基于多特征融合与卷积神经网络的房颤检测

发布时间：2018-04-18 20:40

  本文选题：卷积神经网络 + 多特征融合　； 参考：《激光杂志》2017年05期

【摘要】：为了解决传统的房颤检测算法中P波形态多变而不易提取特征的问题,本文提出了一种基于多特征融合与卷积神经网络结合的房颤检测算法。首先,分别提取单心拍心房活动信号递归矩阵的特征值及相邻两个心拍的心房活动信号的相干谱来得到底层特征;然后,分别采用卷积神经网络对底层特征进行分析;最后,采用决策级融合来改善算法的性能。经MIT-BIH房颤数据库验证,该算法的正确率,灵敏度,特异性分别可达95.62%,99.88%,91.36%。结果表明,该方法能有效解决特征提取困难,泛化能力差的问题。
[Abstract]:In order to solve the problem that P-wave shape is changeable and difficult to extract features in traditional atrial fibrillation detection algorithm, a novel detection algorithm based on multi-feature fusion and convolution neural network is proposed in this paper.Firstly, the eigenvalues of the recurrent matrix of atrial activity signals and the coherent spectrum of the two adjacent atrial activity signals are extracted to obtain the underlying features. Then, convolution neural networks are used to analyze the underlying features.Decision level fusion is used to improve the performance of the algorithm.The accuracy, sensitivity and specificity of the algorithm are 95.62% and 99.88%, respectively. The accuracy, sensitivity and specificity of the algorithm are 99.88% and 91.36%, respectively, verified by the MIT-BIH atrial fibrillation database.The results show that this method can effectively solve the problem of difficult feature extraction and poor generalization ability.
【作者单位】： 河北大学;
【基金】：国家自然科学基金项目(61673158) 河北省自然科学基金项目(F2015201112) 河北省高等学校科学研究重点项目(ZD2015067)
【分类号】：R541.75;TP183
，

本文编号：1769995