Edit Distance for Pulse Diagnosis
In this chapter, by referring to the edit distance with real penalty (ERP) and the recent progress in k-nearest neighbors (KNN) classifiers, we propose two novel ERP-based KNN classifiers. Taking advantage of the metric property of ERP, we first develop a
- PDF / 620,943 Bytes
- 14 Pages / 439.37 x 666.142 pts Page_size
- 81 Downloads / 218 Views
Edit Distance for Pulse Diagnosis
Abstract In this chapter, by referring to the edit distance with real penalty (ERP) and the recent progress in k-nearest neighbors (KNN) classifiers, we propose two novel ERP-based KNN classifiers. Taking advantage of the metric property of ERP, we first develop an ERP-induced inner product and a Gaussian ERP kernel, then embed them into difference-weighted KNN classifiers, and finally develop two novel classifiers for pulse waveform classification. The experimental results show that the proposed classifiers are effective for accurate classification of pulse waveform.
11.1 Introduction Traditional Chinese pulse diagnosis (TCPD) is a convenient, noninvasive, and effective diagnostic method that is widely used in traditional Chinese medicine (TCM) [1]. In TCPD, practitioners feel for the fluctuations in the radial pulse at the styloid processes of the wrist and classify them into the distinct patterns which are related to various syndromes and diseases in TCM. This is a skill which requires considerable training and experience and may produce significant variation in diagnosis results for different practitioners. So in recent years, techniques developed for measuring, processing, and analyzing the physiological signals [2, 3] have been considered in quantitative TCPD research as a way to improve the reliability and consistency of diagnoses [4–6]. Since then, much progress has been received: a range of pulse signal acquisition systems have been developed for various pulse analysis tasks [7–9]; a number of signal preprocessing and analysis methods have been developed in pulse signal denoising, baseline rectification [10], and segmentation [11]; many pulse feature extraction approaches have been proposed by using various time-frequency analysis techniques [12–14]; many classification methods have been studied for pulse diagnosis [15, 16] and pulse waveform classification [17–19]. Pulse waveform classification aims to assign a traditional pulse pattern to a pulse waveform according to its shape, regularity, force, and rhythm [1]. However, because of the complicated intra-class variation in pulse patterns and the inevitable © Springer Nature Singapore Pte Ltd. 2018 D. Zhang et al., Computational Pulse Signal Analysis, https://doi.org/10.1007/978-981-10-4044-3_11
217
218
11 Edit Distance for Pulse Diagnosis
influence of local time shifting in pulse waveforms, it has remained a difficult problem for automatic pulse waveform classification. Although researchers have developed several pulse waveform classification methods such as artificial neural network [18, 21, 22], decision tree [20], and wavelet network [23], most of them are only tested on small datasets and usually cannot achieve satisfactory classification accuracy. Recently, various time series matching methods, e.g., dynamical time warping (DTW) [24] and edit distance with real penalty (ERP) [25], have been applied for time series classification. Motivated by the success of time series matching techniques, we suggest util
Data Loading...
