基于多域特征和GATv2网络的癫痫发作预测方法

doi:10.3785/j.issn.1008-973X.2026.08.015

浙江大学学报(工学版)

2026, Vol. 60

Issue (8): 1760-1769 DOI: 10.3785/j.issn.1008-973X.2026.08.015

计算机技术

基于多域特征和GATv2网络的癫痫发作预测方法

韩哲1,2(

),孟庆芳1,2,*(

),张强3,张相龙1,2,赵亚欧1,2

1. 济南大学信息科学与工程学院，山东济南 250022
2. 济南大学山东省泛在智能计算重点实验室，山东济南 250022
3. 济南市晶恒电子有限责任公司，山东济南 250014

Seizure prediction method based on multi-domain feature and GATv2 network

Zhe HAN1,2(

),Qingfang MENG1,2,*(

),Qiang ZHANG3,Xianglong ZHANG1,2,Yaou ZHAO1,2

1. School of Information Science and Engineering, University of Jinan, Jinan 250022, China
2. Shandong Key Laboratory of Ubiquitous Intelligent Computing, University of Jinan, Jinan 250022, China
3. Jinan Jingheng Electronics Limited Company, Jinan 250014, China

全文: PDF(1818 KB) HTML

摘要：

为了降低癫痫患者的发作伤害风险并优化治疗策略，提出基于多域特征融合和图注意力网络的癫痫发作预测方法. 提取时域、频域和复杂网络域的特征，采用新颖的三元注意力机制对特征进行深度融合，构建节点特征表示. 建模电极间的空间拓扑关系生成邻接矩阵，节点特征矩阵与邻接矩阵联合输入GATv2模型进行深度分析，实现癫痫发作的精准预测. 基于CHB-MIT数据集的实验结果表明，所提方法在癫痫发作预测任务中的分类准确率达到97.58%.

关键词： 癫痫发作预测; 脑电图; 复杂网络; 多域特征; 图注意力网络

Abstract:

To reduce seizure-related injury risks for seizure patients and optimize treatment strategies, an seizure prediction method based on multi-domain feature fusion and a graph attention network was proposed. Multi-dimensional features were extracted from the time domain, the frequency domain, and the complex network domain. A novel triple attention mechanism was employed for deep feature fusion, constructing node feature representations. Additionally, by modeling the spatial topological relationships between electrodes, an adjacency matrix was generated. The node feature matrix, together with the generated adjacency matrix, was then fed into a GATv2 model for deep analysis, achieving accurate seizure prediction. Experimental results on the CHB-MIT dataset demonstrate that the proposed method achieves a classification accuracy of 97.58% in seizure prediction tasks.

Key words: seizure prediction EEG complex network multi-domain feature graph attention network

收稿日期: 2025-07-18 出版日期: 2026-07-16

CLC:

TP 391

基金资助: 山东省自然科学基金资助项目（ZR2024MF124）；济南大学学科交叉融合建设项目 2023（XKJC-202308）；济南市高校创新团队项目（2019GXRC015）.

通讯作者: 孟庆芳 E-mail: 202321100407@stu.ujn.edu.cn;ise_mengqf@ujn.edu.cn

作者简介: 韩哲（2000—），男，硕士生，从事时间序列研究. orcid.org/0009-0009-2859-4828. E-mail：202321100407@stu.ujn.edu.cn

	服务
	把本文推荐给朋友
	加入引用管理器
	E-mail Alert
	作者相关文章
	韩哲
	孟庆芳
	张强
	张相龙
	赵亚欧

引用本文:

韩哲,孟庆芳,张强,张相龙,赵亚欧. 基于多域特征和GATv2网络的癫痫发作预测方法[J]. 浙江大学学报(工学版), 2026, 60(8): 1760-1769.

Zhe HAN,Qingfang MENG,Qiang ZHANG,Xianglong ZHANG,Yaou ZHAO. Seizure prediction method based on multi-domain feature and GATv2 network. Journal of ZheJiang University (Engineering Science), 2026, 60(8): 1760-1769.

链接本文:

https://www.zjujournals.com/eng/CN/10.3785/j.issn.1008-973X.2026.08.015 或 https://www.zjujournals.com/eng/CN/Y2026/V60/I8/1760

图 1 所提癫痫发作预测方法的整体工作流程

表 1 时域的统计特征

图 2 基于加权邻域图的脑电信号复杂网络构建流程图

图 3 三元注意力机制结构图

表 2 所提癫痫发作预测方法在CHB-MIT数据集上的评估结果

表 3 所提癫痫发作预测方法在TUSZ数据集上的评估结果

表 4 跨被试者实验结果

表 5 不同预测方法的评估结果对比

表 6 不同学习方法的评估结果比较

表 7 多维特征对模型性能的影响

表 8 衰减参数对模型性能的影响

表 9 模块消融实验结果

3	NISO G, ROMERO E, MOREAU J T, et al Wireless EEG: a survey of systems and studies[J]. NeuroImage, 2023, 269: 119774 doi: 10.1016/j.neuroimage.2022.119774
4	ZHANG H, ZHOU Q Q, CHEN H, et al The applied principles of EEG analysis methods in neuroscience and clinical neurology[J]. Military Medical Research, 2023, 10 (1): 67 doi: 10.1186/s40779-023-00502-7
5	XIANG J, LI Y, WU X, et al Synchronization-based graph spatio-temporal attention network for seizure prediction[J]. Scientific Reports, 2025, 15: 4080 doi: 10.1038/s41598-025-88492-5
6	LI C, DENG Z, SONG R, et al EEG-based seizure prediction via model uncertainty learning[J]. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 2023, 31: 180- 191 doi: 10.1109/TNSRE.2022.3217929
7	GAO Y, CHEN X, LIU A, et al Pediatric seizure prediction in scalp EEG using a multi-scale neural network with dilated convolutions[J]. IEEE Journal of Translational Engineering in Health and Medicine, 2022, 10: 4900209 doi: 10.1109/jtehm.2022.3144037
8	张培玲, 候康基于多尺度自适应残差网络的癫痫检测方法[J]. 中国医学物理学杂志, 2025, 42 (3): 381- 387 ZHANG Peiling, HOU Kang Epilepsy detection method based on multi-scale adaptive residual network[J]. Chinese Journal of Medical Physics, 2025, 42 (3): 381- 387 doi: 10.3969/j.issn.1005-202X.2025.03.015
9	张喜珍, 张晓莉, 吕洋, 等基于2D-CNN和Cox-Stuart早停机制的癫痫预测模型[J]. 中国医学物理学杂志, 2025, 42 (1): 82- 94 ZHANG Xizhen, ZHANG Xiaoli, LÜYang, et al Epilepsy prediction model based on 2D-CNN and Cox-Stuart early stopping mechanism[J]. Chinese Journal of Medical Physics, 2025, 42 (1): 82- 94
10	SHARMILA A, GEETHANJALI P Evaluation of time domain features on detection of epileptic seizure from EEG signals[J]. Health and Technology, 2020, 10 (3): 711- 722 doi: 10.1007/s12553-019-00363-y
11	SHAKER H, LI J, KOBAYASHI M, et al Is high-frequency activity at seizure onset inhibitory? A stereoelectroencephalographic study of motor cortex seizures[J]. Annals of Neurology, 2024, 95 (6): 1127- 1137 doi: 10.1002/ana.26883
12	DONG X, HE L, LI H, et al Deep learning based automatic seizure prediction with EEG time-frequency representation[J]. Biomedical Signal Processing and Control, 2024, 95: 106447 doi: 10.1016/j.bspc.2024.106447
1	MAIMAITI B, MENG H, LV Y, et al An overview of EEG-based machine learning methods in seizure prediction and opportunities for neurologists in this field[J]. Neuroscience, 2022, 481: 197- 218 doi: 10.1016/j.neuroscience.2021.11.017
2	SHOKA A A E, DESSOUKY M M, EL-SAYED A, et al EEG seizure detection: concepts, techniques, challenges, and future trends[J]. Multimedia Tools and Applications, 2023, 82 (27): 42021- 42051 doi: 10.1007/s11042-023-15052-2
13	YU X, WU Y, MENG F, et al A review of graph and complex network theory in water distribution networks: mathematical foundation, application and prospects[J]. Water Research, 2024, 253: 121238 doi: 10.1016/j.watres.2024.121238
14	LIU Y, ZHOU H, GUAN M, et al Scalp EEG-based automatic detection of epileptiform events via graph convolutional network and bi-directional LSTM co-embedded broad learning system[J]. IEEE Signal Processing Letters, 2023, 30: 448- 452 doi: 10.1109/LSP.2023.3263433
15	CHEN X, ZHENG Y, NIU Y, et al. Epilepsy classification for mining deeper relationships between EEG channels based on GCN [C]// Proceedings of the International Conference on Computer Vision, Image and Deep Learning. Chongqing: IEEE, 2020: 701–706.
16	WANG J, GAO R, ZHENG H, et al SSGCNet: a sparse spectra graph convolutional network for epileptic EEG signal classification[J]. IEEE Transactions on Neural Networks and Learning Systems, 2024, 35 (9): 12157- 12171 doi: 10.1109/TNNLS.2023.3252569
17	HUANG Z, SU L, WU J, et al Rock image classification based on EfficientNet and triplet attention mechanism[J]. Applied Sciences, 2023, 13 (5): 3180 doi: 10.3390/app13053180
18	CHANG W, LIU W SAT-GATv2: a dynamic attention-based graph neural network for solving Boolean satisfiability problem[J]. Electronics, 2025, 14 (3): 423 doi: 10.3390/electronics14030423
19	MAIWALD T, WINTERHALDER M, ASCHENBRENNER-SCHEIBE R, et al Comparison of three nonlinear seizure prediction methods by means of the seizure prediction characteristic[J]. Physica D: Nonlinear Phenomena, 2004, 194 (3/4): 357- 368 doi: 10.1016/j.physd.2004.02.013
20	ZHANG Y, GUO Y, YANG P, et al Epilepsy seizure prediction on EEG using common spatial pattern and convolutional neural network[J]. IEEE Journal of Biomedical and Health Informatics, 2020, 24 (2): 465- 474 doi: 10.1109/JBHI.2019.2933046
21	YANG X, ZHAO J, SUN Q, et al An effective dual self-attention residual network for seizure prediction[J]. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 2021, 29: 1604- 1613 doi: 10.1109/TNSRE.2021.3103210
22	GUO L, YU T, ZHAO S, et al CLEP: contrastive learning for epileptic seizure prediction using a spatio-temporal-spectral network[J]. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 2023, 31: 3915- 3926 doi: 10.1109/TNSRE.2023.3322275
23	JEMAL I, MEZGHANI N, ABOU-ABBAS L, et al An interpretable deep learning classifier for epileptic seizure prediction using EEG data[J]. IEEE Access, 2022, 10: 60141- 60150 doi: 10.1109/ACCESS.2022.3176367
24	ZHU R, PAN W X, LIU J X, et al Epileptic seizure prediction via multidimensional Transformer and recurrent neural network fusion[J]. Journal of Translational Medicine, 2024, 22 (1): 895 doi: 10.1186/s12967-024-05678-7

[1]	刘世锋,苏建宁,张书涛,王世杰,邱凯,杨文瑾. 基于双层复杂网络的产品架构设计核心驱动因素识别[J]. 浙江大学学报(工学版), 2026, 60(8): 1650-1661.
[2]	王春丽,高玉鑫,李金絮. 基于频段特异性时空频融合的听觉注意力解码分析[J]. 浙江大学学报(工学版), 2026, 60(4): 887-895.
[3]	姚梓豪,贾海蓉,李雅荣,陈桂军. 基于多尺度注意力时序编码网络的语音诱发脑电解码[J]. 浙江大学学报(工学版), 2026, 60(4): 896-905.
[4]	董镇滔,徐暟敏,万清颖,刘晓菲,申昊,李书涵,奇格奇. 基于交通事件短视频资源的多模态情绪特征分析[J]. 浙江大学学报(工学版), 2025, 59(4): 661-668.
[5]	陈文强,王东丹,朱文英,汪勇杰,王涛. 基于时空图注意力网络的车辆多模态轨迹预测模型[J]. 浙江大学学报(工学版), 2025, 59(3): 443-450.
[6]	杨延璞,孟文昊,伍智泓,刘嘉玲,卓玥鸣. 过头作业上肢肌间耦合及协同[J]. 浙江大学学报(工学版), 2025, 59(11): 2269-2276.
[7]	杨硕,娄旭,刘硕,李佳睿,王磊. 基于非参数脑网络的帕金森亚型临床差异分析[J]. 浙江大学学报(工学版), 2025, 59(11): 2451-2458.
[8]	李宜轩,李颖,肖倩,王灵月,尹宁,杨硕. 不同情绪错误记忆的脑电微状态功能网络分析[J]. 浙江大学学报(工学版), 2025, 59(1): 49-61.
[9]	舒晴,刘喜平,谭钊,李希,万常选,刘德喜,廖国琼. 基于依存关系图注意力网络的SQL生成方法[J]. 浙江大学学报(工学版), 2024, 58(5): 908-917.
[10]	马飞,委笑琳,孙启鹏,刘擎,苟慧艳. 考虑互补效应的城市群多模式客运网络鲁棒性[J]. 浙江大学学报(工学版), 2024, 58(2): 388-398.
[11]	周逸凡,张灵维,周正东,蔡智,袁梦瑶,袁晓曦,杨泽毅. 基于注意力机制和深度学习的群体语言想象脑电信号分类[J]. 浙江大学学报(工学版), 2024, 58(12): 2540-2546.
[12]	张书涛,杨志强,王世杰,刘世锋,张凡,周爱民. 基于复杂网络的用户社区子群需求聚类模型[J]. 浙江大学学报(工学版), 2023, 57(9): 1843-1855.
[13]	张灵维,周正东,许云飞,王嘉文,吉文韬,宋泽峰. 基于特征融合的语言想象脑电信号分类[J]. 浙江大学学报(工学版), 2023, 57(4): 726-734.
[14]	苏丰龙,景宁. 基于关系聚合的时序知识图谱表示学习[J]. 浙江大学学报(工学版), 2023, 57(2): 235-242.
[15]	陈成,张皞,李永强,冯远静. 关系生成图注意力网络的知识图谱链接预测[J]. 浙江大学学报(工学版), 2022, 56(5): 1025-1034.

Viewed

Full text

Abstract

Cited

Shared

Discussed