【作者】 刘彬；

【导师】 郑杨；

【作者基本信息】 吉林大学 ， 内科学， 2014， 博士

【摘要】 随着城市居民生活方式、工作节奏和饮食规律的变化，心脏疾病已成为现代社会的高发病症，早期识别、正确诊断和及时治疗对降低心血管疾病的死亡率至关重要。心电图有着数百年应用的历史，随着心电图仪器的发展日趋完善，在心血管疾病的检测过程中具有不可替代的作用。简单、快速、价格低廉，有着较高的临床应用价值。但是，心电图检测也存在一些缺陷，例如，会因为操作者的手法、电极的位置等不同因素影响其结果，常常造成同一人出现不完全相同的心电图。正因为检测的差异性，病态心电图种类的复杂性，造成了结果判断的差异性和难辨性，需要具有大量的知识储备及丰富临床经验的专业医师进行分析诊断。对于某些心血管疾病如心瓣膜病和轻度心室肥厚等疾病时，心电图检测通常没有异常改变；对于心血管硬化、部分房颤等心血管疾病，心电图只能作为辅助诊断的指标。虽然心电图的变化会随着时间和部位的不同而变化，但其一些内部的变化是有生物学特征的。因此，利用现代计算技术克服传统心电的弱点，对复杂海量的心电图信息进行生物信息学处理,如何系统挖掘与心血管疾病诊断和生物识别相关的信息，使其具有更加准确性，更高实时性，易被识别的特征性，从而实现心电图可自动分类的现代诊断意义，这也是我们寻求多学科交叉合作研究的要点，也是以往医学与计算生物学研究无法实现的，是解决并提速解决问题的捷径。研究的目的是希望将人体心脏运动过程中产生的微弱电信号，既周期平衡的动态信号，进行特征提取，建立不同人、不同时间的心电信号即心电图标准化的方法。这种方法的特点能把同一个人的心电图相似化(TEC)，不管他们是正常人还是心血管疾病患者。而在两个不同个体之间，心电图永远不会相同。研究结果显示：①通过PTB数据库的试验证明，健康人群的心电数据在一定的时段内是动态不变的可以作为身份识别的生理信号；②通过QT数据库91个不同种类心脏疾病患者的身份识别方面的试验证明，大部分人的心电信号，除去某些心脏极不稳定的情况，可以在一定的时段内保持动态平稳；③试验证明心电归一化算法是心电信号的标准化过程，是心电信号特征提取的重要部分，对心电身份识别算法具有极强的辅助作用；④心电信号因其复杂性、动态唯一特性，我们认为可以作为身份识别的工具，在一定范围内使用可以确保其准确性。大规模使用的话，还需要大样板案例的验证；⑤研究反映出了心电生物学特征的复杂性，该方法的建立可在临床案例的基础上提供一个准确的生物学特征模板；⑥试验证明：CVB3/MKP株具有心肌毒性，分子进化分析显示CVB3/MKP与武汉分离株CVB3/Macocy（JQ040513）及CVB3分离株HM138916处在同一个簇内，为进一步完善CVB3/MKP病毒进化和种系发育关系提供了必要的依据。本文在心电信号特征识别的基础上，对单周期归一化心电信号进行身份识别的研究和心电拟合分类研究。采用了PTB、QT两个心电数据库进行分析。在身份识别PTB数据库方面，对52个健康人心电识别正确率可达100%， QT数据库对91个心脏病人心电识别正确率达90.1%。在心电分类方面，对PTB数据库中的50个心肌梗死患者心电图和52个健康对照心电图进行单周期归一化的心电信号的疾病分类方法研究，研究发现效果比较好的分类器有贝叶斯网络、多层感知器、随机森林分类器。经试验证明随机森林分类器的分类效果比较优秀，为了防止过拟合现象的产生，我们采用了限制决策树层次的方法，使得分类效果比较理想。我们将心电分类法应用于在心肌梗死诊断、信息安全和疾病分类领域。将心电应用于监控个人健康信息，降低可能出现的疾病风险。本文的主要贡献及创新性如下：从心电图的形成机理及结构特性出发，对心电信号特征识别、心电归一化中的关键技术进行了系统研究：（1）实施了临床不同病历心电图采集和建库。对国外心电据库PTB、MIT-BIH、AFPDB、SVDB、TWADB等进行了全面的分析。明晰心电图标准化需依赖测试心电信号数据库，数据库的适合与否决定测试与算法的优劣。如果数据库中各类疾病心电图图像过于单一，可导致算法具有片面性。（2）研究实现了基于心电图标准化的心电归一化方法及提取特征技术的建立，即分别采用检测算法去测心电信号的QRS波、提取信号的电压振幅、不同时间段电信号特征点如RQ、RS、PQ、PS特征值。（3）利用多种分类器和分析方法如PCA等进行分类。这种方法的优点在于，结合了临床医学的有关诊断技巧，可以更符合人类心脏电生理特点，有效的节省存储空间，并且对各组特征群体进一步优化，减少无用信息，实验结果显示，可达到很高识别率。（4）心电图自动分类方法的准确率达到了百分之九十五以上，运行时间仅用几秒到十几秒。研究提高了心电图分类的速度和准确率，可针对心脏疾病进行快速和准确地诊断，具有重要的诊断意义和实用价值。（5）试验证明了心电归一化算法是实现心电信号的标准化必须过程，是心电信号特征提取的重要部分，对心电身份识别具有极强的辅助作用。（6）心电归一化主要应用生物识别技术，去除心电图测量时的人为误差因素。将心电图标准化，使每位被测量者的心电图不会因为测量者的不同等影响因素而发生变化。生物特征识别技术是近年最有发展前景的一项新兴技术,随着社会的需求，研究技术不断成熟，应用领域不断扩大,享受生物识别技术的快捷与便利是现代科学的体现，如果能在体格检查或生活必需时进行，将为人类医学的进步提供支撑，生物信息学促进IT、BT(生物技术)双赢，学科交叉是项目顺利进行的必由之路。更多还原

【Abstract】 Along with the change of citizen’s lifestyle, the pace of work and diet rules, heartdisease has become a high-incident ailment in modern social conditions. Hence, earlyrecognition, accurate diagnosis and timely treatment are very important to decreasethe mortality of cardiovascular disease. With the improvement of electrocardiogramsand hundreds years usage of it, electrocardiograms has been irreplaceable in detectionof cardiovascular disease. This simple, rapid and cheap technology has high values onclinical application. However, detects still exist. For instance, the proficiency ofoperator, the site of electrodes and several other reasons would affect the accuracy ofdetection results, which may lead to attaining different results of the same person. Sothe professional doctors with rich clinical experiences are needed, because of thedetection differences, the various kinds of pathological electrocardiograms, the greatvariability and the difficulty for the judgment. Then, for cardiovascular diseases suchas valvulopathy and mild ventricular hypertrophy, the results of electrocardiogramusually have no changes. And for cardiovascular diseases such as cardiovascularsclerosis and Parts of atrial fibrillation (AF), the results of electrocardiogram can onlyused as an indicator of auxiliary diagnosis. Although the result of electrocardiogramwould change along with the detection time and the site of electrodes, same internalchanges have a common biological characteristic. Hence, the bioinformaticsprocessing on numerous ECG data would increase the accuracy and the ability ofidentification of ECG, and the study of ECG automatic classification methods becomemore and more important.In our study, we aimed to build an ECG normalized method for ECG dataattained from different persons at various times, by extracting the characteristics ofthe weak electrical signals in the heart movements, which were cycle balanceddynamic signals. The major feature of this method was that it could normalize theECG data of the same people, no matter healthy or not the person was, and the ECG data of a person was never same as others. The results showed that,①The ECG dataof healthy persons were stable in a period of time, which could be used asphysiological signals for identification, through the examination using the ECG dataobtained from PTB database.②The ECG date of most people except some seriousheart unstable situation, could maintain dynamically stable in a certain period of time,through the identification test of persons suffering cardiovascular diseases using theECG data got from QT database.③Our study showed that ECG normalizedalgorithm was a process of ECG signal data standardization. ECG normalizedalgorithm was an important step in the extracting of ECG signal characteristics, whichhad a strong supporting role for ECG identification algorithm.④We deemed thatECG signals could be used as a tool for identification because of its complex anddynamic unique characteristics, which may maintain a high accuracy within a certainlimits. Before large-scale usage, a large sample validation still needed.⑤The studyshowed the complexity of the biological characteristics of ECG. Establishment of themethod describe above would provide an accurate template of the biologicalcharacteristics based on clinical samples.⑥Tests prove: CVB3/MKP strains withmyocardial toxicity and found CVB3/MKP、CVB3/Macocy and CVB3/HM138916strains of myocarditis caused by extremely close genetic relationship. To furtherimprove the CVB3/MKP virus evolution and phylogenetic relationship provides thenecessary basis for the development.Based on the ECG normalization algorithm, the article investigated theidentification of single cycle normalization ECG signal and all kinds of aspects ofECG fitting. It also adapted the method which analyses the ECG database about PTBand QT. The recognition accuracy of52healthy participants in the PTB database canachieve the100%. The recognition accuracy of91participants in the QT database canachieve90.1%. This research explores the PTB database which includes the ECGabout50MI patients and52health control participants in order to classify the singlecycle normalization ECG signal. During the experiment, we found that the moreeffective classifiers, including Bayesian network, multilayer perceptron, random forest classifier. According to the experiment, it demonstrates that the random forestclassifier is more excellent and we use the method of limiting decision tree to avoidthe over fitting phenomenon. We will use this ECG classification method in the fieldsof diagnosing MI, information assurance and classifying the diseases. It also can beused for monitoring personal health information and reducing the potential diseaserisk.In this paper, the main contributions and innovative work is as follows: the paperstarts from the formation mechanism and structural characteristics ofelectrocardiogram (ECG), majors in the key technologies of ECG normalization, andput forward some new ideas and algorithms:(1) Using ECG and other equipment foracquiring a large number of clinical ECG and constructing a database. We conducteda comprehensive analysis of foreign ECG database, for example PTB, MIT-BIH,AFPDB, SVDB, TWADB and so on. ECG standardization relies on the database ofECG signals, selecting the appropriate library can test out the advantages anddisadvantages of the algorithm. If ECG images of various disease are single lead topartial algorithm.(2) Achieve the ECG normalization method and feature extractiontechnology based on the standardization of electrocardiogram, the detectionalgorithm were respectively used to measure QRS waves in ECG signals, extract thesignal voltage amplitude, the different time period signal feature points such as RQ,RS, PQ, PS eigenvalues.(3) Using multiple classifiers classify and analysis methodssuch as PCA. The advantage of this approach is that a combination of relevant clinicaldiagnostic techniques can be more in line with the human heart electrophysiologicalcharacteristics, effectively save storage space, further optimize characteristics of eachgroup, and reduce the useless information. Experimental results showed that canachieve high recognition rate.(4) The accuracy of the automatic classification methodabout ECG can achieve over95%and the running time is just from several seconds toten or over ten seconds. This algorithmic increases the ECG classification speed andaccuracy. It also can diagnose the heart diseases quickly and accurately. This methodis of great importance of diagnosing and practical value.(5) The test demonstrated ECG normalization algorithm is a necessary step to achieve ECG standardization, isan important part of the ECG feature extraction, and has a strong supporting role foridentification of the ECG.(6) The ECG normalization mainly uses biometrictechnology, removing the human error factor of ECGmeasurement.Through normalization of ECG data, the ECG data of different tested persons would not be changed because of various factors.Biometric identification technology will become a new technology with greatprospects in21stcentury. As the development and application of this technology,everybody will enjoy the convenience of biological identification technology.Everyone can complete the requirements of biological identification and diseasesclassification via ECG testing, during a physical examination or when it is necessary.更多还原