Detecting epileptic seizures with electroencephalogram via a context-learning model

BMC Medical Informatics and Decision Making

Research

Xiaowei Jia¹  Aidong Zhang¹  Guangxu Xun¹ 

[1]Department of Computer Science and Engineering, SUNY at Buffalo, Buffalo, USA

关键词: Context learning;    Epileptic seizure;    Electroencephalogram;    Feature extraction;   

DOI  :  10.1186/s12911-016-0310-7

来源: Springer

【 摘 要 】

BackgroundEpileptic seizure is a serious health problem in the world and there is a huge population suffering from it every year. If an algorithm could automatically detect seizures and deliver the patient therapy or notify the hospital, that would be of great assistance. Analyzing the scalp electroencephalogram (EEG) is the most common way to detect the onset of a seizure. In this paper, we proposed the context-learning based EEG analysis for seizure detection (Context-EEG) algorithm.MethodsThe proposed method aims at extracting both the hidden inherent features within EEG fragments and the temporal features from EEG contexts. First, we segment the EEG signals into EEG fragments of fixed length. Second, we learn the hidden inherent features from each fragment with a sparse auto-encoder and thus the dimensionality of the original data is reduced. Third, we translate each EEG fragment to an EEG word so that a continuous EEG signal is converted to a sequence of EEG words. Fourth, by analyzing the context information of EEG words, we learn the temporal features for EEG signals. And finally, we concatenate the hidden features and the temporal features together to train a binary classifier which can be used to detect the onset of an epileptic sezure.Results4302 EEG fragments from four different patients are used to train and test our model. An error rate of 22.93 % is achieved by our model as a general, non-patient specific seizure detector. The error rate of our model is averagely 16.7 % lower than the other baseline models. Receiver operating characteristics (ROC curve) and area under curve (AUC) confirm the effectiveness of our model. Furthermore, we discuss the extracted features and how to reconstruct the original data based on the extracted features, as well as the parameter sensitivity.ConclusionsGiven a EEG fragment, by extracting high-quality features (the hidden inherent features and temporal features) from the EEG signals, our Context-EEG model is able to detect the onset of a seizure with high accuracy in real time.

【 授权许可】

CC BY   
© The Author(s) 2016

【 图 表 】

12864_2017_4179_Article_IEq34.gif

12864_2016_3425_Article_IEq2.gif

12894_2015_Article_85_TeX2GIF_IEq1.gif

12864_2017_4179_Article_IEq31.gif