Epilepsy detection with artificial neural network based on as-fabricated neuromorphic chip platform

Abstract

Epilepsy is a serious neurological condition caused by a sudden abnormality of brain neurons. An accurate epilepsy detection based on electroencephalogram (EEG) signals can provide vital information for diagnosis and treatment. In this study, we propose a lightweight automatic epilepsy detection system with artificial neural network based on our as-fabricated neuromorphic chip. The proposed system utilizes a neural network model to achieve high-accuracy detection without the need for epilepsy-related prior knowledge. The model uses a filter module and a convolutional neural network to preprocess the raw EEG signal and uses a long short-term memory recurrent neural network and a fully connected network as the classifier. In the examination, the classification accuracy of the normal cases and seizures approaches 99.10%, and the accuracy of the normal cases, and interictal and seizure cases can reach 94.46%. This design provides possible epilepsy detection in wearable or portable devices.

Detection of Epileptic Seizures from EEG Signals Using Machine Learning Classifiers

“…The CNN is a powerful tool that utilizes deep learning algorithm to extract features from signals [22][23][24]. It also offers the attribute of signal classification.…”

“…Recently, deep learning algorithms based CNN have found widespread applications in the classification and detection of various signals [22][23][24]. The utmost advantage of using CNN is that it offers the enormous capability to extract features from the signals.…”

“…The utmost advantage of using CNN is that it offers the enormous capability to extract features from the signals. In such scheme, raw EEG signals can directly be applied to the CNN for highly-accurate detection of epileptic seizures [22][23][24]. For instances, the 2-class classification of EEG signals can be accomplished with accuracies of 99.86% [22], 100% [23] and 97.03% using deep learning algorithm based CNN.…”

“…In such scheme, raw EEG signals can directly be applied to the CNN for highly-accurate detection of epileptic seizures [22][23][24]. For instances, the 2-class classification of EEG signals can be accomplished with accuracies of 99.86% [22], 100% [23] and 97.03% using deep learning algorithm based CNN. For 3-class classification of EEG signals, CNN classifiers can provide accuracies of 94.46% [22] and 97.07% [23], which are much better than 91.70% accuracy obtained with multiscale entropy based SVM [6].…”

“…For instances, the 2-class classification of EEG signals can be accomplished with accuracies of 99.86% [22], 100% [23] and 97.03% using deep learning algorithm based CNN. For 3-class classification of EEG signals, CNN classifiers can provide accuracies of 94.46% [22] and 97.07% [23], which are much better than 91.70% accuracy obtained with multiscale entropy based SVM [6]. For 5-class classification, the accuracy for using CNN have been reported to be 95.84% in the research work [23], which is also comparable to the classification accuracy of 96% obtained with octal pattern and wavelet transform based KNN [25].…”