基于小波变换的心电波形分类及冠心病自动诊断

【作者】 熊敏；

【导师】 刘雄飞；

【作者基本信息】 中南大学 ， 电子科学与技术， 2011， 硕士

【摘要】 冠心病是当今危害人类健康的主要心血管疾病之一,动态心电图是目前临床上诊断心血管疾病的重要手段,利用人工智能技术对心电信号进行准确的分析一直是国内外研究的热点,而ST段形态的准确分类又是其中的一项重要技术,对提高冠心病自动诊断系统的性能起关键性作用。本文采用更为精确的心电信号检测定位方法,结合自适应神经模糊推理系统建立数学图形分类模型,并对模型进行改进,准确地实现了对冠心病病发初期病理特征的自动诊断,主要的创新性工作如下：1、利用多孔算法的小波变换对于不规则离散信号采样点的无抽取平移不变性和三次B样条函数的高阶平滑特性,把三次B样条小波嵌入到多孔算法的小波变换中,对QRS波群的特征点进行精确的检测定位,通过实验仿真数据验证,该方法的准确率达到了99.8%；2、根据ST段各种形态的数学特征,本文定义四个参数：曲线类型参数d,偏移电平参数c,直线倾斜方向参数κ,曲线凹凸方向参数p,然后采用自适应神经模糊推理系统建立数学图形分类模型,通过对系统输入参数的判断,并对判断结果进行组合,完成对ST段的形态判别,通过实验仿真数据验证,该方法的准确率达到了92%以上：3、根据上述定义的四个参数,结合冠心病病发初期的病理特征,对冠心病病发初期的自动诊断过程进行建模,然后采用共轭梯度法对该系统进行改进,通过对每次反向调整运算的权重矢量的大小和方向的计算,来确定权重的最优值,从而提高系统的运算速度和收敛速度,也提高了冠心病病发初期自动诊断的准确率和精确性。更多还原

【Abstract】 Nowadays, Coronary Heart Disease (CHD) is one of main cardiovascular diseases which are harmful for human health. Dynamic electrocardiogram is an currently important method of clinical diagnosis for cardiovascular disease. It is a popular research which use artificial intelligence technology to analyze ECG accurately at home and abroad. However, the accurate classification of the ST section is the most important technology of it, which plays a critical role to improve the performance of the automatic diagnosis system of CHD.In this paper, it adopted a more accurate localization method of ECG detection which used ANFIS (adaptive neuron-fuzzy unference system) to establish mathematical model of graphics classification, and then the improved model was built. The automatic diagnosis for pathologic characteristics of early CHD was realized. And, the main innovative works are shown as follows:1, Using no extraction translation invariance of wavelet transformation with ATrous for the irregularity discrete signal sampling points and the high order smooth characteristics of the cubic B-spline, the cubic B-spline wavelet was embedded into wavelet transform with ATrous to detect and position accurately for the characteristics of QRS wave points. The simulation results verified that this method accuracy can reach 99.8%.2, According to the mathematical characteristics of various forms in ST section, four parameters were defined:curve type parameter d, offset level parameter c, Linear sloping direction parameter k, and curve bump direction parameter p. The ANFIS was used to establish the mathematical graphics classification model, and then the input parameters was judged and combined with the judgments to complete the discrimination for the ST section’s forms. The simulation results proved that the method’s accuracy can reach 92% above.3, According to the above four parameters and pathologic characteristics of early CHD, the model for the automatic diagnosis was improved by the conjugate gradient method. The size and direction of the weight vector calculation in each reverse was worked out to adjust operation and determine the optimal value of the weights, and the operation speed and the convergence speed of the system were optimized. Finally, the accuracy in automatic diagnosis of early CHD was improved.更多还原