一种基于多层感知器的房颤心电图检测方法

Multilayer perceptron-based method for atrial fibrillation ECG detection

目的:提出一种基于多层感知器(MLP)的新型房颤识别算法。方法:首先设计一种新型自适应的R波阈值检测算法,然后以R波位置和幅度为特征,MLP为分类器进行正常/房颤心电图识别。MLP的网络参数采用深层置信网络预训练算法进行初始化,最后用误差反向传播算法对MLP网络权重进行调整。结果:在单通道心电图数据集上对正常、房颤心电信号进行分类,本研究方法的灵敏度达96.00%,特异性为84.18%,平均识别率为90.09%。结论:这种基于MLP的心电识别算法准确率高、计算复杂度较低,可为房颤的智能诊断提供一种新方法。

Objective To propose an atrial fibrillation(AF)recognition method based on multilayer perceptron(MLP).Methods Firstly,a novel R-wave detection algorithm based on adaptive threshold was designed,and then with the location and amplitude of R-wave as features,MLP was used as classifier to recognize the normal/AF electrocardiogram(ECG).The network parameters of MLP were initialized by deep belief network pre-training algorithm.Finally,the weights of MLP network were tuned by error back-propagation(BP)algorithm.Results The sensitivity,specificity and average recognition rate of the proposed method for the classification of normal and AF ECG signals on a single-channel ECG database were96.00%,84.18%and 90.09%,respectively.Conclusion The proposed algorithm based on MLP which has high accuracy and lower computation complexity can be a new method for the intelligent diagnosis of AF.