中医脉诊信号的无监督聚类分析研究

Unsupervised clustering analysis of human-pulse signal in traditional Chinese medicine

随着中医客观化工作的推进,脉诊技术也越来越走向客观化和仪器化。然而,如何对仪器所检测和收集到的信息进行解读,却还是回到了原来脉诊诊断主观化的问题上。因为传统的机器学习方法,依赖于对大量的脉诊数据进行标注。但是在临床诊断和教学中,医生与医生之间对于脉象的体会不同,会导致他们对病人脉象的区分标注不同。在对比了多种特征提取方法和聚类方案之后,提出了一个较好的无监督脉诊客观化方法,在双树复小波变换(DTCWT)对数据进行预处理的基础上,以梅尔倒谱系数(MFCC)进行特征提取,在中医专家对数据进行标注之前,先根据信号的特征,使用Fuzzy c-means(FCM)聚类算法进行粗线条的分类,使得在此基础之上,可以开展进一步的细化分类研究。实验结果表明:该方法可取得较好的分类效果,为中医脉诊提供了进一步客观化的依据。

With the development of a more objective basis for traditional Chinese medicine （TCM）, objectivity and in-strumentation are growing trends in pulse-taking techniques. However, choosing an objective method for interpreting thedata collected by newly developed TCM diagnostic machines is a recurring issue in the move toward objective pulse-taking diagnosis. Traditional machine learning methods rely heavily on annotated pulse-diagnosis data; however, inTCM practice, different doctors make different annotations based on their different experiences in pulse manifestation.After comparing various feature extraction methods and clustering schemes, in this paper, we propose an improved un-supervised human-pulse identification approach. In this method, we use the dual-tree complex wavelet transform（DTCWT） to preprocess data and Mel-frequency cepstral coefficients （MFCCs） to extract features. Before the data areannotated by TCM experts, we applied the fuzzy c-means （FCM） clustering algorithm to the signal features to classifythick lines, after which further detailed classifications can be made. The experimental results show that excellent classi-fication effects can be obtained by this method, which provides an objective basis for TCM pulse diagnosis.