Pulse waveform classification using support vector machine with Gaussian time warp edit distance kernel.

Area of Science:

• Biomedical Engineering
• Traditional Chinese Medicine
• Signal Processing

Background:

• Quantitative research in traditional Chinese pulse diagnosis benefits from signal processing.
• Automatic classification of pulse waveforms is challenging due to intraclass variations.

Purpose of the Study:

• To address the challenge of automatic pulse waveform classification.
• To propose a novel method using the Time Warp Edit Distance (TWED) metric within a Support Vector Machines (SVM) framework.

Main Methods:

• Utilized the elastic metric, Time Warp Edit Distance (TWED).
• Implemented a Gaussian TWED kernel function for Support Vector Machines (SVM).
• Evaluated the Gaussian TWED kernel SVM (GTWED-SVM) on a dataset of 2470 pulse waveforms across five patterns.

Main Results:

• The GTWED-SVM method demonstrated a lower average error rate compared to existing methods.
• Achieved improved accuracy in classifying distinct pulse waveform patterns.

Conclusions:

• The proposed GTWED-SVM method is effective for automatic pulse waveform classification.
• This approach offers a promising advancement for quantitative analysis in traditional Chinese pulse diagnosis.