Soft computing approach for pulse pattern classification system to objectification of Korean medicine diagnosis

Abstract

The radial pulse is a biosignal that can reflect physiological conditions or pathological problems of the human body by indicating various functions of the cardiovascular system. Korean medicine doctors (KMDs) make a pulse diagnosis on the basis of pulse patterns, which are categorized into 27 pulse types according to the pulse characteristics. However, there are no clear gold standards, and the diagnosis depends only on the subjective method of KMDs. In most previous studies, the pulse pattern diagnosed by KMDs has been regarded as the absolute truth, which could be a false assumption, and complex pulse patterns, which are mixtures of various pulse patterns, have not been considered. In addition, fragmented studies based on evaluations with only one analysis index or only waveform features have been performed, and only a few pulse patterns have been classified with small and unrefined datasets. The aim of this study is to develop a pulse pattern classification (PPC) system based on soft computing to objectification of Korean medicine. In particular, fuzzy theory, which is a method for studying uncertainty and difficult quantification issues, could be the most appropriate method for pulse pattern analysis. Expert knowledge and linguistic information can be applied to the fuzzy rules, and a membership grade for each pulse pattern can also be determined through fuzzy analysis. First, signal processing was performed for a large amount of pulse data, and analysis indices reflecting the rate, depth, strength, width, sharpness, tension and rhythm were developed to quantify the basic pulse properties. In addition, a new analysis index reflecting the pulse smoothness was developed using harmonic analysis based on the pulse morphology variability and was verified. Subsequently, various indices corresponding to each pulse pattern were clustered using unsupervised learning, and a membership function of the fuzzy system based on the clustering results was constructed. Fuzzy theory was applied through fuzzification, rule evaluation and defuzzification steps to classify the pulse patterns within predefined pulse groups. Finally, an integrated PPC system based on soft computing was developed, and when pulse data were imported to the system, the degree of membership for various pulse patterns was calculated, and information about complex pulse patterns was obtained. In addition, a mathematically robust model reflecting a nonlinear relationship was constructed and validated. This study can contribute to the quantification and objectification of Korean medicine diagnoses.