Su Jianyun, Huang Zhentao, Ma Yahong, Shi Hangyu, Yang Yuyao, Xi Min, Li Bei
Neurosurgery Department, Affiliate Children's Hospital of Xi'an Jiaotong University, Xi'an, 710003, Shaanxi Province, China.
Xi'an Key Laboratory of High Pricision Industrial Intelligent Vision Measurement Technology, School of Electronic Information, Xijing University, Xi'an, 710123, China.
Sci Rep. 2025 Aug 30;15(1):31993. doi: 10.1038/s41598-025-17256-y.
In the field of neuroscience, epilepsy is a chronic non-communicable brain disease that affects approximately 50 million people worldwide. Electroencephalography (EEG) has become a key tool in detecting and characterizing human neurological diseases such as epilepsy. This rapid and accurate diagnosis allows doctors to provide timely and effective treatment to patients, significantly reducing the frequency of future seizures and the risk of related complications, which is crucial for ensuring the long-term health and quality of life of patients. Currently, deep learning technologies, particularly Convolutional Neural Networks (CNNs) and Long Short-Term Memory Networks (LSTMs), have demonstrated significant improvements in accuracy across various fields. However, CNNs exhibit limitations in perceiving global dependencies, while LSTMs face challenges such as gradient vanishing in long sequences. This paper proposes a novel EEG recognition model, the Epileptic EEG Detection and Recognition Model based on Multiple Attention Mechanisms and Spatiotemporal Feature Fusion (MASF). MASF consists of a hybrid attention mechanism, Transformer encoder, and dot-product attention mechanism, which directly interprets the epileptic state from the raw EEG signals, thereby eliminating the need for extensive data preprocessing and feature extraction. It is worth noting that our method achieved an accuracy of 94.19% and 72.50% on the CHB-MIT and Bonn University datasets, respectively, in ten-fold cross-validation tests. In conclusion, the MASF method for epileptic seizure ictal detection based on EEG signals demonstrates significant potential in accelerating diagnosis and improving patient prognosis, especially since it achieves high accuracy without the need for extensive data preprocessing or feature extraction. The source code and dataset can be obtained from https://github.com/Xhuangzhentao/MASF-Model-.git .
在神经科学领域,癫痫是一种慢性非传染性脑部疾病,全球约有5000万人受其影响。脑电图(EEG)已成为检测和表征癫痫等人类神经系统疾病的关键工具。这种快速准确的诊断使医生能够为患者提供及时有效的治疗,显著降低未来癫痫发作的频率和相关并发症的风险,这对于确保患者的长期健康和生活质量至关重要。目前,深度学习技术,特别是卷积神经网络(CNN)和长短期记忆网络(LSTM),在各个领域的准确性上都有显著提高。然而,CNN在感知全局依赖性方面存在局限性,而LSTM在长序列中面临梯度消失等挑战。本文提出了一种新颖的脑电识别模型,即基于多重注意力机制和时空特征融合的癫痫脑电检测与识别模型(MASF)。MASF由混合注意力机制、Transformer编码器和点积注意力机制组成,它直接从原始脑电信号中解读癫痫状态,从而无需进行大量的数据预处理和特征提取。值得注意的是,在十折交叉验证测试中,我们的方法在CHB - MIT和波恩大学数据集上分别达到了94.19%和72.50%的准确率。总之,基于脑电信号的MASF癫痫发作期检测方法在加速诊断和改善患者预后方面显示出巨大潜力,特别是因为它无需大量数据预处理或特征提取就能实现高精度。源代码和数据集可从https://github.com/Xhuangzhentao/MASF - Model-.git获取。