Scholarlink

<p>More than 50 million people are affected by the non-communicable brain disease called as epilepsy seizure. In neuroscience studies, intelligent techniques for categorizing irregular and harmless epileptic diagnoses are crucial resources. Electroencephalogram (EEG) data show superior temporal qualities in the detection of epilepsy because it offers real-time insights into the state of the disease compared to Computed Tomography and Magnetic Resonance Imaging. Multiple seizures are a symptom of brain epilepsy. Medication or surgery is not used to treat 30% of epileptic patients. The pre-ictal area is the region of the brain that exhibits unusual activity just before a seizure occurs. If the disease can be detected at an earlier stage, the potential seizures can be easily controlled by utilizing the right medicines. Therefore, several epilepsy seizure detection approaches have been proposed yet, it is not sufficient for handling noisy and redundant EEG signals, resulting more mistaken results. It needs more training period due to the presence of varying frequency characteristics of EEG signals and it generate minimal convergence speed, leading to misdiagnosis. In order to minimizing these complexities, an efficient model is introduced in this research work for detecting seizure via EEG. From the typical resources, the EEG signal is gathered and it is given to the feature extraction performance. Here, the spectral, spatial, and temporal features are retrieved and these features are multiplied with appropriate weights that are optimally selected through the Enhanced Giant Trevally Optimizer. The weighted fused feature attained from this process is considered as feature set 1. Further, feature set 2 is taken from the spectrogram images that are attained by transforming the EEG signal. Then, both feature set 1 and feature set 2 are fed into the developed Hybrid Convolution (1D–2D) based Residual Attention Network with Recurrent Neural Network (HC-RAN-RNN) for detecting epileptic seizure. Moreover, several performance metrics are used to effectively estimate the overall performance. The detection performances of the suggested model are contrasted with other conventional mechanisms. Here, the designed framework has achieved 94% accuracy, 95% sensitivity, 93% specificity, 94% precision, and 94% F1-score value in dataset 1.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Epilepsy Seizure Detection from EEG Signal Using Hybrid Convolution (1D–2D) Based Residual Attention Network with Recurrent Neural Network

  • 期刊论文

  • 发表日期：

    2025年11月7日

查看全文
  • S. Sageengrana,
  • W. Ancy Breen,
  • V. Sabapathi,
  • K. Sundara Velrani,
  • T. Bernatin

摘要

More than 50 million people are affected by the non-communicable brain disease called as epilepsy seizure. In neuroscience studies, intelligent techniques for categorizing irregular and harmless epileptic diagnoses are crucial resources. Electroencephalogram (EEG) data show superior temporal qualities in the detection of epilepsy because it offers real-time insights into the state of the disease compared to Computed Tomography and Magnetic Resonance Imaging. Multiple seizures are a symptom of brain epilepsy. Medication or surgery is not used to treat 30% of epileptic patients. The pre-ictal area is the region of the brain that exhibits unusual activity just before a seizure occurs. If the disease can be detected at an earlier stage, the potential seizures can be easily controlled by utilizing the right medicines. Therefore, several epilepsy seizure detection approaches have been proposed yet, it is not sufficient for handling noisy and redundant EEG signals, resulting more mistaken results. It needs more training period due to the presence of varying frequency characteristics of EEG signals and it generate minimal convergence speed, leading to misdiagnosis. In order to minimizing these complexities, an efficient model is introduced in this research work for detecting seizure via EEG. From the typical resources, the EEG signal is gathered and it is given to the feature extraction performance. Here, the spectral, spatial, and temporal features are retrieved and these features are multiplied with appropriate weights that are optimally selected through the Enhanced Giant Trevally Optimizer. The weighted fused feature attained from this process is considered as feature set 1. Further, feature set 2 is taken from the spectrogram images that are attained by transforming the EEG signal. Then, both feature set 1 and feature set 2 are fed into the developed Hybrid Convolution (1D–2D) based Residual Attention Network with Recurrent Neural Network (HC-RAN-RNN) for detecting epileptic seizure. Moreover, several performance metrics are used to effectively estimate the overall performance. The detection performances of the suggested model are contrasted with other conventional mechanisms. Here, the designed framework has achieved 94% accuracy, 95% sensitivity, 93% specificity, 94% precision, and 94% F1-score value in dataset 1.