Quantifying blood-oxygen saturation measurement error in motion contaminated pulse oximetry signals

Abstract

Oxygen saturation measurements from pulse oximetry (SpO2) can be unreliable in the presence of motion artifacts. While pulse oximetry is a crucial measurement in controlled environments, such as surgery or intensive care, its vulnerability to motion artifacts has slowed its adoption in wearable continuous monitoring devices. Measurement error can cause errors or delays in clinical decision-making. In remote monitoring applications, pulse oximeters should report measurement confidence along with SpO2 to help clinicians make decisions about the validity of alarm conditions. This paper seeks to relate signal quality to SpO2 measurement confidence. In this study, clean photoplethysmograph (PPG) signals were collected from a pulse oximeter and contaminated with motion artifact. A range of linear combinations of signal and artifact were generated and SpO2 measurements were calculated. Since true SpO2 remained constant, measurement variation was caused solely by signal contamination. Unacceptably high measurement error was found below the 15-20 dB signal to noise ratio (SNR) range. Two models based on Additive White Gaussian Noise (AWGN) were evaluated for their similarity to the motion artifact data. The first had identical noise on both red and infrared PPG signals; the second has uncorrelated noise. Both models successfully predicted negative measurement bias at low SNR, but only the second predicted the observed measurement variance.