The circadian profile of epilepsy improves seizure forecasting

A Comparative Study on the Prediction of Occupational Diseases in China with Hybrid Algorithm Combing Models

Occupational disease is a huge problem in China, and many workers are under risk. Accurate forecasting of occupational disease incidence can provide critical information for prevention and control. Therefore, in this study, five hybrid algorithm combing models were assessed on their effectiveness and applicability to predict the incidence of occupational diseases in China. The five hybrid algorithm combing models are the combination of five grey models (EGM, ODGM, EDGM, DGM, and Verhulst) and five state-of-art machine learning models (KNN, SVM, RF, GBM, and ANN). The quality of the models were assessed based on the accuracy of model prediction as well as minimizing mean absolute percentage error (MAPE) and root-mean-squared error (RMSE). Our results showed that the GM-ANN model provided the most precise prediction among all the models with lowest mean absolute percentage error (MAPE) of 3.49% and root-mean-squared error (RMSE) of 1076.60. Therefore, the GM-ANN model can be used for precise prediction of occupational diseases in China, which may provide valuable information for the prevention and control of occupational diseases in the future.

Seizure self-prediction in a randomized controlled trial of stress management

OBJECTIVE

Using electronic diaries as part of a randomized controlled trial of stress reduction for epilepsy, we evaluated factors associated with successful seizure self-prediction.

METHODS

Adults with medication-resistant focal epilepsy were recruited from 3 centers and randomized to treatment with progressive muscle relaxation or control focused attention. An 8-week baseline was followed by 12 weeks of double-blind treatment. Twice daily, participants rated the likelihood of a seizure in the next 24 hours on a 5-point scale from very unlikely to almost certain, along with mood, premonitory symptoms, stress ratings, and seizure counts. We analyzed the association of mood, premonitory symptoms, stress, and circadian influences on seizure self-prediction.

RESULTS

Sixty-four participants completed the trial (3,126 seizures). Diary entry adherence was >82%. Participant self-prediction was associated with seizure occurrence at 6, 12, and 24 hours (p < 0.0001). Odds ratio (OR) of seizure prediction increased systematically with participants' prediction of seizure likelihood (p < 0.0001, all levels of prediction and all time intervals). For the 12-hour prediction window, median specificity for seizure prediction was 0.94 and negative predictive value 0.94; median sensitivity was 0.10 and positive predictive value 0.13. A subset of 13 participants (20% of sample) met criteria for good predictors (median OR for seizure prediction 5.25). Mood, stress, premonitory symptoms, seizure time, and randomized group were not associated with seizure occurrence.

CONCLUSION

In this prospective study, participants' prediction of a high probability of seizure was significantly associated with subsequent seizure occurrence within 24 hours. Future studies should focus on understanding factors that drive self-prediction.

CLINICALTRIALSGOV IDENTIFIER

NCT01444183.

Evaluation of machine learning methods for seizure prediction in epilepsy

Epilepsy affects about 50 million people worldwide of which one third is refractory to medication. An automated and reliable system that warns of impending seizures would greatly improve patient’s quality of life by overcoming the uncertainty and helplessness due to the unpredicted events. Here we present new seizure prediction results including a performance comparison of different methods. The analysis is based on a new set of intracranial EEG data that has been recorded in our working group during presurgical evaluation. We applied two different methods for seizure prediction and evaluated their performance pseudoprospectively. The comparison of this evaluation with common statistical evaluation reveals possible reasons for overly optimistic estimations of the performance of seizure forecasting systems.

Seizure prediction - ready for a new era

Epilepsy is a common disorder characterized by recurrent seizures. An overwhelming majority of people with epilepsy regard the unpredictability of seizures as a major issue. More than 30 years of international effort have been devoted to the prediction of seizures, aiming to remove the burden of unpredictability and to couple novel, time-specific treatment to seizure prediction technology. A highly influential review published in 2007 concluded that insufficient evidence indicated that seizures could be predicted. Since then, several advances have been made, including successful prospective seizure prediction using intracranial EEG in a small number of people in a trial of a real-time seizure prediction device. In this Review, we examine advances in the field, including EEG databases, seizure prediction competitions, the prospective trial mentioned and advances in our understanding of the mechanisms of seizures. We argue that these advances, together with statistical evaluations, set the stage for a resurgence in efforts towards the development of seizure prediction methodologies. We propose new avenues of investigation involving a synergy between mechanisms, models, data, devices and algorithms and refine the existing guidelines for the development of seizure prediction technology to instigate development of a solution that removes the burden of the unpredictability of seizures.

Early prediction of epileptic seizures using a long-term recurrent convolutional network

BACKGROUND

A seizure prediction system can detect seizures prior to their occurrence and allow clinicians to provide timely treatment for patients with epilepsy. Research on seizure prediction has progressed from signal processing analyses to machine learning. However, most prediction methods are hand-engineered and have high computational complexity, increasing the difficulty of obtaining real-time predictions. Some forecasting and early warning methods have achieved good results in the short term but have low applicability in practical situations over the long term.

NEW METHODS

First, electroencephalogram (EEG) time series were converted into two-dimensional images for multichannel fusion. A feasible method, a long-term recurrent convolutional network (LRCN), was proposed to create a spatiotemporal deep learning model for predicting epileptic seizures. The convolutional network block was used to automatically extract deep features from the data. The long short-term memory (LSTM) block was incorporated into learning a time sequence for identifying the preictal segments. New network settings and a postprocessing strategy were proposed in the seizure prediction model.

RESULTS

The deep seizure prediction model achieved an accuracy of 93.40%, prediction sensitivity of 91.88% and specificity of 86.13% in segment-based evaluations. For the event-based evaluations, 164 seizures were predicted. The proposed method provides high sensitivity and a low false prediction rate (FPR) of 0.04 F P/h.

COMPARISON WITH EXISTING METHODS

We employed different methods, including the LRCN, deep learning and traditional machine learning methods, and compared them using the same data in this paper. Overall, the LRCN offers approximately 5-9% increased sensitivity and specificity.

CONCLUSION

This study describes the LRCN network for analyzing EEG data to predict epileptic seizures, thereby enabling the implementation of an early warning system that detects epileptic seizures before they occur in clinical applications.

Precursors of seizures due to specific spatial-temporal modifications of evolving large-scale epileptic brain networks

Knowing when, where, and how seizures are initiated in large-scale epileptic brain networks remains a widely unsolved problem. Seizure precursors – changes in brain dynamics predictive of an impending seizure – can now be identified well ahead of clinical manifestations, but either the seizure onset zone or remote brain areas are reported as network nodes from which seizure precursors emerge. We aimed to shed more light on the role of constituents of evolving epileptic networks that recurrently transit into and out of seizures. We constructed such networks from more than 3200 hours of continuous intracranial electroencephalograms recorded in 38 patients with medication refractory epilepsy. We succeeded in singling out predictive edges and predictive nodes. Their particular characteristics, namely edge weight respectively node centrality (a fundamental concept of network theory), from the pre-ictal periods of 78 out of 97 seizures differed significantly from the characteristics seen during inter-ictal periods. The vast majority of predictive nodes were connected by most of the predictive edges, but these nodes never played a central role in the evolving epileptic networks. Interestingly, predictive nodes were entirely associated with brain regions deemed unaffected by the focal epileptic process. We propose a network mechanism for a transition into the pre-seizure state, which puts into perspective the role of the seizure onset zone in this transition and highlights the necessity to reassess current concepts for seizure generation and seizure prevention.

Epilepsyecosystem.org: crowd-sourcing reproducible seizure prediction with long-term human intracranial EEG

Accurate seizure prediction will transform epilepsy management by offering warnings to patients or triggering interventions. However, state-of-the-art algorithm design relies on accessing adequate long-term data. Crowd-sourcing ecosystems leverage quality data to enable cost-effective, rapid development of predictive algorithms. A crowd-sourcing ecosystem for seizure prediction is presented involving an international competition, a follow-up held-out data evaluation, and an online platform, Epilepsyecosystem.org, for yielding further improvements in prediction performance. Crowd-sourced algorithms were obtained via the 'Melbourne-University AES-MathWorks-NIH Seizure Prediction Challenge' conducted at kaggle.com. Long-term continuous intracranial electroencephalography (iEEG) data (442 days of recordings and 211 lead seizures per patient) from prediction-resistant patients who had the lowest seizure prediction performances from the NeuroVista Seizure Advisory System clinical trial were analysed. Contestants (646 individuals in 478 teams) from around the world developed algorithms to distinguish between 10-min inter-seizure versus pre-seizure data clips. Over 10 000 algorithms were submitted. The top algorithms as determined by using the contest data were evaluated on a much larger held-out dataset. The data and top algorithms are available online for further investigation and development. The top performing contest entry scored 0.81 area under the classification curve. The performance reduced by only 6.7% on held-out data. Many other teams also showed high prediction reproducibility. Pseudo-prospective evaluation demonstrated that many algorithms, when used alone or weighted by circadian information, performed better than the benchmarks, including an average increase in sensitivity of 1.9 times the original clinical trial sensitivity for matched time in warning. These results indicate that clinically-relevant seizure prediction is possible in a wider range of patients than previously thought possible. Moreover, different algorithms performed best for different patients, supporting the use of patient-specific algorithms and long-term monitoring. The crowd-sourcing ecosystem for seizure prediction will enable further worldwide community study of the data to yield greater improvements in prediction performance by way of competition, collaboration and synergism.10.1093/brain/awy210_video1awy210media15817489051001.

Can absence seizures be predicted by vigilance states?: Advanced analysis of sleep-wake states and spike-wave discharges' occurrence in rats

Spike-wave discharges (SWDs) are the main manifestation of absence epilepsy. Their occurrence is dependent on the behavioral state, and they preferentially occur during unstable vigilance periods. The present study investigated whether the occurrence of SWDs can be predicted by the preceding behavioral state and whether this relationship is different between the light and the dark phases of the 24-h day. Twenty-four-hour (12:12 light/dark phases) electroencephalographic (EEG) recordings of 12 Wistar Albino Glaxo, originally bred in Rijswijk (WAG/Rij) rats, a well-known genetic model of absence epilepsy, were analyzed and transformed into sequences of 2-s length intervals of the following 6 possible states: active wakefulness (AW), passive wakefulness (PW), deep slow-wave sleep (DSWS), light slow-wave sleep (LSWS), rapid eye movement (REM) sleep, and SWDs, given discrete series of categorical data. Probabilities of all transitions between states and Shannon entropy of transitions were calculated for the light and dark phases separately and statistically analyzed. Common differences between the light and the dark phases were found regarding the time spent in AW, LSWS, DSWS, and SWDs. The most probable transitions were that AW was preceded and followed by PW and vice versa regardless of the phase of the photoperiod. A similar relationship was found for light and deep slow-wave sleep. The most probable transitions to and from SWDs were AW and LSWS, respectively, with these transition likelihoods being consistent across both circadian phases. The second most probable transitions around SWDs appeared more variable between light and dark. During the light phase, SWDs occurred around PW and participated exclusively in sleep initiation; in the dark phase, SWDs were seen on both, ascending and descending steps towards and from sleep. Conditional Shannon entropy showed that AW and DSWS are the most predictable events, while the possible prediction horizon of SWDs is not larger than 4 s and despite the higher occurrence of SWDs in the dark phase, did not differ between phases. It can be concluded that although SWDs show a stable, strong circadian rhythm with a peak in number during the dark phase, their occurrence cannot be reliably predicted by the preceding behavioral state, except at a very short time base.

Fractal and Multifractal Properties of Electrographic Recordings of Human Brain Activity: Toward Its Use as a Signal Feature for Machine Learning in Clinical Applications

The quantification of brain dynamics is essential to its understanding. However, the brain is a system operating on multiple time scales, and characterization of dynamics across time scales remains a challenge. One framework to study such dynamics is that of fractal geometry; and currently there exist several methods for the study of brain dynamics using fractal geometry. We aim to highlight some of the practical challenges of applying fractal geometry to brain dynamics—and as a putative feature for machine learning applications, and propose solutions to enable its wider use in neuroscience. Using intracranially recorded electroencephalogram (EEG) and simulated data, we compared monofractal and multifractal methods with regards to their sensitivity to signal variance. We found that both monofractal and multifractal properties correlate closely with signal variance, thus not being a useful feature of the signal. However, after applying an epoch-wise standardization procedure to the signal, we found that multifractal measures could offer non-redundant information compared to signal variance, power (in different frequency bands) and other established EEG signal measures. We also compared different multifractal estimation methods to each other in terms of reliability, and we found that the Chhabra-Jensen algorithm performed best. Finally, we investigated the impact of sampling frequency and epoch length on the estimation of multifractal properties. Using epileptic seizures as an example event in the EEG, we show that there may be an optimal time scale (i.e., combination of sampling frequency and epoch length) for detecting temporal changes in multifractal properties around seizures. The practical issues we highlighted and our suggested solutions should help in developing robust methods for the application of fractal geometry in EEG signals. Our analyses and observations also aid the theoretical understanding of the multifractal properties of the brain and might provide grounds for new discoveries in the study of brain signals. These could be crucial for the understanding of neurological function and for the developments of new treatments.

Responsive neurostimulation: Candidates and considerations

Responsive neurostimulation (RNS) has recently emerged as a safe and effective treatment for some patients with medically refractory focal epilepsy who are not candidates for surgical resection. Responsive neurostimulation involves an implanted neurostimulator and intracranial leads that detect incipient seizures and respond with electrical counterstimulation. Over 1800 patients have been treated with RNS since its FDA approval in 2013. Despite its widespread use, however, RNS presents distinct challenges for clinicians. What types of patients are most well-suited for treatment with RNS? Given the availability of two other neurostimulation modalities, vagus nerve stimulation (VNS) and thalamic deep brain stimulation (DBS), what patient characteristics favor or disfavor RNS? Once RNS candidates are identified, lead placement presents another challenge. Unlike VNS and thalamic DBS, which both involve prespecified electrode locations, RNS involves intracranial strip and/or depth electrodes that can be flexibly configured based on knowledge of the seizure onset zone. The efficacy of RNS may depend on optimal lead configuration, but there are few resources to guide clinicians in formulating lead placement strategies. Here, we address these challenges, first by reviewing clinical trial data supporting the safety and efficacy of RNS. Then, through a series of clinical vignettes from our center, we provide a framework for RNS patient selection. For each clinical scenario, we illustrate typical strategies for RNS lead placement. We outline considerations for choosing among available neurostimulation devices based on their intrinsic features. For example, a unique feature of RNS is that the neurostimulator provides chronic electrocorticography (ECoG), which has powerful diagnostic potential. We highlight emerging applications of chronic ECoG, and we discuss how the limitations of RNS will inform development of next-generation devices.

Gauging seizure risk

The current paradigm for treatment of epilepsy begins with trials of antiepileptic drugs, followed by evaluation for resective brain surgery in drug-resistant patients. If surgery is not possible or fails to control seizures, some patients benefit from implanted neurostimulation devices. In addition to their therapeutic benefit, some of these devices have diagnostic capability enabling recordings of brain activity with unprecedented chronicity. Two recent studies using different devices for chronic EEG (i.e., over months to years) yielded convergent findings of daily and multiday cycles of brain activity that help explain seizure timing. Knowledge of these patient-specific cycles can be leveraged to gauge and forecast seizure risk, empowering patients to adopt risk-stratified treatment strategies and behavioral modifications. We review evidence that epilepsy is a cyclical disorder, and we argue that implanted monitoring devices should be offered earlier in the treatment paradigm. Chronic EEG would allow pharmacologic treatments tailored to days of high seizure risk-here termed chronotherapy-and would help characterize long timescale seizure dynamics to improve subsequent surgical planning. Coupled with neuromodulation, the proposed approach could improve quality of life for patients and decrease the number ultimately requiring resective surgery. We outline challenges for chronic monitoring and seizure forecasting that demand close collaboration among engineers, neurosurgeons, and neurologists.

Circadian rhythm and epilepsy

Advances in diagnostic technology, including chronic intracranial EEG recordings, have confirmed the clinical observation of different temporal patterns of epileptic activity and seizure occurrence over a 24-h period. The rhythmic patterns in epileptic activity and seizure occurrence are probably related to vigilance states and circadian variation in excitatory and inhibitory balance. Core circadian genes BMAL1 and CLOCK, which code for transcription factors, have been shown to influence excitability and seizure threshold. Despite uncertainties about the relative contribution of vigilance states versus circadian rhythmicity, including circadian factors such as seizure timing improves sensitivity of seizure prediction algorithms in individual patients. Improved prediction of seizure occurrence opens the possibility for personalised antiepileptic drug-dosing regimens timed to particular phases of the circadian cycle to improve seizure control and to reduce side-effects and risks associated with seizures. Further studies are needed to clarify the pathways through which rhythmic patterns of epileptic activity are generated, because this might also inform future treatment options.

Automated unsupervised behavioral state classification using intracranial electrophysiology

OBJECTIVE

Automated behavioral state classification in intracranial EEG (iEEG) recordings may be beneficial for iEEG interpretation and quantifying sleep patterns to enable behavioral state dependent neuromodulation therapy in next generation implantable brain stimulation devices. Here, we introduce a fully automated unsupervised framework to differentiate between awake (AW), sleep (N2), and slow wave sleep (N3) using intracranial EEG (iEEG) only and validated with expert scored polysomnography.

APPROACH

Data from eight patients undergoing evaluation for epilepsy surgery (age [Formula: see text], three female) with intracranial depth electrodes for iEEG monitoring were included. Spectral power features (0.1-235 Hz) spanning several frequency bands from a single electrode were used to classify behavioral states of patients into AW, N2, and N3.

MAIN RESULTS

Overall, classification accuracy of 94%, with 94% sensitivity and 93% specificity across eight subjects using multiple spectral power features from a single electrode was achieved. Classification performance of N3 sleep was significantly better (95%, sensitivity 95%, specificity 93%) than that of the N2 sleep phase (87%, sensitivity 78%, specificity 96%).

SIGNIFICANCE

Automated, unsupervised, and robust classification of behavioral states based on iEEG data is possible, and it is feasible to incorporate these algorithms into future implantable devices with limited computational power, memory, and number of electrodes for brain monitoring and stimulation.

Circadian and circaseptan rhythms in human epilepsy: a retrospective cohort study

BACKGROUND

Epilepsy has long been suspected to be governed by cyclic rhythms, with seizure rates rising and falling periodically over weeks, months, or even years. The very long scales of seizure patterns seem to defy natural explanation and have sometimes been attributed to hormonal cycles or environmental factors. This study aimed to quantify the strength and prevalence of seizure cycles at multiple temporal scales across a large cohort of people with epilepsy.

METHODS

This retrospective cohort study used the two most comprehensive databases of human seizures (SeizureTracker [USA] and NeuroVista [Melbourne, VIC, Australia]) and analytic techniques from circular statistics to analyse patients with epilepsy for the presence and frequency of multitemporal cycles of seizure activity. NeuroVista patients were selected on the basis of having intractable focal epilepsy; data from patients with at least 30 clinical seizures were used. SeizureTracker participants are self selected and data do not adhere to any specific criteria; we used patients with a minimum of 100 seizures. The presence of seizure cycles over multiple time scales was measured using the mean resultant length (R value). The Rayleigh test and Hodges-Ajne test were used to test for circular uniformity. Monte-Carlo simulations were used to confirm the results of the Rayleigh test for seizure phase.

FINDINGS

We used data from 12 people from the NeuroVista study (data recorded from June 10, 2010, to Aug 22, 2012) and 1118 patients from the SeizureTracker database (data recorded from Jan 1, 2007, to Oct 19, 2015). At least 891 (80%) of 1118 patients in the SeizureTracker cohort and 11 (92%) of 12 patients in the NeuroVista cohort showed circadian (24 h) modulation of their seizure rates. In the NeuroVista cohort, patient 8 had a significant cycle at precisely 1 week. Two others (patients 1 and 7) also had approximately 1-week cycles. Patients 1 and 4 had 2-week cycles. In the SeizureTracker cohort, between 77 (7%) and 233 (21%) of the 1118 patients showed strong circaseptan (weekly) rhythms, with a clear 7-day period. Between 151 (14%) and 247 (22%) patients had significant seizure cycles that were longer than 3 weeks. Seizure cycles were equally prevalent in men and women, and peak seizure rates were evenly distributed across all days of the week.

INTERPRETATION

Our results suggest that seizure cycles are robust, patient specific, and more widespread than previously understood. They align with the accepted consensus that most epilepsies have some diurnal influence. Variations in seizure rate have important clinical implications. Detection and tracking of seizure cycles on a patient-specific basis should be standard in epilepsy management practices.

FUNDING

Australian National Health and Medical Research Council.

Seizure susceptibility and infraslow modulatory activity in the intracranial electroencephalogram

OBJECTIVE

Studies of infraslow amplitude modulations (<0.15 Hz) of band power time series suggest that these envelope correlations may form a basis for distant spatial coupling in the brain. In this study, we sought to determine how infraslow relationships are affected by antiepileptic drug (AED) taper, time of day, and seizure.

METHODS

We studied intracranial electroencephalographic (icEEG) data collected from 13 medically refractory adult epilepsy patients who underwent monitoring at Yale-New Haven Hospital. We estimated the magnitude-squared coherence (MSC) at <0.15 Hz of traditional EEG frequency band power time series for all electrode contact pairs to quantify infraslow envelope correlations between them. We studied, first, hour-long background icEEG epochs before and after AED taper to understand the effect of taper. Second, we analyzed the entire record for each patient to study the effect of time of day. Finally, for each patient, we reviewed the clinical record to find all seizures that were at least 6 hours removed from other seizures and analyzed infraslow envelope MSC before and after them.

RESULTS

Infraslow envelope MSC increased slightly, but significantly, after AED taper, and increased on average during the night and decreased during the day. It was also increased significantly in all frequency bands up to 3 hours preseizure and 1 hour postseizure as compared to background icEEG (61 seizures studied). These changes occurred for both daytime and nighttime seizures (28 daytime, 33 nighttime). Interestingly, there was significant spatial variability to these changes, with the seizure onset area peaking at 3 hours preseizure, then showing progressive desynchronization from 3 hours preseizure to 1 hour postseizure.

SIGNIFICANCE

Infraslow envelope analysis may be used to understand long-term changes over the course of icEEG monitoring, provide unique insight into interictal electrophysiological changes related to ictogenesis, and contribute to the development of novel seizure forecasting algorithms.

Degenerate time-dependent network dynamics anticipate seizures in human epileptic brain

Epileptic seizures are known to follow specific changes in brain dynamics. While some algorithms can nowadays robustly detect these changes, a clear understanding of the mechanism by which these alterations occur and generate seizures is still lacking. Here, we provide crossvalidated evidence that such changes are initiated by an alteration of physiological network state dynamics. Specifically, our analysis of long intracranial electroencephalography (iEEG) recordings from a group of 10 patients identifies a critical phase of a few hours in which time-dependent network states become less variable ("degenerate"), and this phase is followed by a global functional connectivity reduction before seizure onset. This critical phase is characterized by an abnormal occurrence of highly correlated network instances and is shown to be particularly associated with the activity of the resected regions in patients with validated postsurgical outcome. Our approach characterizes preseizure network dynamics as a cascade of 2 sequential events providing new insights into seizure prediction and control.

Understanding and predicting the generation of seizures in epileptic patients is fundamental to improving the quality of life of the more than 1% of the world population who suffer from this illness. Although seizure prediction has made important advances over the last decade, there is a lack of understanding of the common principles explaining the transitions that brain activity undergoes before a seizure. In this study, we characterized this transition from a novel perspective grounded on the mathematical analysis of continuous recordings inside the brains of epileptic patients over several days using depth electrodes. We show that the critical period preceding a seizure unfolds in a two-stage process. It begins with a phase of several hours when the highly correlated activity in the preceding days is altered, and it proceeds with a second, shorter phase of decrease in global connectivity before the seizure onset. Furthermore, our analysis reveals that these global alterations are more locally manifested in areas that are selected for surgical treatment. Our study suggests that preseizure activity might follow global stereotyped dynamics that could be targeted and modulated to prevent epileptic seizures.

Epileptic Seizure Prediction Using Big Data and Deep Learning: Toward a Mobile System

BACKGROUND

Seizure prediction can increase independence and allow preventative treatment for patients with epilepsy. We present a proof-of-concept for a seizure prediction system that is accurate, fully automated, patient-specific, and tunable to an individual's needs.

METHODS

Intracranial electroencephalography (iEEG) data of ten patients obtained from a seizure advisory system were analyzed as part of a pseudoprospective seizure prediction study. First, a deep learning classifier was trained to distinguish between preictal and interictal signals. Second, classifier performance was tested on held-out iEEG data from all patients and benchmarked against the performance of a random predictor. Third, the prediction system was tuned so sensitivity or time in warning could be prioritized by the patient. Finally, a demonstration of the feasibility of deployment of the prediction system onto an ultra-low power neuromorphic chip for autonomous operation on a wearable device is provided.

RESULTS

The prediction system achieved mean sensitivity of 69% and mean time in warning of 27%, significantly surpassing an equivalent random predictor for all patients by 42%.

CONCLUSION

This study demonstrates that deep learning in combination with neuromorphic hardware can provide the basis for a wearable, real-time, always-on, patient-specific seizure warning system with low power consumption and reliable long-term performance.

Seizure Forecasting from Idea to Reality. Outcomes of the My Seizure Gauge Epilepsy Innovation Institute Workshop

The Epilepsy Innovation Institute (Ei²) is a new research program of the Epilepsy Foundation designed to be an innovation incubator for epilepsy. Ei² research areas are selected based on community surveys that ask people impacted by epilepsy what they would like researchers to focus on. In their 2016 survey, unpredictability was selected as a top issue regardless of seizure frequency or severity. In response to this need, Ei² launched the My Seizure Gauge challenge, with the end goal of creating a personalized seizure advisory system device. Prior to moving forward, Ei² convened a diverse group of stakeholders from people impacted by epilepsy and clinicians, to device developers and data scientists, to basic science researchers and regulators, for a state of the science assessment on seizure forecasting. From the discussions, it was clear that we are at an exciting crossroads. With the advances in bioengineering, we can utilize digital markers, wearables, and biosensors as parameters for a seizure-forecasting algorithm. There are also over a thousand individuals who have been implanted with ambulatory intracranial EEG recording devices. Pairing up peripheral measurements to brain states could identify new relationships and insights. Another key component is the heterogeneity of the relationships indicating that pooling findings across groups is suboptimal, and that data collection will need to be done on longer time scales to allow for individualization of potential seizure-forecasting algorithms.