Period Segmentation for Wrist Pulse Signal Based on Adaptive Cascade Thresholding and Machine Learning

Abstract

Wrist pulse signal has been regarded as a physical health indicator for a long history in Traditional Chinese Medicine (TCM). The quantized pulse diagnosis by using the signal processing and pattern recognition technology is introduced to take over the traditional subjective judgments in recent years, and it's attracting more and more attention. However, the previous researches with pulse pre-processing mainly concentrate on the denoising and baseline wander correction procedure. The evaluation criterion isn't associated with the feature analysis, and the performance with shape classification doesn't give any contributions to the pulse diagnosis. Moreover, the signals are processed in a simulated environment by adding disturbance manually. In this paper, we propose a period segmentation method based on adaptive cascade thresholding and machine learning for extracting the information within single period. It's a novel pre-processing stage and the pulse data collected in real conditions for practical usage is analyzed. The experiments show that our method is significant in the pulse pre-processing stage and improves the accuracy for the disease classification between healthy subjects and diabetes.