基于隐马尔可夫心动周期的心音分割算法  

Heart sound segmentation algorithm based on hidden markov cardiac cycle

作　　者：许亚楠 张显飞[1] 赵治栋[2] XU Yanan;ZHANG Xianfei;ZHAO Zhidong(School of Electronic Information,Hangzhou Dianzi University,Hangzhou 310018,China;School of Cyberspace Security,Hangzhou Dianzi University,Hangzhou 310018,China)

机构地区：[1]杭州电子科技大学电子信息学院,杭州310018 [2]杭州电子科技大学网络空间安全学院,杭州310018

出　　处：《杭州电子科技大学学报（自然科学版）》2024年第5期20-30,共11页Journal of Hangzhou Dianzi University：Natural Sciences

基　　金：浙江省公益技术研究计划项目(LGG21F020002)。

摘　　要：心血管疾病是全球范围内死亡人数最多的疾病之一,其早期筛查与精准诊断对降低患病率和死亡率具有重要的社会价值。心音图(phonocardiogram,PCG)作为人体重要的生理信号,包含了丰富的心脏病理信息,能客观反映出心脏和心血管系统的健康状态,是心血管疾病诊断的重要数据源。心音分割是心音信号病理分析的基础,对病理特征的提取和疾病诊断具有决定性作用。而现有的心音分割算法往往需配合同步输入的心电信号才能达到精确的分割效果,对纯心音信号的分割效果不佳。因此提出了基于隐马尔可夫心动周期(Hidden Markov Cardiac Cycle,HMCC)的心音分割算法,实现对纯PCG信号的精准分割。首先利用基线校准和小波去噪实现信号预处理;其次提出基于希尔伯特变换的心音包络提取方法,结合心动周期规律定位峰值与S1、S2的对应关系;进一步提出改进的隐马尔可夫算法,更新心音的初始状态分布,优化维特比算法进行心音区间持续时间的计算。244条心音数据的同基准和国内外研究对比实验表明该心音分割算法,实现了精度分数为97.23%、分割准确率为97.32%的分割定位,为基于PCG的心血管疾病智能辅助诊断分析提供了高质量的分割数据源。Cardiovascular disease is one of the most deadly diseases in the world,and its early screening and accurate diagnosis have important social value for reducing morbidity and mortality.As an important physiological signal of human body,phonocardiogram(PCG) contains rich cardiac pathological information,can objectively reflect the health status of heart and cardiovascular system,and is an important data source for diagnosis of cardiovascular diseases.Heart sound segmentation is the basis of heart sound signal pathological analysis,and plays a decisive role in the extraction of pathological features and disease diagnosis.However,prevailing heart sound segmentation algorithms often need to cooperate with the synchronous input ECG signal to achieve accurate segmentation effect,and the segmentation effect of pure heart sound signal is not good.Therefore,a heart sound segmentation algorithm based on Hidden Markov Cardiac Cycle(HMCC) is proposed to achieve accurate segmentation of pure PCG signals.Firstly,baseline calibration and wavelet denoising are used to realize signal preprocessing.Secondly,the heart sound envelope extraction based on Hilbert transform is proposed,and the corresponding relationship between peak value and S1 and S2 is located by combining with the law of cardiac cycle.An improved hidden Markov algorithm is further proposed to update the initial state distribution of heart sounds and optimize the Viterbi algorithm to calculate the duration of heart sound interval.Through the same benchmark of 244 heart sound data and comparative experiments at home and abroad,the segmentation positioning with accuracy score of 97.23% and segmentation accuracy rate of 97.32% has been achieved,providing a high-quality segmentation data source for PCG-based intelligent auxiliary diagnosis and analysis of cardiovascular diseases.

关 键 词：心音图 心音分割 隐马尔可夫 心动周期 持续时间 

分 类 号：TN911.7[电子电信—通信与信息系统]