面向人体脉搏的特征提取及病理诊断方法

Feature Extraction and Pathological Diagnosis Methods for Human Pulse

采用集合经验模态分解的功率归一化倒谱系数(Power Normalized Cepstral Coefficients base on EEMD,EPNCC)作为人体脉搏时域的补充特征,把多周期人体脉搏信号的时域及EPNCC特征进行融合后,作为卷积神经网络的输入,开展人体脉搏特征的提取、识别及分类研究。采用从MIT-BIH-MIMIC数据库得到的呼吸衰竭、肺水肿、心源性休克三种临床脉搏信号,借助上述方法开展了实验研究,实验结果表明,脉搏特征识别及分类准确率达到95.7%,识别及分类效果较好。

Using the Power Normalized Cepstral Coefficients base on EEMD(EPNCC)of Ensemble Empirical Mode Decomposition as a supplementary feature of the human pulse time domain,the time-domain and EPNCC features of multi cycle human pulse signals are fused as inputs to the Convolutional Neural Networks to carry out research on the extraction,recognition,and classification of human pulse features.Experimental research is conducted using three clinical pulse signals,namely respiratory failure,pulmonary edema,and cardiogenic shock,obtained from the MIT-BIH-MIMIC database.The experimental results shows that the accuracy of pulse feature recognition and classification reaches 95.7%,and the recognition and classification effect is good.