Modeling and Detection of Epileptic Seizures using Multi-modal Data Construction and Analysis Evrim Acar Canan Aykut-Bingol Haluk Bingol Rasmus Bro Anthony L. Ritaccio Bülent Yener The identification of epileptic seizures significantly relies on monitoring and visual analysis of large amounts of multichannelelectroencephalographic (EEG) signals. With a goal of automating this time-consuming and subjective task, we develop apatient-speci?c seizure recognition model for multi-channel scalpEEG signals.We differentiate between seizure and non-seizure periods byrepresenting multi-channel EEG signals using a set of features from both time and frequency domains. Our contributions arethreefold: First, we rearrange multi-channel EEG recordings asa third-order tensor called an Epilepsy Feature Tensor withmodes: time epochs, features and channels. Second, we model the Epilepsy Feature Tensor using a multi-linear discriminant analysis based on Multi-linear Partial Least Squares, which isthe generalization of Partial Least Squares regression to tensors.This two-step approach facilitates the analysis of EEG datafrom multiple channels represented by several features from different domains. Third, our multi-modal approach enables us to understand the differences between seizures of different patients by finding a subset of features capturing the seizure characteristics of each patient. We evaluate the performance of our model considering both sensitivity and specificity. Our results based on the analysis of 29 seizures from 8 patients demonstrate that multiway analysis of an Epilepsy Feature Tensor can detect patient-specific seizures with g-means (geometric mean of sensitivity and specificity) ranging between 77%-97%. Furthermore, we compare our model with a two-way model and demonstrate that our multi-modal approachcan improve a two-way analysis approach in terms of detecting and understanding epileptic seizures. Department of Computer Science, Rensselaer Polytechnic Institute cs-08-02
Modeling and Detection of Epileptic Seizures using Multi-modal Data Construction and Analysis
Evrim Acar
Canan Aykut-Bingol
Haluk Bingol
Rasmus Bro
Anthony L. Ritaccio
Bülent Yener
The identification of epileptic seizures significantly relies on monitoring and visual analysis of large amounts of multichannelelectroencephalographic (EEG) signals. With a goal of automating this time-consuming and subjective task, we develop apatient-speci?c seizure recognition model for multi-channel scalpEEG signals.We differentiate between seizure and non-seizure periods byrepresenting multi-channel EEG signals using a set of features from both time and frequency domains. Our contributions arethreefold: First, we rearrange multi-channel EEG recordings asa third-order tensor called an Epilepsy Feature Tensor withmodes: time epochs, features and channels. Second, we model the Epilepsy Feature Tensor using a multi-linear discriminant analysis based on Multi-linear Partial Least Squares, which isthe generalization of Partial Least Squares regression to tensors.This two-step approach facilitates the analysis of EEG datafrom multiple channels represented by several features from different domains. Third, our multi-modal approach enables us to understand the differences between seizures of different patients by finding a subset of features capturing the seizure characteristics of each patient. We evaluate the performance of our model considering both sensitivity and specificity. Our results based on the analysis of 29 seizures from 8 patients demonstrate that multiway analysis of an Epilepsy Feature Tensor can detect patient-specific seizures with g-means (geometric mean of sensitivity and specificity) ranging between 77%-97%. Furthermore, we compare our model with a two-way model and demonstrate that our multi-modal approachcan improve a two-way analysis approach in terms of detecting and understanding epileptic seizures.
Department of Computer Science, Rensselaer Polytechnic Institute
cs-08-02