Seizure forecasting using minimally invasive, ultra-long-term subcutaneous electroencephalography: Individualized intrapatient models

Prospective validation of a seizure diary forecasting falls short

OBJECTIVE

Recently, a deep learning artificial intelligence (AI) model forecasted seizure risk using retrospective seizure diaries with higher accuracy than random forecasts. The present study sought to prospectively evaluate the same algorithm.

METHODS

We recruited a prospective cohort of 46 people with epilepsy; 25 completed sufficient data entry for analysis (median = 5 months). We used the same AI method as in our prior study. Group-level and individual-level Brier Skill Scores (BSSs) compared random forecasts and simple moving average forecasts to the AI.

RESULTS

The AI had an area under the receiver operating characteristic curve of .82. At the group level, the AI outperformed random forecasting (BSS = .53). At the individual level, AI outperformed random in 28% of cases. At the group and individual level, the moving average outperformed the AI. If pre-enrollment (nonverified) diaries (with presumed underreporting) were included, the AI significantly outperformed both comparators. Surveys showed most did not mind poor-quality LOW-RISK or HIGH-RISK forecasts, yet 91% wanted access to these forecasts.

SIGNIFICANCE

The previously developed AI forecasting tool did not outperform a very simple moving average forecasting in this prospective cohort, suggesting that the AI model should be replaced.

Concept-drifts adaptation for machine learning EEG epilepsy seizure prediction

Seizure prediction remains a challenge, with approximately 30% of patients unresponsive to conventional treatments. Addressing this issue is crucial for improving patients' quality of life, as timely intervention can mitigate the impact of seizures. In this research field, it is critical to identify the preictal interval, the transition from regular brain activity to a seizure. While previous studies have explored various Electroencephalogram (EEG) based methodologies for prediction, few have been clinically applicable. Recent studies have underlined the dynamic nature of EEG data, characterised by data changes with time, known as concept drifts, highlighting the need for automated methods to detect and adapt to these changes. In this study, we investigate the effectiveness of automatic concept drift adaptation methods in seizure prediction. Three patient-specific seizure prediction approaches with a 10-minute prediction horizon are compared: a seizure prediction algorithm incorporating a window adjustment method by optimising performance with Support Vector Machines (Backwards-Landmark Window), a seizure prediction algorithm incorporating a data-batch (seizures) selection method using a logistic regression (Seizure-batch Regression), and a seizure prediction algorithm with a dynamic integration of classifiers (Dynamic Weighted Ensemble). These methods incorporate a retraining process after each seizure and use a combination of univariate linear features and SVM classifiers. The Firing Power was used as a post-processing technique to generate alarms before seizures. These methodologies were compared with a control approach based on the typical machine learning pipeline, considering a group of 37 patients with Temporal Lobe Epilepsy from the EPILEPSIAE database. The best-performing approach (Backwards-Landmark Window) achieved results of 0.75 ± 0.33 for sensitivity and 1.03 ± 1.00 for false positive rate per hour. This new strategy performed above chance for 89% of patients with the surrogate predictor, whereas the control approach only validated 46%.

Personalized strategies of neurostimulation: from static biomarkers to dynamic closed-loop assessment of neural function

Despite considerable advancement of first choice treatment (pharmacological, physical therapy, etc.) over many decades, neurological disorders still represent a major portion of the worldwide disease burden. Particularly concerning, the trend is that this scenario will worsen given an ever expanding and aging population. The many different methods of brain stimulation (electrical, magnetic, etc.) are, on the other hand, one of the most promising alternatives to mitigate the suffering of patients and families when conventional treatment fall short of delivering efficacious treatment. With applications in virtually all neurological conditions, neurostimulation has seen considerable success in providing relief of symptoms. On the other hand, a large variability of therapeutic outcomes has also been observed, particularly in the usage of non-invasive brain stimulation (NIBS) modalities. Borrowing inspiration and concepts from its pharmacological counterpart and empowered by unprecedented neurotechnological advancement, the neurostimulation field has seen in recent years a widespread of methods aimed at the personalization of its parameters, based on biomarkers of the individuals being treated. The rationale is that, by taking into account important factors influencing the outcome, personalized stimulation can yield a much-improved therapy. Here, we review the literature to delineate the state-of-the-art of personalized stimulation, while also considering the important aspects of the type of informing parameter (anatomy, function, hybrid), invasiveness, and level of development (pre-clinical experimentation versus clinical trials). Moreover, by reviewing relevant literature on closed loop neuroengineering solutions in general and on activity dependent stimulation method in particular, we put forward the idea that improved personalization may be achieved when the method is able to track in real time brain dynamics and adjust its stimulation parameters accordingly. We conclude that such approaches have great potential of promoting the recovery of lost functions and enhance the quality of life for patients.

Comparison between epileptic seizure prediction and forecasting based on machine learning

Epilepsy affects around 1% of the population worldwide. Anti-epileptic drugs are an excellent option for controlling seizure occurrence but do not work for around one-third of patients. Warning devices employing seizure prediction or forecasting algorithms could bring patients new-found comfort and quality of life. These algorithms would attempt to detect a seizure's preictal period, a transitional moment between regular brain activity and the seizure, and relay this information to the user. Over the years, many seizure prediction studies using Electroencephalogram-based methodologies have been developed, triggering an alarm when detecting the preictal period. Recent studies have suggested a shift in view from prediction to forecasting. Seizure forecasting takes a probabilistic approach to the problem in question instead of the crisp approach of seizure prediction. In this field of study, the triggered alarm to symbolize the detection of a preictal period is substituted by a constant risk assessment analysis. The present work aims to explore methodologies capable of seizure forecasting and establish a comparison with seizure prediction results. Using 40 patients from the EPILEPSIAE database, we developed several patient-specific prediction and forecasting algorithms with different classifiers (a Logistic Regression, a 15 Support Vector Machines ensemble, and a 15 Shallow Neural Networks ensemble). Results show an increase of the seizure sensitivity in forecasting relative to prediction of up to 146% and in the number of patients that displayed an improvement over chance of up to 300%. These results suggest that a seizure forecasting methodology may be more suitable for seizure warning devices than a seizure prediction one.

Prospective validation of a seizure diary forecasting falls short

OBJECTIVE

Recently, a deep learning AI model forecasted seizure risk using retrospective seizure diaries with higher accuracy than random forecasts. The present study sought to prospectively evaluate the same algorithm.

METHODS

We recruited a prospective cohort of 46 people with epilepsy; 25 completed sufficient data entry for analysis (median 5 months). We used the same AI method as in our prior study. Group-level and individual-level Brier Skill Scores (BSS) compared random forecasts and simple moving average forecasts to the AI.

RESULTS

The AI had an AUC of 0.82. At the group level, the AI outperformed random forecasting (BSS=0.53). At the individual level, AI outperformed random in 28% of cases. At the group and individual level, the moving average outperformed the AI. If pre-enrollment (non-verified) diaries (with presumed under-reporting) were included, the AI significantly outperformed both comparators. Surveys showed most did not mind poor quality LOW-RISK or HIGH-RISK forecasts, yet 91% wanted access to these forecasts.

SIGNIFICANCE

The previously developed AI forecasting tool did not outperform a very simple moving average forecasting this prospective cohort, suggesting that the AI model should be replaced.

Seizure forecasting: Where do we stand?

A lot of mileage has been made recently on the long and winding road toward seizure forecasting. Here we briefly review some selected milestones passed along the way, which were discussed at the International Conference for Technology and Analysis of Seizures-ICTALS 2022-convened at the University of Bern, Switzerland. Major impetus was gained recently from wearable and implantable devices that record not only electroencephalography, but also data on motor behavior, acoustic signals, and various signals of the autonomic nervous system. This multimodal monitoring can be performed for ultralong timescales covering months or years. Accordingly, features and metrics extracted from these data now assess seizure dynamics with a greater degree of completeness. Most prominently, this has allowed the confirmation of the long-suspected cyclical nature of interictal epileptiform activity, seizure risk, and seizures. The timescales cover daily, multi-day, and yearly cycles. Progress has also been fueled by approaches originating from the interdisciplinary field of network science. Considering epilepsy as a large-scale network disorder yielded novel perspectives on the pre-ictal dynamics of the evolving epileptic brain. In addition to discrete predictions that a seizure will take place in a specified prediction horizon, the community broadened the scope to probabilistic forecasts of a seizure risk evolving continuously in time. This shift of gears triggered the incorporation of additional metrics to quantify the performance of forecasting algorithms, which should be compared to the chance performance of constrained stochastic null models. An imminent task of utmost importance is to find optimal ways to communicate the output of seizure-forecasting algorithms to patients, caretakers, and clinicians, so that they can have socioeconomic impact and improve patients' well-being.

Artificial intelligence in epilepsy phenotyping

Artificial intelligence (AI) allows data analysis and integration at an unprecedented granularity and scale. Here we review the technological advances, challenges, and future perspectives of using AI for electro-clinical phenotyping of animal models and patients with epilepsy. In translational research, AI models accurately identify behavioral states in animal models of epilepsy, allowing identification of correlations between neural activity and interictal and ictal behavior. Clinical applications of AI-based automated and semi-automated analysis of audio and video recordings of people with epilepsy, allow significant data reduction and reliable detection and classification of major motor seizures. AI models can accurately identify electrographic biomarkers of epilepsy, such as spikes, high-frequency oscillations, and seizure patterns. Integrating AI analysis of electroencephalographic, clinical, and behavioral data will contribute to optimizing therapy for patients with epilepsy.

Perceived seizure risk in epilepsy: Chronic electronic surveys with and without concurrent electroencephalography

OBJECTIVE

Previous studies suggested that patients with epilepsy might be able to forecast their own seizures. This study aimed to assess the relationships between premonitory symptoms, perceived seizure risk, and future and recent self-reported and electroencephalographically (EEG)-confirmed seizures in ambulatory patients with epilepsy in their natural home environments.

METHODS

Long-term e-surveys were collected from patients with and without concurrent EEG recordings. Information obtained from the e-surveys included medication adherence, sleep quality, mood, stress, perceived seizure risk, and seizure occurrences preceding the survey. EEG seizures were identified. Univariate and multivariate generalized linear mixed-effect regression models were used to estimate odds ratios (ORs) for the assessment of the relationships. Results were compared with the seizure forecasting classifiers and device forecasting literature using a mathematical formula converting OR to equivalent area under the curve (AUC).

RESULTS

Fifty-four subjects returned 10 269 e-survey entries, with four subjects acquiring concurrent EEG recordings. Univariate analysis revealed that increased stress (OR = 2.01, 95% confidence interval [CI] = 1.12-3.61, AUC = .61, p = .02) was associated with increased relative odds of future self-reported seizures. Multivariate analysis showed that previous self-reported seizures (OR = 5.37, 95% CI = 3.53-8.16, AUC = .76, p < .001) were most strongly associated with future self-reported seizures, and high perceived seizure risk (OR = 3.34, 95% CI = 1.87-5.95, AUC = .69, p < .001) remained significant when prior self-reported seizures were added to the model. No correlation with medication adherence was found. No significant association was found between e-survey responses and subsequent EEG seizures.

SIGNIFICANCE

Our results suggest that patients may tend to self-forecast seizures that occur in sequential groupings and that low mood and increased stress may be the result of previous seizures rather than independent premonitory symptoms. Patients in the small cohort with concurrent EEG showed no ability to self-predict EEG seizures. The conversion from OR to AUC values facilitates direct comparison of performance between survey and device studies involving survey premonition and forecasting.

Forecasting seizure likelihood from cycles of self-reported events and heart rate: a prospective pilot study

BACKGROUND

Seizure risk forecasting could reduce injuries and even deaths in people with epilepsy. There is great interest in using non-invasive wearable devices to generate forecasts of seizure risk. Forecasts based on cycles of epileptic activity, seizure times or heart rate have provided promising forecasting results. This study validates a forecasting method using multimodal cycles recorded from wearable devices.

METHOD

Seizure and heart rate cycles were extracted from 13 participants. The mean period of heart rate data from a smartwatch was 562 days, with a mean of 125 self-reported seizures from a smartphone app. The relationship between seizure onset time and phases of seizure and heart rate cycles was investigated. An additive regression model was used to project heart rate cycles. The results of forecasts using seizure cycles, heart rate cycles, and a combination of both were compared. Forecasting performance was evaluated in 6 of 13 participants in a prospective setting, using long-term data collected after algorithms were developed.

FINDINGS

The results showed that the best forecasts achieved a mean area under the receiver-operating characteristic curve (AUC) of 0.73 for 9/13 participants showing performance above chance during retrospective validation. Subject-specific forecasts evaluated with prospective data showed a mean AUC of 0.77 with 4/6 participants showing performance above chance.

INTERPRETATION

The results of this study demonstrate that cycles detected from multimodal data can be combined within a single, scalable seizure risk forecasting algorithm to provide robust performance. The presented forecasting method enabled seizure risk to be estimated for an arbitrary future period and could be generalised across a range of data types. In contrast to earlier work, the current study evaluated forecasts prospectively, in subjects blinded to their seizure risk outputs, representing a critical step towards clinical applications.

FUNDING

This study was funded by an Australian Government National Health & Medical Research Council and BioMedTech Horizons grant. The study also received support from the Epilepsy Foundation of America's 'My Seizure Gauge' grant.

Removing artefacts and periodically retraining improve performance of neural network-based seizure prediction models

The development of seizure prediction models is often based on long-term scalp electroencephalograms (EEGs) since they capture brain electrical activity, are non-invasive, and come at a relatively low-cost. However, they suffer from major shortcomings. First, long-term EEG is usually highly contaminated with artefacts. Second, changes in the EEG signal over long intervals, known as concept drift, are often neglected. We evaluate the influence of these problems on deep neural networks using EEG time series and on shallow neural networks using widely-used EEG features. Our patient-specific prediction models were tested in 1577 hours of continuous EEG, containing 91 seizures from 41 patients with temporal lobe epilepsy who were undergoing pre-surgical monitoring. Our results showed that cleaning EEG data, using a previously developed artefact removal method based on deep convolutional neural networks, improved prediction performance. We also found that retraining the models over time reduced false predictions. Furthermore, the results show that although deep neural networks processing EEG time series are less susceptible to false alarms, they may need more data to surpass feature-based methods. These findings highlight the importance of robust data denoising and periodic adaptation of seizure prediction models.

Perceived seizure risk in epilepsy â€" Chronic electronic surveys with and without concurrent EEG

Objective

Previous studies suggested that patients with epilepsy might be able to fore-cast their own seizures. We sought to assess the relationships of premonitory symptoms and perceived seizure risk with future and recent self-reported and EEG-confirmed seizures in the subjects living with epilepsy in their natural home environments.

Methods

We collected long-term e-surveys from ambulatory patients with and without concurrent EEG recordings. Information obtained from the e-surveys included medication compliance, sleep quality, mood, stress, perceived seizure risk and seizure occurrences preceding the survey. EEG seizures were identified. Univariate and multivariate generalized linear mixed-effect regression models were used to estimate odds ratios (ORs) for the assessment of the relationships. Results were compared with device seizure forecasting literature using a mathematical formula converting OR to equivalent area under the curve (AUC).

Results

Sixty-nine subjects returned 12,590 e-survey entries, with four subjects acquiring concurrent EEG recordings. Univariate analysis revealed increased stress (OR = 2.52, 95% CI = [1.52, 4.14], p < 0.001) and decreased mood (0.32, [0.13, 0.82], 0.02) were associated with increased relative odds of future self-reported seizures. On multivariate analysis, previous self-reported seizures (4.24, [2.69, 6.68], < 0.001) were most strongly associated with future self-reported seizures, and high perceived seizure risk (3.30, [1.97, 5.52], < 0.001) remained significant when prior self-reported seizures were added to the model. No significant association was found between e-survey responses and subsequent EEG seizures.

Significance

It appears that patients may tend to self-forecast seizures that occur in sequential groupings. Our results suggest that low mood and increased stress may be the result of previous seizures rather than independent premonitory symptoms. Patients in the small cohort with concurrent EEG showed no ability to self-predict EEG seizures. The conversion from OR to AUC values facilitates direct comparison of performance between survey and device studies involving survey premonition and forecasting.

Key points

Long-term e-surveys data and concurrent EEG signals were collected across three study sites to assess the ability of the patients to self-forecast their seizures.Patients may tend to self-forecast self-reported seizures that occur in sequential groupings.Factors, such as mood and stress, may not be independent premonitory symptoms but may be the consequence of recent seizures.No ability to self-forecast EEG confirmed seizures was observed in a small cohort with concurrent EEG validation.A mathematic relation between OR and AUC provides a means to compare forecasting performance between survey and device studies.