先天性心脏病心音库构建与心音特征提取方法研究

Congenital heart disease (CHD) is one of most serious cardiovascular diseases for children, especially in altiplano area. Data records showed about 0.7% - 1% babies were born with CHD and among them 60% would die in one year after. It has become the first 'killer' of infants and children. It also made a disaster and heavy load for patient families and society. The most common way for early diagnosis of CHD is by heart auscultation. But it is very difficult for physician to handle. Doppler echocardiogram could accurately diagnose a CHD. But it cost a lot and is only available in some big hospital at large city. The most farmer families cannot afford it. The primary diagnosing CHD by heart auscultation is not really easy and accurate. It depends on the physician's clinical experience which contains considerable difficulties for them to take it. Heart sound provides useful information on the structural integrity and function of heart valves. It also provides valuable physiological and pathological information of heart and blood vessels. PCG (phonocardiogram) is a record of heart sound. It was useful in clinical in last century. But heart sound is a nonstationary and complex signal. It is hardly finished to analyse a heart sound by using PCG in the time domain. In this study, time-frequency analysis methods, such as short-time Fourier transform, continuous wavelet transform, discrete wavelet decomposition, wavelet packet, MP (matching pursuit method), EMD（The Empirical Mode Decomposition）, and so on, will be used to discompose the hear sound signals for feature extraction of CHD. Our research group has focused on this study for 6 years. It is a joint team including experts from digital signal processing and cardiovascular areas. We got a little fund (RMB 60000.0) from NSFC in 2006 for 1 year. Some research results have been gotten. In our previous study, about 200 cases of heat sound were recorded, including normal and CHD heart sounds. The heart sounds were analysed by using the time-frequency methods mentioned above. Some features of CHD heart sound were extracted. The correct recognition ratio was 70% from a small database. Also about 10 research papers have been published. Though some active results have been gotten, the research does not finish. The feature extracted needs to be test by using more cases CHD patients. The analysis of CHD heart sound and new feature extraction will be taken continuously. A database of CHD heart sound should to be set. We need more and more case of CHD heart sound. Our research goal is to search an effective way for early diagnosis of CHD. We do not want to replace the Doppler echocardiogram by our method. It is just an assistant way to increase the correct ratio when physician auscultate heart for primary diagnosis of CHD. It will be not only helpful for early diagnosis and basis CHD research, but also be propitious for CHD patients, especially in small city, farm, and west altiplano.

先天性心脏病（先心病）是我国青少年的高发性疾病，高原尤为严重。其初诊主要靠心脏听诊，专业性极强，基层医生难以掌握。心音为非平稳信号，对其分析研究具有重要的科学意义和临床应用价值。本课题旨在建立先心病心音数据库、采用数字信号处理技术，例如：小波分析、EMD、MP等时频分析法，对病例心音进行分析研究、特征提取，并应用模式识别、人工神经网络等技术作分类识别。目前国内外尚无此类心音数据库，类似研究也颇少。我们前期的研究已获得了一些积极结果，采集心音202例，从中提取到了部分能反映相关生理病理信息的特征，对小样本的正确识别率达70%。但心音数据还十分有限，没有足够的数据，无法完成对先心病心音的最终分析。对已提取的特征需更多的病例来验证，现有分析方法、算法也有待改进，需提取更有效的特征，让心脏听诊与分析同步进行，探索便捷有效的先心病早期临床诊断技术，帮助基层医生、服务先心病基础研究和广大西部农村患者。

先天性心脏病CHD (congenital heart disease)，简称先心病，我国每年出生的婴儿约1%患有先心病，给患者家庭和社会带来巨大的不幸和深重的负担。先心病的初诊主要依靠心脏听诊，此项工作对临床经验和专业知识要求很高，基层的临床医生往往难以掌握。云南省是先心病的高发区，每年的先心病的筛查基本上是依靠省级医疗机构的专家小分队深入地州、山区进行，目前，地州还无能力承担，改进先心病初诊方法已经迫在眉睫。随着数字技术、互联网技术的发展，采用现代数字信号处理技术，采集先心病病例心音，建立病例数据库，对先心病病例心音进行分析、特征提取，进而完成辅助诊断，并进一步在云端通过大数据处理，对先心病进行远程诊断、病例分析、研究先心病的分布、发病机制等。项目旨在研发一种便携式先心病辅助诊断仪，及相应的先心病心音特征提取、识别算法，借助大数据处理，服务于先心病筛查。项目完成后，将使乡村卫生院有能力自行承担先心病初诊筛查工作，更好地为云南省先心病的防治发挥作用。先心病初诊主要依靠心脏听诊，这表明病例心音中含有相应的特征信息。如何提取先心病心音固有的特征并进行分类识别，就是本课题需要研究的内容。但由于国内外无现成的先心病病例数据库，因而采集病例心音、建立心音数据库也是研究工作的一个重点。从学术的角度看，本课题的研究工作实际上是要解决一个机器辅助听诊，以及模式识别与人工智能的问题。具体主要研究内容如下：1）采集先心病病例心音并构建心音样本数据库；2）研发心音（PCG）、心电（ECG）采集系统；3）对先心病病例心音进行分析研究，研究有效的分析算法； 4）研究如何将上述成果具体应用到临床中，研究先心病初诊辅助诊断技术。.项目主要进展和成果：1）已经采集建立了一个包括常见的房间隔缺损（ASD）、室间隔缺损（VSD）、动脉导管未闭（PDA）等在内的各类先心病例心音数据库，累计心音数据样本1001例。2）改进并完善了自行设计的便携式的心音（PCG）、心电（ECG）信号采集及分析系统。3）对先心病病例心音进行分析研究，目前的分析算法，对病例心音的分类识别率已经达到76.7%。4）深入研究了先心病初诊辅助诊断技术，目前处于实验室验证阶段。5）已经发表学术论文13篇，申请专利1项，培养研究生10名，其中，博士研究生1名。课题组将继续努力，争取尽早完成研究并用于临床，为云南先心病的防治发挥作用。