Predicting Seizures in Intracranial EEG Data Using Diffusion Maps
Presenting Author: Dominique Duncan
Collaborating Authors: Ronen Talmon, Ronald R. Coifman, Hitten P. Zaveri
Often medication or surgery are not viable options for patients with epilepsy, thus it is important to find a reliable tool to predict seizures. This way the patient may be warned at least a few minutes prior to the seizure and take the necessary precautions. Finding an accurate predictor of seizures has become a major focus of research during the last few decades.
The goal of this study is to predict a seizure from intracranial EEG (icEEG) data. A novel approach is proposed that capitalizes on the diffusion map framework, which was recently presented and is considered to be one of the current leading manifold learning methods. Diffusion mapping provides dimensionality reduction of the data as well as pattern recognition that may be used to distinguish different states of the patient, for example, resting and preseizure. Based on diffusion maps, a new nonlinear independent component analysis (ICA) algorithm is developed to construct coordinates that generate efficient geometric representations of the complex underlying data structures.
The algorithm is tested on icEEG data recorded from several electrodes from patients being evaluated for possible epilepsy surgery at the Yale-New Haven Hospital. Artifacts are removed from the data prior to the analysis. Preliminary results show that the proposed approach provides a distinction between resting and preseizure states.
Technology for Sensing the Brain and Controlling Seizures
Presenting Author: Hitten P. Zaveri
Collaborating Authors: Ronnie Dhaher, Tore Eid, Lawrence J. Hirsch, Dennis D. Spencer
This presentation describes innovative technologies which are being brought to bear on the direct sensing of the brain in epilepsy and intervention to control seizures. This research is being conducted through two related projects being pursued in collaboration with ITN Energy Systems (Littleton, CO) and the Department of Electrical and Computer Engineering at the University of North Carolina at Charlotte. These projects address the battery free wireless transmission of intracranial EEGs (icEEGs), and the fault-tolerant monitoring of brain activity. In the first project we have designed, fabricated and tested a prototype 64 channel brain implantable device for the wireless transmission of icEEGs. The device allows digital icEEG acquisition and transmission through a standard infra-red (IR) data link and has the potential to perform electrical stimulation. The device can be powered by an embedded battery, wired external power or battery free power through a radio frequency (RF) power link. Bench-top, ex-vivo and in-vivo rat evaluations of the 64-channel wireless icEEG device demonstrate proof-of-principle for an implantable solution to sense, condition, amplify, digitize and wirelessly transmit multi-channel icEEGs. The second project stems from an argument for the inclusion of fault-tolerance in brain implantable devices to extend their dependability. We focus on multielectrode arrays (MEAs) and propose two redundancy based solutions. The first solution uses rows or columns of spare modules to replace faulty modules within a MEA. The second solution uses space redundancy with local reconfiguration. Different fault-tolerant solutions with varying degrees of redundancy and the equivalent graph models for these solutions are described. A maximum matching algorithm is described to match faulty primary to functioning spare modules for MEA reconfiguration. The results of our analysis demonstrate that a considerable improvement in MEA dependability can be achieved with a well-designed increase in redundancy.
Testing Multi-Modal Attention and Awareness with Intracranial EEG
Presenting Author: Paul Guillod
Collaborating Authors: Nicole Tsai, William Chen, Leisel Martin, Mark Youngblood, Ryan Aronberg, William Walker, Andrew Engell, Jason Gerrard, Dennis D. Spencer, Gregory McCarthy, Hal Blumenfeld
Modern neuroimaging allows researchers to probe the mechanisms that give rise to consciousness. This has yielded great insight into how the brain manages and attends to incoming sensory data. Studies have already shown signatures of conscious perception through synchronized, long-range, high frequency oscillations; however, the precise timing and neural correlates behind perception remain elusive.
Our study investigates mechanisms of conscious perception by presenting faint, simultaneous auditory (beeps) and visual (circular gratings) stimuli and measuring the neural response. Stimuli intensities are first calibrated such that subjects report perceiving stimuli half the time. They then run through ~1200 trials that ask them to focus on one of the stimuli (circle or beep) each trial. The stimuli are briefly shown simultaneously and subjects subsequently report which side and whether they perceived one of the two stimuli. They are most often asked about the cued stimulus.
By presenting stimuli at the threshold of perception, we can group events by subjective awareness during analysis. Directing subject's focus on visual or auditory stimuli allows us to dissociate attention from awareness related effects. Furthermore, the incorporation of two sensory modalities can demonstrate how localized and generalized neural correlates of perception are by sense.
So far we have run the task on over 30 subjects and 1 intracranial EEG patient with plans to add scalp EEG combined with fMRI. Recent task performance results can be seen in figure 1 below.
Many patients with epilepsy are unable to perceive and respond to sensory input during seizures. A greater knowledge of the neural circuitry involved in perception will allow us to eventually diagnose and gear therapies in such a way to avoid disruption of those vital circuits.
Limbic Seizures Decrease Subcortical Arousal in Brainstem Cholinergic and Thalamic Neurons
Presenting Author: Joshua E. Motelow
Collaborating Authors: Geoffrey Liu, Hyun Seung Lee, Victoria Chu, Abhijeet Gummadavelli, Asht M. Mishra, Robert N. Sachdev, Basavaraju Sanganahalli, Moran Furman, Dario Englot, Fahmeed Hyder, Hal Blumenfeld
The mechanism of impaired consciousness during partial limbic seizures is not understood. Cortical intracranial EEG recordings in epilepsy patients demonstrate high-frequency, poly-spike seizure activity in the temporal lobe but simultaneous low frequency oscillations across the neocortex. These low-frequency ictal neocortical oscillations are similar to those seen in sleep or deep anesthesia, which provides a clue to the mechanism underlying loss of consciousness. We have developed a rodent model of complex partial limbic seizures in which an electrically induced hippocampal seizure causes the frontal cortex to convert from high-frequency to low-frequency slow oscillations. Blood oxygen level dependent (BOLD) fMRI data during these seizures show increased signal in hippocampus, septal nuclei and hypothalamus associated with widespread BOLD signal decreases in cortex, thalamus, and brainstem. During seizures, multiunit electrophysiology recordings show increased neuronal activity in hypothalamus, decreased activity in brainstem cholinergic nuclei and spindle-like activity in thalamic relay nuclei. Because the brainstem cholinergic nuclei are heterogeneous, juxtacellular recordings were conducted from single labeled cholinergic and non-cholinergic neurons. Cholinergic neurons decrease their firing dramatically during seizures (during which the cortex converts to slow oscillations) while non-cholinergic neurons show mixed behavior. At the same time, we have recorded decreased levels of choline (as a proxy for acetylcholine) in the cortex and thalamus. These data suggest a mechanism for the transition from consciousness to loss of consciousness: (1) hippocampal seizures propagate to limbic structures including the septal nuclei and hypothalamus, (2) descending inhibition from the septal nuclei and hypothalamus depress subcortical arousal systems, and (3) suppression of ascending arousal systems such as the brainstem cholinergic nuclei lead to cortical rhythms normally present during non-REM sleep.
Sudden Unexpected Death in the Multicenter Study of Epilepsy Surgery Outcomes
Presenting Author: Cel Ezeani
Collaborating Authors: Friedman D, Detyniecki K, Hamid H, Spencer DD, Hirsch LJ, Devinsky O
Background: Sudden unexpected death in epilepsy (SUDEP) is death occurring in people with epilepsy in the absence of a known structural cause. Although the frequency of SUDEP varies depending on severity of epilepsy and other factors, the risk is known to be more than 20 times higher than that in the general population. In a seven-center, prospective study of resective epilepsy surgery, we examined the frequency of SUDEP over a five-year follow-up period.
Methods: Patients aged 12 years and above were enrolled at time of referral for epilepsy surgery, and underwent standardized evaluation, treatment, and follow-up procedures. SUDEP cases were classified as definite, probable, or possible. 386 postsurgical and 127 non-surgical patients who had follow-up data were analyzed. In 1648.75 total person-years of follow-up, we examined the effects of seizure freedom, history of generalized convulsions, and resective surgery with respect to predicting an outcome of SUDEP.
Results: Among the 386 post-surgical patients, there were 19 deaths (5%; 12 per 1000 PYs); 5 of these deaths were attributed to definite (1), probable (1) or possible (3) SUDEP (1%; 3 per 1000 PYs). 1 out of the 182 seizure free patients died from SUDEP (0.6%; 1.3 per 1000 PYs), compared to 4 of 204 non-seizure free patients (2%; 4.5 per 1000 PYs). 4 of 97 patients who had persistent generalized tonic-clonic (GTC) seizures after surgery died from SUDEP (4%; 10 per 1000 PYs). On the other hand, only 1 death was due to SUDEP out of 289 patients who did not have persistent GTC seizures (Fisher's Exact Test, p=0.015). We also found that 2 of 6 deaths in the non-surgical group were due to SUDEP (1 probable and 1 possible) although when compared with the post-surgical group, this finding was not significant (See table).
Conclusion: In our cohort of patients, persistent GTCs after epilepsy surgery appear to have an effect on incidence of SUDEP. Although there was a lower rate of SUDEP in the seizure free group compared to the non-seizure free group, this did not reach significance.
Definite or Probable
Definite, Probable or possible SUDEP
Surgery, not seizure free
Surgery, seizure free
1/204 = 0.49%
2/204 = 0.98%
4/204 = 2%
1/182 = 0.6%
Surgery, still has GTCs
Surgery, no GTCs
1/97 = 1%
2/97 = 2%
4/97 = 4%
1/289 = 0.4%
1/386 = 0.3%
5/386 = 1%
Pre-op GTCs, then surgery
Pre-op GTCs, no surgery
4/282 = 1.4%
1/92 = 1.1%
Ictal and Interictal Attention Performance in Childhood Absence Epilepsy Patients
Presenting Author: Robert Kim
Collaborating Authors: J. Rodríguez-Fernández, J. Guo, S. Jhun, W. Xiao, H. Mistry, N. Michiro, RT. Constable, H. Blumenfeld
Childhood absence seizures are characterized by brief impaired consciousness and generalized 3-4 Hz spike-wave discharges on the electroencephalogram. Prior studies have shown impaired attention performance during absence seizures.Here we investigate relationships between variably impaired ictal attention and: (1) task difficulty; (2) seizure duration; (3) behavioral time course during the ictal period; and (4) interictal attention deficits. We tested patients with a diagnosis of childhood absence epilepsy. Each subject underwent either a Repetitive Tapping Task (RTT), or a more difficult Continuous Performance Task (CPT). Out of a total of 90 children, 34 had absence seizures during testing (235 seizures during RTT and 291 during CPT). Mean seizure duration was 5.09±0.17 s (mean±SD). Mean ictal omission rate was 73% for CPT and 51% for RTT, compared to the mean interictal omission rate for CPT 18.66±1.8 and RTT 22.68±0.6. Interestingly, we observed a higher omission rate in seizures longer than 10 seconds (CPT omission rate 89.97±1.9 and RTT omission rate 79.15±2.4) than in less than 5 seconds seizures (CPT omission rate 51.8±4.5 vs RTT omission rate 33.33±3.4). Clear outliers exist in CPT and RTT where short seizures have severely impaired performance and longer seizures have relatively spared performance. For both CPT and RTT, omissions were least severe toward the end of the seizure (last 1s of the seizure): CPT and RTT omission rates were 49.72±5 and 26.4±4.3 respectively. For RTT, performance was also relatively spared on average during the first ~1s of the seizure (performance 55.39±4.9). Also we observed an increased variability in reaction times for both CPT and RTT, and increased duration and variability in inter-response intervals for RTT during the ictal period. These data suggest that task difficulty and task timing have important effects on attention performance during absence seizures. Ictal attention impairment may be related to interictal deficits, suggesting chronic dysfunction in attention networks. This investigation will help us to understand the neural mechanisms of attention deficits observed in CAE patients.
IV Ketamine for the Treatment of Refractory Status Epilepticus: A Retrospective Multicenter Study
Presenting Author: Nicolas Gaspard
Collaborating Authors: B. Foreman, L.M. Judd, J.N. Brenton, B.M. McCoy, A. Al-Otaibi, R. Kilbride, I Sanchez Fernandez, S. Samuel, A. Zakaria, G.P. Kalamangalam, T. Loddenkemper, C.D. Hahn, H.P. Goodkin, J. Claassen, L.J. Hirsch, S.M. LaRoche
Rationale: The treatment of refractory status epilepticus (RSE) is notoriously difficult. Evidence from animal models suggests that inhibition of glutamate antagonists might be more efficient in the late stage of status epilepticus than GABA agonists.Ketamine is an NMDA receptor antagonist occasionally used in this indication. The rationale of this study is to review the use of ketamine in RSE, including doses, efficacy and adverse events.
Methods: We conducted a retrospective multicenter study involving 6 academic centers. Medical records and EEG reports were reviewed. We included children and adults. Continuous EEG was used in all but one case.
Results: We identified 22 patients (11 female) and 23 episodes of RSE treated with ketamine. Mean age was 28+/- 4 years (range: 7 months-74 years). The most common etiologies of RSE were CNS infection (5), inflammatory disease (4), malformation of cortical development (4) and genetic disorder (3). Status epilepticus was primary generalized in 6 cases and secondary generalization occurred in 12 cases. Five cases had purely electrographic status epilepticus while the remaining patients had associated clinical manifestations, including generalized (4), partial (3) and subtle (7) motor activity. In most cases, ketamine was introduced at least 1 week after the onset of SE. An initial loading dose was commonly used (usually 1.5-2.5 mg/kg), followed by continuous infusion (usual initial rate of 0 .1-0.5 mg/kg/h). The maximal infusion rate was most frequently 1-2mg/kg/h, although in 4 patients it was superior to 5mg/kg/h. The duration of the treatment with ketamine ranged from 2 to 24 days. Three patients developed cardiorespiratory depression and two required mechanical ventilation. All 3 were concomitantly treated with benzodiazepines and/or barbiturates. Vasopressors were increased in 8 patients but only one developed significant hypotension. No patient demonstrated signs of raised intracranial pressure. Three patients developed severe acidosis that was attributed to shock and acute kidney injury in 2 cases. The third patient developed a syndrome similar to the Propofol Infusion Syndrome (PRIS), while on high dose of ketamine (7.5mg/kg/h) and midazolam. Status epilepticus was permanently controlled in 10 cases. All 10 were alive at discharge from the hospital but significantly impaired. Ketamine was the last drug to be used in 3 of them. The others received at least one other additional drug after ketamine, most commonly benzodiazepine (7), phenytoin/fosphenytoin (6) and topiramate (4). Length of stay in the ICU ranged from 13 to 244 days.
Conclusion: Assessing the efficacy of ketamine in RSE is difficult as it is used in the late stage of refractory SE and with a wide range of doses. This study indicates that it is associated with a low incidence of serious adverse events, although we describe for the first time its association with a syndrome similar to PRIS. Further study is needed before recommendation can be made to extend its usage and this multicenter effort will be pursued.
Relationship between Depression, Anxiety, and Quality of Life Outcomes Post Epilepsy Surgery: A Prospective Multicenter Study
Presenting Author: Hamada Hamid
Rationale: People with seizures have a lower quality of life (QOL) compared to people without. Depression and anxiety scores strongly correlate with QOL scores in cross-sectional studies. However, the role of depression and anxiety symptoms on QOL outcomes after epilepsy surgery has not been explored prospectively.
Methods: The design, measures, and subject recruitment of the Epilepsy Surgery Multicenter Study has been detailed elsewhere. Briefly, 7 tertiary epilepsy centers enrolled 396 patients and completed a comprehensive diagnostic workup that included a comprehensive medical history and physical exam; neuropsychological, neuroimaging and neurophysiology testing; and a psychiatric as well as quality of life evaluation. Subjects were evaluated prior to surgery, then at 3, 6, 12, 48, and 60 months after surgery. Standardized, assessments included The Quality of Life in Epilepsy Inventory-89 (HRQOL) and Beck Depression (BDI) and Anxiety (BAI) Inventories. Seizure outcome was classified into one of four categories: "excellent" for subjects seizure free (and no auras) for all five years, "good" for two consecutive years but not all five, "fair" if subjects were seizure free for one year but never two consecutive years, and "poor" if subjects never had a one year period of seizure freedom.
Statistical Analysis: The mixed-model repeated-measures analysis was used to analyze overall HRQOL score and four dimension subscores (cognitive distress, physical health, mental health, epilepsy-targeted) association with depression, anxiety, seizure outcome, seizure history, with overall HRQOL score and the four dimension subscores, respectively, over time. The model included gender, race, education, duration of seizure history, laterality of seizure focus, resection location, BDI and BAI scores, and time as fixed effect as well as random intercept and slope.
Results: All four subscores of QOL improved over time (p<0.0001). Excellent and good seizure control groups both have significant positive impact on the overall QOL compared to the fair and poor seizure control group; interestingly, there is no difference in change in overall QOL over time between the fair and poor groups. The time and seizure control interaction was marginal significant (p=0.0606). Subjects who had left sided resections shown significant lower rate of improvement in overall QOL score compared to those with right sided resection. The BDI and BAI score are both highly and negatively associated with overall QOL; increases in BDI and BAI score are associated with decreased overall QOL score. This association appeared to be driven by the cognitive subscore and the association was not significant in the other subscores. Duration of seizure history, gender, race, education and temporal versus extratemporal resection did not show significant association with overall QOL.
Conclusion: Depression and anxiety are strongly and independently associated with worse QOL post epilepsy surgery. Management of mood and anxiety is a critical component to post-surgical care.
Distinct Contribution of Different Interneuron Subtypes to Network Stability in Vivo
Presenting Author: Mitra Miri
Collaborating Authors: M. Miri, J.A. Cardin
GABAergic inhibition is critical for regulation of excitation and is thought to maintain neural network stability through a precise balancing process. Loss or dysfunction of inhibitory interneurons is thought to lead to abnormal activity patterns and ultimately to seizure initiation. Two major interneuron classes are the Parvalbumin-expressing (PV), fast-spiking cells and Somatostatin-expressing (SOM), low-threshold spiking cells. Due to differing physiology, biophysical properties, and synaptic targeting, they may contribute differently to ongoing computations in the surrounding local network. Previous studies have examined preferential interneuron vulnerability leading to or resulting from seizures, but there the roles of specific sources of inhibition in seizure generation remain unclear. Using combined optical and electrophysiological tools, we can investigate network dynamics during the development of seizures by monitoring the activity of individual interneurons and synchronization between multiple hippocampal cell types.
Here we used two transgenic mouse lines: PV-Cre and SOM-Cre. Using AAV-DIO ChR2-mCherry, we selectively targeted optogenetic tools to PV or SOM interneurons in CA1. During experiments, we used stereotrode arrays to extracellularly record the activity of many simultaneous cells in hippocampal CA1 in animals lightly anesthetized with ketamine/xylazine. We identified targeted ChR2-expressing PV and SOM interneurons in the recorded population by stimulating with brief pulses of blue light at 473nm. Light pulses were continued throughout the experiment to track identified neurons without altering the pattern of spontaneous activity. Using a pharmacological model of seizure induction (PTZ), we performed measurements of neural activity during three periods: 1) baseline 2) early preictal and 3) late preictal. We first assessed the recruitment of local excitatory neurons by ictal activity. We further assessed the temporal pattern of identified interneuron activity during each period and the relationship between interneuron and excitatory neuron activity.
To examine the relationship between local network activity and interneuron spiking, we calculated the spike-triggered LFP average for each cell during each period. We find that PV+ cells have a lower spike probability in response to blue light pulses that progressively decreases from baseline to late preictal stages. Additionally, we find that average local synaptic activity increases surrounding spike times of identified RS cells during both early and late preictal periods, as compared to baseline. These preliminary results demonstrate that these combined approaches provide a powerful set of tools for dissecting network activity during seizure initiation.