Hinnerk Feldwisch-Drentrup

Joining the benefits: Combining epileptic seizure prediction methods

Purpose:  In recent years, a variety of methods developed in the field of linear and nonlinear time series analysis have been used to obtain reliable predictions of epileptic seizures. Because individual methods for seizure prediction so far have shown statistical significance but insufficient performance for clinical applications, we investigated possible improvements by combining algorithms capturing different aspects of electroencephalogram (EEG) dynamics.

The EPILEPSIAE database: an extensive electroencephalography database of epilepsy patients.

From the very beginning the seizure prediction community faced problems concerning evaluation, standardization, and reproducibility of its studies. One of the main reasons for these shortcomings was the lack of access to high‐quality long‐term electroencephalography (EEG) data. In this article we present the EPILEPSIAE database, which was made publicly available in 2012. We illustrate its content and scope. The EPILEPSIAE database provides long‐term EEG recordings of 275 patients as well as extensive metadata and standardized annotation of the data sets. It will adhere to the current standards in the field of prediction and facilitate reproducibility and comparison of those studies. Beyond seizure prediction, it may also be of considerable benefit for studies focusing on seizure detection, basic neurophysiology, and other fields.

EPILAB: A software package for studies on the prediction of epileptic seizures

A Matlab®-based software package, EPILAB, was developed for supporting researchers in performing studies on the prediction of epileptic seizures. It provides an intuitive and convenient graphical user interface. Fundamental concepts that are crucial for epileptic seizure prediction studies were implemented. This includes, for example, the development and statistical validation of prediction methodologies in long-term continuous recordings. Seizure prediction is usually based on electroencephalography (EEG) and electrocardiography (ECG) signals. EPILAB is able to process both EEG and ECG data stored in different formats. More than 35 time and frequency domain measures (features) can be extracted based on univariate and multivariate data analysis. These features can be post-processed and used for prediction purposes. The predictions may be conducted based on optimized thresholds or by applying classifications methods such as artificial neural networks, cellular neuronal networks, and support vector machines. EPILAB proved to be an efficient tool for seizure prediction, and aims to be a way to communicate, evaluate, and compare results and data among the seizure prediction community.

Slow modulations of high-frequency activity (40–140 Hz) discriminate preictal changes in human focal epilepsy

Recent evidence suggests that some seizures are preceded by preictal changes that start from minutes to hours before an ictal event. Nevertheless an adequate statistical evaluation in a large database of continuous multiday recordings is still missing. Here, we investigated the existence of preictal changes in long-term intracranial recordings from 53 patients with intractable partial epilepsy (in total 531 days and 558 clinical seizures). We describe a measure of brain excitability based on the slow modulation of high-frequency gamma activities (40–140 Hz) in ensembles of intracranial contacts. In prospective tests, we found that this index identified preictal changes at levels above chance in 13.2% of the patients (7/53), suggesting that results may be significant for the whole group (p < 0.05). These results provide a demonstration that preictal states can be detected prospectively from EEG data. They advance understanding of the network dynamics leading to seizure and may help develop novel seizure prediction algorithms.

Identification of preseizure states in epilepsy: a data-driven approach for multichannel EEG recordings

The retrospective identification of preseizure states usually bases on a time-resolved characterization of dynamical aspects of multichannel neurophysiologic recordings that can be assessed with measures from linear or non-linear time series analysis. This approach renders time profiles of a characterizing measure – so-called measure profiles – for different recording sites or combinations thereof. Various downstream evaluation techniques have been proposed to single out measure profiles that carry potential information about preseizure states. These techniques, however, rely on assumptions about seizure precursor dynamics that might not be generally valid or face the statistical problem of multiple testing. Addressing these issues, we have developed a method to preselect measure profiles that carry potential information about preseizure states, and to identify brain regions associated with seizure precursor dynamics. Our data-driven method is based on the ratio S of the global to local temporal variance of measure profiles. We evaluated its suitability by retrospectively analyzing long-lasting multichannel intracranial EEG recordings from 18 patients that included 133 focal onset seizures, using a bivariate measure for the strength of interactions. In 17/18 patients, we observed S to be significantly correlated with the predictive performance of measure profiles assessed retrospectively by means of receiver-operating-characteristic statistics. Predictive performance was higher for measure profiles preselected with S than for a manual selection using information about onset and spread of seizures. Across patients, highest predictive performance was not restricted to recordings from focal areas, thus supporting the notion of an extended epileptic network in which even distant brain regions contribute to seizure generation. We expect our method to provide further insight into the complex spatial and temporal aspects of the seizure generating process.

The role of high-quality EEG databases in the improvement and assessment of seizure prediction methods

Initially, seizure prediction was based on the analysis of brief EEG segments preceding clinically manifest seizures. Whereas such approaches suggested that the sensitivities of various EEG-derived features in predicting seizures were high, the inclusion of longer interictal periods and the combined assessment of sensitivity and specificity and the application of statistical validation methods have put into question the validity of such claims. We here show that the duration of EEG on which analyses are based and the number of seizures assessed negatively correlate with the reported sensitivities of prediction studies. Methodological aspects of seizure prediction are discussed in the framework of currently existing databases and of the newly established European Union database. This article is part of a Supplemental Special Issue entitled The Future of Automated Seizure Detection and Prediction.

Anticipating the unobserved: Prediction of subclinical seizures

Subclinical seizures (SCS) have rarely been considered in the diagnosis and therapy of epilepsy and have not been systematically analyzed in studies on seizure prediction. Here, we investigate whether predictions of subclinical seizures are feasible and how their occurrence may affect the performance of prediction algorithms. Using the European database of long-term recordings of surface and invasive electroencephalography data, we analyzed the data from 21 patients with SCS, including in total 413 clinically manifest seizures (CS) and 3341 SCS. Based on the mean phase coherence we investigated the predictive performance of CS and SCS. The two types of seizures had similar prediction sensitivities. Significant performance was found considerably more often for SCS than for CS, especially for patients with invasive recordings. When analyzing false alarms triggered by predicting CS, a significant number of these false predictions were followed by SCS for 9 of 21 patients. Although currently observed prediction performance may not be deemed sufficient for clinical applications for the majority of the patients, it can be concluded that the prediction of SCS is feasible on a similar level as for CS and allows a prediction of more of the seizures impairing patients, possibly also reducing the number of false alarms that were in fact correct predictions of CS. This article is part of a Supplemental Special Issue entitled The Future of Automated Seizure Detection and Prediction.

Seizure prediction in epilepsy: From circadian concepts via probabilistic forecasting to statistical evaluation

Seizure prediction performance is hampered by high numbers of false predictions. Here we present an approach to reduce the number of false predictions based on circadian concepts. Based on eight representative patients we demonstrate that this approach increases the performance considerably. The fraction of patients for whom we found a significant seizure prediction performance was increased from 25% to 38% by accounting for circadian dependencies.

A common strategy and database to compare the performance of seizure prediction algorithms

A reliable algorithm for the timely prediction of epileptic seizures would be a milestone in epilepsy research. Prediction performances have so far been determined using retrospective data assessment, leaving open the question as to whether they prove statistically significant and clinically useful under prospective conditions. To this aim, a Seizure Prediction Competition has been set up. Here, the background and the details of this competition are described.

Brainatic: A System for Real-Time Epileptic Seizure Prediction

A new system developed for real-time scalp EEG-based epileptic seizure prediction is presented, based on real time classification by machine learning methods, and named Brainatic. The system enables the consideration of previously trained classifiers for real-time seizure prediction. The software facilitates the computation of 22 univariate measures (features) per electrode, and classification using support vector machines (SVM), multilayer perceptron (MLP) neural networks and radial basis functions (RBF) neural networks. Brainatic was able to operate in real-time on a dual Intel® AtomTM netbook with 2GB of RAM, and was used to perform the clinical and ambulatory tests of the EU project EPILEPSIAE.