基于序列分解和符号化的心率失常信号分析方法研究

发布时间：2018-06-12 10:29

本文选题：心室纤颤 + 持续性心动过速　；参考：《济南大学》2017年硕士论文

【摘要】：心血管疾病是威胁人类健康的疾病之一。近年来,心血管疾病的发病率逐渐上升,严重危及人们的生命安全。心血管疾病患者越来越多,越来越趋于年轻化。心血管疾病最严重的临床表现就是心脏猝死,如果患者心脏猝死后不能得到及时有效的治疗,将会失去生命。正因如此,多国医药卫生部门、研究中心都对此进行研究。经研究发现,大多数心脏猝死都是由心室纤颤(Ventricular Fibrillation,VF)或者持续性心动过速(Ventricular Tachycardia,VT)恶化导致的。针对VF与VT,临床医治方案是不同的。如果患者心室纤颤发作,必须马上进行除颤,这是目前效果最佳的治疗方案。如果患者是持续性室性心动过速,需要及时正确的进行药物治疗,以达到降低转为心室纤颤或者心脏猝死发生率的目的。如果VT被诊断为VF,患者将遭受不必要的电击,对心脏造成创伤。如果VF被诊断为VT,没有及时进行除颤,患者将会发生心脏猝死。目前检测算法能很好的区分窦性心律和VT、VF,但是VT与VF的检测还在持续研究中。根据目前对心电信号的研究,可将ECG视为非线性时间序列的范畴。用非线性动力学理论来分析ECG具有明显的优势。本文以非线性动力学为理论依据对符号时间序列分析方法进行了分析研究,并结合时间序列分解算法提出了基于序列分解和符号时间序列分析方法的VT/VF检测新算法,其中包括基于EMD与符号时间序列分析方法的新算法和基于小波分析与符号时间序列分析方法的新算法。然后利用数据集对提出的算法进行实验验证,结合符号时间序列理论对实验结果进行分析与比较。实验表明EMD结合符号时间序列分析方法其VT/VF的识别率为97.82%,小波分析结合符号时间序列分析方法其VT/VF的识别率为99.5%。并针对时间序列分析方法中的二值粗粒化不足对其改进。通过实验对比了改进之后算法与改进之前的变化。针对VT/VF的识别度以及算法执行时间对改进算法和未改进算法做了对比试验。实验表明,在VT/VF识别度基本一致的情况下,改进算法相比未改进算法相比,计算量减少,改进算法的时间缩减了30多倍。而且针对小样本时,改进算法会获得更高的VT/VF识别率。换言之,在实时性要求较高的监护系统中,改进算法能对输入信号更快的做出响应。对算法进行改进后,计算量大大减小,算法执行时间缩短,适合用于监护仪、自动体外除颤仪(AED)植和入式除颤仪(ICD)等实时监测预警和自动除颤设备。最后从样本时间与符号化等级两方面对改进算法与单独的符号时间序列分析方法进行了对比实验,实验表明改进算法无论是在样本时间还是符号化等级相比单独使用符号时间序列分析方法具有更好的识别性能,说明了基于序列分解算法及符号时间序列分析算法的融合算法的有效性。本文主要创新点:第一,提出了基于经验模态分解序列分解方法与符号时间序列分析方法的VT/VF检测新算法,研究表明该方法对VT/VF具有较高的识别度。第二,提出了基于小波分析与符号时间序列分析方法的VT/VF检测新算法,研究表明适当的信号采样率有利于符号时间序列分析方法挖掘序列的本质特征,通过小波分解得到的分解序列提取符号熵对VT/VF相比EMD具有更高的识别度。最后,分析了符号时间序列分析方法现有的不足,对过粗粒化问题进行了改进,并从多方面对改进的算法进行了分析,实验表明,改进算法在保持对VT/VF识别率基本不变的情况下其计算量降低,并且针对小样本具有更高的识别率。
[Abstract]:Cardiovascular disease is one of the diseases that threaten human health. In recent years, the incidence of cardiovascular disease is increasing, which seriously endangers the life safety of people. More and more patients with cardiovascular diseases are becoming younger. The most serious clinical manifestation of cardiovascular disease is the sudden cardiac death, if the sudden death of the heart can not be obtained in time. Effective treatment will lose life. Because of this, many national medical departments, research centers have studied it. Most sudden cardiac deaths have been found to be caused by the deterioration of Ventricular Fibrillation (VF) or Ventricular Tachycardia (VT). For VF and VT, a clinical treatment plan. If the patient is persistent ventricular tachycardia, the patient needs to be treated in time and correctly in order to reduce the rate of ventricular fibrillation or sudden cardiac death. If VT is diagnosed as VF, the patient will suffer. If the VF is diagnosed as VT and does not defibrillate in time, the patient will have a sudden cardiac death. The current detection algorithm can distinguish the sinus rhythm and VT, VF, but the detection of VT and VF is still in a continuous study. According to the current study of ECG signals, ECG can be considered as a nonlinear time series. In this paper, the analysis of ECG has obvious advantages by using nonlinear dynamics theory. In this paper, the method of symbolic time series analysis is analyzed based on nonlinear dynamics, and a new VT/VF detection algorithm based on sequence decomposition and symbol time series analysis is proposed, which includes the time series decomposition algorithm. A new algorithm based on EMD and symbolic time series analysis method and a new algorithm based on wavelet analysis and symbolic time series analysis. The experimental results are analyzed and compared with the symbolic time series theory. The experimental results show that EMD combined with symbolic time series analysis method. The recognition rate of VT/VF is 97.82%. The recognition rate of VT/VF is 99.5%. with the method of wavelet analysis and symbolic time series analysis. The improvement of the two value coarse graining in the time series analysis method is improved. The changes of the algorithm and the improvement before the improved algorithm are compared. The recognition degree of the VT/ VF and the time of the algorithm execution are modified. Compared with the unimproved algorithm, the experiment shows that, when the VT/VF recognition degree is basically consistent, the improved algorithm is less computational than that of the unimproved algorithm, and the time of the improved algorithm is reduced by 30 times. Moreover, the improved algorithm will get a higher VT/VF recognition rate for small samples. In other words, the real-time requirement is better than that of the improved algorithm. In the high monitoring system, the improved algorithm can respond to the input signal faster. After improving the algorithm, the calculation amount is greatly reduced and the execution time of the algorithm is shortened. It is suitable for monitor, automatic defibrillator (AED) plant and ICD defibrillator (ICD) and other real-time monitoring and defibrillator equipment. The two aspects of the improved algorithm are compared with the single symbol time series analysis method. The experiment shows that the improved algorithm has better recognition performance, which is based on the sequence decomposition algorithm and the symbolic time series analysis, not only in the sample time or the symbolic level, but also in the separate use of symbolic time series analysis. The main innovation points in this paper are as follows: first, a new VT/VF detection algorithm based on the empirical mode decomposition sequence decomposition method and the symbolic time series analysis method is proposed. The research shows that the method has a high recognition degree to VT/VF. Second, the VT/VF based on the small wave analysis and the symbol time sequence analysis method is proposed. The new algorithm is detected. The research shows that the appropriate signal sampling rate is beneficial to the symbolic time series analysis method to excavate the essential characteristics of the sequence. The extraction of the symbol entropy by the decomposition sequence obtained by the wavelet decomposition has a higher recognition degree to the VT/VF compared with the EMD. Finally, the shortcomings of the symbolic time series analysis method are analyzed, and the problem of over coarse granulation is analyzed. The improvement is carried out and the improved algorithm is analyzed from many aspects. The experiment shows that the improved algorithm reduces the amount of calculation and has a higher recognition rate for the small sample in the case of keeping the VT/VF recognition rate basically unchanged.
【学位授予单位】：济南大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R541.7;TN911.6

【参考文献】