Scholarlink

<p>This paper presents DOFANet, a novel deep learning architecture for real-time seizure prediction using EEG signals. The proposed model integrates three key modules: Dual-Octave Convolution for multiscale temporal-spectral feature extraction, Zero-Shortening to preserve signal boundaries, and a Fusion Attention Block to emphasize informative EEG regions. Preprocessed EEG data is segmented and used to extract handcrafted time and frequency domain features, which are fed into DOFANet for binary classification. The framework incorporates clinically relevant intervals—Seizure Prediction Horizon and Seizure Occurrence Period—to support timely alerting. Comprehensive experiments on CHB-MIT and Siena datasets validate the effectiveness, robustness, and generalizability of the model, achieving up to 99.58% accuracy, 99.62% sensitivity, and 99.55% F1-score on CHB-MIT, and 99.05% accuracy, 98.90% sensitivity, and 98.75% F1-score on Siena. Cross-dataset validation confirms the robustness and generalizability of the proposed model, highlighting its potential for integration into clinical and embedded seizure monitoring systems.</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

Efficient EEG-based seizure prediction using DOFANet: a deep learning approach

  • 期刊论文

  • 发表日期：

    2025年7月25日

查看全文
  • Prabhat Kumar Upadhyay,
  • Priyaranjan Kumar,
  • Sudhansu Kumar Mishra,
  • Monica Bhutani

摘要

This paper presents DOFANet, a novel deep learning architecture for real-time seizure prediction using EEG signals. The proposed model integrates three key modules: Dual-Octave Convolution for multiscale temporal-spectral feature extraction, Zero-Shortening to preserve signal boundaries, and a Fusion Attention Block to emphasize informative EEG regions. Preprocessed EEG data is segmented and used to extract handcrafted time and frequency domain features, which are fed into DOFANet for binary classification. The framework incorporates clinically relevant intervals—Seizure Prediction Horizon and Seizure Occurrence Period—to support timely alerting. Comprehensive experiments on CHB-MIT and Siena datasets validate the effectiveness, robustness, and generalizability of the model, achieving up to 99.58% accuracy, 99.62% sensitivity, and 99.55% F1-score on CHB-MIT, and 99.05% accuracy, 98.90% sensitivity, and 98.75% F1-score on Siena. Cross-dataset validation confirms the robustness and generalizability of the proposed model, highlighting its potential for integration into clinical and embedded seizure monitoring systems.