Search

EP-4736759-A2 - APPLIED DATA QUALITY METRICS FOR PHYSIOLOGICAL MEASUREMENTS

EP4736759A2EP 4736759 A2EP4736759 A2EP 4736759A2EP-4736759-A2

Abstract

A model of data quality is derived for physiological monitoring with a wearable device by comparing data from the wearable device to concurrent data acquisition from a ground truth device such as a chest strap or electrocardiography (EKG) heart rate monitor. With this comparative data, a machine learning model or the like may be derived to prospectively evaluate data quality based on the data acquisition context, as determined, for example, by other sensor data and signals from the wearable device.

Inventors

  • CAPODILUPO, John, Vincenzo
  • TAVAKOLI, Behnoosh
  • GHANNAD-REZAIE, MOSTAFA

Assignees

  • Whoop, Inc.

Dates

Publication Date
20260506
Application Date
20181017

Claims (12)

  1. A method comprising: obtaining calibrated heart rate data from a number of subjects; obtaining uncalibrated heart rate data from the number of subjects concurrently with the calibrated heart rate data using one or more physiological monitors; obtaining feature data from one or more sensors of the one or more physiological monitors concurrently with the calibrated heart rate data and the uncalibrated heart rate data, wherein the feature data characterizes a data acquisition context for the uncalibrated heart rate data obtained from the one or more physiological monitors; associating a quality metric for the uncalibrated heart rate data with the feature data, wherein the quality metric indicates whether, for each measurement sample, the uncalibrated heart rate data is within a predetermined threshold of the calibrated heart rate data; creating a quality estimator engine used to evaluate a likelihood of the uncalibrated heart rate data from the one or more sensors of the one or more physiological monitors being accurate based on the feature data; receiving second uncalibrated heart rate data and second feature data from a second physiological monitor, wherein the second feature data characterizes the data acquisition context for the second uncalibrated heart rate data; determining a probability using the quality estimator engine that the second uncalibrated heart rate data is accurate for a window of measurements based on the second feature data over a distribution of values for the calibrated heart rate data within the window of measurements; and associating the probability with the window as a measure of quality for the uncalibrated heart rate data within the window.
  2. The method of claim 1 wherein the feature data includes at least one signal used to estimate a heart rate using the uncalibrated heart rate data.
  3. The method of claim 1 wherein the feature data includes at least one value derived from a signal from one of the one or more physiological monitors.
  4. The method of claim 1 wherein the feature data includes a signal derived from a motion sensor of one of the one or more physiological monitors .
  5. The method of claim 1 wherein the quality metric is a one when the uncalibrated heart rate data is within the predetermined threshold of the uncalibrated heart rate data and a zero when the uncalibrated heart rate data is not within the predetermined threshold of the uncalibrated heart rate data.
  6. The method of claim 1 wherein creating the quality estimator engine includes training a machine learning random forest to estimate the quality metric for a set of feature data.
  7. The method of claim 1 further comprising providing feedback to a user concerning an adjustment to the second physiological monitor based on the measure of quality.
  8. The method of claim 7 wherein the adjustment includes a change in a position of the second physiological monitor.
  9. The method of claim 7 wherein the adjustment includes a change in a tension of a band for the second physiological monitor.
  10. The method of claim 1 further comprising: characterizing a user of the second physiological monitor; identifying a subset of the number of subjects similar to the user; and associating the quality metric for the uncalibrated heart rate data with the feature data for the subset of the number of subjects similar to the user.
  11. A computer program product comprising computer executable code embodied in a computer readable medium that, when executing on a one or more computing devices, performs the steps of any of claims 1 to 10.
  12. A system comprising one or more computing devices and a computer memory including instructions which, when executed on the one or more computing devices, perform the steps of any of claims 1 to 10.

Description

CROSS-REFERENCE TO RELATED APPLICATIONS This application claims priority to U.S. Provisional Patent Application No. 62/573,683 filed on October 17, 2017, the entire content of which is hereby incorporated by reference. This application is also related to the following commonly-owned patent applications, each of which is hereby incorporated by reference in its entirety: U.S. Patent Application No. 14/290,065 filed on May 29, 2014, and U.S. Patent Application No. 15/265,761 filed on September 14, 2016 (now U.S. Patent No. 9,596,997). BACKGROUND Wearable devices can be used for physiological monitoring. While convenient, these devices are susceptible to a variety of different types of errors. For example, data quality may vary according to where a device is positioned on the body, whether the device is securely strapped to the body, the type of activity a user is engaged in, and so forth. There remains a need for real-time, data-driven assessments of data quality to accompany physiological data acquisition from wearable monitoring devices. SUMMARY A model of data quality is derived for physiological monitoring with a wearable device by comparing data from the wearable device to concurrent data acquisition from a ground truth device such as a chest strap or electrocardiography (EKG) heart rate monitor. With this comparative data, a machine learning model or the like may be derived to prospectively evaluate data quality based on the data acquisition context, as determined, for example, by other sensor data and signals from the wearable device. In one aspect, a method disclosed herein includes obtaining calibrated heart rate data from a number of subjects using one or more chest strap type sensors; obtaining uncalibrated heart rate data from the number of subjects concurrently with the calibrated heart rate data using one or more physiological monitors of a wrist-worn photoplethysmography type; obtaining feature data from one or more sensors of the one or more physiological monitors of the wrist-worn photoplethysmography type concurrently with the calibrated heart rate data and the uncalibrated heart rate data, the feature data characterizing a plurality of features of a data acquisition context for a corresponding one of the physiological monitors of the wrist-worn photoplethysmography type; associating a quality metric for the uncalibrated heart rate data with the feature data based on whether, for each data acquisition context, the uncalibrated heart rate data is within a predetermined threshold of the calibrated heart rate data; creating a quality estimator engine to evaluate a likelihood of the uncalibrated heart rate data being accurate based on the feature data; receiving second uncalibrated heart rate data and second feature data from a second physiological monitor of the wrist-worn photoplethysmography type; determining a probability that the second uncalibrated heart rate data is accurate for a window of measurements by calculating a conditional probability that the second uncalibrated heart rate data is accurate based on the second feature data over a distribution of values for the calibrated heart rate data within the window of measurements based on the quality estimator engine; and associating the probability with the window as a measure of quality for the uncalibrated heart rate data within the window. The feature data may include at least one signal used to estimate a heart rate using the uncalibrated heart rate data. The feature data may include at least one value derived from a signal from one of the physiological monitors of the wrist-worn photoplethysmography type. The feature data may include a signal derived from a motion sensor of one of the physiological monitors of the wrist-worn photoplethysmography type. The quality metric may be a one when the uncalibrated heart rate data is within the predetermined threshold of the uncalibrated heart rate data and a zero when the uncalibrated heart rate data is not within the predetermined threshold of the uncalibrated heart rate data. Creating the quality estimator engine may include training a machine learning random forest to estimate the quality metric for a set of feature data. The predetermined threshold may include a number of beats per minute for a heart rate. The method may further include providing feedback to a user concerning an adjustment to the second physiological monitor based on the measure of quality. The adjustment may include a change in a position of the second physiological monitor. The adjustment may include a change in a tension of a band for the second physiological monitor. The method may further include characterizing a user of the second physiological monitor, identifying a subset of the number of subjects similar to the user, and associating the quality metric for the uncalibrated heart rate data with the feature data for the subset of the number of subjects similar to the user. In another aspect, a method disclosed herein