Estimation of respiratory rate from principal components of photoplethysmographic signals

Abstract

Continuous monitoring of respiratory activity is mandatory in clinical, high risk situations such as ambulatory monitoring, intensive care, stress tests and sleep disorder investigations. Extraction of surrogate respiratory activity from electrocardiogram (ECG), blood pressure (BP) and photoplethysmographic (PPG) signals will potentially eliminate the use of additional sensor intended to record respiration. Principal Component Analysis (PCA) is a simple and standard non-parametric mathematical tool for extracting relevant information from complex data sets. In this paper, PCA is exploited for extraction of surrogate respiratory activity from PPG signals. The respiratory induced intensity variations (RIIV) of PPG signal are described by coefficients of computed principal components. Singular value ratio (SVR) trend is used to find the periodicity, which is one of the key parameters in forming the data sets for PCA. Test results on MIMIC data base clearly indicated a strong correlation between the extracted and actual respiratory signals. The evaluated similarity measures, both in time (RCC-Relative Correlation Coefficient) and frequency (MSC-Magnitude Squared Coherence) domains and calculated accuracy demonstrated the fact that respiratory signal is present in the form of first principal component.