Spatiotemporal Network Based on GCN and BiGRU for Seizure Detection

IEEE Journal of Biomedical and Health Informatics, Journal Year: 2024, Volume and Issue: 28(4), P. 2037 - 2046

Published: Jan. 4, 2024

As an important tool for detecting and diagnosing epilepsy, multi-channel EEG records the neuronal activities of different brain regions. Visual identification abnormal signals poses challenges, making use artificial intelligence techniques automated seizure detection inevitable trend. However, existing methods often overlook spatial relationship between channels, which can't take full advantage network structure. In this paper, we design end-to-end spatiotemporal architecture based on Graph Convolutional Networks (GCN) Bidirectional Gated Recurrent Units (BiGRU) to efficiently model dependence temporal dynamics EEG. Firstly, original are preprocessed by applying wavelet transform temporal-frequency analysis. The Pearson correlation matrix is computed specific frequency bands GCN utilized extract features channels. Then, these sent into BiGRU capture relationships. Finally, decisions achieved using fully connected layers multi-level decision rules implemented provide final results. proposed method validated CHB-MIT dataset, achieving 98.85% sensitivity, 95.83% specificity, 97.35% accuracy, 97.4% F1-score, 97.33% AUC. This fusions multiple characteristics in spatial-temporal-frequency domains improve performance promising result demonstrates that superior or par with methods.

Language: Английский

Combining EEG Features and Convolutional Autoencoder for Neonatal Seizure Detection

International Journal of Neural Systems, Journal Year: 2024, Volume and Issue: 34(08)

Published: April 19, 2024

Neonatal epilepsy is a common emergency phenomenon in neonatal intensive care units (NICUs), which requires timely attention, early identification, and treatment. Traditional detection methods mostly use supervised learning with enormous labeled data. Hence, this study offers semi-supervised hybrid architecture for detecting seizures, combines the extracted electroencephalogram (EEG) feature dataset convolutional autoencoder, called Fd-CAE. First, various features time domain entropy are to characterize EEG signal, helps distinguish epileptic seizures subsequently. Then, unlabeled fed into autoencoder (CAE) training, effectively represents by optimizing loss between input output features. This unsupervised process can better combine optimize from After that, pre-trained encoder part of model used further data obtain its low-dimensional representation achieve classification. performed on collected at University Helsinki Hospital, has high discriminative ability detect an accuracy 92.34%, precision 93.61%, recall rate 98.74%, F1-score 95.77%, respectively. The results show that CAE beneficial characterization signals, proposed Fd-CAE method significantly improves classification performance.

Language: Английский

A hybrid network based on multi-scale convolutional neural network and bidirectional gated recurrent unit for EEG denoising

Neuroscience, Journal Year: 2025, Volume and Issue: unknown

Published: Feb. 1, 2025

Language: Английский