Machine learning techniques for arterial pressure waveform analysis.

Area of Science:

• Biomedical Engineering
• Cardiovascular Physiology
• Machine Learning Applications

Background:

• Arterial Pressure Waveform (APW) analysis offers insights into arterial wall integrity and stiffness.
• Current APW analysis methods often process hemodynamic parameters individually, neglecting interdependencies in pulse morphology.
• A novel approach is needed to evaluate the holistic nature of APW for improved diagnostic accuracy.

Purpose of the Study:

• To develop and evaluate a machine learning framework for analyzing APW morphology.
• To assess the interdependencies of hemodynamic parameters within the APW using vectorized features.
• To compare the efficacy of machine learning algorithms against traditional single-parameter analysis, such as Augmentation Index (AIx).

Main Methods:

• Data acquisition using a custom non-invasive electromechanical device from 50 subjects.
• Extraction of morphological attributes from key APW features (onset, Systolic Peak, Point of Inflection, Dicrotic Wave).
• Application of four machine learning algorithms (Random Forest, BayesNet, J48, RIPPER) on pre-processed, vectorized features for classification.

Main Results:

• Random Forest achieved high classification accuracy (96.95%) and AUC (0.961) on the ROC curve.
• Validation tests revealed a correlation between high-risk labels from the multi-parametric approach and positive AIx values.
• The machine learning approach demonstrated superior performance compared to traditional single-parameter analysis.

Conclusions:

• Machine learning algorithms effectively analyze vectorized APW features for improved hemodynamic assessment.
• This multi-parametric approach enhances the understanding of arterial pulse dynamics and arterial stiffness.
• The proposed methodology offers a robust alternative to conventional APW analysis, mitigating risks associated with single-parameter failures.