Peripheral arterial disease classification using machine learning and multi-point photoplethysmography.

Area of Science:

• Biomedical Engineering
• Medical Informatics
• Cardiovascular Diagnostics

Background:

• Peripheral arterial disease (PAD) diagnosis can be challenging.
• Photoplethysmography (PPG) signals offer a non-invasive method for vascular assessment.
• Machine learning (ML) holds potential for improving diagnostic accuracy in PAD.

Purpose of the Study:

• To classify peripheral arterial disease (PAD) using machine learning (ML) based on single-point and multi-point photoplethysmography (PPG) signals.
• To identify the most informative feature categories for PAD classification.
• To evaluate if multi-point PPG enhances diagnostic accuracy by capturing segmental vascular differences.

Main Methods:

• Collected PPG data from 60 subjects at 12 anatomical sites on both feet.
• Analyzed single-point (36 features) and multi-point (336 features) data, including clinical measures, pulse rate variability (PRV), time-domain, frequency-domain, and APG ratios.
• Applied four ML classifiers (Random Forest, XGBoost, SVM, Logistic Regression) and permutation importance analysis.

Main Results:

• The multi-point XGBoost model achieved the highest test accuracy at 91.67%.
• Single-point analysis consistently yielded lower accuracy across all tested classifiers.
• Multi-point analysis highlighted time-domain and PRV features as dominant contributors to PAD classification accuracy.

Conclusions:

• Multi-point PPG analysis significantly enhances PAD detection through segmental vascular assessment.
• This approach captures features related to systemic hemodynamic and autonomic function, improving classification.
• Integrating spatially distributed PPG signals with interpretable ML, like XGBoost, offers a promising tool for early PAD screening and intervention.