脉搏波的频域特征提取与自动识别技术

Spectrum Feature Extraction and Automatic Recognition of Pulse Waves

脉搏波的频谱蕴含丰富的病理信息,但其复杂的频谱计算和分类是临床应用的瓶颈之一.本文引入模式识别技术,建立了心血管疾病的自动识别专家系统,为脉搏波频谱分析在临床中的应用开辟了新的研究思路.首先采用小波变换在多分辨率层次上提取脉搏波的频域特征,不仅获得了各个频带的谱能分量,而且得到了频谱分布参数小波熵;然后采用贝叶斯判别分析法建立自动识别模型,对频域特征进行分类.临床采集了30例冠心病人和30例正常人的脉搏波信号,对识别模型进行了训练,最后对模型进行了交互验证.结果表明,该识别模型对冠心病人的识别准确率为83.3%,对正常人的识别准确率为70.0%.该方法具有较好的识别效果,为脉搏波自动识别技术的发展提供了借鉴.

The spectrum of pulse waves contains abundant pathological information, yet its complicated frequency-domain calculation and taxonomy are one of the bottlenecks in clinic application. In this paper, pattern recognition technique was introduced and an automatic recognition system of cardiovascular diseases was established, which provides a new research approach for pulse wave frequency-domain analysis in clinic application. First, wavelet transform was used to extract spectrum features of pulse waves, including the spectrum energy of each frequency component and the complexity parameter of spectrum distribution, i.e. the wavelet entropy. Then the automatic recognition model was built up based on Bayes discriminant analysis to classify spectrum features. The pulse waves of 30 normal subjects and 30 subjects with coronary disease were collected to train the recognition model, which was then evaluated by crossvalidation. The results show that the correct recognition rate of the model can reach 83.3% for patients with cardiovascular diseases and 70. 0% for the normal. So the proposed model that integrates spectrum analysis and pattern recognition is satisfactory in recognition of cardiovascular diseases and has supplied insight into the automatic recognition technique of pulse waves.