Classification of patterns of EEG synchronization for seizure prediction

Abstract

Objective: Research in seizure prediction from intracranial EEG has highlighted the usefulness of bivariate measures of brainwave synchronization. Spatio-temporal bivariate features are very high-dimensional and cannot be analyzed with conventional statistical methods. Hence, we propose state-of-the-art machine learning methods that handle high-dimensional inputs. Methods: We computed bivariate features of EEG synchronization (cross-correlation, nonlinear interdependence, dynamical entrainment or wavelet synchrony) on the 21-patient Freiburg dataset. Features from all channel pairs and frequencies were aggregated over consecutive time points, to form patterns. Patient-specific machine learning-based classifiers (support vector machines, logistic regression or convolutional neural networks) were trained to discriminate interictal from preictal patterns of features. In this explorative study, we evaluated out-of-sample seizure prediction performance, and compared each combination of feature type and classifier. Results: Among the evaluated methods, convolutional networks combined with wavelet coherence successfully predicted all ouf-of-sample seizures, without false alarms, on 15 patients, yielding 71% sensitivity and 0 false positives. Conclusions: Our best machine learning technique applied to spatio-temporal patterns of EEG synchronization outperformed previous seizure prediction methods on the Freiburg dataset. Significance: By learning spatio-temporal dynamics of EEG synchronization, pattern recognition could capture patient-specific seizure precursors. Further investigation on additional datasets should include the seizure prediction horizon.

A review of channel selection algorithms for EEG signal processing

“…Mirowski et al [20] presented a method based on computing bivariate features of EEG synchronization. These features are cross correlation, non-linear independence, dynamic entrainment, and wavelet synchrony.…”

“…They used time aggregation for features before classification. Chang et al [22] proposed a channel selection method to reduce the feature pattern size produced from Mirowski et al work [20] for seizure prediction. Their work can be classified as a wrapper technique with a pre-specified subset of channels.…”

“…These frequency bands are delta band (0-4 Hz), theta band (3.5-7.5 Hz), alpha band (7.5-13 Hz), beta band (13)(14)(15)(16)(17)(18)(19)(20)(21)(22)(23)(24)(25)(26), and gamma band [4]. Delta waves are related to the deep sleep state.…”

Energy Distribution of EEG Signal Components by Wavelet Transform

“…Recent applications of the WT and Neural Network (NN) to engineering-medical problems can be found in several studies that refer primarily to signal processing and classification in different medical areas. Several authors used WT in different ways to analyze EEG signals and combined WT and NN in the process of classification Guo et al, 2010;Leung et al, 2009;Mirowski et al, 2009;Zandi et al, 2008).…”

Correlation between EEG channels for epileptic seizure prediction

“…Вже існує велика кількість методів прогнозування епілептичних нападів [12,9], але вони мають цілий ряд недо-ліків, основним з яких є висока обчислювальна складність цих методів, що наразі практично унеможливлює їх використання в реальних сис-темах прогнозування. В роботі [8] було виявле-но зниження синхронізації між каналами сигналу перед епілептичним нападом в сигналі ЕЕГ, са-ме тому міри синхронізації представляють най-більший інтерес серед ознак для прогнозування епілептичних нападів. Метою даної роботи є до-слідження можливості використання кореляції між каналами сигналу ЕЕГ для прогнозування епілептичних нападів.…”