一种用于心电图分类的改进神经网络算法  

An Improved Neural Network Algorithm for ECG Classification

作　　者：尘昌华 乔风娟 李彬[2] CHEN Chang-hua;QIAO Feng-juan;LI Bin(Shanghai Open University Fengxian Branch,Shanghai 201499,China;School of Mathematics and Statistics,Qilu University of Technology(Shandong Academy of Sciences),Jinan 250353,China)

机构地区：[1]上海开放大学奉贤分校,上海201499 [2]齐鲁工业大学(山东省科学院)数学与统计学院,山东济南250353

出　　处：《计算机技术与发展》2023年第1期178-186,共9页Computer Technology and Development

基　　金：山东省高等学校青创科技支持计划(2019KJN011);山东省自然科学基金项目(ZR2020MF097);济南市“新高校20条”项目(2021GXRC100)。

摘　　要：心血管疾病的死亡率在所有疾病中居于首位,心电图能够反映人体的电信号活动情况,它已成为医生用来诊断心血管疾病的重要依据。随着计算机辅助ECG诊断技术的快速发展,深度学习方法已能够实现ECG信号的特征提取和分类。为了较好地提高ECG信号的分类识别率和处理效率,该文提出了一种新的心电图分类方法。首先,对原始数据进行去噪,提出了基于经验小波变换(EWT)的提升小波阈值去噪方法。然后,重构经过提升小波阈值去噪技术处理过的模态分量。在训练过程中,设计了基于局部感受野的极限学习机(ELM-LRF)和双向长短时记忆网络(BLSTM)结合的神经网络算法,并利用注意力机制优化该算法,提出了LRF-BLSTM-Attention模型。最后,在CCDD和MIT-BIH数据集上对提出算法的性能进行验证,准确率分别达到86.12%和99.87%,证明了该算法在临床心血管疾病智能诊断中的实用性。与其他模型相比,该模型的收敛速度更快,收敛的损失值更小。The mortality of cardiovascular disease ranks first among all diseases. ECG can reflect the activity of human electrical signals, which has become an important basis for doctors to diagnose cardiovascular diseases. With the rapid development of computer-aided ECG diagnosis technology, deep learning method has been able to realize the feature extraction and classification of ECG signals. In order to better improve the classification and recognition rate and processing efficiency of ECG signals, a new ECG classification method is proposed. First of all, the original data is denoised and an improved wavelet threshold denoising method based on empirical wavelet transform(EWT) is proposed, and then the modal components processed by lifting wavelet threshold denoising technology are reconstructed.In the training process, a neural network algorithm combining local receptive fields based extreme learning machine(ELM-LRF) and bidirectional long short-term memory network(BLSTM) is designed, and the attention mechanism is introduced to optimize the model. Finally, the performance of the proposed algorithm is verified on CCDD and MIT-BIH Arrhythmia Datasets. The classification accuracy rates have reached 86.12% and 99.87%,respectively, proving the algorithm’s practicability and effectiveness in clinical intelligent diagnosis of cardiovascular diseases. Compared with other models, the convergence speed of such model is faster and the loss of convergence is smaller.

关 键 词：极限学习机 局部感受野 双向长短时记忆网络 注意力机制 心电图 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]