Using high-dimensional features for high-accuracy pulse diagnosis.

Area of Science:

• Biomedical Engineering
• Traditional Chinese Medicine
• Data Science

Background:

• Accurate pulse diagnosis traditionally relies on extensive clinical experience.
• Computer-aided methods offer faster physiological condition assessment but often use low-dimensional features, potentially losing subtle information.
• Existing methods may overlook crucial details in pulse waveform analysis.

Purpose of the Study:

• To develop a novel high-dimensional pulse classification method for improved diagnostic accuracy.
• To retain and analyze subtle pulse information often missed by low-dimensional approaches.
• To enhance the precision of computer-aided pulse diagnosis.

Main Methods:

• Extraction of 71 high-dimensional pulse features from time, spatial, and frequency domains.
• Application of Principal Component Analysis (PCA) for dimensionality reduction and feature selection.
• Training artificial neural networks for the classification of 10 distinct pulse types.

Main Results:

• Principal Component Analysis (PCA) captured 95% of the total variance in pulse data.
• The high-dimensional classification method achieved a peak accuracy of 98.2%.
• Key discriminators identified include pulse energy, local characteristics, main frequency, and waveform complexity.

Conclusions:

• High-dimensional feature analysis retains more comprehensive pulse information compared to traditional low-dimensional methods.
• The developed method significantly improves the accuracy of computer-aided pulse diagnosis.
• This approach offers a more robust tool for objective pulse assessment in clinical settings.