基于小波包分解的复杂心音信号分段定位与特征提取研究

发布时间：2018-11-22 20:24

【摘要】：先天性心脏病(简称先心病)属于出生缺陷疾病。通过手术即可治愈,并且早发现、早治疗将大大提高其治愈率。但先心病的早期症状不明显,其诊断分为初诊和确诊,初诊阶段主要是听诊,易受主观经验影响而出现漏诊,耽误治疗的黄金时间。心音图含有大量的心脏信息,对心音图进行数字信号处理,可大大改善这一情况,提高疾病的诊断效率。本文从心音的产生机制、信号时频特性,以及基本的心音处理流程出发,包括预处理、心音分割、特征提取,对心音信号进行了定量和定性分析。在预处理阶段,提出了针对心音分析的简化小波包多分辨率分解算法,得到了心音能量在不同频率段的分布,有效地将噪声与心音、生理及病理杂音分离开后；利用心音与杂音的频率分布关系,定性地将各小波包系数划分为超低、低、中和高频4个频段,并使用归一化香浓能量包络提取算法,来计算上述4个频段的信号包络；然后根据自相关相关原理,通过相乘与相加对这四个包络信号进行突出与压缩,得到心音包络和标尺包络。其中,心音包络具有信号的总体包络信息,而标尺包络对能量较强的奇异点进行了放大,对能量较弱的毛刺进行了削弱。在心音分割阶段,本文提出了新的包络提取策略,利用心音包络和标尺包络,结合心音基本时域特征,实现了心音的自适应分段定位,且无需心电等其他参考信息。该方法从时域和频域两大角度出发,得到的包络信息较完整。对正常和异常共计50例复杂心音的定位分段准确率达95%以上。新的方法不仅效率高,同时有效地降低了传统心音分段方法仅从峰值点处左右移动固定长度,来计算心音边界位置而存在的偏移风险,减少误判等错误的出现。在特征提取方面,本文进行了杂音检测和心力储备评估两方面的心音特征提取,将郭兴明教授提出的心脏储备指标(S1/S2、D/S、HR)作为先心病无创检测的辅助指标,发现异常心音的D/S评级较低,HR偏高。此外,杂音的出现是先心病最主要的病症,为此本文计算了收缩期、舒张期各期能量分数,发现在中、高频段早、中、晚各期的能量分数中,异常心音要高于正常心音,这有力地提示了病理杂音的出现；其次,发现低频段生理杂音较多,且容易与该频段的病理杂音混淆,同时正常的第三心音和第四心音,容易与舒张期中、晚期杂音混淆,至此,本文从频率和能量的角度,对正常、异常,以及不同疾病程度的50例心音进行研究,提取到了有效的心音信号特征,为临床心音分析应用的开发打下了基础。
[Abstract]:Congenital heart disease (CHD) is a birth defect disease. Surgical treatment can be cured, and early detection, early treatment will greatly improve its cure rate. But the early symptoms of congenital heart disease are not obvious, its diagnosis is divided into initial diagnosis and diagnosis, the first diagnosis stage is mainly auscultation, easy to be affected by subjective experience and missed diagnosis, delaying the prime time of treatment. Cardiac phonogram contains a lot of heart information. Digital signal processing can greatly improve this situation and improve the diagnosis efficiency of disease. In this paper, the mechanism of heart sound generation, the time-frequency characteristic of the signal, and the basic flow of heart sound processing, including pretreatment, heart sound segmentation, feature extraction, and quantitative and qualitative analysis of heart sound signal are analyzed quantitatively and qualitatively. In the preprocessing stage, a simplified wavelet packet multi-resolution decomposition algorithm for heart sound analysis is proposed. The distribution of heart sound energy in different frequency bands is obtained, and the noise is effectively separated from heart sound, physiological and pathological murmur. According to the frequency distribution of heart sound and murmur, the wavelet packet coefficients are qualitatively divided into four bands: ultra low, middle and high frequency, and the normalized energy envelope extraction algorithm is used to calculate the signal envelope of the above four bands. According to the autocorrelation principle, the four envelope signals are extruded and compressed by multiplying and adding, and the heart sound envelope and the scale envelope are obtained. The heart sound envelope has the total envelope information of the signal, while the scale envelope amplifies the singularity point with stronger energy and weakens the burr with weaker energy. In the phase of heart sound segmentation, a new envelope extraction strategy is proposed in this paper. With the use of heart sound envelope and scale envelope, combined with the basic characteristics of heart sound in time domain, the adaptive segmentation of heart sound is realized, and other reference information such as ECG are not required. The envelope information obtained by this method is relatively complete from time domain and frequency domain. The accuracy of segmental localization of 50 cases of normal and abnormal heart sounds was over 95%. The new method not only has high efficiency, but also effectively reduces the shift of fixed length from the peak point to the left and right in the traditional heart sound segmentation method, so as to calculate the deviation risk of the heart sound boundary position and reduce the occurrence of errors such as misjudgment. In the aspect of feature extraction, the murmur detection and assessment of cardiac force reserve were carried out in this paper. The cardiac reserve index (S1 / S2D / SHR) proposed by Professor Guo Xingming was used as the auxiliary index for noninvasive detection of congenital heart disease. The D / S rating of abnormal heart sounds was lower and HR was higher. In addition, the emergence of murmur is the most important disease of congenital heart disease. Therefore, the energy fractions of systolic and diastolic phases are calculated. It is found that abnormal heart sounds are higher than normal heart sounds in the early, middle and late stages of middle, high frequency. This strongly indicates the appearance of pathological murmur; Secondly, we found that there are many physiological murmur in low frequency band, and it is easy to be confused with pathological murmur in this frequency band. At the same time, the normal third heart sound and fourth heart sound are easily confused with middle and late diastolic murmur. So far, from the angle of frequency and energy, 50 cases of heart sounds with normal, abnormal and different degree of disease were studied, and the effective characteristics of heart sounds were extracted, which laid a foundation for the development of clinical analysis and application of heart sounds.
【学位授予单位】：云南大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：R725.4;TN911.7

【参考文献】