Classification of patterns of EEG synchronization for seizure prediction

Piotr Mirowski, Deepak Madhavan, Yann LeCun, Ruben Kuzniecky

Research output: Contribution to journalArticle

188 Citations (Scopus)

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 out-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.

Original languageEnglish (US)
Pages (from-to)1927-1940
Number of pages14
JournalClinical Neurophysiology
Volume120
Issue number11
DOIs
StatePublished - Nov 1 2009

Fingerprint

Electroencephalography
Seizures
Brain Waves
Logistic Models
Learning
Research
Datasets
Machine Learning

Keywords

  • Classification
  • Feature extraction
  • Machine learning
  • Neural networks
  • Pattern recognition
  • Seizure prediction

ASJC Scopus subject areas

  • Sensory Systems
  • Neurology
  • Clinical Neurology
  • Physiology (medical)

Cite this

Classification of patterns of EEG synchronization for seizure prediction. / Mirowski, Piotr; Madhavan, Deepak; LeCun, Yann; Kuzniecky, Ruben.

In: Clinical Neurophysiology, Vol. 120, No. 11, 01.11.2009, p. 1927-1940.

Research output: Contribution to journalArticle

Mirowski, P, Madhavan, D, LeCun, Y & Kuzniecky, R 2009, 'Classification of patterns of EEG synchronization for seizure prediction', Clinical Neurophysiology, vol. 120, no. 11, pp. 1927-1940. https://doi.org/10.1016/j.clinph.2009.09.002
Mirowski, Piotr ; Madhavan, Deepak ; LeCun, Yann ; Kuzniecky, Ruben. / Classification of patterns of EEG synchronization for seizure prediction. In: Clinical Neurophysiology. 2009 ; Vol. 120, No. 11. pp. 1927-1940.
@article{26401b8628ab4786957ced29633dfe9d,
title = "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 out-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.",
keywords = "Classification, Feature extraction, Machine learning, Neural networks, Pattern recognition, Seizure prediction",
author = "Piotr Mirowski and Deepak Madhavan and Yann LeCun and Ruben Kuzniecky",
year = "2009",
month = "11",
day = "1",
doi = "10.1016/j.clinph.2009.09.002",
language = "English (US)",
volume = "120",
pages = "1927--1940",
journal = "Clinical Neurophysiology",
issn = "1388-2457",
publisher = "Elsevier Ireland Ltd",
number = "11",

}

TY - JOUR

T1 - Classification of patterns of EEG synchronization for seizure prediction

AU - Mirowski, Piotr

AU - Madhavan, Deepak

AU - LeCun, Yann

AU - Kuzniecky, Ruben

PY - 2009/11/1

Y1 - 2009/11/1

N2 - 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 out-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.

AB - 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 out-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.

KW - Classification

KW - Feature extraction

KW - Machine learning

KW - Neural networks

KW - Pattern recognition

KW - Seizure prediction

UR - http://www.scopus.com/inward/record.url?scp=70350721621&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=70350721621&partnerID=8YFLogxK

U2 - 10.1016/j.clinph.2009.09.002

DO - 10.1016/j.clinph.2009.09.002

M3 - Article

C2 - 19837629

AN - SCOPUS:70350721621

VL - 120

SP - 1927

EP - 1940

JO - Clinical Neurophysiology

JF - Clinical Neurophysiology

SN - 1388-2457

IS - 11

ER -