Techniques for estimating skin tone using a machine learning model

Patent metadata

Field	Value
Publication number	US-12502123-B2
Application number	US-202318320844-A
Country	US
Kind code	B2
Filing date	May 19, 2023
Priority date	May 19, 2023
Publication date	Dec 23, 2025
Grant date	Dec 23, 2025

Abstract

Methods, systems, and devices for estimating skin tone using a machine learning (ML) model are described. A wearable device may transmit light using a light-emitting component, and generate a signal based on the light received at a photodetector. In some examples, a strength of the received signal may be maintained within a signal strength band. The wearable device may identify a data set pair based on the signal, the data set pair including the signal strength band and a transmit power parameter that corresponds to the signal strength band. The data set pair may be inputted into an ML model and the ML model may output a skin tone metric for the user. The estimated skin tone metric may then be used to adjust measurement parameters used by the wearable device to improve a quality of physiological data, or validate algorithms used by the wearable device across skin tones.

First claim

What is claimed is:

1 . A method for estimating skin tone associated with a user, comprising:
transmitting light associated with a first wavelength using one or more light-emitting components of a wearable device associated with the user; generating a signal based at least in part on the light received at one or more photodetectors of the wearable device, wherein a strength of the signal received at the one or more photodetectors is maintained within one or more signal strength bands throughout one or more time intervals; identifying one or more data set pairs based at least in part on the signal, wherein the one or more data set pairs comprise a signal strength band of the one or more signal strength bands and a transmit power parameter of the one or more light-emitting components that corresponds to the signal strength band during a respective time interval of the one or more time intervals; inputting the one or more data set pairs into a machine learning model; and determining, using the machine learning model, a skin tone metric associated with the user based at least in part on the one or more data set pairs.

2 . The method of claim 1 , further comprising:
transmitting second light associated with a second wavelength using the one or more light-emitting components of the wearable device; generating a second signal based at least in part on the second light received at the one or more photodetectors, wherein a strength of the second signal received at the one or more photodetectors is maintained within one or more additional signal strength bands throughout one or more additional time intervals; identifying one or more additional data set pairs based at least in part on the second signal, wherein the one or more additional data set pairs comprise a second signal strength band of the one or more additional signal strength bands and a second transmit power parameter of the one or more light-emitting components that corresponds to the second signal strength band during a respective additional time interval of the one or more additional time intervals; and inputting the one or more additional data set pairs into the machine learning model, wherein the skin tone metric is determined using the machine learning model based at least in part on the one or more additional data set pairs.

3 . The method of claim 1 , wherein the one or more data set pairs comprise a first data set pair associated with a first time interval and a second data set pair associated with a second time interval, the first data set pair comprises a first signal strength band of the signal during the first time interval and a first transmit power parameter used during the first time interval, and the second data set pair comprises a second signal strength band of the signal during the second time interval and a second transmit power parameter used during the second time interval.

4 . The method of claim 1 , further comprising:
inputting, to the machine learning model, one or more additional data set pairs associated with one or more additional users, wherein the one or more additional data set pairs comprise pairs of additional signal strength bands and corresponding additional transmit power parameters associated with additional wearable devices of the one or more additional users, wherein the skin tone metric is determined using the machine learning model based at least in part on the one or more additional data set pairs.

5 . The method of claim 1 , further comprising:
determining one or more measurement parameters associated with the user, the wearable device, or both, based at least in part on the skin tone metric; and acquiring physiological data from the user via the wearable device based at least in part on the one or more measurement parameters.

6 . The method of claim 5 , wherein the one or more measurement parameters comprise a power level associated with the one or more light-emitting components, a burn time associated with the one or more light-emitting components, an algorithm for analyzing the light received by the one or more photodetectors, or any combination thereof.

7 . The method of claim 1 , further comprising:
selectively adjusting one or more transmit power parameters of the one or more light-emitting components to maintain the strength of the signal received at the one or more photodetectors within the one or more signal strength bands, wherein identifying the one or more data set pairs is based at least in part on selectively adjusting the one or more transmit power parameters.

8 . The method of claim 1 , further comprising:
acquiring physiological data from the user via the wearable device, the physiological data comprising motion data; and identifying a satisfaction of a trigger condition for determining the skin tone metric based at least in part on the motion data failing to satisfy a threshold level of motion, wherein transmitting the light, generating the signal, identifying the one or more data set pairs, and determining the skin tone metric using the machine learning model are initiated based at least in part on identifying the satisfaction of the trigger condition.

9 . The method of claim 8 , further comprising:
identifying the satisfaction of the trigger condition based at least in part on the motion data failing to satisfy the threshold level of motion during a time period that the user is asleep, during a time period that an ambient light level is less than some threshold level of ambient light, or both.

10 . The method of claim 1 , further comprising:
inputting, to the machine learning model, an indication of one or more distances between the one or more light-emitting components and the one or more photodetectors, wherein determining the skin tone metric is based at least in part on the indication of the one or more distances.

11 . The method of claim 10 , wherein the indication of the one or more distances comprises a size of the wearable device.

12 . The method of claim 1 , wherein the transmit power parameter of the one or more data set pairs comprises one or more statistical parameters associated with a current that is provided to the one or more light-emitting components to generate the corresponding signal strength band.

13 . The method of claim 1 , further comprising:
determining a baseline skin tone metric associated with the user based at least in part on additional light transmitted by the one or more light-emitting components and received by the one or more photodetectors; and determining an orientation of the wearable device relative to the user based at least in part on a comparison between the skin tone metric and the baseline skin tone metric.

14 . The method of claim 13 , further comprising:
selectively adjusting an activation state of one or more sensors associated with the wearable device based at least in part on the orientation, wherein the one or more sensors comprise the one or more light-emitting components, the one or more photodetectors, additional sensors, or any combination thereof; and acquiring physiological data associated with the user via the wearable device based at least in part on selectively adjusting the activation state of the one or more sensors.

15 . The method of claim 13 , further comprising:
generating, via the wearable device, a user device associated with the wearable device, or both, an instruction for the user to adjust the orientation of the wearable device.

16 . The method of claim 1 , wherein the skin tone metric comprises a Fitzpatrick scale metric.

17 . The metho

Assignees

• Oura Health Oy

Inventors

• Zhang Xi
• Noujaim Cristina
• Thigpen Nina

Classifications

• A61B5/6826 Primary

  Finger · CPC title

• A61B5/0059

  using light, e.g. diagnosis by transillumination, diascopy, fluorescence (photoacoustic A61B5/0093; optical measurement of heart rate A61B5/02416; optical measurement of blood flow A61B5/0261; optical measurement of analytes A61B5/1455) · CPC title

• A61B5/7267

  involving training the classification device · CPC title

• A61B5/441 Primary

  Skin evaluation, e.g. for skin disorder diagnosis · CPC title