添加链接 注册    登录
link之家
链接快照平台
  • 输入网页链接，自动生成快照
  • 标签化管理网页链接
相关文章推荐
悲伤的汉堡包  ·  java word ...·  1 月前    · 
性感的大象  ·  MySql查询，聚合函数，分组，分页，排序等 ...·  3 月前    · 
焦虑的双杠  ·  excel数据处理一：巧妙使用openpyx ...·  8 月前    · 
开朗的路灯  ·  大模型遭泄两月后，Meta意外变赢家_手机新浪网·  1 年前    · 
骑白马的皮带  ·  怎么排查 Spring Boot ...·  1 年前    · 
link之家  ›  基于深度学习并行网络模型的心律失常分类方法 - PMC
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8526312/
帅气的荔枝
1 年前
The .gov means it’s official. Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site. The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely. As a library, NLM provides access to scientific literature. Inclusion in an NLM database does not imply endorsement of, or agreement with, the contents by NLM or the National Institutes of Health. Learn more about our disclaimer.
Nan Fang Yi Ke Da Xue Xue Bao. 2021 Aug 31; 41(9): 1296–1303.
Chinese. doi: 10.12122/j.issn.1673-4254.2021.09.02
PMCID: PMC8526312
PMID: 34658342

Language: Chinese | English

基于深度学习并行网络模型的心律失常分类方法

An arrhythmia classification method based on deep learning parallel network model

甘 屹 , 1, 2, * 施 俊丞 , 1 高 丽 , 3, 4 and 何 伟铭 1, 2

甘 屹

上海理工大学机械工程学院，上海 200093, School of Mechanical Engineering, University of Shanghai for Science and Technology, Shanghai 200093, China 日本中央大学理工学部精密工学 科，日本 东京 112-0003, Faculty of Science and Engineering, Chuo University, Tokyo 112-0003, Japan

Find articles by 甘 屹

施 俊丞

上海理工大学机械工程学院，上海 200093, School of Mechanical Engineering, University of Shanghai for Science and Technology, Shanghai 200093, China

Find articles by 施 俊丞

高 丽

上海理工大学图书馆，上海 200093, Library, University of Shanghai for Science and Technology, Shanghai 200093, China 上海理工大学光电与计算机学院，上海 200093, School of Optical-Electrical and Computer Engineering, University of Shanghai for Science and Technology, Shanghai 200093, China

Find articles by 高 丽

何 伟铭

上海理工大学机械工程学院，上海 200093, School of Mechanical Engineering, University of Shanghai for Science and Technology, Shanghai 200093, China 日本中央大学理工学部精密工学 科，日本 东京 112-0003, Faculty of Science and Engineering, Chuo University, Tokyo 112-0003, Japan 上海理工大学机械工程学院，上海 200093, School of Mechanical Engineering, University of Shanghai for Science and Technology, Shanghai 200093, China 日本中央大学理工学部精密工学 科，日本 东京 112-0003, Faculty of Science and Engineering, Chuo University, Tokyo 112-0003, Japan 上海理工大学图书馆，上海 200093, Library, University of Shanghai for Science and Technology, Shanghai 200093, China 上海理工大学光电与计算机学院，上海 200093, School of Optical-Electrical and Computer Engineering, University of Shanghai for Science and Technology, Shanghai 200093, China

其中， X 0 , X 1 , …, X n 分别为第0, 1, …, n 层的输出特征；[]代表级联运算； H n 代表第 n 层的非线性变换函数。

过渡层包含批量归一化、Relu函数、卷积层和池化层（Pooling）。由于密集连接使得网络参数增多，通常将池化层中的通道数减少一半来减少特征数量，提高模型运行速度。

1.2.2. 双向长短时记忆网络

长短时记忆网络（LSTM）是一种特殊的循环神经网络，主要解决序列数据的长期依赖问题。如（ 图 4 ）所示，LSTM中的输入门( i t )、输出门( O t )以及遗忘门( f t )可以大幅度提高模型处理序列数据的能力。LSTM的输入包括当前时刻的输入 x t ，上一时刻记忆单元的状态 c t -1 以及上一时刻的输出 h t -1 ，经过tanh、 σ （Sigmoid）等函数的运算可以得到当前时刻的输出 h t 以及当前时刻记忆单元的状态 c t 。

An external file that holds a picture, illustration, etc. Object name is nfykdxxb-41-9-1296-4.jpg

长短时记忆网络

Long short-term memory network.

LSTM能够提取时序信号的特征，但是无法同时捕捉前后时刻波形特征之间的联系，BiLSTM有效解决了这一问题。（ 图 5 ）为BiLSTM结构，图中前向LSTM和后向LSTM构成BiLSTM的前向和后向传输层，其中 equation M2 和 equation M3 分别为当前时刻前向和后向传输层的输出值。在BiLSTM中输入 x t ，经过传输层等一系列计算就能得到输出 y t 。

An external file that holds a picture, illustration, etc. Object name is nfykdxxb-41-9-1296-5.jpg

双向长短时记忆网络

Bidirectional long short-term memory network.

1.2.3. 高效通道注意力网络

ECANet（ 图 6 ）主要是给重要特征赋权重值，提升模型对心电信号重要特征（P波、QRS波群、T波的宽度和幅值以及RR间期的长度等特征）的提取能力。其工作过程：首先使用全局平均池化层（GAP）把输入特征( x )变为单个数值；然后在不降低维度的情况下使用卷积核尺寸为 k 的一维卷积来生成通道权重；最后通道权重通过乘法加权到输入特征上，得到输出特征( equation M4 )。

An external file that holds a picture, illustration, etc. Object name is nfykdxxb-41-9-1296-6.jpg

高效通道注意力网络

Efficient channel attention network.

1.2.4. 并行网络分类模型结构

首先基于深度学习网络模型，采用并行分支结构，分别输入小尺度心拍和大尺度心拍。然后每个分支中包含2个DenseNet和1个BiLSTM，并且在这两个网络后面加入1个ECANet。为了缓解过拟合，在每个分支的末尾分别引入全局平均池化层（GAP）和丢弃层（Dropout）。最后将两个分支提取的特征进行融合，使用Softmax函数实现心律失常的4分类。在训练模型的过程中，batch_size为128，epoch为50，使用Adam优化器，学习率为0.001，最大池化层（MaxPool1D）的步长和池化核尺寸都为2，使用交叉熵损失函数实现对模型的训练 [ 34 ] 。模型的其余超参数见（ 表 1 ），并行网络分类模型结构见（ 图 7 ）。

1

模型的超参数

Hyperparameters of the model

Type of network Number of Conv1D Number of Conv1D kernels *size Output size 1 Output size 2
Conv1D means 1-dimensional convolutional layer with step size 1; Dropout parameter is set to 0.5; Concatenate means stitching operation; FC means fully connected layer.
Inputs - - 300*1 550*1
- Conv1D*1 8*32 300*8 550*8
DenseNet1 Conv1D*3 16*32 300*16 550*16
Conv1D*1 28*32 300*28 550*28
ECANet1 Conv1D*1 1*3 150*28 275*28
DenseNet2 Conv1D*2 32*32 150*32 275*32
Conv1D*1 46*32 150*46 275*46
ECANet2 Conv1D*1 1*5 75*46 138*46
BiLSTM - - 75*256 138*256
ECANet3 Conv1D*1 1*5 75*256 138*256
GAP+Dropout - - 256 256
Concatenate+FC - - 128 -
Softmax - - 4 -
An external file that holds a picture, illustration, etc. Object name is nfykdxxb-41-9-1296-7.jpg

基于深度学习的心律失常并行网络分类模型

Parallel network classification model for arrhythmia based on deep learning.

2. 结果

2.1. 数据集

本文采用MIT-BIH心律失常数据库中MLII导联的数据评估心律失常的分类性能。MLII导联中包含48条患者记录，采样频率为360 Hz。由于4条患者记录质量较低，因此只对剩下44条患者记录采用1.1节中的方法进行预处理，一共得到100 634个样本，其中N、S、V、F这4类样本个数分别为90 045、2780、7007、802。本文采用留出法 [ 35 ] 验证模型性能，具体方法是将100 634个样本以3∶1∶1的比例划分为训练集、验证集和测试集 [ 36 ] 。为了缓解类别不平衡问题，采用SMOTE算法对训练集中的S类和F类进行数据增强，使其数量与V类数量一致，因此数据增强后，训练集、验证集和测试集数量分别为66 509、20 127和20 127。

2.2. 实验平台及评价指标

实验平台的CPU为i5-10500K，操作系统为Window10，显卡为RTX2080Ti，内存为11G，实验基于Python3.7和Tensorflow2.1实现。

通常使用混淆矩阵（CM）表示模型的分类结果 [ 37 , 38 ] ，混淆矩阵的每一行代表真实类别，每一列代表预测类别，表示如下：

其中， n ij 代表类别为 i 的样本的数量被归类为类别为 j 的数量； K 代表混淆矩阵的维度。

分类模型总体性能的评价指标，通常使用总体准确率（OverallAccuracy）、平均灵敏度（Average Sensitivity）和平均特异性（Average Specificity） [ 16 ] ，计算公式如下：

其中，总体准确率代表被正确分类的样本占所有样本的比例；平均灵敏度代表所有类别灵敏度的平均值，平均灵敏度越高说明总体漏诊率越低；平均特异性代表所有类别特异性的平均值，平均特异性越高说明总体误诊率越低 [ 32 ] 。

2.3. 实验1：多种并行网络模型分类性能对比

实验1的数据集采用测试集。在并行输入小尺度心拍和大尺度心拍的情况下，对DenseNet，BiLSTM，ECANet三种网络模型组合性能进行对比，其混淆矩阵见 表 2 ，其中，“1”代表DenseNet；“2”代表BiLSTM；“3”代表ECANet。

2

4种并行网络分类模型的混淆矩阵

Confusion matrix of four parallel network classification models

Model 2+3 1+3 1+2 1+2+3
N S V F N S V F N S V F N S V F
Predictive value N 17850 44 21 19 17 885 12 14 23 17 854 38 31 11 17 876 37 10 11
S 59 491 4 1 93 454 8 0 35 513 7 0 22 531 2 0
Actual value V 22 5 1426 16 21 2 1432 14 15 3 1434 17 12 4 1437 16
F 12 0 7 150 7 0 14 148 11 0 6 152 8 0 7 154

表 2 给出4种并行网络分类模型的混淆矩阵，结合公式（2）~（5）可知，BiLSTM+ECANet的总体准确率、平均灵敏度和平均特异性分别为98.96%、93.46%、98.84%。DenseNet+ECANet的总体准确率、平均灵敏度和平均特异性分别为98.97%、91.65%、98.51%。DenseNet + BiLSTM的总体准确率、平均灵敏度和平均特异性分别为99.14%、94.86%、99.16%。DenseNet+ECANet+ BiLSTM（本文所提模型）的总体准确率、平均灵敏度和平均特异性分别为99.36%、96.08%、99.41%。实验1还给出本文所提模型测试一次测试集所需的时间为17 s。

2.4. 实验2：不同心拍输入方式下，各分类模型性能对比

实验2的数据集采用测试集。实验2在实验1的基础上，将本文所提的并行网络分类模型（DenseNet+ BiLSTM+ECANet）分别与机器学习并行分类模型和采用单个心拍输入方式的深度学习分类模型进行性能对比（ 表 3 ）。其中，PCA指主成分分析，DTW指动态时间归整，ELM指极限学习机。

3

不同心拍输入方式下，各分类模型性能比较

Performance comparison of each classification model under different heartbeat input methods

Model structure Input Evaluation index
Overall accuracy (%) Average sensitivity (%) Average specificity (%)
PCA+DTW+SVM [ 12 ] Single heartbeat and RR interval 97.80 88.83 -
CNN+ELM [ 14 ] Single heartbeat 98.51 94.41 98.45
CNN [ 15 ] Single heartbeat 99.06 89.85 98.28
CNN+LSTM [ 16 ] Single heartbeat 98.93 94.43 99.33
DenseNet+ECANet+BiLSTM Small-scale heartbeat and large-scale heartbeat 99.36 96.08 99.41

表 3 中，文献[ 12 ]采用并行输入结合机器学习分类模型进行分类，总体准确率达到97.80%。文献[ 14 - 16 ]均采用单个心拍输入结合深度学习分类模型进行分类，总体准确率和平均特异性达到98%以上，平均灵敏度基本都超过90%。本文所提模型的总体准确率、平均灵敏度和平均特异性同实验1。

2.5. 实验3：并行网络分类模型收敛性能分析

实验3的数据集采用训练集和验证集，以验证实验1中提出的4种并行网络分类模型的收敛性能。为基于4种并行网络分类模型的准确率与损失率的收敛图（ 图 8 ）。

An external file that holds a picture, illustration, etc. Object name is nfykdxxb-41-9-1296-8.jpg

准确率和损失率的收敛图

Convergence graph of accuracy and loss.

显示在开始训练模型时，4种并行网络分类模型基于训练集和验证集的准确率和损失率曲线变化很快（ 图 8 ）。DenseNet+BiLSTM+ECANet（本文所提模型）基于训练集和验证集的准确率曲线上升速度最快，当epoch大于40后，两条曲线趋于平稳，基本实现收敛，并且没有出现严重的过拟合 图 8 准确率曲线。其余3种模型基于训练集和验证集的准确率曲线上升速度较慢，并且都存在一定程度的波动。本文所提模型基于训练集和验证集的损失率曲线下降最快，并且在epoch大于40后，也基本实现收敛，没有出现严重的过拟合 图 8 损失率曲线。除此之外，实验3还给出本文所提模型训练一次训练集所需的时间为41 s。

3. 讨论

在并行网络分类模型中，BiLSTM + ECANet和DenseNet+ECANet的总体准确率和平均特异性都没有超过99%，相比之下，DenseNet+BiLSTM的在总体准确率和平均特异性得到提高，但是平均灵敏度不高，这会增加疾病漏诊的风险。相比以上3种模型，DenseNet+ BiLSTM+ECANet（本文所提模型）的总体准确率、平均灵敏度和平均特异性最高，分别达到99.36%、96.08%、99.41%，表明DenseNet通过对心电信号波形特征的重利用，有助于进一步挖掘隐藏特征，提高模型的总体准确率；ECANet通过增加权重的方式，增强心电信号中P波、QRS波群、T波的幅值和宽度等重要特征，从而提高模型的平均灵敏度；BiLSTM的使用，解决了DenseNet受卷积核尺寸的限制使其不能充分捕捉心电信号波形特征前后联系的问题，提升了模型的平均特异性，实验证明DenseNet，BiLSTM，ECANet的组合改善了模型的分类性能。

虽然采用需要人工特征提取的机器学习分类模型，但是通过使用并行输入的方式使其总体准确率达到97.80% [ 12 ] 。本文所提模型都采用深度学习网络模型进行分类，总体准确率都超过98%，与前人报道一致 [ 14 - 16 ] 。但本文所提模型的平均灵敏度和平均特异性更高。主要原因本文采用并行输入，并行输入不仅可以提取单个心拍的波形特征，还能提取RR间期的波形特征，从而丰富模型提取的特征数量，进一步证明本文所提模型能够实现心律失常的有效分类。

本文所提模型基于训练集和验证集的准确率与损失率曲线都实现收敛，随着epoch的增加，准确率最早达到99%，损失率最早达到0.3，表明收敛速度较快。同时，本文所提模型每次训练时间为41s，有些模型每次训练时间分别为118 s和150 s [ 17 , 36 ] ，三者对比表明，本文所提模型可快速准确识别不同种类心律失常的特征。

综上所述，本文提出基于深度学习的心律失常并行网络分类模型实现对N、S、V、F这4种心律失常的分类。该模型并行输入小尺度心拍和大尺度心拍以提取单个心拍和RR间期的波形特征。运用DenseNet充分提取心电信号波形特征，较好地捕捉到心电信号波形中的隐藏特征。使用BiLSTM获得心电信号波形的前后依赖关系。ECANet的引入，增强对重要波形特征的识别能力，减少模型参数数量，提高了模型运行速度。基于MIT-BIH心律失常数据库的对比实验表明，模型在总体准确率、平均灵敏度，平均特异性这3个指标上取得了较为满意的结果，并且训练时间较短，因此适合用于可穿戴式心电设备等领域。后续研究可尝试将基于深度学习的心律失常并行网络分类模型移植到便携式心电监护设备上，实现心律失常便捷快速临床诊断。

Funding Statement

国家自然科学基金（51375314）

Supported by National Natural Science Foundation of China (51375314)

References

1. Shi HT, Wang HR, Zhang F, et al. Inter-patient heartbeat classi- fication based on region feature extraction and ensemble classifier. Biomed Signal Process Control. 2019; 51 :97–105. doi: 10.1016/j.bspc.2019.02.012.
[Shi HT, Wang HR, Zhang F, et al. Inter-patient heartbeat classi- fication based on region feature extraction and ensemble classifier [J]. Biomed Signal Process Control, 2019, 51: 97-105.] [ CrossRef ] [ Google Scholar ]
2. Wang HR, Shi HT, Lin K, et al. A high-precision arrhythmia classification method based on dual fully connected neural network. Biomed Signal Process Control. 2020; 58 :101874. doi: 10.1016/j.bspc.2020.101874.
[Wang HR, Shi HT, Lin K, et al. A high-precision arrhythmia classification method based on dual fully connected neural network [J]. Biomed Signal Process Control, 2020, 58: 101874.] [ CrossRef ] [ Google Scholar ]
3. Chang KC, Hsieh PH, Wu MY, et al. Usefulness of machine learningbased detection and classification of cardiac arrhythmias with 12- lead electrocardiograms. Can J Cardiol. 2021; 37 (1):94–104. doi: 10.1016/j.cjca.2020.02.096.
[Chang KC, Hsieh PH, Wu MY, et al. Usefulness of machine learningbased detection and classification of cardiac arrhythmias with 12- lead electrocardiograms[J]. Can J Cardiol, 2021, 37(1): 94-104.] [ PubMed ] [ CrossRef ] [ Google Scholar ]
4. Hao HQ, Liu M, Xiong P, et al. Multi-lead model-based ECG signal denoising by guided filter. http://www.onacademic.com/detail/journal_1000041585561699_0919.html . EngApplArtif Intell. 2019; 79 :34–44.
[Hao HQ, Liu M, Xiong P, et al. Multi-lead model-based ECG signal denoising by guided filter[J]. EngApplArtif Intell, 2019, 79: 34-44.] [ Google Scholar ]
5. Marinho LB, Nascimento NDMM, Souza JWM, et al. A novel electrocardiogram feature extraction approach for cardiac arrhythmia classification. Future Gener Comput Syst. 2019; 97 :564–77. doi: 10.1016/j.future.2019.03.025.
[Marinho LB, Nascimento NDMM, Souza JWM, et al. A novel electrocardiogram feature extraction approach for cardiac arrhythmia classification[J]. Future Gener Comput Syst, 2019, 97: 564-77.] [ CrossRef ] [ Google Scholar ]
6. Liu H, Zhao Z, Chen X, et al. Using the VQ-VAE to improve the recognition of abnormalities in short-duration 12-lead electrocardiogram records. Comput Methods Programs Biomed. 2020; 196 :105639. doi: 10.1016/j.cmpb.2020.105639.
[Liu H, Zhao Z, Chen X, et al. Using the VQ-VAE to improve the recognition of abnormalities in short-duration 12-lead electrocardiogram records[J]. Comput Methods Programs Biomed, 2020, 196: 105639.] [ PubMed ] [ CrossRef ] [ Google Scholar ]
7. Martis RJ, Acharya UR, Min LC, et al. Application of higher order cumulant features for cardiac health diagnosisusing ECG signals. http://www.researchgate.net/profile/U_Rajendra_Acharya/publication/237094219_Application_of_higher_order_cumulant_features_for_cardiac_health_diagnosis_using_ECG_signals/links/0deec531c073d24c34000000/Application-of-higher-order-cumulant-features-for-cardiac-health-diagnosis-using-ECG-signals.pdf . Int J Neural Syst. 2013; 23 (4):806–39.
[Martis RJ, Acharya UR, Min LC, et al. Application of higher order cumulant features for cardiac health diagnosisusing ECG signals[J]. Int J Neural Syst, 2013, 23(4): 806-39] [ PubMed ] [ Google Scholar ]
8. Mondéjar-Guerra V, Novo J, Rouco J, et al. Heartbeat classification fusing temporal and morphological information of ECGs via ensemble of classifiers. Biomed Signal Process Control. 2019; 47 :41–8. doi: 10.1016/j.bspc.2018.08.007.
[Mondéjar-Guerra V, Novo J, Rouco J, et al. Heartbeat classification fusing temporal and morphological information of ECGs via ensemble of classifiers[J]. Biomed Signal Process Control, 2019, 47: 41-8.] [ CrossRef ] [ Google Scholar ]
9. Arvanaghi R, Daneshvar S, Seyedarabi H, et al. Fusion of ECG and ABP signals based on wavelet transform for cardiac arrhythmias classification. Comput Methods Programs Biomed. 2017; 151 :71–8. doi: 10.1016/j.cmpb.2017.08.013.
[Arvanaghi R, Daneshvar S, Seyedarabi H, et al. Fusion of ECG and ABP signals based on wavelet transform for cardiac arrhythmias classification[J]. Comput Methods Programs Biomed, 2017, 151: 71-8.] [ PubMed ] [ CrossRef ] [ Google Scholar ]
10. Dias FM, Monteiro HLM, Cabral TW, et al. Arrhythmia classification from single- lead ECG signals using the inter-patient paradigm. Comput Methods Programs Biomed. 2021; 202 :105948. doi: 10.1016/j.cmpb.2021.105948.
[Dias FM, Monteiro HLM, Cabral TW, et al. Arrhythmia classification from single- lead ECG signals using the inter-patient paradigm[J]. Comput Methods Programs Biomed, 2021, 202: 105948.] [ PubMed ] [ CrossRef ] [ Google Scholar ]
11. Raj S, Ray KC. Sparse representation of ECG signals for automated recognition of cardiac arrhythmias. Expert Syst Appl. 2018; 105 :49–64. doi: 10.1016/j.eswa.2018.03.038.
[Raj S, Ray KC. Sparse representation of ECG signals for automated recognition of cardiac arrhythmias[J]. Expert Syst Appl, 2018, 105: 49-64.] [ CrossRef ] [ Google Scholar ]
12. Zhu WL, Chen XH, Wang Y, et al. Arrhythmia recognition and classification using ECG morphology and segment feature analysis. IEEE/ACM Trans Comput Biol Bioinform. 2019; 16 (1):131–8. doi: 10.1109/TCBB.2018.2846611.
[Zhu WL, Chen XH, Wang Y, et al. Arrhythmia recognition and classification using ECG morphology and segment feature analysis [J]. IEEE/ACM Trans Comput Biol Bioinform, 2019, 16(1): 131-8.] [ PubMed ] [ CrossRef ] [ Google Scholar ]
13. Kalidas V, Tamil LS. Detection of atrial fibrillation using discretestate Markov models and Random Forests. Comput Biol Med. 2019; 113 :103386. doi: 10.1016/j.compbiomed.2019.103386.
[Kalidas V, Tamil LS. Detection of atrial fibrillation using discretestate Markov models and Random Forests[J]. Comput Biol Med, 2019, 113: 103386.] [ PubMed ] [ CrossRef ] [ Google Scholar ]
14. Zhou SR, Tan B. Electrocardiogram soft computing using hybrid deep learning CNN-ELM. Appl Soft Comput. 2020; 86 :105778. doi: 10.1016/j.asoc.2019.105778.
[Zhou SR, Tan B. Electrocardiogram soft computing using hybrid deep learning CNN-ELM[J]. Appl Soft Comput, 2020, 86: 105778.] [ CrossRef ] [ Google Scholar ]
15. Wang HR, Shi HT, Chen XJ, et al. An improved convolutional neural network based approach for automated heartbeat classification. J Med Syst. 2019; 44 (2):35. doi: 10.1007/s10916-019-1511-2?utm_source=xmol.
[Wang HR, Shi HT, Chen XJ, et al. An improved convolutional neural network based approach for automated heartbeat classification[J]. J Med Syst, 2019, 44(2): 35.] [ PubMed ] [ CrossRef ] [ Google Scholar ]
16. 张 异凡, 黄 亦翔, 汪 开正, et al. 用于心律失常识别的LSTM和CNN并行组合模型 https://www.cnki.com.cn/Article/CJFDTOTAL-HEBX201910011.htm . 哈尔滨工业大学学报 2019; 51 (10):76–82.
[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.] [ Google Scholar ]
17. Chen AY, Wang F, Liu WH, et al. Multi-information fusion neural networks for arrhythmia automatic detection. Comput Methods Programs Biomed. 2020; 193 :105479. doi: 10.1016/j.cmpb.2020.105479.
[Chen AY, Wang F, Liu WH, et al. Multi-information fusion neural networks for arrhythmia automatic detection[J]. Comput Methods Programs Biomed, 2020, 193: 105479.] [ PubMed ] [ CrossRef ] [ Google Scholar ]
18. Chen C, Hua ZC, Zhang RQ, et al. Automated arrhythmia classification based on a combination network of CNN and LSTM. Biomed Signal Process Control. 2020; 57 :101819. doi: 10.1016/j.bspc.2019.101819.
[Chen C, Hua ZC, Zhang RQ, et al. Automated arrhythmia classification based on a combination network of CNN and LSTM[J]. Biomed Signal Process Control, 2020, 57: 101819.] [ CrossRef ] [ Google Scholar ]
19. Sellami A, Hwang H. A robust deep convolutional neural network with batch-weighted loss for heartbeat classification. Expert Syst Appl. 2019; 122 :75–84. doi: 10.1016/j.eswa.2018.12.037.
[Sellami A, Hwang H. A robust deep convolutional neural network with batch-weighted loss for heartbeat classification[J]. Expert Syst Appl, 2019, 122: 75-84.] [ CrossRef ] [ Google Scholar ]
20. Golrizkhatami Z, Acan A. ECG classification using three-level fusion of different feature descriptors. Expert Syst Appl. 2018; 114 :54–64. doi: 10.1016/j.eswa.2018.07.030.
[Golrizkhatami Z, Acan A. ECG classification using three-level fusion of different feature descriptors[J]. Expert Syst Appl, 2018, 114: 54-64.] [ CrossRef ] [ Google Scholar ]
21. Sharma A, Garg N, Patidar S, et al. Automated pre-screening of arrhythmia using hybrid combination of Fourier-Bessel expansion and LSTM. Comput Biol Med. 2020; 120 :103753. doi: 10.1016/j.compbiomed.2020.103753.
[Sharma A, Garg N, Patidar S, et al. Automated pre-screening of arrhythmia using hybrid combination of Fourier-Bessel expansion and LSTM[J]. Comput Biol Med, 2020, 120: 103753.] [ PubMed ] [ CrossRef ] [ Google Scholar ]
22. 赖 杰伟, 陈 韵岱, 韩 宝石, et al. 基于DenseNet的心电数据自动诊断算法 http://www.j-smu.com/CN/10.12122/j.issn.1673-4254.2019.01.11 . 南方医科大学学报 2019; 39 (1):69–75.
[赖杰伟, 陈韵岱, 韩宝石, 等. 基于DenseNet的心电数据自动诊断算法[J]. 南方医科大学学报, 2019, 39(1): 69-75.] [ PMC free article ] [ PubMed ] [ Google Scholar ]
23. Yildirim O, Baloglu UB, Tan RS, et al. A new approach for arrhythmia classification using deep coded features and LSTM networks. Comput Methods Programs Biomed. 2019; 176 :121–33. doi: 10.1016/j.cmpb.2019.05.004.
[Yildirim O, Baloglu UB, Tan RS, et al. A new approach for arrhythmia classification using deep coded features and LSTM networks[J]. Comput Methods Programs Biomed, 2019, 176: 121- 33.] [ PubMed ] [ CrossRef ] [ Google Scholar ]
24. Yıldırım Ö, Pławiak P, Tan RS, et al. Arrhythmia detection using deep convolutional neural network with long duration ECG signals. Comput Biol Med. 2018; 102 :411–20. doi: 10.1016/j.compbiomed.2018.09.009.
[Yıldırım Ö, Pławiak P, Tan RS, et al. Arrhythmia detection using deep convolutional neural network with long duration ECG signals [J]. Comput Biol Med, 2018, 102: 411-20.] [ PubMed ] [ CrossRef ] [ Google Scholar ]
25. Yan ZL, Zhou J, Wong WF. Energy efficient ECG classification with spiking neural network. Biomed Signal Process Control. 2021; 63 :102170. doi: 10.1016/j.bspc.2020.102170.
[Yan ZL, Zhou J, Wong WF. Energy efficient ECG classification with spiking neural network[J]. Biomed Signal Process Control, 2021, 63: 102170.] [ CrossRef ] [ Google Scholar ]
26. Peimankar A, Puthusserypady S. DENS-ECG: a deep learning approach for ECG signal delineation. Expert Syst Appl. 2021; 165 :113911. doi: 10.1016/j.eswa.2020.113911.
[Peimankar A, Puthusserypady S. DENS-ECG: a deep learning approach for ECG signal delineation[J]. Expert Syst Appl, 2021, 165: 113911.] [ CrossRef ] [ Google Scholar ]
27. Guo LB, Lei BY, Chen WL, et al. Dual attention enhancement feature fusion network for segmentation and quantitative analysis of paediatric echocardiography. Med Image Anal. 2021; 71 :102042. doi: 10.1016/j.media.2021.102042.
[Guo LB, Lei BY, Chen WL, et al. Dual attention enhancement feature fusion network for segmentation and quantitative analysis of paediatric echocardiography[J]. Med Image Anal, 2021, 71: 102042.] [ PubMed ] [ CrossRef ] [ Google Scholar ]
28. Guo MF, Yang NC, You LX. Wavelet-transform based early detection method for short-circuit faults in power distribution networks. Int J Electr Power Energy Syst. 2018; 99 :706–21. doi: 10.1016/j.ijepes.2018.01.013.
[Guo MF, Yang NC, You LX. Wavelet-transform based early detection method for short-circuit faults in power distribution networks[J]. Int J Electr Power Energy Syst, 2018, 99: 706-21.] [ CrossRef ] [ Google Scholar ]
29. 李 彦兵, 杜 兰, 刘 宏伟, et al. 基于微多普勒效应和多级小波分解的轮式履带式车辆分类研究 https://www.cnki.com.cn/Article/CJFDTOTAL-DZYX201304022.htm . 电子与信息学报 2013; 35 (4):894–900.
[李彦兵, 杜兰, 刘宏伟, 等. 基于微多普勒效应和多级小波分解的轮式履带式车辆分类研究[J]. 电子与信息学报, 2013, 35(4): 894-900.] [ Google Scholar ]
30. 尹 丽, 陈 富民, 张 琦, et al. 采用集合经验模态分解和改进阈值函数的心电自适应去噪方法 https://www.cnki.com.cn/Article/CJFDTOTAL-XAJT202001014.htm . 西安交通大学学报 2020; 54 (1):101–7.
[尹丽, 陈富民, 张琦, 等. 采用集合经验模态分解和改进阈值函数的心电自适应去噪方法[J]. 西安交通大学学报, 2020, 54(1): 101-7.] [ Google Scholar ]
31. 李 国权, 李 必禄, 林 金朝, et al. 基于改进EWT的心电信号基线矫正算法 https://www.cnki.com.cn/Article/CJFDTOTAL-YQXB202004018.htm . 仪器仪表学报 2020; 41 (4):156–66.
[李国权, 李必禄, 林金朝, 等. 基于改进EWT的心电信号基线矫正算法[J]. 仪器仪表学报, 2020, 41(4): 156-66.] [ Google Scholar ]
32. 邓 力, 傅 蓉. 基于心拍的端到端心律失常分类 http://www.j-smu.com/CN/10.12122/j.issn.1673-4254.2019.09.11 . 南方医科大学学报 2019; 39 (9):1071–7.
[邓力, 傅蓉. 基于心拍的端到端心律失常分类[J]. 南方医科大学学报, 2019, 39(9): 1071-7.] [ Google Scholar ]
33. Feng S, Keung J, Yu X, et al. Investigation on the stability of SMOTE-based oversampling techniques in software defect prediction. http://www.sciencedirect.com/science/article/pii/S0950584921001257 . Inf Softw Technol. 2021; 139 :106662.
[Feng S, Keung J, Yu X, et al. Investigation on the stability of SMOTE-based oversampling techniques in software defect prediction[J]. Inf Softw Technol, 2021, 139: 106662.] [ Google Scholar ]
34. Rahim T, Hassan SA, Shin SY. A deep convolutional neural network for the detection of polyps in colonoscopy images. http://www.sciencedirect.com/science/article/pii/S1746809421002512 . Biomed Signal Process Control. 2021; 68 :102654.
[Rahim T, Hassan SA, Shin SY. A deep convolutional neural network for the detection of polyps in colonoscopy images[J]. Biomed Signal Process Control, 2021, 68: 102654.] [ Google Scholar ]
35. 柯 丽, 王 丹妮, 杜 强, et al. 基于卷积长短时记忆网络的心律失常分类方法 https://www.cnki.com.cn/Article/CJFDTOTAL-DZYX202008023.htm . 电子与信息学报 2020; 42 (8):1990–8.
[柯丽, 王丹妮, 杜强, 等. 基于卷积长短时记忆网络的心律失常分类方法[J]. 电子与信息学报, 2020, 42(8): 1990-8.] [ Google Scholar ]
36. 熊 慧, 梁 美玲, 刘 近贞. 卷积神经网络混合模型的心律失常分类算法 https://www.cnki.com.cn/Article/CJFDTOTAL-HEBX202102004.htm . 哈尔滨工业大学学报 2021; 53 (2):33–9.
[熊慧, 梁美玲, 刘近贞. 卷积神经网络混合模型的心律失常分类算法[J]. 哈尔滨工业大学学报, 2021, 53(2): 33-9.] [ Google Scholar ]
37. Luque A, Carrasco A, Martín A, et al. The impact of class imbalance in classification performance metrics based on the binary confusion matrix. http://www.onacademic.com/detail/journal_1000042280352099_1fc7.html . Pattern Recognit. 2019; 91 :216–31.
[Luque A, Carrasco A, Martín A, et al. The impact of class imbalance in classification performance metrics based on the binary confusion matrix[J]. Pattern Recognit, 2019, 91: 216-31.] [ Google Scholar ]
38. Ruuska S, Hämäläinen W, Kajava S, et al. Evaluation of the confusion matrix method in the validation of an automated system for measuring feeding behaviour of cattle. http://smartsearch.nstl.gov.cn/paper_detail.html?id=57a565b551783f63ceb9127ab5580c9b . Behav Processes. 2018; 148 :56–62.
[Ruuska S, Hämäläinen W, Kajava S, et al. Evaluation of the confusion matrix method in the validation of an automated system for measuring feeding behaviour of cattle[J]. Behav Processes, 2018, 148: 56-62.] [ PubMed ] [ Google Scholar ]
Articles from Journal of Southern Medical University are provided here courtesy of Editorial Department of Journal of Southern Medical University
 
推荐文章
悲伤的汉堡包  ·  java word XWPFDocument - CSDN文库
1 月前
性感的大象  ·  MySql查询，聚合函数，分组，分页，排序等复杂查询 - 一半人生 - 博客园
3 月前
焦虑的双杠  ·  excel数据处理一：巧妙使用openpyxl提取、筛选数据 - 知乎
8 月前
开朗的路灯  ·  大模型遭泄两月后，Meta意外变赢家_手机新浪网
1 年前
骑白马的皮带  ·  怎么排查 Spring Boot 内存泄露？ - 知乎
1 年前
今天看啥   ·   Py中国   ·   codingpro   ·   藏经阁   ·   小百科   ·   link之家   ·   卧龙AI搜索
删除内容请联系邮箱 2879853325@qq.com
link之家 - 链接快照平台
© 2024 ~ 沪ICP备11025650号