基于一维卷积神经网络的心电信号质量分类

ECG signal quality classification based on one-dimensional Convolutional Neural Network

研究一种基于单导联心电信号质量二分类方法。为了改善传统方法在进行心电信号质量分类下需要手动提取特征的复杂性以及选取规则包含主观性的缺点,基于Tensorflow框架设计了一个一维卷积神经网络,利用MIT-BIH和NSTDB数据库构建训练数据集,通过不断调整网络模型自动学习分类特征,使用2个公开测试集与1个私有测试集验证算法的泛化性,实验结果表明,提出的算法在3个测试集上的平均准确率为96.5%、灵敏性为98.1%和特异性为94.7%。最后,相比于基于传统SVM模型或CNN的方法,本文算法不仅精度较高,而且在未知的数据集上表现较好。研究证明,提出的方法不仅能够避免手动处理海量数据的弊端,而且能够以更客观、更高的准确度实现心电信号质量的分类。

This paper studies a two-classification method based on single lead ECG signal quality.To improve the complexity of manual feature extraction and the subjectivity of rules selection in traditional methods for ECG quality classification,a one-dimensional Convolutional Neural Network is designed based on the Tensorflow framework.The training data set is constructed by using MIT-BIH and NSTDB database,and the classification features are automatically learned by constantly adjusting the network model.Two public test sets and one private test set are used to verify the generalization of the algorithm.The experimental results show that the average accuracy,sensitivity,and specificity of the proposed algorithm on the three test sets are 96.5%,98.1%,and 94.7%.Finally,compared with the methods based on the traditional SVM model or CNN,the algorithm not only has higher accuracy,but also performs well on unknown data sets.The research shows that the proposed method can avoid the disadvantages of manually processing massive data,and realize the classification of ECG signal quality with more objective and higher accuracy.