基于灰色关联度k-近邻法的房性心律失常识别

Atrial Arrhythmia Recognition Based on Grey Correlation and K-nearest Neighbour

目的研究用灰色理论的灰色关联度结合k-近邻法快速准确地识别窦性、房扑和房颤信号。方法将心电信号分成训练集和测试集,首先用多尺度小波将心电信号变换到时频域,然后提取小波系数矩阵的奇异值作为信号的特征向量,将所有训练样本的特征向量作为标准模板,求出测试样本特征向量与标准模板之间的灰关联系数,最后结合k-近邻法对测试样本做出判断。分别用MIT-BIH心律失常数据库和犬心外膜信号数据库来评价提出的基于灰关联度的k-近邻法识别心律失常信号的特异性、敏感性和准确率。结果实验结果表明:和常规灰关联度法、常规k-近邻法、BP神经网络相比,本方法对窦性、房扑和房颤信号有较好的识别性能,且具有识别速度快的优点。结论本方法不需要大量的训练样本,计算简单,能较准确快速地识别窦性、房扑和房颤信号,有望应用于治疗心律失常的可植入装置。

Objective To study the method combining the grey correlation of the grey theory and the k-nearest neighbour to recognize sinus rhythm （SR）, atrial flutter （AFL） and atrial fibrillation （AF）. Methods The electrocardiograms were divided into training data and testing data. Firstly, signals were transformed into time-frequency domain using multi-scale wavelet. Then singular values were extracted from the wavelet coefficient matrix as feature vectors of the signals. With feature vectors of all the training data as normal template, grey correlation coefficients between feature vectors of the testing data and the normal template were calculated. Finally recognition was made using the knearest neighbour. Sensitivity （SE）, specificity（SP） and accuracy （AC） of the method were evaluated for atrial arrhythmia recognition with two databases, the MIT-BIH arrhythmia database and the canine endocardial database. Results Experimental results demonstrated that the proposed method achieved higher recognition performance for SR, AFL or AF with a higher computation speed compared with the traditional gey correlation, the traditional k-nearest neighbour or the back propagation （BP） neural network. Conclusion The proposed method can recognize SR,AFL and AF accurately with a simple computation and small training samples. It is expected to be used in implantable devices for therapy of atrial arrythmias.