Multivariate classification of systemic vascular resistance using photoplethysmography

Area of Science:

• Cardiovascular Physiology
• Medical Informatics
• Biomedical Engineering

Background:

• Systemic vascular resistance (SVR) classification is crucial for diagnosing and prognosing critical conditions.
• Identifying patients with abnormally high or low SVR is of significant clinical value.

Purpose of the Study:

• To develop and evaluate a supervised classifier for SVR categorization using photoplethysmogram (PPG) waveform features.
• To assess the feasibility of non-invasive SVR classification in intensive care unit (ICU) patients.

Main Methods:

• Employed a Bayes' rule-based supervised classifier on 48 ICU patients.
• Utilized features from finger PPG waveforms, heart rate, and mean arterial pressure.
• Compared Gaussian distribution and Parzen window kernel density estimation models, optimizing with Cohen's kappa coefficient.

Main Results:

• The Gaussian model achieved the highest kappa coefficient (κ = 0.57), outperforming the non-parametric model (κ = 0.51).
• Optimal features included PPG variability (LF/HF, MF(NU)) and pulse wave characteristics (pulse width, peak-to-notch time, reflection index, notch time ratio).
• The classifier demonstrated high performance in discriminating low SVR (sensitivity 85%, specificity 86%).

Conclusions:

• A multivariate statistical approach using non-invasive PPG signals is feasible for SVR classification in clinical settings.
• This method offers a simple and accessible way to monitor SVR in critically ill patients.
• The findings support the potential for integrating PPG-based SVR analysis into routine ICU monitoring.