Are PPG sensors less accurate for people with darker skin?

By Dr. Steven LeBoeuf

From time-to-time in the wearable world, the blogosphere is jolted by claims that wearable optical sensors do not work on people with dark skin tones. It reached a new level when a New England Journal of Medicine article prompted the FDA to issue a “safety communication” that pulse oximeters may be less accurate in people with darker skin pigmentation and high melanin counts.

These concerns have been posted in reference to a diverse array of optical sensing methods, such as traditional photoplethysmography, pulse oximetry, and biometric imaging (machine vision). The stated concern is that, through some fundamental mechanism, dark skin tones pose a limitation on optical biometrics sensing that is not being sufficiently addressed by the wearables industry. Or even more wildly, enamored by the irresistible temptation of click-bate, some bloggers have even proposed that wearables are “racist”!

I can say with full conviction that wearables are NOT racist, for multiple reasons that are outside the scope of this blog post. Nonetheless, it is true that extremely dark skin, a subject-specific characteristic, can pose a challenge for the accurate measurement of optically-derived biometrics through the skin.

Of course, subject-specific characteristics are not limited to optical sensors but rather affect all types of sensor modalities. Electrodes used in ECGs, for example, face an array subject-specific challenges, such as oily skin, sensitive (itchy) skin, branch bundle-block, pH-driven electrode corrosion, skin resistance variability, and the like. Similarly, auscultatory sensors are notoriously problematic for obese subjects.

In the face of these diverse subject-specific technical challenges, we often dwell on the potential weaknesses of a sensor modality rather than how these challenges have been, or can be, overcome. Since the first realization that high-melanin may cause limitations in pulse oximetry, roughly 30 years ago, researchers have developed several techniques to eliminate the issue:

  1. One approach has been to implement dynamic light intensity at the microprocessor level, such that darker skin will transmit the same average number of photons as lighter skin as blood pulsates through the blood vessels.
  2. Another approach has been to implement multiwavelength techniques to judge the melanin concentration within subjects and adjust algorithms accordingly.
  3. Yet another approach has been to design the optomechanics of the pulse oximeter to accommodate melanin-dependent optical scattering paths. Several other techniques have been either implemented or proposed by those in industry and academia.

Ironically, the proof that pulse oximeters can be designed to work equally well for all skin tones is presented in the very same article commonly cited in the case against optical sensors! This article, entitled “Dark Skin Decreases the Accuracy of Pulse Oximeters at Low Oxygen Saturation: The Effects of Oximeter Probe Type and Gender”, states that melanin bias exists in pulse oximetry, especially for those with dark skin who are experiencing hypoxia. Yet at the same time, the article definitively states that one particular pulse oximeter (a Nonin finger clip) was found to have a bias of only -0.60%, which was determined to NOT be “statistically significant”. While it is true that other devices tested (including a Nonin disposable pulse oximeter) did show a significant bias, the fact that one device did not only proves that pulse oximeters can be designed bias-free.

The moral of the story here is that wearable optical sensors can be designed to provide accurate assessments for all skin tones. While some device manufacturers may not be implementing best practices to assure their sensors work on all skin tones, many of them are. So when choosing a pulse oximeter (or any other device using optical sensors), be sure to note the brand, the type of sensor (e.g., clip-on or disposable), the device claims, and the specifications provided for dark skin.