Melanin bias in pulse oximetry explained by light source spectral bandwidth

Background: Pulse oximetry uses noninvasive optical measurements of light transmission from each of two sources through vascularised living tissue over the cardiac cycle (SpO2). From those measurements, the relative amount of oxygenated haemoglobin (SaO2) in circulating blood can be deduced. Recent reports have shown that, compared with SaO2 measurements from blood samples, SpO2 measurements are biased erroneously high for patients with dark skin. Methods: We developed a new method, spectrally resolved photoplethysmography (srPPG), to examine how spectral bandwidth affects the transmission of polychromatic light through the fingertip across the cardiac cycle. We measured and recorded the spectral transmission through the fingertip as the O2 concentration in inspired air was reduced. We applied digital spectral filters of two different bandwidths, narrow or broad, to the same srPPG recordings to determine whether SpO2 readings systematically varied for the two bandwidths. The srPPG method also allowed us to measure the fractional amount of melanin in the optical path. The effect of melanin content on the ratio of SpO2 readings for narrow and broad spectral bandwidths was analysed. Results: We hypothesised, based upon the Beer-Lambert law, and then showed experimentally, that the light emission spectra of light-emitting diode light sources, as used in commercial pulse oximeters, result in erroneously high SpO2 measurements for patients having greater melanin concentrations in their skin than those of the subject pool used for instrument calibration. Conclusions: To eliminate melanin bias, pulse oximeters should use much narrower spectral bandwidths than those used in current models.