多导心电算法及应用系统的研制

发布时间：2018-01-29 22:27

本文关键词： 心电图数学形态学稀疏自编码器支持向量室性早搏　出处：《深圳大学》2017年硕士论文　论文类型：学位论文

【摘要】：心血管疾病是导致人类死亡的头号杀手,心电图(ECG)检查是目前心血管疾病预防和诊断的最常用手段之一。为了减小医生工作量,提高医生诊断的准确性,逐渐将心电信号自动分析技术应用到临床心电图诊断。已有的心电算法仍存诸如抗扰、精确分类和智能多导分析等问题,限制其临床的广泛应用。本课题主要围绕”波形检测和特征点定位”、”数字特征和形态特征提取”、”心搏分类和诊断分析”3个方面展开。波形检测和特征点定位的研究:在QRS波群检测方面,提出了一种基于数学形态学的PanTompkin算法,该方法主要优点在于抑制了高大T波,基线漂移以及高频噪声算法的影响。QRS波群起点和终点的定位,提出一种新的低通差分法(NLPD),算法在融合多通道ECG信号的基础之上,通过一种平滑处理,避免了起点和终点差分峰的定位,提高了算法对异常QRS波群检测的准确性。P、T波的检测,本文首次综合传统的低通差分法、梯形面积法以及临床医生常用的切线法,减小了P、T融合、u波等情况下的测量误差,改善了长PR间期导致的P波漏检等。经过MIT-BIH数据库、CSE数据库等验证了本文所提出算法的有效性。心电信号数字特征和形态特征提取的研究:本文对比了常用的参数分析方法—时域、频域、时频域、高阶统计量分析方法,提取了QT间期、PR间期等医生常用参数。QRS等波基本形态分类,相关研究较少,本文首次采用了基于深度学习的方法—稀疏自编码器(SAE)对QRS波群基本形态进行分类。波形分类和诊断分析的研究:主要提出了一种基于特征和支持向量机(SVM)的波形分类算法。该算法所选参数是基于本文所提出的等参数,并结合其他的时域QRS波群参数将心搏分为PVC/N两种类型,其他类型的心律失常疾病分类可以在此基础上进行展开。此外,基于深度学习的方法进行心律失常分析,相关研究较少,本文研究了基于稀疏自编码器(SAE)将心搏分为PVC/N两种。经过MIT-BIH数据库测试分析,验证了本文所提出算法的有效性。本课题搭建了心电信号采集平台,采集了600例不同年龄、性别、疾病种类的临床12导心电信号,经过医生标定,初步建立一个基于中国人的ECG数据库(称为SZU数据库),并对本文所提出的算法进行了验证。该数据库更加适合于基于中国人的ECG自动分析算法的测试和开发。
[Abstract]:Cardiovascular disease is the leading killer of human death. Electrocardiogram (ECG) is one of the most commonly used methods to prevent and diagnose cardiovascular disease in order to reduce the workload of doctors. To improve the accuracy of doctor diagnosis, ECG automatic analysis technology has been gradually applied to clinical ECG diagnosis. Existing ECG algorithms still exist such problems as immunity, accurate classification and intelligent multi-conductance analysis. This paper focuses on "Waveform Detection and feature Point location" and "Digital feature and morphological feature extraction". Research on waveform detection and feature point location: in the aspect of QRS wave group detection, a PanTompkin algorithm based on mathematical morphology is proposed. The main advantage of this method is that it can suppress the influence of high T wave, baseline drift and high frequency noise algorithm. A new low pass difference method (NLPDs) is proposed to locate the starting and ending point of QRS wave group. Based on the fusion of multi-channel ECG signals, the algorithm avoids the location of the starting and end differential peaks by a smoothing process, and improves the accuracy of the algorithm in detecting abnormal QRS waves. In this paper, the traditional low-pass difference method, trapezoidal area method and tangent method commonly used by clinicians are used for the first time to reduce the measurement error in the case of PtT fusion. Improved the P-wave leakage caused by the long PR interval, etc. Passed through the MIT-BIH database. CSE database and so on verify the effectiveness of the proposed algorithm. ECG digital features and morphological features extraction research: this paper compared the commonly used parameter analysis methods-time domain, frequency domain, time-frequency domain. The basic morphological classification of QT interval and PR interval was extracted by high-order statistical analysis, and the basic morphological classification of QRS and so on was seldom studied. For the first time, a method based on depth learning, sparse self-encoder (SAE), is used in this paper. Classification of basic morphology of QRS wave groups. Waveform classification and diagnostic analysis: a new method based on feature and support vector machine (SVM) is proposed. The algorithm selects parameters based on the isoparametric proposed in this paper. Combined with other time-domain QRS wave group parameters, cardiac beats can be divided into two types of PVC/N, and other types of arrhythmia disease classification can be carried out on this basis. There are few studies on arrhythmia analysis based on the method of deep learning. In this paper, based on sparse self-encoder (SAE), cardiac beats are divided into two types of PVC/N, which are tested and analyzed by MIT-BIH database. The validity of the proposed algorithm is verified. In this paper, a ECG acquisition platform is set up, and 600 clinical ECG signals of different age, sex and disease type are collected and calibrated by doctors. Establish a Chinese based ECG database (called SZU database). The proposed algorithm is verified and the database is more suitable for the test and development of ECG automatic analysis algorithm based on Chinese.
【学位授予单位】：深圳大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN911.7;R540.4

【相似文献】