A Study of Machine-Learning Classifiers for Hypertension Based on Radial Pulse Wave.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Cardiovascular Diagnostics

Background:

• Hypertension diagnosis traditionally relies on subjective measures.
• Pulse diagnosis in Traditional Chinese Medicine (TCM) offers dynamic insights but lacks objective quantification.
• Developing objective, data-driven methods for hypertension assessment is crucial.

Purpose of the Study:

• To utilize machine learning for classifying hypertensive and healthy individuals based on pulse wave data.
• To assess hypertension risk by analyzing dynamic changes in pulse wave signals.
• To provide an objective reference for the clinical application of TCM pulse diagnosis.

Main Methods:

• Collected data from 450 hypertensive and 479 healthy cases, including self-reported information and pulse wave data from a custom instrument (PDA-1).
• Developed machine learning models using questionnaire and pulse wave data as inputs.
• Applied K-means clustering to remove noise and analyze the impact on model accuracy, stability, and feature importance.

Main Results:

• Machine learning models demonstrated improved accuracy and Area Under the Curve (AUC) after noise reduction.
• Random Forest achieved 85.33% accuracy and 0.85 AUC; AdaBoost and Gradient Boosting reached 86.41% accuracy and 0.86 AUC.
• Support Vector Machine (SVM) accuracy increased from 79.57% to 83.15%, with AUC improving from 0.79 to 0.83.

Conclusions:

• Pulse wave-based analysis provides a valuable, objective method for hypertension diagnosis.
• Digital pulse wave diagnosis is feasible for dynamic hypertension evaluation.
• This research lays a foundation for TCM in the dynamic assessment of modern diseases and treatment efficacy.