 3-14, patient 2), in which case the HFO-defined ROI was in agreement with one neurologist, or ROIs that were in an area not adjacent to the clinically-defined SOZs of either A or B, such that a poorer outcome occurred despite agreement of the two neurologists (figure 3-13, patient 4). In both cases, this supports the notion that the resected tissue did not contain all of the epileptogenic focus or foci, in accordance with there being areas of seizure-related HFOs identified by our HFO analyses that were not removed. While it is important to note that the number of patients included in this study was small (and therefore the clinical significance of our results is limited in this regard), we propose the idea that the proximity of HFO-defined regions to excised tissue possibly predicts a positive post-surgical

103 86 outcome. Similarly, we speculate that the removal of all areas identified by coherence and/or intensity measures might have resulted in better outcomes, as several studies have demonstrated good seizure-free post-surgical outcomes resulting from HFO-guided resections (Ochi et al., 2007; Jacobs et al., 2010; Wu et al., 2010; Akiyama et al., 2011). This idea is supported by the spatial overlap observed here between strongly cohered electrodes with simultaneous increases in HFO intensity, as HFO intensity (as related to power) has already been shown as an effective biomarker of SOZs in the literature (Crépon et al., 2011). It has also been observed that the removal of tissue generating HFOs was apparently predictive of a better surgical outcome, when compared to the excision of tissue based solely on SOZs (Jacobs et al., 2010). HFO activity has been commonly classified into two frequency bands: ripples ( Hz) and fast ripples ( Hz) (Zijlmans et al., 2012). Furthermore, both normal and pathological ripple and fast ripple activities have been recorded in the human brain (Engel et al., 2009). It has been suggested that faster activities are generated by the hippocampus, while neocortical structures tend to generate HFOs in the ripple range (Jacobs et al., 2008). The seizures studied here were recorded from patients with ETLE, and minimal coherence and intensity increases were observed in activity > 270 Hz. Therefore, when defining seizure-related ROIs for ETLE, targeting ripple activity may be more relevant. Furthermore, HFO features such as the average and maximum coherence, which retain spatial and temporal information, many enhance the detection and delineation of seizure-related ROIs.

104 87 Chapter 4 Extratemporal Lobe Seizure Localization: Comparison of Clinical SOZs, Ictal HFOs and Interictal LFO Activity In clinical settings, EEGs are typically acquired at low sampling frequencies (as low as 200 Hz), which is not conductive to the analysis of fast EEG signal components. Consequently, a large amount of literature, with regards to the interactions of neuronal electrical oscillations, has focused on slower rhythms (i.e. < 100 Hz). Coordinated neuronal activity has been observed within and across brain areas, during both physiological and pathological processes (Schnitzler and Gross, 2005; Uhlhaas and Singer, 2006; Uhlhaas et al., 2009; Valderrama et al., 2012; Ahmed and Cash, 2013). Physiologically entrained oscillations have been proposed to be involved, during both sleep and wakefulness, in memory, perception and attention processes, as well as during the communication between brain regions (Steriade et al., 1993; Buzsaki and Draguhn, 2004; Montgomery et al., 2008; Fell and Axmacher, 2011; Alvarado-Rojas and Le Van Quyen, 2012; Miller, 2013). In addition, correlations between neuronal synchrony and pathological brain states have been observed in individuals with impaired cognition, for several neurological disorders, including epilepsy (Niedermeyer and Lopes da Silva, 2005; Schnitzler and Gross, 2005; Uhlhaas and Singer, 2006).

105 88 Several studies have investigated the coherence of slower rhythms (<100 Hz) during seizures in epileptic patients, demonstrating a correlation between enhanced coherence and the ictal state (Duckrow and Spencer, 1992; Mormann et al., 2000; Le Van Quyen et al., 2001; Uhlhaas and Singer, 2006). A subset of coherence studies have also analyzed low-frequency activity during the interictal EEG, demonstrating both elevated (Mormann et al., 2000; Le Van Quyen et al., 2005; Schevon et al., 2007) and reduced (Chavez et al., 2003) coherence, in select frequency bands, during activity separate from epileptiform disturbances. Whereas numerous groups have explored the coherence of low-frequency activity, in relation to epilepsy, little is known about the relationship between low frequency oscillations (LFOs; <80Hz) and high frequency oscillations (HFOs; Hz). Only recently have high frequency oscillations been recorded in the intracranial EEG (ieeg) of epilepsy patients (Bragin et al., 1999b; Bragin et al., 2002; Staba et al., 2002) and several groups have proposed high-frequency activity to be a probable biomarker of epileptogenicity (Worrell and Gotman, 2011; Jacobs et al., 2012). New studies have also suggested that atypical LFO-HFO interactions characterize epileptic tissue (Alvarado-Rojas et al., 2011; Cotic et al., 2011; Nariai et al., 2011; Nagasawa et al., 2012; Guirgis et al., 2013; Ibrahim et al., 2014). As the seizure state is associated with excessive neuronal entrainment (Engel, 2013), coherence techniques present as an attractive approach for the study of coupling patterns in various frequencies in the epileptic brain. Given that several studies have explored phase coherence in the epileptic brain, in relation to slower rhythms (Duckrow and Spencer, 1992; Mormann et al., 2000; Le Van Quyen et al., 2001; Chavez et al., 2003; Le Van Quyen et al., 2005; Uhlhaas and Singer, 2006; Schevon et al., 2007), the focus of this chapter is the investigation of the relationship between the spatiotemporal coherence patterns of LFOs and HFOs during interictal and ictal activity associated with extratemporal lobe seizures.

106 Characterization of ictal LFOs via wavelet phase coherence The WPC profiles of low frequency oscillations (LFO; <80 Hz) were generated, as outlined in sections and 3.2, for the ictal ieeg segments of all patients. The seizure ieeg patient data were as described in section 2.1 and all the ieeg data sampled at 2000 Hz (Chapter 3) were analyzed. Seizure segments were comprised of a seizure episode, as well as (on average) one minute of ieeg leading up to and following the seizure, to allow for the study of ieeg activity immediately preceding and following seizures (table 4.1). Wavelet phase coherence was applied to every possible combination of electrode contacts from the implanted subdural grids for ictal ieeg segments. In general, during seizures, electrode pairings (with increased coherence) displayed the following coherence patterns: increases in: a) slow activity only, b) fast activity only and c) both slow and fast activity (figures 3-8 and 4-1). However, in contrast to the spatiotemporal coherence patterns observed for ictal HFO activity (in Chapter 3) elevated ictal LFO coherence was not spatially selective, across the implanted grids, when studied across all seven patients. Average LFO WPC was computed to qualitatively characterize the spatiotemporal coherence patterns of LFOs. WPC values in the 2-10 Hz

107 90 Table 4.1: Patient ictal ieeg, SR = 2000 Hz Patient/Seizure No. of channels analyzed Duration of recording 1/1 1/2 1/3 2/1 2/2 3/1 3/2 4/1 4/2 4/3 5/1 5/2 5/3 6/1 6/2 6/3 6/4 6/5 6/6 7/1 7/2 7/ seconds 240 seconds 240 seconds 240 seconds 240 seconds 240 seconds 240 seconds 401 seconds 237 seconds 238 seconds 240 seconds 240 seconds 360 seconds 86 seconds 139 seconds 152 seconds 221 seconds 88 seconds 326 seconds 470 seconds 504 seconds 178 seconds

108 91 Figure 4-1. Increased coherence emerged in low and high frequency bands during seizures. Electrode pairings (with increased coherence) displayed the following coherence patterns: increases in: a) both slow and fast activity (top), b) fast activity only (middle) and c) slow activity only (bottom). Three representative electrode pairings are shown above for patient 3.

109 92 Figure 4-2. Average WPC before, during and following seizure onset for patient 1. Time and frequency averaged global WPC values were computed for each electrode, using all possible electrode pairings, for slow and fast frequency bands: (80-270) Hz and (2-10) Hz, across 1-s windows. Fifteen successive 1-second windows are shown for five segments of the plotted ieeg activity. In patient 1, these mappings highlight only HFO regions of interest in time and space during seizures. No spatial selectivity is visible for LFOs.

110 93 Figure 4-3. Average WPC before, during and following seizure onset for patient 3. Time and frequency averaged global WPC values were computed for each electrode, using all possible electrode pairings, for slow and fast frequency bands: (80-270) Hz and (2-10) Hz, across 1-s windows. Fifteen successive 1-second windows are shown for five segments of the plotted ieeg activity. These mappings highlight both LFO and HFO regions of interest in time and space during seizures for patient 3.

111 94 frequency band were averaged in space (across all electrode combinations), and in time (within 1-second windows), yielding global mean coherence values for each electrode. Fifteen successive windows are shown for various segments of the plotted ieeg activity (before, during and after seizure onset) for patient 1 (figure 4-2) and patient 3 (figure 4-3). Average HFO coherence, for the same ieeg activity, is shown beside the LFO coherence. For patient 3, while average LFO coherence increased and remained highest in a limited area of the grid during seizure activity (figure 4-3), a similar trend was not observed in the remaining patients. 4.2 Characterization of interictal LFOs via wavelet phase coherence The WPC profiles of low frequency oscillations (i.e. <80 Hz) were generated, as outlined in sections and 3.2, for the interictal ieeg segments of all patients. The interictal patient data were as described in section 2.1. All the ieeg data sampled at 2000 Hz (Chapter 3) and additional ieeg data sampled at 200 Hz (patients 2 and 3) were analyzed. Interictal activity was recorded during periods when patients (a) did not experience clinical seizures and (b) were at rest and/or undergoing minimal movement. The results presented in this chapter in regards to interictal observations refer to these ieeg segments (table 4.2). Wavelet phase coherence profiles were calculated for all possible electrode combinations. Strong LFO WPC values were observed during interictal activity in all patients. The WPC profiles for a single electrode pairing from all patients are illustrated in figure 4-4. Average LFO WPC,

112 95 Table 4.2: Patient interictal ieeg, Sampling Rate = 200 Hz and 2000 Hz Patient/interictal segment No. of channels analyzed Duration of recording Time from seizure onset/offset 1/1 1/ seconds 240 seconds 6.4 minutes 14.8 minutes 2/1 2/2 2/ seconds 91 seconds 120 seconds 5.8 hours 1.2 hours 1.7 hours 3/1 3/2 3/3 3/ seconds 60 seconds 240 seconds 300 seconds 1.6 minutes 1.9 minutes 3.2 minutes 1.6 minutes 4/1 4/2 4/3 4/4 4/5 4/ seconds 20 seconds 31 seconds 21 seconds 39 seconds 42 seconds 10.5 hours 6.4 hours 20.6 minutes 4.8 minutes 8.3 hours 8.8 hours 5/ seconds 10.6 minutes 6/1 6/2 6/ seconds 221 seconds 128 seconds sec. to min. sec. to min. sec. to min. 7/1 7/2 7/ seconds 120 seconds 30 seconds 1.2 minute 0.33 minutes 0.7 minutes Sampling rate is indicated by font colour (black = 2000 Hz, blue (bold) = 200 Hz)

113 96 Figure 4-4. Coherence profiles of interictal LFOs. One electrode pairing, exhibiting elevated coherence is shown for each patient. Average LFO WPC (over the entire segment) is shown to the right of each plot.

114 97 calculated over the entire time segment is shown to the right of each plot, highlighting frequency activity in the lower range (< 20 Hz). Figure 4-5 depicts the WPC profiles of two electrode pairings (during interictal activity) from patient 1. While coherence is visible in other frequency bands, the strongest coherence is visible in the slower rhythms (< 20Hz). Furthermore, while electrodes 9 and 10 showed marked coherence in LFO activity, minimal coherence was visible between neighbouring electrodes 23 and 24. Similarly to HFO coherence, it was observed that the pairings which displayed the strongest coherence showed a preference that was not always indicative of the physical distance separating them on the electrode grid. It was observed that the strength of WPC was not dependent on just the distance between the electrode pairing, but whether the electrodes fell within the section of the grid later determined to possess the strongest LFO coherence. Studying the interictal segments from all patients, it was observed that elevated LFO coherence appeared in select electrode clusters, for all patients studied. Electrodes possessing strong LFO interictal coherence (i.e. electrode pairings with mean LFO- WPC values greater than the indicated thresholds in table 4.4) were further explored to elucidate the frequency spread of cohered LFO activity. While the bandwidths of low-frequency activity varied in space and time, the defined LFO bandwidth for each patient were based upon the widest frequency range of LFO activity identified across all electrode clusters possessing strong LFO coherence. In figure 4-6, mean LFO bandwidths are plotted for all patients, for the indicated electrode pairings. Frequencies were averaged in time, over the entire duration of an interictal segment (patient 1:segment 1; patient 2:segment 2; patient 3: segment 4; patient 4: segment 3; patient 5: segment 1, patient 6: segment 2, patient 7: segment 2, from table 4.2). A lowfrequency bandwidth of 5-12 Hz (red outline) was chosen to capture LFO-WPC changes across all patients, except patient 7, where elevated WPC appeared in activity < 5 Hz.

115 98 Figure 4-5. Strong LFO WPC values emerged in select electrode clusters during interictal activity. A: ieeg recorded from electrode 10 during interictal activity of patient 1. B: WPC profile for electrode pairing 9, 10 (E9,E10). C1: WPC from (B) magnified for the 1-30 Hz frequency range. Strong WPC in the lower frequency range was visible for this pairing during interictal activity. C2: WPC profile for E23,E24. Minimal coherence was observed during interictal activity. C3: Average WPC for the electrode pairings from C1 and C2 over the 120 seconds of interictal activity plotted. Strong coherence is visible, centered at 10 Hz, for pairing E9E10 (black-dashed).

116 99 Figure 4-6. Mean WPC (1-80 Hz) during interictal activity. The mean WPC activity for one electrode pairing from each patient is plotted. Mean WPC was averaged in time for one interictal segment for the indicated electrode pairings at right. Frequencies < 30 Hz are expanded at right. Elevated WPC, in the 5-12 Hz frequency band, was observed across all patients, except patient 7, where elevated WPC appeared in activity <5Hz.

117 100 A comprehensive study of all electrode pairings on the implanted subdural grids, during interictal activity, yielded the spatial locations of strongly cohered electrode clusters for all patients. WPC values were averaged across frequency and time to generate a matrix consisting of an average WPC estimate, for each electrode pairing. To isolate LFO activity, the averaging was completed using only the low-frequency bands (5-12 Hz) of each patient. The matrices, for one interictal segment, for each patient are shown in figure 4-7 (left). (Note, as each matrix is symmetric, only half of each matrix is displayed for clarity). The mean seizure LFO WPC values (from the matrices at left) are also plotted as histograms (figure 4-7, middle). Electrode pairings with LFO coherence values greater than the indicated thresholds are highlighted at right (black circles). For all interictal segments and patients, a threshold of +5σ (where is the mean and σ the standard deviation of the mean LFO-WPC values for each interictal segment) highlighted clusters of electrodes with strongly cohered low-frequency (5-12 Hz) activity. The suprathreshold electrodes (for thresholds = +4σ and +5σ) for all interictal segments and all patients are listed in tables 4.3 and Comparison of HFO-defined and LFO-defined ROIs Given the HFO-defined ROIs described in chapter 3 and the LFO-defined ROIs described in the previous section, the next step involved the investigation of the relationship between the spatiotemporal coherence patterns of slow interictal oscillations with those of fast ictal activity during extratemporal lobe seizures.

118 101 Figure 4-7. Mean interictal LFO (5-12 Hz) WPC matrices and histograms. Mean interictal LFO WPC values were calculated for all possible electrode pairings (left column). Matrix values were also plotted as histograms (middle), to identify strongly cohered electrode pairings. Suprathreshold electrodes (i.e. electrodes involved in pairings exhibiting coherence interactions greater than the indicated thresholds) are highlighted on the grids at right.

119 102 Table 4.3 Suprathreshold electrodes for all patients (threshold = +4σ) Patient Interictal segment 1 Suprathreshold Electrodes (threshold = + 4σ) Interictal Interictal segment 3 segment 4 Interictal segment 2 Interictal segment 5 Interictal segment 6 1 1, 5, 6, 9, 10, 13 1, 2, 3, 5, 6, 9, 10, 14, 28, , 24, 28, 32 20, 24, 28, 32 12, 16, 20, 21, 24, 25, 28, , 28, 30, 32 26, 28, 30, , 12, 15, 16 11, 15, 16 11, 12, 15, 16, 19 11, 15 11, 12, 15, 16 3, 7, 11, 12, 15, 16, 25, , 11, 12, 15, 19, 23, 27, , 7, 10, 11, 13, 14, 15, 18, 22, 26, 27, 30, 31 11, 13, 14, 15, 25, 26, 27, 30, 31 3, 4, 7, 11, 15, 16, 20, 26, 27, 30, , 12, 13, 15, 17, 19 13, 17 13,

120 103 Table 4.4 Suprathreshold electrodes for all patients (threshold = +5σ) Patient Interictal segment 1 Suprathreshold Electrodes (threshold = + 5σ) Interictal Interictal Interictal Interictal segment 2 segment 3 segment 4 segment 5 Interictal segment 6 1 1, 6, 9, 10, 13 6, , 24, 28, 32 20, 24, 28, 32 20, 24, 28, , 28, 30, 32 26, 28, 30, , 15 11, 15 11, 15 11, 15 11, 12, 15, 16 3, 7, 11, 12, 15, , 11, 12, 15, 19, 23, 27, , 7,11, 13,14, 15, 18, 22, 26, 30 11, 13, 14, 15, 26, 27, 30, 31 16, 20, 27, , 15, 17, 19 13, 17 13,

121 104 Average LFO WPC was computed, as in section 4.1 to qualitatively characterize the spatiotemporal coherence patterns of LFO interictal activity. WPC values were averaged across the defined HFO and LFO frequency bands, in space (across all electrode combinations) and in time (within 1-second and 2-second windows), yielding global coherence values for each electrode. These average WPC values were arranged in the same spatial layout as the subdural grid electrodes. Average WPC values for electrode contacts from patient 1 are shown in figure 4-8. Consecutive time windows of spatially averaged HFO/LFO coherence are shown for various segments of the plotted ieeg activity (i.e. seizure and non-seizure activity). While it was observed that the mean LFO coherence varied in time and space, the strongest mean LFO coherence persisted in a given cluster of electrodes. Furthermore, HFO coherence increased and remained highest in a similar area of the patient grid during seizure activity. The spatial location of electrodes exhibiting strong LFO interictal coherence and strong ictal HFO coherence from figures 3-9 and 4-7 (suprathreshold electrodes, where average WPC was higher than the indicated thresholds) are marked on the patient grids in figure 4-9. Electrode clusters possessing strong HFO WPC during seizures and strong LFO WPC during interictal activity highlight similar and overlapping areas on the patient grids.

122 105 Figure 4-8. Spatiotemporal patterns of average LFO and HFO coherence during interictal and seizure activity. LFO and HFO WPC values were averaged over the indicated time windows and frequency bands during non-seizure and seizure activity for patient 1. The strongest mean LFO coherence persisted in a given cluster of electrodes (top). Mean HFO coherence increased and remained highest in a similar area of the patient grid during seizure activity.

123 106 dissimilar Figure 4-9. Spatial locations of cohered suprathreshold electrodes during seizure and nonseizure ieeg activity. Electrodes on the patient grids possessing elevated HFO coherence during seizures generally showed a topographical overlap with areas possessing elevated LFO coherence during interictal activity. Ictal HFO intensity is plotted for patient 5, as elevated ictal HFO coherence was not observed.

124 Clinical significance of interictal LFO and Ictal HFO mappings The spatial location of electrodes exhibiting strong LFO interictal coherence (i.e. mean WPC > threshold = +5σ; figure 4-7, table 4.4) are mapped as LFO-defined ROIs in figures 4-10 to Strong HFO ictal coherence (i.e. mean WPC > threshold; figure 3-9) and statistically significant HFO ictal intensity (figure 3-2) are mapped as HFO-defined ROIs in figures 4-11 to The LFO and HFO defined ROIs are mapped onto the clinical SOZ electrodes identified by neurologists A (figure 4-11) and B (figure 4-12) and over the surgically excised electrodes (figure 4-13). The resected electrodes corresponded to the SOZs identified by neurologist A and covered additional electrode contacts, unless the SOZ was found to be in proximity to functional/eloquent cortex. In general, the SOZs defined by both neurologists did not always match. Neurologists A and B marked similar SOZs for patients 3, 4 and 6, yet their SOZs noticeably differed for patients 1, 2 and 7. Furthermore, neurologist B determined that the brain region responsible for seizure onset was not discernible in the ieeg of patient 5, and concluded that the SOZ originated in an area not covered by the implanted grid. As a result, no SOZs are defined for patient 5 in figure It was observed that the LFO/HFO defined ROIs were typically in close proximity or overlapping with the clinically marked SOZs of at least one neurologist.

125 108 Figure Mapping of tissue resection areas and LFO defined ROIs. The best surgical outcome was observed when the resected areas and LFO defined ROIs were in close proximity. For patients with Engel class I and II outcomes, several LFO-defined electrodes were found to lie within and/or proximal to excised tissue, while the electrodes of patients with Engel class III or IV outcomes were observed to lie farther from LFO-defined electrodes.

126 Figure Mapping of clinically marked SOZs of neurologist A with LFO/HFO defined ROIs. 109

127 Figure Mapping of clinically marked SOZs of neurologist B with LFO/HFO defined ROIs. 110

128 Figure Mapping of tissue resection areas with LFO/HFO defined ROIs. 111

129 112 In figure 4-13, post-surgical patient outcomes were ordered from worst to best, in terms of surgical scores (left to right) and ranged from the left side with patient 2 who possessed an Engel class 4 outcome (no worthwhile improvement) to the right end with patients 6 and 7 who both had Engel class 1 outcomes (seizure free). The best clinical outcome was observed when the patient resection areas and HFO/LFO-defined ROIs were in close proximity. Similarly, a poorer surgical outcome was obtained when tissue resection areas and ROIs were incongruent (i.e. patient 2). 4.5 Discussion and Concluding Remarks In this chapter we applied a wavelet phase coherence analysis to investigate whether electrode clusters demonstrating strong LFO coherence, relative to all other electrodes on the implanted grids, resided in tissue areas exhibiting increased ictal HFO coherence and/or intensity. We speculated that regions exhibiting increased LFO and HFO coherence identified local abnormalities that underlie epileptogenic networks. Visually, large-amplitude rhythmic fluctuations are readily apparent in the EEG of epilepsy parents, indicative of an underlying excitatory and coordinated network (Uhlhaas and Singer, 2006). Furthermore, as the seizure state is associated with enhanced neuronal coordination (Engel, 2013), coherence techniques present as an effective approach for the study of coupling patterns in the epileptic brain. We observed clusters of electrodes possessing strongly cohered LFOs during interictal activity. In comparing these more active interictal LFO associated regions of the cortex with the ictal HFO ROIs, it was observed that electrodes on the patient grids possessing elevated HFO

130 113 coherence were typically similar to those possessing elevated LFO coherence during interictal activity. Only patient 7 showed no overlap between LFO/HFO identified regions. When examining the more active LFO-associated regions of the cortex with the clinical data, a good surgical outcome generally resulted for patients in whom the resected tissue (including SOZs identified by neurologist A) was in closer proximity to LFO-defined regions of interest (i.e. patients 3 and 5-7 in figure 4-13). While the HFO ROIs identified for patient 7 were not proximal to the resection zone, and those of patient 3 were proximal, but not adjacent, it is of interest to note that several LFO identified electrodes were excised for both patients. Thus, in this study, it was observed that for cases where the clinical outcome saw improvement (i.e. Engel class 1 and 2), either HFO or LFO alone, or HFO and LFO ROIs (both) were contained within the resected cortex. Traditionally, oscillations in the 10 Hz frequency range were considered an idling rhythm, as they decreased with movement or cognitive changes (Hebert et al.). However, recent studies have suggested that alpha rhythms (8-12 Hz) play an important role in controlling cortical excitability by exerting an inhibitory effect on cortical processing (Sadaghiani et al., 2012). In the epileptic brain, increases in the (4-9 Hz) frequency band have been reported in the EEG of epilepsy patients, relative to normal controls (Zhao et al., 2009). Furthermore, Le Van Quyen et al. (Le Van Quyen et al., 2005) concluded that the synchrony of electrode pairings near the primary epileptogenic zone, in the (4-15) Hz frequency range, increased and decreased before seizure onset. Along the same lines, the presence of high theta activity (i.e. 4-8 Hz) in awake adults is suggestive of abnormal and/or pathological activity (Adeli et al., 2003). As HFOs have been shown to highly localize to SOZs (Jacobs et al., 2008; Crepon et al., 2010; Jacobs et al., 2012) and several studies have concluded that enhanced local phase synchrony is an important correlate of seizure activity (Mormann et al., 2000; van Putten, 2003; Schevon et al., 2007;

131 114 Spencer and Huh, 2008) we speculate that the overlapping spatial regions observed here, which exhibited both increased coherence in ictal HFOs and interictal LFOs in the (5-12 Hz) frequency band, identified local abnormalities that underlie epileptogenic networks. Whereas numerous groups have explored WPC (< 100 Hz) in relation to interictal and ictal activity, the relationship between slow and fast oscillations in ieeg signals recorded from the epileptic brain has yet to be fully explored. Recent studies have proposed the idea that epileptic tissue may be characterized by atypical cross-frequency interactions (Alvarado-Rojas et al., 2011; Cotic et al., 2011; Nariai et al., 2011; Nagasawa et al., 2012; Guirgis et al., 2013; Ibrahim et al., 2014). A new study demonstrated that high-frequency amplitudes ( Hz) were modulated by the phase of slower rhythms (~6-14 Hz) in patient SOZs during the ictal period, while no cross-frequency coupling was observed in the interictal period (Ibrahim et al., 2014). Also, Nariai et al. (Nariai et al., 2011) concluded that interictal HFOs at the seizure onset zones were tightly coupled with slow-wave phases in the (3-10 Hz) frequency range in children with epileptic spasms. The HFO and LFO guided techniques presented in chapters 3 and 4 respectively, show promise as potential epilepsy biomarkers, to complement those already available in the literature and can be used to support the pre-surgical planning phase, when determining SOZs. In this study, it was observed that for cases where the clinical outcome saw improvement (i.e. Engel class 1 and 2), HFO and/or LFO ROIs were contained within the resected cortex. While it is important to note that the number of patients included in this study was small (and therefore the clinical significance of our results is limited in this regard), we propose the idea that the proximity of LFO/HFO defined regions to excised tissue possibly predicts a positive post-surgical outcome. Similarly, we speculate that the removal of all areas identified by HFO/LFO coherence and/or

132 115 HFO intensity measures might have resulted in better outcomes. It was observed here that the SOZs of the treating neurologist (neurologist A) and independent reviewer (neurologist B) were at odds with each other in approximately half of the patients. Thus, it may be noted that if the electroencephalographers are at odds with each other, the HFO/LFO guided techniques may be used as an additional and independent measure for the identification of tissue resection areas, prior to sending the patient for surgery. The positive correlation observed here between HFO/LFO defined ROIs and patient surgical outcomes suggests the ability of both interictal and ictal recordings to contribute to the identification of seizure onset zones in extratemporal lobe seizures. While HFOs present as a new and relatively unexplored epilepsy biomarker, the slower interictal rhythms observed here also offer a practical avenue for further exploration. Firstly, as large numbers of recording electrodes are generally required for the identification of patient SOZs, slower rhythms satisfy present clinical acquisition settings, allowing for low sampling rates and consequently lower volumes of recorded ieeg data. Secondly, slower rhythms are more easily accessible in the scalp EEG, providing a less invasive option for the identification of patient SOZs during the presurgical planning phase. Thirdly, as patients may be removed from medications and/or monitored until ictal activity can be recorded, targeting interictal oscillations would remove the necessity of potentially harmful ictal recordings during the pre-surgical planning phase. And lastly, the low-frequency coherence patterns observed here may have broad clinical applications: they may be relevant as ieeg features for early warning systems for epilepsy patients, as well aid in interventions for seizure abortion. In both instances, by providing a time window during which a seizure can be anticipated patients can be made aware of an impending seizure and/or therapeutic measures can be taken.

133 116 Chapter 5 Results and Future Work 5.1 Coherence in the Epileptic Brain There is an interest in studying cohered brain oscillations by means of phase analyses, as in contrast to classical coherence techniques, phase coherence allows for the separation of phase components from amplitude for a given frequency or frequency range. Standard coherence techniques provide a measure of spectral covariance, but do not separate the effects of amplitude and phase, when measuring the relations between two signals. As a result, the measured coherence values cannot be directly ascribed to changes in either amplitude or phase. The contribution of phase and amplitude to signal synchrony is an important issue when studying EEG signals, as two signals simultaneously recorded in the brain can possess a high degree of phase locking, yet display a parallel disparity in amplitude (Lachaux et al., 1999). Furthermore, as fast brain activities are typically associated with low amplitudes, amplitude independent measures such as phase coherence, present as an effective tool for the study of all frequencies in general and higher frequency activities in particular.

134 117 The coherence of EEG signals has been widely used to characterize the strength of interactions of slower rhythms (i.e. <100 Hz) under normal and epileptic conditions (Uhlhaas et al., 2009), while the phase-locking properties of HFOs in the epileptic brain have yet to be explored. This assessment is important, as it provides a measure of the functional connectivity between two given areas in the brain for a given frequency of interest. High coherence levels are interpreted as evidence of increased connectivity between two measured signals and their underlying neuronal assemblies. Phase relations are being closely examined in relation to epilepsy, as rhythmic activity appears abnormally dominate during seizures (Penfield and Jasper, 1954; Buzsaki and Draguhn, 2004; Uhlhaas and Singer, 2006; Worrell et al., 2012). Several groups have reported increases and decreases in synchrony across distinct frequency bands ( <100 Hz) during seizures (Mormann et al., 2000; Netoff and Schiff, 2002; Le Van Quyen et al., 2003; Mormann et al., 2003; Garcia Dominguez et al., 2005; Uhlhaas and Singer, 2006). Furthermore, numerous studies have concluded that enhanced local phase synchrony is a an important correlate of seizure activity (Mormann et al., 2000; van Putten, 2003; Schevon et al., 2007), and have also provided evidence for a reduction in long-range phase synchronization (Bhattacharya, 2001). Warren et al. (Warren et al., 2010) have concluded that increased synchrony in the SOZ is offset by reduced synchrony in the surrounding cortical regions which bridges the SOZ and surrounding brain tissue. This has been hypothesized to contribute to the functional isolation of epileptic foci and perhaps allow for the development of supra-excitatory activity (Uhlhaas and Singer, 2006). Due to the recent emergence of high frequency oscillations, and the availability of new recording techniques, very fast brain activity (>80Hz) has become a new research focus in the area of seizure genesis. Initially identified in animals, high frequency oscillations (HFOs) have recently

135 118 been recorded in epilepsy patients and proposed as possible novel biomarkers of epileptogenicity. Investigating the coupling characteristics of high frequency activity-those that correlate with the clinical manifestation of seizures-may yield additional insights for delineating seizure onset zones (SOZs) and epileptogenic regions of the brain, as well as provide additional information regarding the commonalities and or differences between normal and epileptic HFO activities. 5.2 Hypothesis Revisited This dissertation studied the spatiotemporal coordinated interactions of very fast electrical brain rhythms, which have yet to be investigated (in this context) in the ieeg of patients during nonseizure and seizure activity. We also examined the relationship between the spatiotemporal coherence patterns of slow and fast neuronal oscillations, which have similarly received minimal investigation. We hypothesized that the spatiotemporal characteristics of low and high frequency electrocorticographic oscillations delineate seizure onset zones in epileptic patients. Specifically, that the phase coherence between subdural grid electrodes captures frequency-dependent pathological neuronal interactions in the epileptic brain. In the next few paragraphs, we discuss the conclusions of this thesis, from which we conclude that the results presented here validate our hypothesis, providing evidence that elevated interictal coherence in the (5-12 Hz) and elevated ictal ( Hz) rhythms are potential epilepsy biomarkers for localizing epileptogenic regions of extratemporal lobe seizures. Furthermore, the conclusions of the thesis can have profound impact not only in the specific clinical context of neurosurgery but also in our understanding of the mechanisms of the epileptic condition and those of transitions to and from ictal states.

136 Elevated HFO coherence marked ictal activity in select electrode clusters. Elevated HFO ( Hz) coherence was observed during seizures, while minimal changes in HFO coherence were visible during interictal activity. Furthermore, strong WPC was not homogeneous, but strongly dependent on the location of the electrode contacts. As patient grids, all together, covered both the right and left hemispheres, as well as all four brain lobes (i.e. frontal, temporal, parietal and occipital), it does not appear that the elevated HFO coherence that was observed was influenced by the underlying functional cortical tissue; rather it appears more likely that the enhanced phase locking, which was associated with the seizure state, was indicative of brain areas involved in pathological phase locking. 2. Elevated ictal HFO coherence coincided with regions exhibiting high ictal HFO intensity. Cortical regions possessing elevated ictal HFO coherence coincided with regions exhibiting high ictal HFO intensity (related to the square-root of the power), relative to all other electrodes on the implanted patient grids. As several groups have recently demonstrated that the surgical removal of regions generating ictal HFO increases correlated with positive seizure-free postsurgical outcomes (Ochi et al., 2007; Jacobs et al., 2010; Wu et al., 2010; Akiyama et al., 2011; Fujiwara et al., 2012), we also calculated HFO ictal intensity changes (where intensity was related to the square-root of the wavelet power). We observed clusters of electrodes possessing both high intensity HFOs and strongly cohered HFOs during seizures, relative to all other electrodes on the implanted grids. The identification of brain areas displaying enhanced local HFO phase synchrony, as well as the observed spatial overlap observed between increased HFO intensity and coherence, support the idea that ictal HFO coherence may also be used to distinguish epileptogenic regions and can complement current clinical methods for localizing SOZs in ETLEs during the presurgical planning phase. It is also important to note that although

137 120 both HFO intensity and HFO coherence identified similar regions on the implanted grids, the coherence measure demonstrated less variability across patients in the choice of threshold, which was used to highlight electrodes displaying significant HFO activity. 3. Elevated ictal HFO coherence coincided with regions exhibiting high interictal LFO coherence. The HFO-defined ROIs were also observed to correlate spatially with areas on the patient grids exhibiting strong low-frequency (5-12 Hz) coherence during interictal activity (in 6/7 patients studied). Similar to HFO activity, LFOs have also been shown to possess pathologic characteristics. The presence of high theta activity (i.e. 4-8 Hz) in awake adults is suggestive of abnormal and/or pathological activity (Adeli et al., 2003; Douw et al., 2010). Furthermore, the presence of alpha rhythms (8-12 Hz), which demonstrated elevated interictal phase locking, has been associated with the control of cortical excitability by exerting an inhibitory effect on cortical processing (Sadaghiani et al., 2012). Several groups have suggested that the loss of critical inhibitory influences, in select local cortical regions, could mediate the transition from normal to pathological oscillations (Crépon et al., 2011; Buzsaki and Silva, 2012). Also, a study by Le Van Quyen et al. concluded that the synchrony of electrode pairings near the primary epileptogenic zone, in the (4-15) Hz frequency range, increased and decreased before seizure onset (Le Van Quyen et al., 2005). Finally, as HFOs have been shown to highly localize to SOZs (Le Van Quyen et al., 2006), we speculate that the overlapping spatial regions observed here, which exhibited both increased coherence in ictal HFOs and interictal LFOs, identified local abnormalities that underlie epileptogenic networks. As such, we suggest that these epileptogenic regions may be mapped by ieeg interictal LFO coherence-outside a seizure episode-and/or by HFO ictal coherence during ictal events.

138 A positive surgical outcome was observed for patients in whom the clinically marked SOZs and/or surgically excised tissue was in close proximity to LFO/HFO coherence highlighted regions of interest. The goal of epilepsy surgery is the complete removal of the epileptogenic zone: "the area of cortex that is necessary and sufficient for initiating seizures and whose removal (or disconnection) is necessary for complete abolition of seizures" (Engel, 1993). The epileptogenic zone is a theoretical concept, as it can only be confirmed postoperatively, if the patient is seizurefree following surgery. Thus, while comparisons to SOZs may provide support for given biomarkers, the golden standard lies in correlating a potential marker with patient clinical outcomes. Several studies have positively correlated measures of connectivity and coherence with clinical outcomes, demonstrating a connection between the removal of cohered cortical regions with seizure control, for slower rhythms (Ortega et al., 2008; Ortega et al., 2011; Antony et al., 2013). In this study, it was observed that for cases where the clinical outcome saw significant improvement (i.e. Engel class 1 and 2), either HFO or LFO alone, or HFO and LFO ROIs (together) were contained within the resected cortex. Thus, we conclude that the results presented here provide evidence that elevated coherence in the (5-12 Hz) and ( Hz) rhythms are potential epilepsy biomarkers for localizing epileptogenic regions of extratemporal lobe seizures. When evaluating the clinical results, it is also important to note that extratemporal seizures are associated with lower seizure-free outcomes after surgery compared to temporal lobe resections (Spencer and Huh, 2008). ETLEs are harder to localize, when compared to TLEs, as extratemporal seizures can originate in any of the other three lobes of the brain, which cover a

139 122 large area of the cortex. Furthermore, as current visual identification of SOZs relies heavily on experts, disagreements between reviewers is not rare. In this study, the SOZs were identified by the treating neurologist (A) and a second neurologist (B), who was blinded to all clinical and surgical outcome information. The SOZs identified by both neurologists were inconsistent in at least half of the patients studied. This highlights the absence of a metric or tool to ensure uniformity of ieeg analysis in epilepsy in an unbiased manner. As such, we propose that targeting the coherence of ictal HFO activity in the ( Hz) frequency range and/or interictal LFO activity in the (5-12 Hz) frequency range, may enhance the detection and delineation of seizure-related ROIs in the ieegs of patients with extratemporal lobe epilepsy (figure 5-1).

140 123 Figure 5-1. Spatial overlap of cohered LFO interictal activity with cohered ictal HFOs highlight seizure related regions of interest in extratemporal lobe epilepsies.