New pulse oximeter adapts to skin color

For years, pulse oximeters have been a common medical tool, providing instant readings of oxygen levels in the blood. These devices function by emitting red and infrared light through or from the skin, with oxygen-rich and oxygen-poor hemoglobin absorbing these wavelengths differently. Advanced software then converts this varying light signal into a measurement of a patient's oxygen saturation.

However, factors such as skin pigmentation, blood flow, and age can impact how well these devices perform, potentially leading to significant errors in oxygen saturation readings, which can be crucial in determining the need for oxygen, hospitalization, or intensive care.

Early in the COVID-19 pandemic, it became apparent that pulse oximeters may not accurately reflect oxygen levels in all patients, particularly those with darker skin. A 2020 study published in the New England Journal of Medicine highlighted that up to 17% of Black patients were misread by pulse oximeters, a rate more than three times higher than white patients.

This prompted the development of new technology aimed at addressing this disparity. The ChromaSense device, created by Valencia Koomson and her team from the electrical and computer engineering department, is designed to measure blood oxygen saturation, heart rate, and respiration rate across a wide range of skin tones.

Unlike conventional pulse oximeters that emit light through the finger, ChromaSense uses a wrist-worn box to measure reflected light from the skin and tissue. The device first measures the user's skin reflectance profile, then adjusts both the emitted light level and signal-processing parameters accordingly. In initial trials involving 50 participants with diverse skin tones, ChromaSense demonstrated an oxygen-saturation measurement accuracy within 1.4% of a standard reference oximeter, meeting FDA performance requirements.

The device was further tested under conditions of varying oxygen levels at the Hypoxia Research Laboratory at the University of California, San Francisco. Under these more stringent conditions, ChromaSense maintained an accuracy within 2.87% of a standard reference oximeter, without any observable bias related to skin tone. This performance was consistent across volunteers with Black, Asian, Hispanic, white, and multiethnic backgrounds.

The technology underpinning devices like ChromaSense is photoplethysmography (PPG), which relies on the principle that blood volume in microvascular blood vessels fluctuates with each heartbeat. When light illuminates tissue, the reflected light pulses in a pattern that reflects arterial blood flow. In pulse oximetry, the intensity of the reflected light at different wavelengths—specifically red and infrared—is measured to infer levels of oxygenated and deoxygenated hemoglobin.

However, melanin, a skin pigment, also absorbs and scatters light, which can weaken the signal or skew the ratio used to calculate oxygen saturation in darker-skinned individuals.

Koomson and her team are exploring the addition of blood pressure monitoring to ChromaSense. They have developed machine-learning models to estimate systolic and diastolic blood pressure from PPG waveforms using data from large healthcare databases. The models account for variations in skin tone, age, and gender by analyzing the waveform's characteristics, such as the width of different parts, to assess arterial stiffness.

In a study involving 2,315 adult ICU patients, the team reported high accuracy rates (up to 90%) in systolic and diastolic blood pressure estimations across various demographic groups. However, this blood pressure monitoring feature is not yet integrated into ChromaSense. Koomson emphasizes the importance of ensuring diverse datasets in training machine-learning models to avoid performance discrepancies among different demographic groups.