基于混合注意力 Transformer 的癫痫检测方法

Epilepsy detection method based on hybrid attention Transformer

摘要

针对现有癫痫检测方法的输入信号为单一特征，无法充分提取多维特征问题，提出一种融合频域、非线性域和时域脑电信号特征的多维特征张量，分别是相位锁相值、互信息和肯德尔相关系数τ_b，并精心设计一个与之相匹配的混合注意力Transformer模型（HA-Trans），该模型巧妙融合空间注意力、交叉注意力和时间注意力机制，充分提取多维脑电特征，最终输入到全连接层实现分类。本文方法在 CHB-MIT 数据集的分类准确率为 99.17%，特异性为 99.03%，敏感性为98.95%，优于现有方法，实现癫痫的有效检测。

Abstract

Given that existing epilepsy detection methods adopt single-dimensional features as inputs and fail to fully extract multidimensional features, this study proposes a multidimensional feature tensor which integrates frequency-, nonlinear-, and time-domain electroencephalogram features: phase locking value, mutual information, and Kendall correlation coefficientτ _b. Accordingly, a hybrid attention Transformer (HA-Trans) model is developed to effectively extract multidimensional EEG featuresvia spatial attention, cross attention, and temporal attention mechanisms, and the extracted features are then input into the fully connected layer for classification. The proposed method achieves a classification accuracy of 99.17%, specificity of 99.03%, and sensitivity of 98.95% on the CHB-MIT dataset, outperforming existing methods and enabling effective epilepsy detection.

关键词

脑电信号 / 癫痫 / 张量 / 混合注意力 / Transformer

Key words

electroencephalogram signal / epilepsy / tensor / hybrid attention / Transformer

参考文献

[1] Xin Q, Hu SH, Liu SQ, et al. An attention—based wavelet convolution neural network for epilepsy EEG classification[J]. IEEE Trans Neural Syst Rehabil Eng, 2022, 30: 957-966.

[2] Yu HS, Yang ZY, Dang WD . Attention mechanism—based network characteristic analysis for major depressive disorder detection[J]. Europhys Lett, 2022, 139(1): 11004.

[3] Ravish DK, Devi S, Krishnamoorthy SG . Wavelet analysis of EEG for seizure detection: coherence and phase synchrony estimation[J]. Biomed Res, 2015, 26(3): 514-524.

[4] Mohammad—Alikhani A, Jamshidpour E, Dhale S, et al. Fault diagnosis of electric motors by a channel—wise regulated CNN and differential of STFT[J]. IEEE Trans Ind Appl, 2025, 61(2): 3066-3077.

[5] Deng LF, Zhao C, Yan XH, et al. A novel approach for bearing fault diagnosis in complex environments using PSO—CWT and SA—FPN[J]. Measurement, 2025, 249: 117027.

[6] Cai XJ, Li DG, Zhang JL, et al. MA—EMD: aligned empirical decomposition for multivariate time—series forecasting[J]. Expert Syst Appl, 2025, 267: 126080.

[7] 梁绪宁, 王思琪, 杨海龙, 等 . 基于自适应张量交换和重算的大模型推理优化[J]. 计算机工程, 2025, 51(10): 27-36.

[8] Liang XN, Wang SQ, Yang HL, et al. Inference optimization for large models based on adaptive tensor swapping and recomputation[J]. Computer Engineering, 2025, 51(10): 27-36.

[9] Wang XY, Huang ZX, Zhu JW, et al. S3—Net: learning spectral—spatio self—similarity for hyperspectral image super—resolution[J]. Neural Netw, 2025, 188: 107490.

[10] Gholamipourbarogh N, Roessner V, Bluschke A, et al. Novel neural activity profiles underlying inhibitory control deficits of clinical relevance in attention—deficit/hyperactivity disorder: insights from electroencephalography tensor decomposition[J/OL]. Biol Psychiatry Cogn Neurosci Neuroimaging. (2025—05—09). https://doi.org/10.1016/j.bpsc.2025.05.001.

[11] 高熠徐 . 基于张量的智慧校园服务体系构建[J]. 信息技术与信息化, 2024(12): 13-17.

[12] Gao YX . Construction of a tensor based smart campus service system[J]. Information Technology and Informatization, 2024(12): 13-17.

[13] 李心雨, 康可涵, 彭冲 . 基于结构化张量学习的多视图聚类[J]. 计算机应用研究, 2025, 42(2): 448-454.

[14] Li XY, Kang KH, Peng C . Multi—view clustering based on structured tensor learning[J]. Application Research of Computers, 2025, 42(2): 448-454.

[15] 黎恒, 冯子扬, 徐韶华, 等 . 基于光纤传感与张量仿射聚类的隧道车辆检测方法[J]. 交通工程, 2024, 24(7): 100-104.

[16] Li H, Feng ZY, Xu SH, et al. A tunnel vehicle detection method based on fiber optic sensing and tensor affine clustering[J]. Journal of Transportation Engineering, 2024, 24(7): 100-104.

[17] Cassidy B, McBride C, Kendrick C, et al. An enhanced harmonic densely connected hybrid transformer network architecture for chronic wound segmentation utilising multi—colour space tensor merging[J]. Comput Biol Med, 2025, 192(Pt A): 110172.

[18] 李响, 艾尔肯·亥木都拉 . 基于CNN与Transformer的脑电解码研究[J/OL]. 微电子学与计算机. (2025—04—16). https://link.cnki.net/urlid/61.1123.tn.20250416.1456.006.

[19] Li X, Arkin H . Research on EEG decoding based on CNN and Transformer[J/OL]. Microelectronics & Computer. (2025—04—16). https://link.cnki.net/urlid/61.1123.tn.20250416.1456.006.

[20] Huang YJ, Chen YC, Xu SM, et al. Self—supervised learning with adaptive frequency—time attention transformer for seizure prediction and classification[J]. Brain Sci, 2025, 15(4): 382.

[21] 周振霄, 王华, 魏德健, 等 . Transformer在医学图像分割中的研究进展[J]. 计算机工程与应用, 2025, 61(20): 54-74.

[22] Zhou ZX, Wang H, Wei DJ, et al. Research progress of transformers in medical image segmentation[J]. Computer Engineering and Applications, 2025, 61(20): 54-74.

[23] 王科平, 索冰倩, 张高鹏, 等 . 基于空频分离交叉注意力Swin Transformer的小目标检测[J]. 激光与光电子学进展, 2025, 62(16): 338-349.

[24] Wang KP, Suo BQ, Zhang GP, et al. Small object detection based on spatial—frequency separated cross—attention Swin Transformer[J]. Laser & Optoelectronics Progress, 2025, 62(16): 338-349.

[25] Janjarasjitt S. Performance of epileptic single—channel scalp EEG classifications using single wavelet—based features[J]. Australas Phys Eng Sci Med, 2017, 40(1): 57-67.

[26] Gao B, Zhou JZ, Yang YY, et al. Generative adversarial network and convolutional neural network—based EEG imbalanced classification model for seizure detection[J]. Biocybern Biomed Eng, 2022, 42(1): 1-15.

[27] Tsiouris ΚΜ, Pezoulas VC, Zervakis M, et al. A long short—term memory deep learning network for the prediction of epileptic seizures using EEG signals[J]. Comput Biol Med, 2018, 99: 24-37.

[28] Harpale V, Bairagi V . An adaptive method for feature selection and extraction for classification of epileptic EEG signal in significant states[J]. J King Saud Univ Comput Inf Sci, 2021, 33(6): 668-676.

[29] He JT, Cui J, Zhang GB, et al. Spatial—temporal seizure detection with graph attention network and bi—directional LSTM architecture[J]. Biomed Signal Process Control, 2022, 78: 103908.

[30] 邓欣, 高红伟, 张家豪, 等 . 基于卷积和注意力机制的癫痫脑电检测模型[J]. 重庆邮电大学学报(自然科学版), 2023, 35(5): 927-934.

[31] Deng X, Gao HW, Zhang JH, et al. Epileptic EEG detection model based on attention mechanism and convolutional network[J]. Journal of Chongqing University of Posts and Telecommunications (Natural Science Edition), 2023, 35(5): 927-934.

[32] Abdulwahhab AH, Abdulaal AH, Thary Al—Ghrairi AH, et al. Detection of epileptic seizure using EEG signals analysis based on deep learning techniques[J]. Chaos Solitons Fractals, 2024, 181: 114700.

[33] Sun Q, Liu YJ, Li SD . Automatic seizure detection using multi—input deep feature learning networks for EEG signals[J]. J Sens, 2024, 2024(1): 8835396.

[34] Hazarika U, Borah BB, Roy S, et al. Automated detection of aberrant episodes in epileptic conditions: leveraging EEG and machine learning algorithms[J]. Bioengineering, 2025, 12(4): 355.

基金资助

国家自然科学基金(52274205)

国家重点研发计划(2018YFB1403303)