Wrist02 -- Reliable Peripheral Oxygen Saturation Readings from Wrist-Worn Pulse Oximeters

TL;DR

This study investigates the reliability of wrist-worn pulse oximeters for measuring blood oxygen saturation, revealing current algorithms are inadequate and proposing a new method that significantly improves accuracy for health monitoring.

Contribution

The paper presents the first analysis of wrist-based Sp02 sensing reliability and introduces extbackslash codename, a novel algorithm that enhances measurement accuracy over existing methods.

Findings

Existing algorithms are ineffective for wrist Sp02 sensing, with over 90% inaccurate readings.

Sensor placement and skin tone significantly affect measurement accuracy.

extbackslash codename reduces error by an order of magnitude while maintaining measurement frequency.

Abstract

Peripheral blood oxygen saturation Sp02 is a vital measure in healthcare. Modern off-the-shelf wrist-worn devices, such as the Apple Watch, FitBit, and Samsung Gear, have an onboard sensor called a pulse oximeter. While pulse oximeters are capable of measuring both Sp02 and heart rate, current wrist-worn devices use them only to determine heart rate, as Sp02 measurements collected from the wrist are believed to be inaccurate. Enabling oxygen saturation monitoring on wearable devices would make these devices tremendously more useful for health monitoring and open up new avenues of research. To the best of our knowledge, we present the first study of the reliability of Sp02 sensing from the wrist. Using a custom-built wrist-worn pulse oximeter, we find that existing algorithms designed for fingertip sensing are a poor match for this setting, and can lead to over 90% of readings being inaccurate and unusable. We further show that sensor placement and skin tone have a substantial effect on the measurement error, and must be considered when designing wrist-worn Sp02 sensors and measurement algorithms. Based on our findings, we propose \codename, an alternative approach for reliable Sp02 sensing. By selectively pruning data, \codename achieves an order of magnitude reduction in error compared to existing algorithms, while still providing sufficiently frequent readings for continuous health monitoring.