一种新的卷积神经网络房颤检测方法

A Novel Detection Method of Atrial Fibrillation Based on Convolutional Neural Network

心房颤动(AF)是一种最为常见的心功能紊乱心脏病,为提高房颤的识别效率和准确率,提出了一种基于卷积神经网络的心电信号分类模型。首先采用双中值滤波器对输入心电信号进行预处理,去除基线漂移;而后提出基于卷积神经网络的R波检测器,其对MIT-BIH心律失常数据库评估结果达到了98.83%的阳性预测率、99.72%的灵敏度和99.54%的准确度;最后提出了一种基于CNN的心电信号分类模型,其对CPSC 2017挑战赛数据库实验结果总体达到96.82%的灵敏度、97.18%的阴性预测率,显示该模型性能良好,具有较强的泛化性与鲁棒性。

Atrial fibrillation(AF)is one of the most common cardiac disorders.To improve the detection accuracy and efficiency of atrial fibrillation,this paper proposes a convolutional neural network(CNN)based ECG classification model for atrial fibrillation detection.Firstly,the double median filters are adopted to remove baseline drift noise of the input electrocardiogram(ECG)signal.Secondly,a CNN based R-wave detector is proposed,which achieves 98.83%positive prediction rate,99.72%sensitivity,and 99.54%accuracy on the MIT-BIH arrhythmia database,respectively.Finally,a CNN based ECG classification model for atrial fibrillation detection is developed.Experimental results demonstrate that the proposed ECG classification model achieves good performance,generalization,and robustness.The proposed method achieves an overall sensitivity of 96.82%and a negative predictive rate of 97.18%on the CPSC 2017 challenge database.