Self-calibrated pulse oximetry algorithm based on photon pathlength change and the application in human freedivers.

Abstract

Pulse oximetry estimates the arterial oxygen saturation of hemoglobin () based on relative changes in light intensity at the cardiac frequency. Commercial pulse oximeters require empirical calibration on healthy volunteers, resulting in limited accuracy at low oxygen levels. An accurate, self-calibrated method for estimating is needed to improve patient monitoring and diagnosis. Given the challenges of calibration at low levels, we pursued the creation of a self-calibrated algorithm that can effectively estimate across its full range. Our primary objective was to design and validate our calibration-free method using data collected from human subjects. We developed an algorithm based on diffuse optical spectroscopy measurements of cardiac pulses and the modified Beer-Lambert law (mBLL). Recognizing that the photon mean pathlength () varies with related absorption changes, our algorithm aligns/fits the normalized (across wavelengths) obtained from optical measurements with its analytical representation. We tested the algorithm with human freedivers performing breath-hold dives. A continuous-wave near-infrared spectroscopy probe was attached to their foreheads, and an arterial cannula was inserted in the radial artery to collect arterial blood samples at different stages of the dive. These samples provided ground-truth via a blood gas analyzer, enabling us to evaluate the accuracy of estimation derived from the NIRS measurement using our self-calibrated algorithm. The self-calibrated algorithm significantly outperformed the conventional method (mBLL with a constant ratio) for estimation through the diving period. Analyzing 23 ground-truth data points ranging from 41% to 100%, the average absolute difference between the estimated and the ground truth is for our algorithm, significantly lower than the observed with the conventional approach. By factoring in the variations in the spectral shape of relative to , our self-calibrated algorithm enables accurate estimation, even in subjects with low levels.