基于深度信念网的心电自动分类

发布时间：2018-06-01 05:10

  本文选题：深度信念网 + 心电节拍分类　； 参考：《计算机工程与设计》2017年05期

【摘要】：提出一种基于深度信念网(deep belief network,DBN)和心电波形采样的心电自动分类算法。对心电信号进行滤波、R波定位后,以QRS波群的180Hz下采样表示心拍形态,结合RR间期特征,使用的DBN共6层,隐藏层神经元数目为30。使用标准数据库对DBN进行训练和测试,结果为平均Se88.6%,平均P~+62.1%,优于现有特征选择方法的结果,基于深度学习的心拍分类算法无需波形特征提取步骤,解决了目前的波形特征对心拍的人间差异没有鲁棒性的问题。
[Abstract]:An automatic ECG classification algorithm based on deep belief network (DBN) and ECG waveform sampling is proposed. After filtered R wave location of ECG signal, the beat morphology was represented by 180Hz downsampling of QRS wave group. Combined with RR interval characteristics, six layers of DBN were used, and the number of neurons in hidden layer was 30. The standard database is used to train and test the DBN. The results show that the average Se88.6 and P62.1 are better than the existing feature selection methods. The beat classification algorithm based on depth learning does not need waveform feature extraction step. It solves the problem that the current waveform features are not robust to the human differences.
【作者单位】： 中国科学院微电子研究所;中国科学院微电子研究所昆山分所;
【基金】：国家自然科学基金项目(61271423)
【分类号】：R540.4;TN911.7
，

本文编号：1962958