Linear and Non-linear Quantification of the Respiratory Sinus Arrhythmia Using Support Vector Machines

Area of Science:

• Cardiology
• Physiology
• Biomedical Engineering

Background:

• Respiratory sinus arrhythmia (RSA) represents cardiorespiratory coupling, with heart rate changes synchronizing with respiration.
• RSA is theorized to involve both linear and nonlinear influences.
• Quantifying nonlinear RSA components may offer improved diagnostic biomarkers for various health conditions.

Purpose of the Study:

• To present a novel framework for quantifying RSA using support vector machines (SVM).
• To differentiate between linear and nonlinear components of cardiorespiratory coupling.
• To assess the framework's efficacy in simulated and real-world clinical data.

Main Methods:

• Utilized multivariate autoregressive models to predict heart rate variability from respiration.
• Employed SVM with kernel selection to isolate linear and nonlinear cardiorespiratory interactions.
• Validated the approach on simulated data and a polysomnography dataset from 110 obstructive sleep apnea patients.

Main Results:

• Simulations demonstrated successful capture of nonlinear components under weak cardiorespiratory coupling.
• Increased coupling strength in simulations led to the identification of predominantly linear interactions.
• Analysis of obstructive sleep apnea patient data indicated a stronger linear than nonlinear cardiorespiratory interaction.

Conclusions:

• The proposed SVM-based framework effectively quantifies RSA, distinguishing linear and nonlinear contributions.
• Cardiorespiratory coupling in obstructive sleep apnea patients appears predominantly linear.
• This method provides a tool for assessing cardiorespiratory dynamics and their clinical relevance.