计算机工程与应用 ›› 2021, Vol. 57 ›› Issue (10): 125-132.DOI: 10.3778/j.issn.1002-8331.2002-0390

• 模式识别与人工智能 • 上一篇    下一篇

基于功率谱密度与卷积神经网络的心音分类

许春冬,辛鹏丽,周静,应冬文   

  1. 1.江西理工大学 信息工程学院,江西 赣州 341000
    2.中国科学院大学 电子电气与通信工程学院,北京 100049
  • 出版日期:2021-05-15 发布日期:2021-05-10

Classification of Heart Sounds Using Power Spectral Density and Convolutional Neural Networks

XU Chundong, XIN Pengli, ZHOU Jing, YING Dongwen   

  1. 1.School of Information Engineering, Jiangxi University of Science and Technology, Ganzhou, Jiangxi 341000, China
    2.School of Electronic, Electronical, and Communication Engineering, University of Chinese Academy of Sciences, Beijing 100049, China
  • Online:2021-05-15 Published:2021-05-10

摘要:

正常与异常心音分类在心血管疾病的筛查中有着重要的作用。建立在无心音分割的基础上,提出了一种基于功率谱密度时频分布特征与卷积神经网络的心音分类方法。该方法采用小波降噪做预处理,通过循环自相关获取心动周期,采用双线性插值法提取维度一致的心动周期功率谱密度时频特征,并送入卷积神经网络进行训练与测试。实验采用Challenge 2016数据集进行训练与测试,测试集的分类精度达到0.847 2,灵敏度和特异性评分达到0.776 3和0.946 3,整体性能良好。与其他算法的对比结果显示,该算法获得了更高的总体评分。

关键词: 心音分类, 功率谱密度, 卷积神经网络, 双线性插值法, 心动周期检测

Abstract:

The classification of normal and abnormal heart sounds plays an important role in the screening of cardiovascular diseases. On the basis of no heart sound segmentation, this paper proposes a heart sound classification method based on power spectral density time-frequency distribution features and convolutional neural network. The proposed method uses wavelet denoising as preprocessing, and uses cyclic autocorrelation to obtain the cardiac cycle. Bilinear interpolation is used to extract the time-frequency features of the power spectrum density of the cardiac cycle with the same dimensions, and then the features are sent to the convolutional neural network for training and testing. The experiment uses the Challenge 2016 data set for training and testing. The classification accuracy of the test set reaches 0.8472, and the sensitivity and specificity scores reach 0.7763 and 0.9463. The comparison with other algorithms shows that the proposed algorithm achieves a higher overall score.

Key words: heart sound classification, power spectral density, convolutional neural network, bilinear interpolation, cardiac cycle detection