摘 要：单一维度的特征检测使现有基于 EEG 的癫痫诊断准确性受到限制。通过将 EEG 转换成格拉姆角场图和小波时频图，构建一种由 2 个 2 维 CNN 和 1 个 DNN 的集成深度学习模型，2 个 2 维 CNN 分别提取格拉姆角场图和小波时频图的特征并融合，将融合特征输出至 DNN 以进行癫痫融合识别。借助波恩大学的脑电数据集测试了该集成深度学习模型的有效性，结果表明，该模型对癫痫 EEG 识别的准确度、特异性以及敏感度分别为 96.5%、95.0% 以及 96.0%，整体识别性能优于传统的单神经网络模型，可为癫痫等疾病的诊断提供更好的辅助功能。
关键词：深度学习；癫痫；卷积神经网络；连续小波变换；格拉姆角场
DOI:10.19850/j.cnki.2096-4706.2022.20.002
基金项目：广西自然科学基金资助项目（2020GXNSFAA159067）； 教育部认知无线电重点实验室基金资助项目（CRKL200102，CRKL200105）
中图分类号：TP391 文献标识码：A 文章编号：2096-4706（2022）20-0006-05
Multi-domain Transformation of EEG and Deep Learning Epilepsy Diagnosis
CHEN Haobin1, GE Wei 2, YANG Chao1, ZHENG Lin1
(1.Guangxi Key Laboratory of Wireless Wideband Communications and Signal Processing, Guilin University of Electronic Technology, Guilin 541004, China; 2.College of Humanities and Management, Guilin Medical University, Guilin 541004, China)
Abstract: Feature detection in a single dimension limits the accuracy of existing EEG-based epilepsy diagnosis. By converting EEG into Gram angle field map and wavelet time-frequency map, an integrated deep learning model consisting of two 2D CNNs and one DNN is constructed. Two 2D CNNs extract and fuse the features of Gram angle field map and wavelet time-frequency map respectively, and output the fused features to DNN for epileptic fusion recognition. The effectiveness of the integrated deep learning model is tested with the EEG dataset of the University of Bonn. The results show that the accuracy, specificity and sensitivity of the model for EEG recognition of epilepsy are 96.5%, 95.0% and 96.0%, respectively. The overall recognition performance is better than the traditional single neural network model, which can provide better auxiliary functions for the diagnosis of epilepsy and other diseases.
Keywords: deep learning; epilepsy; convolutional neural network; continuous wavelet transform; Gram angle field
参考文献：
[1] 秦兵，廖卫平，臼井直敬，等 . 癫痫：全球重点防治的神经精神疾病 [J]. 实用医学杂志，2012，28（13）：2109-2111.陈浩滨，等 : 脑电波信号多域变换与深度学习癫痫诊断 准确率第 20 期 现代信息科技10 2022.10
[2] EMAMI A，KUNII N，MATSUO T，et al. Seizure detection by convolutional neural network-based analysis of scalp electroencephalography plot images [J/OL].NeuroImage：Clinical， 2019，22：101684[2022-05-26].https://linkinghub.elsevier.com/ retrieve/pii/S2213158219300348.
[3] ACHARYA U R，OH S L，HAGIWARA Y，et al. Deep convolutional neural network for the automated detection and diagnosis of seizure using EEG signals [J].Computers in biology and medicine， 2018，100：270-278.
[4] SHANKAR A，KHAING H K，DANDAPAT S， et al. Epileptic seizure classification based on Gramian angular field transformation and deep learning [C]//2020 IEEE Applied Signal Processing Conference （ASPCON）.Kolkata：IEEE，2020：147-151.
[5] 郭怡湘 . 基于机器学习的癫痫脑电信号分类 [D]. 天津：天津大学，2017.
[6] QIAN S. Introduction to time-frequency and wavelet transforms [M].Prentice Hall，2001：304.
[7] 刘晓燕 . 临床脑电图学 [M]. 北京：人民卫生出版化，2006.
[8] MALLAT S. A wavelet tour of signal processing（third edition：the sparse way） [EB/OL].[2022-05-26]. https://www.di.ens. fr/~mallat/papiers/WaveletTourChap1-2-3.pdf.
[9] ZHANG T，CHEN W Z. LMD based features for the automatic seizure detection of EEG signals using SVM [J].IEEE Transactions on Neural Systems and Rehabilitation Engineering， 2017，25（8）：1100-1108.
[10] LIANG P F，DENG C，WU J，et al. Intelligent fault diagnosis of rotating machinery via wavelet transform， generative adversarial nets and convolutional neural network [J/OL].Measurement， 2020，159（C）：107768[2022-05-26].https://schlr.cnki.net/zn/Detail/ urlLink/GARJ2020/SJES74031E6C6A3AA8C740FC8F169D8B89BB?li nkUrl=http%3A%2F%2Fwww.sciencedirect.com%2Fscience%2Farticle %2Fpii%2FS0263224120303067&product=SJES&browseType=.
[11] WANG Z，OATES T. Encoding time series as images for visual inspection and classification using tiled convolutional neural networks [EB/OL].[2022-05-26]. ttps://medium.com/analytics-vidhya/ encoding-time-series-as-images-b043becbdbf3.
[12] ANDRZEJAK R G，LEHNERTZ K，MORMANN F，et al. Indications of nonlinear deterministic and finite-dimensional structures in time series of brain electrical activity： Dependence on recording region and brain state [J/OL].Physical Review E： Statistical， Nonlinear & Soft Matter Physics，2001，64（6）：061907[2022-05-26].https:// journals.aps.org/pre/abstract/10.1103/PhysRevE.64.061907.
[13] XIANG J，LI C G，LI H F，et al. The detection of epileptic seizure signals based on fuzzy entropy [J].Journal of neuroscience methods，2015，243：18-25.
作者简介：陈浩滨（1994—），男，汉族，广东普宁人，硕士研究生在读，主要研究方向：生物电信号处理；通讯作者：葛微（1979—），女，汉族，辽宁沈阳人，副教授，博士，主要研究方向：大数据并行计算；杨超（1988—），男，汉族，陕西西安人，讲师，博士，主要研究方向：超宽带通信、杂波抑制和雷达通信一体化；郑霖（1973—），男，汉族，安徽黄山人，教授，博士，主要研究方向：无线通信信号处理、MIMO、超宽带通信和雷达通信一体化。