WristO2: Reliable Peripheral Oxygen Saturation Readings from Wrist-Worn Pulse Oximeters

Caleb Phillips, Daniyal Liaqat, Moshe Gabel, Eyal de Lara

Abstract: Peripheral blood oxygen saturation (SpO <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sub> ) is a vital health signal with many clinical applications. Modern wrist-worn devices, such as the Apple Watch, FitBit, and Samsung Gear, have pulse oximeter sensors, making them theoretically capable of measuring SpO <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sub> . However, current techniques for SpO <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sub> measurements using pulse oximeter sensors are based on readings taken from the fingertip. Readings collected from the wrist are unreliable and often inaccurate, due to motion and insufficient skin contact. Enabling accurate oxygen saturation monitoring on wearable devices would allow continuous health monitoring and open up new avenues of research. In this work, we explore the reliability of SpO <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sub> measurements from the wrist. Using a custom wrist-worn pulse oximeter, we find that existing algorithms used in traditional fingertip SpO <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sub> sensors are a poor match for taking measurements from the wrist and can lead to over 90% of readings being inaccurate. We further show that skin tone, IMU sensors, and user-level calibration affect measurement error, and must be considered when designing wrist-worn SpO <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sub> sensors and measurement algorithms. Next, based on our findings, we propose WristO <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sub> , an alternative approach for reliable SpO <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sub> sensing. By selectively pruning unreliable data, WristO <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sub> achieves an order of magnitude reduction in error compared to existing algorithms, while still providing sufficiently frequent readings for continuous health monitoring.