AUTHOR=Zeng Wei , Shan Liangmin , Su Bo , Du Shaoyi 

TITLE=Epileptic seizure detection with deep EEG features by convolutional neural network and shallow classifiers

JOURNAL=Frontiers in Neuroscience

VOLUME=Volume 17 - 2023

YEAR=2023

URL=https://www.frontiersin.org/journals/neuroscience/articles/10.3389/fnins.2023.1145526

DOI=10.3389/fnins.2023.1145526

ISSN=1662-453X

ABSTRACT=Automatic detection of epileptic seizure becomes urgent in clinical application since it could significantly reduce the burden for the care of patients suffering from intractable epilepsy.  Electroencephalography (EEG) signals record the brain's electrical activity and contain rich information about brain dysfunction. They commonly serve as a non-invasive and low-cost tool for the onset detection of epileptic seizure Visual screening of EEG recordings is tedious and subjective which requires significant improvements.  The objective of this study is to develop an automatic classification method for epileptic seizure detection from EEG recordings. A deep neural network (DNN) model is used on the raw data during the feature extraction of the EEG inputs. Deep feature maps obtained from hierarchically placed layers in convolution neural network are fed to various shallow classifiers for the anomaly detection. Principal component analysis (PCA) technique is used to reduce the high dimensions of feature maps. Finally, in order to evaluate the robustness of our proposed method, two famous epilepsy datasets namely Bonn dataset and EEG Epilepsy dataset, are used for verification. The datasets are significantly different in terms of data acquisition, clinical protocols, digital storages and signal qualities, making it challenging to process and analyze. By using 10-fold cross-validation style, experimental results demonstrate that the proposed deep features with shallow classifiers yield highest performance with accuracy of approximately  100% for binary and multi-class classification on the two datasets. The results not only manifest our proposal outperforms other state-of-the-art approaches, but also indicate that our method has the potential to be deployed in clinical settings.