Bruno Direito

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.

Epileptic seizure predictors based on computational intelligence techniques

The ability of computational intelligence methods to predict epileptic seizures is evaluated in long-term EEG recordings of 278 patients suffering from pharmaco-resistant partial epilepsy, also known as refractory epilepsy. This extensive study in seizure prediction considers the 278 patients from the European Epilepsy Database, collected in three epilepsy centres: Hôpital Pitié-là-Salpêtrière, Paris, France; Universitätsklinikum Freiburg, Germany; Centro Hospitalar e Universitário de Coimbra, Portugal. For a considerable number of patients it was possible to find a patient specific predictor with an acceptable performance, as for example predictors that anticipate at least half of the seizures with a rate of false alarms of no more than 1 in 6 h (0.15 h⁻¹). We observed that the epileptic focus localization, data sampling frequency, testing duration, number of seizures in testing, type of machine learning, and preictal time influence significantly the prediction performance. The results allow to face optimistically the feasibility of a patient specific prospective alarming system, based on machine learning techniques by considering the combination of several univariate (single-channel) electroencephalogram features. We envisage that this work will serve as benchmark data that will be of valuable importance for future studies based on the European Epilepsy Database.

Modeling epileptic brain states using EEG spectral analysis and topographic mapping.

Changes in the spatio-temporal behavior of the brain electrical activity are believed to be associated to epileptic brain states. We propose a novel methodology to identify the different states of the epileptic brain, based on the topographic mapping of the time varying relative power of delta, theta, alpha, beta and gamma frequency sub-bands, estimated from EEG. Using normalized-cuts segmentation algorithm, points of interest are identified in the topographic mappings and their trajectories over time are used for finding out relations with epileptogenic propagations in the brain. These trajectories are used to train a Hidden Markov Model (HMM), which models the different epileptic brain states and the transition among them. Applied to 10 patients suffering from focal seizures, with a total of 30 seizures over 497.3h of data, the methodology shows good results (an average point-by-point accuracy of 89.31%) for the identification of the four brain states--interictal, preictal, ictal and postictal. The results suggest that the spatio-temporal dynamics captured by the proposed methodology are related to the epileptic brain states and transitions involved in focal seizures.

Output regularization of SVM seizure predictors: Kalman Filter versus the “Firing Power” method

Two methods for output regularization of support vector machines (SVMs) classifiers were applied for seizure prediction in 10 patients with long-term annotated data. The output of the classifiers were regularized by two methods: one based on the Kalman Filter (KF) and other based on a measure called the “Firing Power” (FP). The FP is a quantification of the rate of the classification in the preictal class in a past time window. In order to enable the application of the KF, the classification problem was subdivided in a two two-class problem, and the real-valued output of SVMs was considered. The results point that the FP method raise less false alarms than the KF approach. However, the KF approach presents an higher sensitivity, but the high number of false alarms turns their applicability negligible in some situations.

A Realistic Seizure Prediction Study Based on Multiclass SVM

A patient-specific algorithm, for epileptic seizure prediction, based on multiclass support-vector machines (SVM) and using multi-channel high-dimensional feature sets, is presented. The feature sets, combined with multiclass classification and post-processing schemes aim at the generation of alarms and reduced influence of false positives. This study considers 216 patients from the European Epilepsy Database, and includes 185 patients with scalp EEG recordings and 31 with intracranial data. The strategy was tested over a total of 16,729.80[Formula: see text]h of inter-ictal data, including 1206 seizures. We found an overall sensitivity of 38.47% and a false positive rate per hour of 0.20. The performance of the method achieved statistical significance in 24 patients (11% of the patients). Despite the encouraging results previously reported in specific datasets, the prospective demonstration on long-term EEG recording has been limited. Our study presents a prospective analysis of a large heterogeneous, multicentric dataset. The statistical framework based on conservative assumptions, reflects a realistic approach compared to constrained datasets, and/or in-sample evaluations. The improvement of these results, with the definition of an appropriate set of features able to improve the distinction between the pre-ictal and nonpre-ictal states, hence minimizing the effect of confounding variables, remains a key aspect.

Combining Energy and Wavelet Transform for Epileptic Seizure Prediction in an Advanced Computational System

Seizure prediction in epileptic patients will allow a deep improvement in their quality of life. In the paper a new method using energy relative measures in wavelet coefficients is proposed and tested in several patients. The results show the potential of the technique, but also its limitations, stressing the needs for further work in a larger number of patients, using multimodal information and an advanced database with a large features set to be used in seizure prediction An advanced computational framework is under development, using multisensorial information to build a large set of features to be used in a classification system supporting seizure prediction. This system is composed of two main parts the algorithms base and the database, briefly described.

Optimized feature subsets for epileptic seizure prediction studies

The reduction of the number of EEG features to give as inputs to epilepsy seizure predictors is a needed step towards the development of a transportable device for real-time warning. This paper presents a comparative study of three feature selection methods, based on Support Vector Machines. Minimum-Redundancy Maximum-Relevance, Recursive Feature Elimination, Genetic Algorithms, show that, for three patients of the European Database on Epilepsy, the most important univariate features are related to spectral information and statistical moments.

Towards Personalized Neural Networks for Epileptic Seizure Prediction

Seizure prediction for untreatable epileptic patients, one of the major challenges of present neuroinformatics researchers, will allow a substantial improvement in their safety and quality of life. Neural networks, because of their plasticity and degrees of freedom, seem to be a good approach to consider the enormous variability of physiological systems. Several architectures and training algorithms are comparatively proposed in this work showing that it is possible to find an adequate network for one patient, but care must be taken to generalize to other patients. It is claimed that each patient will have his (her) own seizure prediction algorithms.

Epileptic seizure prediction based on a bivariate spectral power methodology

The spectral power of 5 frequently considered frequency bands (Alpha, Beta, Gamma, Theta and Delta) for 6 EEG channels is computed and then all the possible pairwise combinations among the 30 features set, are used to create a 435 dimensional feature space. Two new feature selection methods are introduced to choose the best candidate features among those and to reduce the dimensionality of this feature space. The selected features are then fed to Support Vector Machines (SVMs) that classify the cerebral state in preictal and non-preictal classes. The outputs of the SVM are regularized using a method that accounts for the classification dynamics of the preictal class, also known as “Firing Power” method. The results obtained using our feature selection approaches are compared with the ones obtained using minimum Redundancy Maximum Relevance (mRMR) feature selection method. The results in a group of 12 patients of the EPILEPSIAE database, containing 46 seizures and 787 hours multichannel recording for out-of-sample data, indicate the efficiency of the bivariate approach as well as the two new feature selection methods. The best results presented sensitivity of 76.09% (35 of 46 seizures predicted) and a false prediction rate of 0.15-1.

Feature selection in high dimensional EEG features spaces for epileptic seizure prediction

Abstract Digital signal processing of Electroencephalogram (EEG) can support the diagnosis and alarming for the benefit of humans. About one third of all epileptic patients suffer from refractory epilepsy; seizure prediction based on the EEG information content is an area of intense activity since at least twenty years. In this paper we analyze the high dimensional feature space created by a variety of feature extraction methods for prediction of epileptic seizures. We combined features selection algorithm minimum redundancy maximum relevance (mRMR) and Support Vector Machines (SVMs) architectures to study the best features set for seizure prediction. We present the comparison between the classification results obtained by a feature set composed by 147 features and a reduced set based on the first 20-ranked features using mRMR scores. We critically discuss the composition of the feature subset. The results suggest some patient specificity in features and channel selection. The best models lead us to hypothesize the preference for wider preictal periods.