Wrist Pulse Waveform Feature Extraction and Dimension Reduction with Feature Variability Analysis

Time-domain feature analysis on wrist pulse waveform is common in Traditional Chinese Medical (TCM) engineering and diagnosis modernization. A derivative-based method on the automated time-domain feature extraction of wrist pulse waveform is proposed in this paper with the consideration of some practical issues. Variability analysis is performed on the features extracted from the pulse waveform trends. The dimension of pattern, i.e. vector of features, is then reduced by the cross-correlation analysis on the variability of features. A real classification case, with dataset including pulse waveform from 20 healthy person and 50 persons with cardiovascular disease, is performed based on pattern containing both reduced features and other combinations. Comparison results show that features are properly selected and the classification performance is acceptable.