DB-SMOTE及多层堆叠用于心律失常识别

doi:10.19665/j.issn1001-2400.2021.04.018

摘要/Abstract

摘要：

为提高心电图的识别效果,尤其是少数类疾病的识别准确率,设计了基于聚类插值过采样算法和多层堆叠模型的心律失常识别方法。由于合成少数类过采样算法忽略了少数类数据的类内不平衡性,提出了聚类插值过采样算法生成少数类边界样本。该算法利用有噪空间密度聚类算法把少数类数据分为多个簇并滤去噪声样本,以各簇的边界数据为主体生成新样本,并利用t分布随机邻近嵌入算法可视化分析样本生成情况。因单一的分类器性能不能满足需求,故使用多层堆叠模型组合多个不同分类器来进行识别。多层堆叠模型分为两层:第一层基模型K近邻、极端梯度提升树和梯度提升树将特征F映射为F',第二层逻辑回归模型识别特征F'。用MITBIH数据集进行检测,上述识别方法的分类准确率达99.66%,能较大地提升少数类样本的识别效果,因此该方法能够有效地用于心律失常识别。

关键词: 心电图, 聚类插值过采样算法, t分布随机近邻嵌入, 堆叠模型

Abstract:

In order to improve the recognition performance of the electrocardiogram,especially the recognition accuracy of minor diseases,this paper proposes the electrocardiogram recognition architecture based on the DB-SMOTE algorithm and multi-layered stacking model.The DB-SMOTE algorithm is proposed to solve the problem because the classical oversampling SMOTE algorithm ignores the intra-class unbalance of minority-class data.The new algorithm utilizes DBSCAN clustering to divide the data of minority classes into multiple clusters and filter out the noise samples,takes the boundary data of each cluster as the main body to generate new samples,and analyzes visually by tSNE.The performance of a single classifier cannot meet the requirements,so a multi-layered stack classification is used for identification,which is divided into two parts:the first is based on KNN,Xgboost and GBDT,and the feature F is mapped to F';the second part of the model is to identify the feature F'.This architecture has a 99.66% accuracy rate in identifying the electrocardiogram and can improve the recognition accuracy of minor diseases well,so it can be used to identify arrhythmias effectively.

Key words: electrocardiogram, DB-SMOTE, tSNE, stacking

中图分类号:

TP181

王波,邓科. DB-SMOTE及多层堆叠用于心律失常识别[J]. 西安电子科技大学学报, 2021, 48(4): 136-143.

WANG Bo,DENG Ke. DB-SMOTE and multi-layer stacking used for Arrhythmia recognition[J]. Journal of Xidian University, 2021, 48(4): 136-143.

图/表 12

图1

图2

图3

图4

表1

图5

图6

图7

图8

表2

表3

表4

参考文献 17

[1]	YANG H, WEI Z. Arrhythmia Recognition and Classification Using Combined Parametric and Visual Pattern Features of ECG Morphology[J]. IEEE Access, 2020, 8(99):47103-47117. doi: 10.1109/Access.6287639
[2]	张金玲, 崔彤, 刘为斌. 基于自适应时间窗函数优化的心电特征波形识别[J]. 北京邮电大学学报, 2019, 42(04):82-88.
	ZHANG Jinling, CUI Tong, LIU Weibin. ECG Characteristic Waveform Recognition Based on Adaptive Time Window Function Optimization[J]. Journal of Beijing University of Posts and Telecommunications, 2019, 42(04):82-88.
[3]	KIRANYAZ S, INCE T, GABBOUJ M. Real-Time Patient-Specific ECG Classification by 1-D Convolutional Neural Networks[J]. IEEE Transactions on Biomedical Engineering, 2016, 63(3):664-675. doi: 10.1109/TBME.2015.2468589
[4]	ZHU W L, CHEN X H, WANG Y, et al. Arrhythmia Recognition and Classification Using ECG Morphology and Segment Feature Analysis[J]. IEEE/ACM Transactions on Computational Biology & Bioinformatics, 2019, 16(1):131-138.
[5]	吕卫, 邓为贤, 褚晶辉, 等. 采用S变换特征选择方法的心律失常分类[J]. 数据采集与处理, 2018, 33(02):306-316.
	LÜ Wei, DENG Weixian, CHU Jinghui, et al. Classification of Arrhythmia Using S Transform Feature Selection Method[J]. Data Collection and Processing, 2018, 33(02):306-316.
[6]	庞彦伟, 李潇, 梁金升, 等. 基于卷积神经网络的第一导联心电图心拍分类[J]. 天津大学学报:自然科学与工程技术版, 2018, 51(10):997-1004.
	PANG Yanwei, LI Xiao, LIANG Jinsheng, et al. Classification of The First Lead Electrocardiogram Based on Convolutional Neural Network[J]. Journal of Tianjin University:Natural Science and Engineering Technology Edition, 2018, 51(10):997-1004.
[7]	PAN J, TOMOKINS W J. A Real-Time QRS Detection Algorithm[J]. IEEE Transactions on Biomedical Engineering, 1985, 32(3):230-236.
[8]	DOMAZET E, GUSEV M. Improving the QRS Detection for One-channel ECG Sensor[J]. Technology&Health Care:Official Journal of the European Society for Engineering and Medicine, 2019, 27(6):623-642.
[9]	CHAWLA N V, BOWYER K W, HALL L O, et al. SMOTE:Synthetic Minority Over-Sampling Technique[J]. Journal of Artificial Intelligence Research, 2002, 16(1):321-357. doi: 10.1613/jair.953
[10]	ZHANG A, GAO F, YANG M, et al. A New Rule Reduction and Training Method for Extended Belief Rule Base Based on DBSCAN Algorithm[J]. International Journal of Approximate Reasoning, 2020, 119:20-39. doi: 10.1016/j.ijar.2019.12.016
[11]	GISBRECHT A, SCHULZ A, HAMMER B. Parametric Nonlinear Dimensionality Reduction Using Kernel t-SNE[J]. Neurocomputing, 2015, 147(5) :71-82. doi: 10.1016/j.neucom.2013.11.045
[12]	左磊, 产秀秀, 禄晓飞, 等. 时频域分解海面回波及慢速微弱目标检测[J]. 西安电子科技大学学报, 2019, 46(5):84-90.
	ZUO Lei, CHAN Xiuxiu, LU Xiaofei, et al. Decomposing Sea Echoes in The Time-Frequency Domain and Detecting A Slow-Moving Weak Target in The Sea Clutter[J]. Journal of Xidian University, 2019, 46(5):84-90.
[13]	欧阳宁, 王先傲, 蔡晓东, 等. 结合小波与递归神经网络的低分辨率人脸识别[J]. 西安电子科技大学学报, 2019, 46(6):95-101.
	OUYANG Ning, WANG Xian’ao, CAI Xiaodong, et al. Low Resolution Face Recognition Method Based on Wavelet and Recursive Neural Networks[J]. Journal of Xidian University, 2019, 46(6):95-101.
[14]	BOUAZIZ F, BOUTANA D, OULHADJ H. Diagnostic of ECG Arrhythmia Using Wavelet Analysis and K-Nearest Neighbor Algorithm[C]//Proceedings of The 2018 International Conference on Applied Smart Systems (ICASS).Medea:Algeria, 2018:1-6.
[15]	张异凡, 黄亦翔, 汪开正, 等. 用于心律失常识别的LSTM和CNN并行组合模型[J]. 哈尔滨工业大学学报, 2019, 51(10):76-82.
	ZHANG Yifan, HUANG Yixiang, WANG Kaizheng, et al. LSTM and CNN Combined Model for Arrhythmia Recognition[J]. Journal of Harbin Institute of Technology, 2019, 51(10):76-82.
[16]	SHARMA M, SINGH S, KUMAR A, et al. Automated Detection of Shockable and Non-Shockable Arrhythmia Using Novel Wavelet-based ECG Features[J]. Computers in Biology and Medicine, 2019, 115(8):81-90.
[17]	查雪帆, 杨丰, 吴俣南, 等. 结合迁移学习与深度卷积网络的心电分类研究[J]. 中国医学物理学杂志, 2018, 35(11):1307-1312.
	ZHA Xuefan, YANG Feng, WU Yunan, et al. A Study of ECG Classification Based on Transfer Learning and Deep Convolutional Networks[J]Chinese Journal of Medical Physics, 2018, 35(11):1307-1312.

类别	N	LBBB	RBBB	PVC	APB
训练集	5 000	4 000	3 600	3 500	1 500+1 101
测试集	5 000	4 068	3 654	3 534	1 045

分类	正类	负类
正类	TP	FP
负类	FN	TN

参数	KNN	GBDT	Xgboost	KNN+DB-SMOTE	GBDT+DB-SMOTE	Xgboost+DB-SMOTE	多层堆叠
F-value	0.988 9	0.989 4	0.989 7	0.990 3	0.991 5	0.991 3	0.994 3
G-mean	0.988 7	0.989 7	0.988 6	0.990 4	0.991 6	0.991 6	0.994 4

文献	分类器方法	准确率/%
文中方法	多层分类器堆叠	99.66
文中深度学习方法1	VGG13网络	98.70
文中深度学习方法2	LSTM网络	99.08
FATIHA B^[14]方法	KNN	98.70
张异凡^[15]方法	LSTM;CNN	99.11
SHARMA M^[16]方法	小波分解;支持向量机	98.90
查雪帆^[17]方法	Alexnet	98.00