Systems and methods for determining physiological parameters using photoplethysmography

Abstract

An example method for determining a physiological measurement from a physiological sensor includes receiving a physiological signal from the physiological sensor; inputting the physiological signal into a trained machine learning model; and outputting, using the trained machine learning model, an algorithm from a set of algorithms, where the algorithm is an algorithm configured to estimate a physiological measurement based on the physiological signal. An example system includes a wearable device comprising a plurality of physiological sensors; and a computing device operably coupled to the wearable device, where the computing device is configured to: receive a plurality of physiological signals measured by the plurality of physiological sensors; input the plurality of physiological signals into a trained machine learning model; and outputting, using the trained machine learning model, a physiological measurement.

Description

SYSTEMS AND METHODS FOR DETERMINING PHYSIOLOGICAL PARAMETERSUSING PHOTOPLETHYSMOGRAPHYCROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application claims the benefit of U.S. provisional patent application No. 63 / 586,663 filed on September 29, 2023, U.S. provisional patent application No. 63 / 621,879 filed on January 17, 2024, U.S. provisional patent application No. 63 / 634,145 filed on April 15, 2024, and U.S. provisional patent application No. 63 / 634,150 filed on April 15, 2024, which are expressly incorporated herein by reference in their entireties.BACKGROUND

[0002] A photoplethysmogram (PPG) is an optical measure of blood volume changes in tissue. In a PPG device, light illuminates a subject’s skin, and the amount of light transmitted or reflected is measured. By measuring the amount of light transmitted or reflected, the volume of blood in the subject’s tissue can be estimated. This volume of blood can be used to estimate different physiological parameters. One such parameter of the circulatory system is blood pressure.

[0003] Blood pressure is commonly measured at a known position relative to the patient’s heart. Movement and / or positioning of a patient’s body can impact the measurement of blood pressure.

[0004] Improvements to the estimates of physiological parameters, including blood pressure, using can improve patient health by allowing more accurate non-invasive measurements of blood pressure.SUMMARY

[0005] In some aspects, implementations of the present disclosure include a method for determining a physiological measurement from a physiological sensor, including: receiving a physiological signal from the physiological sensor; inputting the physiological signal into a trained machine learning model; and outputting, using the trained machine learning model, an algorithm from a set of algorithms.

[0006] In some aspects, implementations of the present disclosure include a method, further including determining a physiological measurement based on the algorithm and the physiological signal.

[0007] In some aspects, implementations of the present disclosure include a method, wherein the physiological sensor is a photoplethysmography (PPG) sensor.

[0008] In some aspects, implementations of the present disclosure include a method, wherein the trained machine learning model is a recurrent neural network (RNN).

[0009] In some aspects, implementations of the present disclosure include a method, wherein the trained machine learning model is a long short term memory (LSTM) network.

[0010] In some aspects, implementations of the present disclosure include a method, wherein the physiological measurement is a measurement of blood pressure.

[0011] In some aspects, implementations of the present disclosure include a method, further including inputting a patient health state into the trained machine learning model.

[0012] In some aspects, implementations of the present disclosure include a method, wherein the patient health state includes an activity profile of a patient.

[0013] In some aspects, implementations of the present disclosure include a method, wherein the patient health state includes physiological parameters of a patient.

[0014] In some aspects, implementations of the present disclosure include a method, wherein the physiological parameters of the patient include at least one of age, gender, weight, ethnicity, or health history.

[0015] In some aspects, implementations of the present disclosure include a method, further including inputting an environmental data into the trained machine learning model.

[0016] In some aspects, implementations of the present disclosure include a method, wherein the environmental data includes at least one of location, altitude, weather condition, time zone changes or temperature.

[0017] In some aspects, implementations of the present disclosure include a method for determining a physiological measure from a plurality of physiological sensors, including: receiving a plurality of physiological signals from a plurality of physiological sensors; inputting the plurality of physiological signals into a trained machine learning model; and outputting, using the trained machine learning model, a physiological measurement.

[0018] In some aspects, implementations of the present disclosure include a method, wherein the plurality of physiological signals include a photoplethysmography (PPG) signal.

[0019] In some aspects, implementations of the present disclosure include a method, wherein the plurality of physiological sensors includes a photoplethysmography (PPG) sensor.

[0020] In some aspects, implementations of the present disclosure include a method, further including inputting each of the plurality of physiological signals into a plurality of corresponding sensor-specific trained machine learning models.

[0021] In some aspects, implementations of the present disclosure include a method, wherein the trained machine learning model is a recurrent neural network.

[0022] In some aspects, implementations of the present disclosure include a method, wherein the trained machine learning model is a long short term memory (LSTM) network.

[0023] In some aspects, implementations of the present disclosure include a method, wherein the physiological measurement is a measurement of blood pressure.

[0024] In some aspects, implementations of the present disclosure include a method, further including inputting a patient health state into the trained machine learning model.

[0025] In some aspects, implementations of the present disclosure include a method, wherein the patient health state includes an activity profile of a patient.

[0026] In some aspects, implementations of the present disclosure include a method, wherein the patient health state includes physiological parameters of a patient.

[0027] In some aspects, implementations of the present disclosure include a method, wherein the physiological parameters of the patient include at least one of age, gender, weight, ethnicity, or health history.

[0028] In some aspects, implementations of the present disclosure include a method, further including inputting an environmental data into the trained machine learning model.

[0029] In some aspects, implementations of the present disclosure include a method, wherein the environmental data includes at least one of location, altitude, weather condition, time zone changes, or temperature.

[0030] In some aspects, implementations of the present disclosure include a system including: a wearable device including a plurality of physiological sensors; and a computing device operably coupled to the wearable device, wherein the computing device includes at least one processor and memory, the memory having computerexecutable instructions stored thereon that, when executed by the at least one processor, cause the at least one processor to: receive a plurality of physiological signals measured by the plurality of physiological sensors; input the plurality of physiological signals into a trained machine learning model; and outputting, using the trained machine learning model, a physiological measurement.

[0031] In some aspects, implementations of the present disclosure include a system, further including a portable electronic device operably coupled to the wearable device, wherein the portable electronic device is configured to: receive the plurality of physiological signals from the plurality of physiological sensors; and transmit the plurality of physiological signals to the computing device.

[0032] In some aspects, implementations of the present disclosure include a system, wherein the computing device is a cloud server.

[0033] In some aspects, implementations of the present disclosure include a system, wherein the computing device is further configured to: transmit the physiological measurement to the portable electronic device.

[0034] In some aspects, implementations of the present disclosure include a system, wherein the portable electronic device is configured to display the physiological measurement on a graphical user interface (GUI).

[0035] In some aspects, implementations of the present disclosure include a method including: receiving a plurality of PPG signals; receiving a plurality of corresponding ground-truth blood pressure measurements; receiving a plurality of corresponding selected blood pressure algorithms wherein the plurality of corresponding selected blood pressure algorithms each provide an estimate of blood pressure based on a respective PPG signal and wherein a corresponding selected blood pressure algorithm for each respective PPG signal provides an estimate of blood pressure that is closest to the corresponding ground-truth blood pressure for the respective PPG signal; creating a training set including the plurality of PPG signals and a respective selected blood pressure algorithm for the plurality of PPG signals; and training a machine learning model using the training set, wherein the trained machine learning model is configured to identify a corresponding selected blood pressure algorithm for a patient based on a test PPG signal measured from the patient.

[0036] In some aspects, implementations of the present disclosure include a method, wherein the method further includes receiving sets of calibration data values, wherein each of the sets of calibration data values correspond to at least one PPG signal from the plurality of PPG signals, and wherein the training set includes the sets of calibration data values.

[0037] In some aspects, implementations of the present disclosure include a method, wherein the calibration data values include a plurality of patient health states.

[0038] In some aspects, implementations of the present disclosure include a method, wherein the calibration data values include a plurality of activity profiles.

[0039] In some aspects, implementations of the present disclosure include a method, wherein the patient health state includes physiological parameters of the patient.

[0040] In some aspects, implementations of the present disclosure include a method, wherein the physiological parameters of the patient include age, gender, weight, ethnicity, or health history.

[0041] In some aspects, implementations of the present disclosure include a method, wherein the calibration data values further include a plurality of environmental data values input into the trained machine learning model.

[0042] In some aspects, implementations of the present disclosure include a method, wherein the environmental input includes location data, altitude, time zone changes or temperature.

[0043] In some aspects, implementations of the present disclosure include a method, wherein the trained machine learning model is a recurrent neural network.

[0044] In some aspects, implementations of the present disclosure include a method, wherein the trained machine learning model is a long short term memory (LSTM) model.

[0045] In some aspects, implementations of the present disclosure include a method including: receiving optical data from an optical sensor; receiving calibration data; deploying a trained machine learning model, wherein the trained machine learning model is configured to output an algorithm for estimating blood pressure based on optical data; selecting, by the trained machine learning model, a selected algorithm from a plurality of algorithms based on the calibration data and the optical data, wherein the selected algorithm is the algorithm chosen to be closest to ground truth blood pressure for the optical data and calibration data; and estimating blood pressure based on the optical data, the calibration data, and the selected algorithm.

[0046] In some aspects, implementations of the present disclosure include a method, wherein the plurality of algorithms each correspond to a physical state of a patient, and wherein the method further includes determining a disease state of the patient based on the selected algorithm.

[0047] In some aspects, implementations of the present disclosure include a method, wherein the trained machine learning model is a recurrent neural network.

[0048] In some aspects, implementations of the present disclosure include a method, wherein the trained machine learning model is a long short term memory (LSTM) model.

[0049] In some aspects, implementations of the present disclosure include a method, wherein the calibration data includes a plurality of patient health states.

[0050] In some aspects, implementations of the present disclosure include a method, wherein the patient health state includes physiological parameters of the patient.

[0051] In some aspects, implementations of the present disclosure include a method, wherein the physiological parameters of the patient include age, gender, weight, ethnicity, or health history.

[0052] In some aspects, implementations of the present disclosure include a method, wherein the calibration data includes a plurality of activity profiles.

[0053] In some aspects, implementations of the present disclosure include a method, wherein the calibration data further includes a plurality of environmental data values, an environmental input into the trained machine learning model.

[0054] In some aspects, implementations of the present disclosure include a method, wherein the environmental input includes location data, altitude, time zone changes or temperature.

[0055] In some aspects, implementations of the present disclosure include a method including: receiving optical data from an optical sensor; receiving calibration data; deploying a trained machine learning model, wherein the trained machine learning model is configured to output an algorithm for estimating blood pressure based on optical data; selecting, by the trained machine learning model, a selected algorithm from a plurality of algorithms based on the calibration data and the optical data, wherein the selected algorithm is the algorithm chosen to estimate a respective blood pressure closest to a ground truth blood pressure for the optical data and calibration data; receiving, second optical data from the optical sensor; receiving second calibration data; and training the trained machine learning model using the second optical data and the second calibration data.

[0056] In some aspects, implementations of the present disclosure include a method, wherein the trained machine learning model is an LTSM model.

[0057] In some aspects, implementations of the present disclosure include a method, wherein the calibration data includes a plurality of patient health states.

[0058] In some aspects, implementations of the present disclosure include a method wherein the patient health state includes physiological parameters of the patient.

[0059] In some aspects, implementations of the present disclosure include a method, wherein the physiological parameters of the patient include age, gender, weight, ethnicity, or health history.

[0060] In some aspects, implementations of the present disclosure include a method, wherein the calibration data includes a plurality of activity profiles.

[0061] In some aspects, implementations of the present disclosure include a method, wherein the calibration data further includes a plurality of environmental data values, an environmental input into the trained machine learning model.

[0062] In some aspects, implementations of the present disclosure include a method including: receiving a plurality of sensor signals; receiving a plurality of corresponding ground-truth physiological measurements; receiving a plurality of corresponding selected physiological estimation algorithms wherein the plurality of corresponding selected physiological estimation algorithms each provide an estimate of a physiological measurement based on a respective sensor signal and wherein a selected physiological estimation algorithm for each respective sensor signal provides an estimate of the physiological measurement that is closest to the corresponding ground-truth physiological measurement for the respective sensor signal; creating a training set including the plurality of sensor signals and a respective selected physiological measurement algorithm for the plurality of sensor signals; and training a machine learning model using the training set, wherein the trained machine learning model is configured to identify a corresponding selected physiological estimation algorithm for a patient based on a test sensor signal measured from the patient.

[0063] In some aspects, implementations of the present disclosure include a method, wherein the method further includes receiving an input from a patient.

[0064] In some aspects, implementations of the present disclosure include a method, wherein the patient input includes an indication of breathlessness.

[0065] In some aspects, implementations of the present disclosure include a method, wherein the physiological estimation algorithm is configured to estimate at least one of: pulmonary capillary wedge pressure and left atrial pressure.

[0066] In some aspects, implementations of the present disclosure include a method, wherein the physiological estimation algorithm is configured to estimate imaging data.

[0067] In some aspects, implementations of the present disclosure include a method, wherein the physiological estimation algorithm is configured to estimate left ventricular ejection fraction derived from ultrasound, RV volume or LV volumes.

[0068] In some aspects, implementations of the present disclosure include a method, wherein the physiological estimation algorithm is configured to estimate blood glucose.

[0069] In some aspects, implementations of the present disclosure include a method, wherein the physiological estimation algorithm is configured to estimate heart rate.

[0070] In some aspects, implementations of the present disclosure include a method, wherein the plurality of sensor signals includes measurements of B-Type Natriuretic Peptide (BNP).

[0071] In some aspects, implementations of the present disclosure include a method, wherein the plurality of sensor signals includes data from at least one of: electrocardiogram (ECG) electrodes and sensors, temperature sensors, SpO2 (oxygen saturation) sensors, impedance cardiography (ICG) electrodes, accelerometers and / or gyroscopes, ultrasonic transducers, skin conductance sensors, phonocardiogram sensors, ballistocardiography sensors, piezoelectric / piezoresistive sensors, ultrasound sensors, arterial tonometry sensors, and / or capacitive respiration sensors.

[0072] In some aspects, implementations of the present disclosure include a computer- implemented method including: accessing a plurality of training PPG waveform data; accessing a plurality of corresponding ground truth body position measurements; accessing a plurality of corresponding ground truth blood pressure measurements; creating a training data set including the plurality of training PPG waveform data, the corresponding ground truth body position measurements, and the corresponding ground truth blood pressure measurements; and training a machine learning model, using the training data set, to estimate a gravity contribution in each respective input training PPG waveform, wherein the gravity contribution corresponds to a hydrostatic pressure of blood relative to a heart.

[0073] In some aspects, implementations of the present disclosure include a computer- implemented method, wherein estimating the respective gravity contribution for each of the respective input training PPG waveforms includes estimating a column pressure caused by gravity at each of the plurality of corresponding ground truth body position measurements.

[0074] In some aspects, implementations of the present disclosure include a computer- implemented method, wherein the corresponding ground truth body position measurements include respective estimates of a height dimension at which respective training PPG data were gathered.

[0075] In some aspects, implementations of the present disclosure include a computer- implemented method, wherein determining the respective gravity contributions for each of the plurality of PPG data includes estimating a density of the blood.

[0076] In some aspects, implementations of the present disclosure include a computer- implemented method, wherein the estimated density is a constant.

[0077] In some aspects, implementations of the present disclosure include a computer- implemented method, wherein the training data set further includes a plurality of activity profiles.

[0078] In some aspects, implementations of the present disclosure include a computer- implemented method, wherein the training data set further includes a plurality of patient health states.

[0079] In some aspects, implementations of the present disclosure include a computer- implemented, wherein the plurality of patient health states include physiological parameters of the patients.

[0080] In some aspects, implementations of the present disclosure include a computer- implemented, wherein the physiological parameters of the patient include age, gender, weight, ethnicity, or health history.

[0081] In some aspects, implementations of the present disclosure include a computer- implemented, wherein the training data set further includes a plurality of environmental data values input into the trained machine learning model.

[0082] In some aspects, implementations of the present disclosure include a computer- implemented, wherein the environmental data values include location data, altitude, time zone changes or temperature.

[0083] In some aspects, implementations of the present disclosure include a computer- implemented method including: accessing a plurality of training PPG waveform data; accessing a plurality of corresponding ground truth blood pressure measurements; accessing a plurality of physiological parameters corresponding to each of the ground truth blood pressure measurements; and training a machine learning model, using the training PPG waveform data, the ground truth blood pressure measurements, and the plurality of physiological parameters, to output an estimated body position corresponding to the ground truth blood pressure measurements based on a respective training PPG waveform.

[0084] In some aspects, implementations of the present disclosure include a computer- implemented method, wherein training the machine learning model includes identifying patterns in combinations of the plurality of physiological parameters corresponding to respective ground truth blood pressure measurements.

[0085] In some aspects, implementations of the present disclosure include a computer- implemented method, wherein the physiological parameters include subject body dimensions, subject blood composition, subject blood density, subject vascular anatomy, and subject heart position.

[0086] In some aspects, implementations of the present disclosure include a computer- implemented method wherein outputting the estimated body position includes estimating a gravity contribution to the respective training PPG waveform by calculating an estimated height dimension at which the ground truth blood pressure measurements were gathered.

[0087] In some aspects, implementations of the present disclosure include a computer- implemented method, wherein calculating the estimated height dimension includes utilizing subject body dimensions at which the ground truth blood pressure measurements were gathered.

[0088] In some aspects, implementations of the present disclosure include a computer- implemented method of estimating blood pressure, the method including: receiving an input PPG signal from a PPG sensor located on a body part of a subject; receiving a trained blood pressure prediction machine learning model; receiving a trained blood pressure correction model; estimating, using the blood pressure prediction machine learning model, a blood pressure estimate; estimating, using the trained blood pressure correction model, blood pressure correction data; and determining, based on the blood pressure correction data and the blood pressure estimate, a corrected blood pressure.

[0089] In some aspects, implementations of the present disclosure include a computer- implemented method, wherein the blood pressure correction model is configured to correct the estimated blood pressure for a position of the PPG sensor.

[0090] In some aspects, implementations of the present disclosure include a computer- implemented method, wherein the blood pressure correction model is configured to correct the estimated blood pressure for a hydrostatic pressure of gravity.

[0091] In some aspects, implementations of the present disclosure include a computer- implemented method, wherein the trained blood pressure correction model includes a model trained.

[0092] In some aspects, implementations of the present disclosure include a method including: receiving a plurality of PPG signals; receiving a plurality of corresponding calibration data for each of the plurality of PPG signals; receiving a plurality of corresponding ground-truth blood pressure measurements for each of the plurality of PPG signals; receiving a plurality of corresponding selected blood pressure algorithms wherein the plurality of corresponding selected blood pressure algorithms each provide an estimate of blood pressure based on a respective PPG signal and wherein a corresponding selected blood pressure algorithm for each respective PPG signal provides an estimate of blood pressure that is closest to the corresponding ground-truth bloodpressure for the respective PPG signal; creating a training set including the plurality of PPG signals and a respective selected blood pressure algorithm for the plurality of PPG signals; and training a machine learning model using the training set, wherein the trained machine learning model is configured to identify a corresponding selected blood pressure algorithm for a patient based on a test PPG signal measured from the patient.

[0093] In some aspects, implementations of the present disclosure include a method, wherein the corresponding calibration data includes temperature data.

[0094] In some aspects, implementations of the present disclosure include a method, wherein the PPG signals are each measured from a corresponding vascular bed, and wherein the temperature data includes pressure measured at the corresponding vascular bed of the corresponding PPG signal.

[0095] In some aspects, implementations of the present disclosure include a method, wherein the corresponding calibration data includes pressure data.

[0096] In some aspects, implementations of the present disclosure include a method, wherein the PPG signals are each measured from a corresponding vascular bed, and wherein the pressure data includes pressure measured at the corresponding vascular bed of the corresponding PPG signal.

[0097] In some aspects, implementations of the present disclosure include a method, wherein the trained machine learning model is a recurrent neural network.

[0098] In some aspects, implementations of the present disclosure include a method, wherein the trained machine learning model is a long short term memory (LSTM) model.

[0099] In some aspects, implementations of the present disclosure include a method including: receiving optical data from an optical sensor; receiving calibration data; deploying a trained machine learning model, wherein the trained machine learning model is configured to output an algorithm for estimating blood pressure based on optical data; selecting, by the trained machine learning model, a selected algorithm from a plurality of algorithms based on the calibration data and the optical data, wherein the selected algorithm is the algorithm chosen to be closest to ground truth blood pressure for the optical data and calibration data; and estimating blood pressure based on the optical data, the calibration data, and the selected algorithm.

[0100] In some aspects, implementations of the present disclosure include a method, wherein the optical sensor is disposed over a vascular bed, and the calibration data includes a temperature value of the vascular bed.

[0101] In some aspects, implementations of the present disclosure include a method, wherein the optical sensor is disposed over a vascular bed, and the calibration data includes a pressure value on the vascular bed.

[0102] In some aspects, implementations of the present disclosure include a method, wherein the trained machine learning model is a recurrent neural network.

[0103] In some aspects, implementations of the present disclosure include a method, wherein the trained machine learning model is a long short term memory (LSTM) model.

[0104] In some aspects, implementations of the present disclosure include a system including: a wearable device including an optical sensor and a physiological sensor; and a computing device operably coupled to the wearable device, wherein the computing device includes at least one processor and memory, the memory having computerexecutable instructions stored thereon that, when executed by the at least one processor, cause the at least one processor to: receive an optical signal from the optical sensor; receive a corresponding physiological signal measured by the physiological sensor; input the optical signal and the corresponding physiological signal into a trained machine learning model; and output, using the trained machine learning model, a physiological measurement.

[0105] In some aspects, implementations of the present disclosure include a system, wherein the physiological sensor includes a temperature sensor.

[0106] In some aspects, implementations of the present disclosure include a system, wherein the physiological sensor includes a pressure sensor.

[0107] In some aspects, implementations of the present disclosure include a system, wherein the physiological sensor includes a pressure sensor and a temperature sensor.

[0108] In some aspects, implementations of the present disclosure include a system, wherein the pressure sensor includes a piezoelectric pressure sensor.

[0109] In some aspects, implementations of the present disclosure include a system, wherein the physiological sensor and the optical sensor are configured to measure the same vascular bed of a patient.

[0110] In some aspects, implementations of the present disclosure include a system, further including a wristband configured to position the physiological sensor and the optical sensor adjacent to a single vascular bed.

[0111] In some aspects, implementations of the present disclosure include a system, further including a portable electronic device operably coupled to the wearable device, wherein the portable electronic device is configured to: receive the correspondingphysiological signal from the physiological sensor; and transmit the corresponding physiological signal to the computing device.

[0112] It should be understood that the above-described subject matter may also be implemented as a computer-controlled apparatus, a computer process, a computing system, or an article of manufacture, such as a computer-readable storage medium.

[0113] Other systems, methods, features and / or advantages will be or may become apparent to one with skill in the art upon examination of the following drawings and detailed description. It is intended that all such additional systems, methods, features and / or advantages be included within this description and be protected by the accompanying claims.BRIEF DESCRIPTION OF THE DRAWINGS

[0114] The components in the drawings are not necessarily to scale relative to each other. Like reference numerals designate corresponding parts throughout the several views.

[0115] FIG. 1 A illustrates an example method for selecting an algorithm based on a physiological signal, according to implementations of the present disclosure.

[0116] FIG. IB illustrates an example method for outputting an algorithm from a set of algorithms based on a plurality of physiological signals, according to implementations of the present disclosure.

[0117] FIG. 2A illustrates a system including a wearable device including physiological sensors and configured to output an estimate of a physiological parameter based on a physiological sensor signal, according to implementations of the present disclosure.

[0118] FIG. 2B illustrates a system including a wearable device including an optical sensor and position sensor, and configured to output an estimate of a physiological parameter based on a PPG signal, according to implementations of the present disclosure.

[0119] FIG. 2C illustrates a system including a wearable device including a PPG sensor and configured to output an estimate of a physiological parameter based on a PPG signal, according to implementations of the present disclosure.

[0120] FIG. 2D illustrates a wearable device including a PPG sensor, pressure sensor, and temperature sensor disposed over a vascular bed of a subject.

[0121] FIG. 3A illustrates an example arterial waveform recorded in a healthy subject.

[0122] FIG. 3B illustrates an example arterial waveform for a subject with aortic valvular stenosis.

[0123] FIG. 4 illustrates an example computing device.

[0124] FIG. 5 illustrates an example block diagram of a system according to implementations of the present disclosure.

[0125] FIG. 6A illustrates an example method for training a machine learning model to select a blood pressure estimation algorithm, according to implementations of the present disclosure.

[0126] FIG. 6B illustrates an example method for deploying a trained machine learning model to estimate blood pressure based on optical data, according to implementations of the present disclosure.

[0127] FIG. 6C illustrates an example method for further training a trained machine learning model to estimate blood pressure based on optical data, according to implementations of the present disclosure.

[0128] FIG. 6D illustrates an example method of training a machine learning model to select a physiological measurement algorithm for a patient based on a PPG signal measure from the patient, according to implementations of the present disclosure.

[0129] FIG. 7 A illustrates an example method for training a machine learning model to estimate an effect of gravity on a PPG waveform and / or blood pressure estimate, according to implementations of the present disclosure.

[0130] FIG. 7B illustrates an example method for deploying a trained machine learning model to a body position based on a PPG signal, according to implementations of the present disclosure.

[0131] FIG. 7C illustrates an example method of determining a corrected blood pressure by using a trained machine learning model to estimate blood pressure correction data, according to implementations of the present disclosure.

[0132] FIG. 8 A illustrates an example method for training a machine learning model to select a blood pressure estimation algorithm, according to implementations of the present disclosure.

[0133] FIG. 8B illustrates an example method for deploying a trained machine learning model to estimate blood pressure based on optical data and calibration data, according to implementations of the present disclosure.

[0134] FIG. 9 illustrates an example method for estimating blood pressure using a machine learning model to select a blood pressure estimation algorithm from a waveform library, including calibration, according to implementations of the present disclosure.

[0135] FIG. 10 illustrates an example method of estimating a blood pressure, according to implementations of the present disclosure.

[0136] FIG. 11 illustrates an example method of estimating a blood pressure, according to implementations of the present disclosure.

[0137] FIG. 12 illustrates an example block diagram of a system for acquiring optical data, according to implementations of the present disclosure.

[0138] FIG. 13 illustrates an example method of estimating blood pressure using optical inputs, according to implementations of the present disclosure.DETAILED DESCRIPTION

[0139] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art. Methods and materials similar or equivalent to those described herein can be used in the practice or testing of the present disclosure. As used in the specification, and in the appended claims, the singular forms “a,” “an,” “the” include plural referents unless the context clearly dictates otherwise. The term “comprising” and variations thereof as used herein is used synonymously with the term “including” and variations thereof and are open, nonlimiting terms. The terms “optional” or “optionally” used herein mean that the subsequently described feature, event or circumstance may or may not occur, and that the description includes instances where said feature, event or circumstance occurs and instances where it does not. Ranges may be expressed herein as from "about" one particular value, and / or to "about" another particular value. When such a range is expressed, an aspect includes from the one particular value and / or to the other particular value. Similarly, when values are expressed as approximations, by use of the antecedent "about," it will be understood that the particular value forms another aspect. It will be further understood that the endpoints of each of the ranges are significant both in relation to the other endpoint, and independently of the other endpoint. While implementations will be described for estimating blood pressure using PPG sensors and signals, it should be understood that implementations of the present disclosure include systems, devices, and methods for estimating and / or determining any combination physiological measurements

[0140] As used herein, the terms "about" or "approximately" when referring to a measurable value such as an amount, a percentage, and the like, is meant to encompass variations of ±20%, ±10%, ±5%, or ±1% from the measurable value.

[0141] Implementations of the present disclosure provide improvements to estimating physiological parameters using one or more measurements of the circulatory system and models that relate the one or more measurements to the physiological parameters.Implementations of the present disclosure include systems and methods for selecting models that relate measurements to different physiological parameters, and using those models to output reliable estimates of different physiological parameters.

[0142] Models of the circulatory system can be used to estimate one or more physiological parameters of a patient based on measured physiological parameters of that patient. For example, measurements of electrical activity can be used to estimate heart rate using a model that relates electrical activity to estimated heart rate. As another example, measurements of blood volume can be used to estimate blood pressure using a model that relates blood volume to blood pressure.

[0143] Estimating physiological parameters of a patient based on a model of the circulatory system poses numerous challenges. Modeling the circulatory system is computationally challenging. Wide demographic variations are present in the population, ranging from young and healthy individuals, to older or less healthy individuals, as well as individuals with unique medical conditions. Moreover medical devices that measure and estimate physiological parameters are often required to output highly accurate results across many different combinations of age, health, and other demographics. Models can be made that are fit to specific groups of patients, to patients with specific conditions, or patients in specific situations. However, this raises additional challenges in selecting the correct models to estimate physiological parameters for each patient, adjusting those models over time, and determining when to use different models for the same patient.

[0144] Medical devices, (e.g., photoplethysmography-based blood pressure measurement devices) can benefit from systems and methods that identify and use customized models of the circulatory system. Implementations of the present disclosure include systems and methods to estimate a subject’s blood pressure based on the user’s specific attributes (e.g., age, health conditions, and other demographic information). These systems can benefit patients by providing efficient and non-invasive monitoring of blood pressure and other measures of cardiovascular health.

[0145] As non-limiting examples, implementations of the present disclosure include the following example improvements to systems, devices, and methods for estimating physiological measurements:

[0146] Improvements to selecting algorithms for estimating blood pressure from PPG signals: Existing methods of estimating blood pressure from PPG signals are limited by the use of a generic / generalized algorithm that is used for all patients, and / or rely on users to determine what algorithm should be used to estimate a blood pressure from a PPG signal. Implementations of the present disclosure overcome the limits ofthese conventional methods by providing multiple algorithms that are configured for estimating blood pressure for different types of patients (e.g., patients with different amounts of arterial stiffness, heart conditions, etc.) and selecting the appropriate algorithm for each patient using a trained machine learning model (e.g., a classifier or other model).

[0147] Improvements to environmental and / or health-state calibration for estimating physiological measurements from PPG signals: Implementations of the present disclosure include improvements to systems and methods for using calibration data (e.g., patient health state, environmental conditions, and / or additional physiological data) to select algorithms for estimating physiological measurements (e.g., blood pressure) from PPG signals. Conventional systems may use simple, generalized calibrations to blood pressure estimates. Implementations of the present disclosure allow for more accurate / precise calibrations to be performed by using calibration data to train a machine learning model to select an algorithm of a set of algorithms for estimating the physiological measurement. Alternatively or additionally, implementations of the present disclosure can include machine learning models trained to calibrate an estimated physiological measurement (for example, a physiological measurement obtained using a machine learning model to select an algorithm from a set of algorithms, and using the selected algorithm to estimate the physiological parameter based on a sensor input).

[0148] Improvements to position-based calibration for estimating blood pressure from PPG signals: Implementations of the present disclosure include trained machine learning models, systems, methods, and devices that can detect position, orientation, and / or motion of an individual’s body. The example implementations described herein can be used to provide accurate continuous measurements of blood pressure in wearable devices, where the measurements of blood pressure are adjusted for the detected position, orientation, and / or motion of the wearer of the wearable device.

[0149] Implementations of the present disclosure include systems and methods to train machine learning models to estimate physiological measurements of a subject (e.g., blood pressure) based on sensor data (e.g., PPG or optical sensor data). It should be understood that a photoplethysmogram (PPG) sensor is a type of optical sensor, so that references to “optical sensors” herein include PPG sensors, but it should also be understood that optical sensors includes sensors that receive any kind of optical data, in and are not limited to PPG sensors. The machine learning models can be further calibrated based on the subject’s (e.g., a user of a wearable device) attributes (e.g., age, health conditions, other demographic information, etc.) as well as environmental conditions (e.g.,temperature, time of year, altitude, etc.). These systems can benefit patients by providing efficient and non-invasive monitoring of blood pressure and other physiological measurements.

[0150] Improvements to temperature and pressure calibration for estimating blood pressure using PPG signals: Implementations of the present disclosure include systems and methods to train machine learning models to estimate physiological measurements of a subject (e.g., blood pressure) based on sensor data (e.g., PPG or optical sensor data). It should be understood that a photoplethysmogram (PPG) sensor is a type of optical sensor, so that references to “optical sensors” herein include PPG sensors, but it should also be understood that optical sensors includes sensors that receive any kind of optical data, in and are not limited to PPG sensors. The machine learning models can be further calibrated based on the subject’s (e.g., a user of a wearable device) attributes (e.g., age, health conditions, other demographic information, etc.) as well as environmental conditions (e.g., temperature, time of year, altitude, etc.). These systems can benefit patients by providing efficient and non-invasive monitoring of blood pressure and other physiological measurements.

[0151] Example challenges of modeling the circulatory system include environmental conditions around an optical sensor including pressure and / or temperature. When an optical sensor or a wearable device including an optical sensor is touching a subject’s skin, the optical sensor can compress the vascular bed that is measured by the sensor. The compression of the vascular bed can affect the shape of a PPG waveform measured by the optical sensor, and cause the PPG waveform to change in ways that do not reflect physiological changes in the subject. For example, pressure on the vascular bed could reduce the amplitude of the patient’ s pulse and / or lengthen the pulse transit time to the vascular bed, without affecting the patient’s physiological condition (e.g., the patient’s ground-truth blood pressure may stay the same as a wearable device is tightened or loosened, but the PPG signal measured at that device could change along with the blood pressure determined based on the PPG signal).

[0152] Additionally, different vascular beds can behave differently. For example, implementations of the present disclosure include systems configured to measure vascular beds of the finger, fingertip, ear, and wrist, all of which may have different waveforms which correspond to the same blood pressure. Accordingly, implementations of the present disclosure can be used to accurately estimate blood pressure using sensors (e.g., PPG sensors) that can be positioned in any locations on the body. Optionally, the sensors of the present disclosure can be implemented using wireless sensors (e.g.,wireless fingertip sensors). Implementations of the present disclosure can improve over conventional cuff-based PPG sensors that can require blood pressure to be measured at the upper arm, with the arm in a fixed position.

[0153] The effects of temperature, pressure, and other environmental factors can be considered “artifacts” or “errors” of sensing that are caused by the structure or limits of the sensor itself and not present in the subject being sensed. For example, when a subject’s wrist is heated or cooled near the optical sensor, it can affect the PPG signal measured by the sensor, without that change in PPG signal representing a change in the patient’s health (e.g., blood pressure). Likewise, pressing or releasing the optical sensor from the skin can also create sensor artifacts by affecting the pressure on the vascular bed being measured by an optical sensor, but those pressure changes do not correspond to changes in the patient’s health. Accordingly, systems and methods described herein can be used to remove or reduce the effects of sensor artifacts.

[0154] As another example, when the temperature increases in area of subject’s skin being sensed by the optical sensor, the vascular resistance in that area of the subject may decrease. As non-limiting examples, reduced vascular resistance can increase pulse transit time detected by the optical sensor and reduce the amplitude of the PPG signal, without affecting the patient’s ground-truth blood pressure or other health parameters. For example, the wearable device may include a wristband that heats up the subject’s wrist over time, without heating up the rest of the subject’s body.

[0155] The systems and methods described herein can include systems and methods that can accurately determine blood pressure based on PPG signals by compensating for environmental conditions (e.g., the temperature and / or pressure changes described herein). The systems and methods described herein can further include systems and methods for training machine learning models to estimate or determine blood pressure based on PPG by training the machine learning model using data including environmental conditions, PPG signals, and corresponding ground-truth blood pressure measurements, for example.

[0156] Definitions:

[0157] PPG Sensors: A photoplethysmogram (PPG) sensor can include one or more light sources (e.g., LEDs) and / or one or more light sensors (e.g., photodiodes, photoresistors, imaging sensors, etc.). By illuminating tissue of a patient (e.g., the surface of the skin) and measuring the absorbed light, a PPG sensor can measure light absorbance in the tissue. It should be understood that a (PPG) sensor is a type of optical sensor, so that references to “optical sensors” herein include PPG sensors, but it should also beunderstood that optical sensors includes sensors that receive any kind of optical data, in and are not limited to sensors that collect PPG measurements. As described herein, PPG sensors can be configured to take physiological measurements including blood volume, blood volume changes, and / or oxygen saturation.

[0158] PPG Signals: A PPG signal as used herein can refer to any measurement of light absorbance in any tissue. PPG signals can include measures of blood volume, blood volume changes, and / or oxygen saturation, for example.

[0159] Wearable Device(s): A device configured to be worn on / around any body part. Example types of wearable devices contemplated by the present disclosure include rings, wristbands, adhesive patches / straps / bandages, clamps (e.g., finger clamps), chest straps, etc. Example locations where wearable devices can be attached include any limb (wrists, fingers, fingertips, hands, feet, legs, arms), and / or any location on a patient’s head, neck, and / or torso, according to various implementations of the present disclosure.Additionally, the present disclosure contemplates that a wearable device and / or system including a wearable device can optionally be implemented using multiple sensors configured to be attached to different locations on the patient’s body (e.g., a first sensor configured to be worn on a fingertip, and a second sensor configured to be worn on the wrist). Additionally, the present disclosure contemplates that wearable devices can include computing devices that may not be located in the same position as the sensor(s). For example, a wearable device configured as a fingertip clamp or adhesive bandage may be in operable communication with a computing device that is not attached to the fingertip of the patient (e.g., in a wrist band or portable electronic device)

[0160] Physiological Measurement: Physiological measurements include any measurement of a patient’s physiology. Non-limiting examples of physiological measurements include blood pressure, body temperature, pulse rate, breathing rate, ECG (electrocardiography), ICG (impedance cardiography), galvanic skin response (GSR), skin conductance, motion sensors (e.g., accelerometers), etc. Physiological measurements can be stored as physiological data that can be processed using the computer- implemented methods described herein.

[0161] Physiological Signals and Waveform(s): Physiological signals can include any signals form any sensor configured to perform a physiological measurement. For example, if the sensors is a PPG sensor, the physiological signals can include the outputs of one or more photodiodes (or any other light / imaging sensor) used to measure light absorption for the PPG measurement. “Physiological waveforms” or “waveforms” refer to time-varying physiological signals. Physiological signals and / or waveforms as usedherein can further refer to features of the physiological signals and / or their waveforms (e.g., peak amplitude, and / or frequency).

[0162] Ground-Truth Physiological Measurement: As used herein, a ground-truth physiological measurement is a physiological measurement used as a ground truth value to train the machine learning models described herein. Optionally, ground truth physiological measurements can be collected using conventional / standardized instruments. For example, a blood pressure cuff (sphygmomanometer) and / or invasive measurement of the pressure of an arterial line can be used to record ground truth blood pressure values that can be used as training data for the machine learning models described herein.

[0163] Physiological Sensor(s): Any sensor configured to measure the physiology of a patient. Example physiological sensors include: optical sensors, PPG sensors, electrocardiogram (ECG) electrodes and sensors (including any number of leads), temperature sensors, SpO2 (oxygen saturation) sensors, impedance cardiography (ICG) electrodes, accelerometers and / or gyroscopes, ultrasonic transducers, skin conductance sensors, phonocardiogram sensors, ballistocardiography sensors, piezoelectric / piezoresistive sensors, ultrasound sensors, arterial tonometry sensors, B- Type Natriuretic Peptide (BNP) sensors, and / or capacitive respiration sensors.

[0164] Patient Health State or Health State: As used herein, the “patient health state” or “health state” can include any information related to the health or demographics of a patient. For example, the patient health state can include a disease state (e.g., the presence of, absence of, and / or progression of a disease). Alternatively or additionally, the patient health state can include an activity profile of the patient. As yet another example, the patient health state can include physiological parameters of the patient including age, gender, weight, body mass index (BMI), medical history, ethnicity, health history, any other demographic information, and / or any combination thereof.

[0165] Activity Profile: As used herein, the activity profile can include any or all of: how active the patient is, physical activity, food intake profile(s), liquid intake profile(s) and / or when the patient is active.

[0166] Environmental Data: As used herein, “environmental data” can refer to any information about the environment that the patient is in, or may experience. Non-limiting examples of environmental data include location, altitude, weather condition(s), time zone changes or temperature, and / or any combination thereof.

[0167] Artificial Intelligence and Machine Learning

[0168] The term “artificial intelligence” is defined herein to include any technique that enables one or more computing devices or comping systems (i.e., a machine) to mimic human intelligence. Artificial intelligence (Al) includes, but is not limited to, knowledge bases, machine learning, representation learning, and deep learning. The term “machine learning” is defined herein to be a subset of Al that enables a machine to acquire knowledge by extracting patterns from raw data. Machine learning techniques include, but are not limited to, logistic regression, support vector machines (SVMs), decision trees, Naive Bayes classifiers, and artificial neural networks. The term “representation learning” is defined herein to be a subset of machine learning that enables a machine to automatically discover representations needed for feature detection, prediction, or classification from raw data. Representation learning techniques include, but are not limited to, autoencoders. The term “deep learning” is defined herein to be a subset of machine learning that that enables a machine to automatically discover representations needed for feature detection, prediction, classification, etc. using layers of processing. Deep learning techniques include, but are not limited to, artificial neural network or multilayer perceptron (MLP).

[0169] Machine learning models include supervised, semi-supervised, and unsupervised learning models. In a supervised learning model, the model learns a function that maps an input (also known as feature or features) to an output (also known as target or target) during training with a labeled data set (or dataset). In an unsupervised learning model, the model learns a function that maps an input (also known as feature or features) to an output (also known as target or target) during training with an unlabeled data set. In a semi- supervised model, the model learns a function that maps an input (also known as feature or features) to an output (also known as target or target) during training with both labeled and unlabeled data.

[0170] Neural Networks

[0171] An artificial neural network (ANN) is a computing system including a plurality of interconnected neurons (e.g., also referred to as “nodes”). This disclosure contemplates that the nodes can be implemented using a computing device (e.g., a processing unit and memory as described herein). The nodes can be arranged in a plurality of layers such as input layer, output layer, and optionally one or more hidden layers. An ANN having hidden layers can be referred to as deep neural network or multilayer perceptron (MLP). Each node is connected to one or more other nodes in the ANN. For example, each layer is made of a plurality of nodes, where each node is connected to all nodes in the previous layer. The nodes in a given layer are not interconnected with one another, i.e., the nodesin a given layer function independently of one another. As used herein, nodes in the input layer receive data from outside of the ANN, nodes in the hidden layer(s) modify the data between the input and output layers, and nodes in the output layer provide the results. Each node is configured to receive an input, implement an activation function (e.g., binary step, linear, sigmoid, tanH, or rectified linear unit (ReLU) function), and provide an output in accordance with the activation function. Additionally, each node is associated with a respective weight. ANNs are trained with a dataset to maximize or minimize an objective function. In some implementations, the objective function is a cost function, which is a measure of the ANN’s performance (e.g., error such as LI or L2 loss) during training, and the training algorithm tunes the node weights and / or bias to minimize the cost function. This disclosure contemplates that any algorithm that finds the maximum or minimum of the objective function can be used for training the ANN.Training algorithms for ANNs include, but are not limited to, backpropagation. It should be understood that an artificial neural network is provided only as an example machine learning model. This disclosure contemplates that the machine learning model can be any supervised learning model, semi-supervised learning model, or unsupervised learning model. Optionally, the machine learning model is a deep learning model. Machine learning models are known in the art and are therefore not described in further detail herein.

[0172] A convolutional neural network (CNN) is a type of deep neural network that has been applied, for example, to image analysis applications. Unlike a traditional neural networks, each layer in a CNN has a plurality of nodes arranged in three dimensions (width, height, depth). CNNs can include different types of layers, e.g., convolutional, pooling, and fully-connected (also referred to herein as “dense”) layers. A convolutional layer includes a set of filters and performs the bulk of the computations. A pooling layer is optionally inserted between convolutional layers to reduce the computational power and / or control overfitting (e.g., by downsampling). A fully-connected layer includes neurons, where each neuron is connected to all of the neurons in the previous layer. The layers are stacked similar to traditional neural networks. GCNNs are CNNs that have been adapted to work on structured datasets such as graphs.

[0173] Other Supervised Learning Models :

[0174] A logistic regression (LR) classifier is a supervised classification model that uses the logistic function to predict the probability of a target, which can be used for classification. LR classifiers are trained with a data set (also referred to herein as a “dataset”) to maximize or minimize an objective function, for example a measure of theLR classifier’s performance (e.g., error such as LI or L2 loss), during training. This disclosure contemplates that any algorithm that finds the minimum of the cost function can be used. LR classifiers are known in the art and are therefore not described in further detail herein.

[0175] An Naive Bayes’ (NB) classifier is a supervised classification model that is based on Bayes’ Theorem, which assumes independence among features (i.e., presence of one feature in a class is unrelated to presence of any other features). NB classifiers are trained with a data set by computing the conditional probability distribution of each feature given label and applying Bayes’ Theorem to compute conditional probability distribution of a label given an observation. NB classifiers are known in the art and are therefore not described in further detail herein.

[0176] A k-NN classifier is a supervised classification model that classifies new data points based on similarity measures (e.g., distance functions). k-NN classifiers are trained with a data set (also referred to herein as a “dataset”) to maximize or minimize an objective function, for example a measure of the k-NN classifier’s performance, during training. This disclosure contemplates that any algorithm that finds the maximum or minimum of the objective function can be used. k-NN classifiers are known in the art and are therefore not described in further detail herein.

[0177] An majority voting ensemble is a meta-classifier that combines a plurality of machine learning classifiers for classification via majority voting. In other words, the majority voting ensemble’s final prediction (e.g., class label) is the one predicted most frequently by the member classification models. Majority voting ensembles are known in the art and are therefore not described in further detail herein.

[0178] Example Implementations: Wearable devices

[0179] FIGS. 2A-2D illustrate example wearable devices according to implementations of the present disclosure. As shown in FIGS. 2A-2D, implementations of the present disclosure include systems, methods, and devices for performing physiological measurements. Any or all of the methods described herein with reference to FIGS. 1 A, 1 B , 6A- 11, 12, and 13 , (or any combination of those methods) can be performed using the hardware described with reference to FIGS. 2A-2D, for example.

[0180] As shown in FIG. 2A, an example system 200a can include a wearable device 210 including one or more physiological sensors 220. The physiological sensors 220 can include any of the physiological sensors described herein, for example with reference to FIGS. 1A and IB.

[0181] The example system can further include a computing device 250 operably coupled to the wearable device 210. It should be understood that the computing device 250 and wearable device 210 can be coupled using any combination of wireless and / or wired connections, including Bluetooth, Wi-Fi, cellular / LTE, etc. Additionally, it should be understood that the computing device can be part of the wearable device in some implementations of the present disclosure. Alternatively or additionally, the computing device can be separate from the wearable device (e.g., a server). Optionally, the computing device 250 can be a cloud server.

[0182] The computing device 250 can include any of the elements shown and described with reference to the computing device illustrated in FIG. 4.

[0183] The computing device 250 can include at least one processor and memory (not shown), where the includes computer-executable instructions that, cause the processor to implement any of the methods described herein (e.g., the methods 100, 150 described with reference to FIG. 1 A and FIG. IB). For example, the computing device 250 can be configured to (A) select an algorithm to estimate a physiological measurement of the patient; (B) estimate the physiological measurement of the patient using the algorithm; and / or (C) estimate a physiological measurement of the patient using a machine learning model.

[0184] For example, the system 200a can be configured to receive a plurality of physiological signals measured by the plurality of physiological sensors 220.

[0185] The system 200a can further be configured to input the plurality of physiological signals into a trained machine learning model 255 and output, using the trained machine learning model 255, a physiological measurement. Optionally, the computing device and / or wearable device can include a display 260. The display can be configured to receive the physiological measurement and display the physiological measurement to the user.

[0186] In some implementations, the system 200a can further include a portable electronic device 230 operably coupled to the wearable device 210, where the portable electronic device 230 is configured to: receive the plurality of physiological signals from the plurality of physiological sensors 220; and transmit the plurality of physiological signals to the computing device 250. Optionally, the computing device 250 can transmit the physiological measurements to the portable electronic device 230. In some implementations of the present disclosure, the trained machine learning model 255 can be included in the portable electronic device 230.

[0187] In some implementations, the portable electronic device 230 can include a display 270, and the portable electronic device 230 can be configured to display the physiological measurement on a graphical user interface (GUI).

[0188] It should be understood that, in some implementations, both the computing device 250 and portable electronic device 230 can be the same computing device, and / or that the portable electronic device and / or computing device can be part of the same structure as the wearable device. For example, the wearable device can optionally be implemented as a smart wristband that includes the physiological sensors 220, and a single computing device and single display are configured to both perform the functions of the portable electronic device 230 described herein and the functions of the computing device 250 described herein. Alternatively or additionally, the portable electronic device 230 and / or computing device 250 can be implemented as a single computing device in operable communication with the wearable device 210. For example, a smartphone or smartwatch can implement both the portable electronic device 230 and computing device 250, and be in operable communication with the wearable device 210 to receive physiological data from the physiological sensors.

[0189] In some implementations, the system 200a can further include a portable electronic device 230 operably coupled to the wearable device 210, where the portable electronic device 230 is configured to: receive the plurality of physiological signals from the plurality of physiological sensors 220; and transmit the plurality of physiological signals to the computing device 250. Optionally, the computing device 250 can transmit the physiological measurements to the portable electronic device 230.

[0190] In some implementations, the portable electronic device 230 can include a display 270, and the portable electronic device 230 can be configured to display the physiological measurement on a graphical user interface (GUI).

[0191] Implementations of the present disclosure include improvements to sensing body position and / or using body position to estimate blood pressure using PPG signals.

[0192] Implementations of the present disclosure can be used in combination with systems and methods for determining blood pressure. A system or method that uses a physiological signal collected at a location on a patient’s body may not be accurate when the location on the patient’s body moves relative to the heart. For example, a wearable device including a wristband may collect a PPG signal to estimate blood pressure based on the PPG signal collected at a patient’s wrist. The magnitude of the blood pressure can be calibrated based on a secondary measurement of blood pressure, for example a blood pressure cuff. Optionally, the calibration using the secondary measurement of bloodpressure can be performed when the wearable device is at a known position (e.g., when the arm is lowered, raised, positioned horizontally to the ground, etc.).

[0193] When the wristband is raised or lowered relative to the heart, the apparent blood pressure of the person may appear to change due to the change in measurement position. Therefore a wearable device that is not calibrated for position can show changes in blood pressure that are due to position, not due to underlying physiological phenomena (e.g., cardiovascular conditions or disease).

[0194] In an example implementation of the present disclosure, the wearable device is a wristband configured to estimate blood pressure based on PPG signals collected on a user’s wrist. The wearable device can detect instantaneous body position changes or movement, for example the velocity, acceleration, and / or position of the body of the wearer.

[0195] Based on the body position changes, the wearable device can estimate a proportion of the blood pressure at any position that is due to the position of the wearable device. For example, if the wearable device is lowered relative to the heart, the wearable device can estimate the increase in blood pressure due to the wearable device being below the heart. If the wearable device is raised relative to the heart, the wearable device can estimate a decrease in blood pressure that is due to the wearable device being lowered relative to the heart.

[0196] In some implementations of the present disclosure, the wearable device can estimate the position, orientation, acceleration and / or velocity of the wearable device based on properties of a sensed physiological waveform. For example, body motion can affect a PPG waveform or signal differently than the ordinary motion of blood in the circulatory system. The effect of body motion on a PPG signal or other physiological signal can be considered motion artifacts. By detecting the motion artifacts in the signal, the motion of the body can be detected and / or estimated, and a standardized measure of blood pressure (e.g., the blood pressure that would be measured at the brachial artery if the arm was at a resting position) can be measured, regardless of whether the measurement is taken at the brachial artery or whether the arm is in the correct position relative to the heart.

[0197] Optionally, implementations of the present disclosure can use machine learning models to identify characteristic artifacts in a physiological signal (e.g., a PPG signal) that are caused by the position, orientation, or motion of the body in order to compensate for the effect of the position, orientation, or motion of the body and output a standardized measure of blood pressure. Alternatively or additionally, implementations of the presentdisclosure can include trained machine learning models that estimate an amount of motion, velocity, or acceleration of a body or body part based on PPG signals. Alternatively or additionally, implementations of the present disclosure include trained machine learning models that can estimate a standardized blood pressure based on a physiological signal or waveform that includes motion artifacts. For example, the trained machine learning model can be trained based on datasets including physiological signals collected at rest and with various amounts of known motion.

[0198] With reference to FIG. 2B, an example system 200b can include a wearable device 210 including one or more physiological sensors 220. The physiological sensors 220 can include any of the physiological sensors described herein, including, for example, PPG sensors.

[0199] Still with reference to FIG. 2B, implementations of the present disclosure can optionally include a position sensor 224 in addition to the features shown / described with reference to FIG. 2B. In some implementations of the present disclosure, the methods described herein can be implemented as computer-implemented methods, or using computer-readable media. For example, in the system 200b shown in FIG. 2B, the methods (and computer-implemented methods) of training and deploying machine learning models described herein can be implemented using the computing device 250 and / or the portable electronic device 230.

[0200] An example system 200b can include a wearable device 210 including one or more physiological sensors 220. The physiological sensors 220 can include any of the physiological sensors described herein, including, for example, PPG sensors.

[0201] In some implementations of the present disclosure, the physiological sensors 220 of the wearable device 210 can further include one or more position sensors 224 configured to measure position, orientation, and / or motion of the wearable device 210. Non- limiting examples of position sensors 224 include any combination of inertial measurement units, accelerometers, gyroscopes, and compasses. Optionally the physiological sensors 220 can be configured to measure position, orientation, and / or motion relative to a heart or any other body part of the wearer of the wearable device 210.

[0202] As described herein, the position, orientation, and / or motion of the wearable device 210 can cause changes in a measured blood pressure at the wearable device 210. The present disclosure contemplates that by measuring the position, orientation, and / or motion of the wearable device the blood pressure output by the wearable device can beadjusted, for example to make the output blood pressure value equivalent to the blood pressure value that would be measured at the brachi l artery.

[0203] In the example system shown in FIG. 2B, the physiological sensors 220 include an optical sensor 222. The example system can further include a computing device 250 operably coupled to the wearable device 210. It should be understood that the computing device 250 and wearable device 210 can be coupled using any combination of wireless and / or wired connections, including Bluetooth, Wi-Fi, cellular / LTE, etc. Additionally, it should be understood that the computing device can be part of the wearable device in some implementations of the present disclosure. Alternatively or additionally, the computing device can be separate from the wearable device (e.g., a server). Optionally, the computing device 250 can be a cloud server.

[0204] The computing device 250 can include any of the elements shown and described with reference to the computing device 400 illustrated in FIG. 4.

[0205] The computing device 250 can include at least one processor and memory (not shown), where the includes computer-executable instructions that, cause the processor to implement any of the methods described herein (e.g., the methods described with reference to FIGS. 7A-7C).

[0206] For example, the system 200b can be configured to receive a plurality of physiological signals measured by the plurality of physiological sensors 220.

[0207] The system 200b can further be configured to input the plurality of physiological signals into a trained machine learning model 255 and output, using the trained machine learning model 255, a physiological measurement. Optionally, the computing device and / or wearable device can include a display 260. The display can be configured to receive the physiological measurement and display the physiological measurement to the user. As a non-limiting example, the trained machine learning model 255 can be trained and / or deployed using the methods described in FIGS. 7A and 7C.

[0208] In some implementations, the system 200b can further include a portable electronic device 230 operably coupled to the wearable device 210, where the portable electronic device 230 is configured to: receive the plurality of physiological signals from the plurality of physiological sensors 220; and transmit the plurality of physiological signals to the computing device 250. Optionally, the computing device 250 can transmit the physiological measurements to the portable electronic device 230. In some implementations of the present disclosure, the trained machine learning model 255 can be included in the portable electronic device 230.

[0209] The example system 200b can be configured to perform any of the methods described with reference to FIG. 7A-7C, for example by using a trained machine learning model to output a calibration value for the blood pressure estimated based on a PPG signal, and / or using a trained machine learning model to output a blood pressure estimate that compensates for motion, position, and / or orientation of the location where the PPG signal is collected, configured measure a PPG signal by the optical sensor 222 and output a calibrated blood pressure value based on the position determined by the position sensor. Optionally, the calibrated blood pressure value can be determined by using a machine learning model trained according to the methods described herein.

[0210] In some implementations, the portable electronic device 230 can include a display 270, and the portable electronic device 230 can be configured to display physiological measurements including the blood pressure and / or corrected blood pressure measurements described herein on a graphical user interface (GUI).

[0211] With reference to FIG. 2C, implementations of the present disclosure include systems for performing physiological measurements using wearable devices. In some implementations of the present disclosure, the methods 800, 820 shown in FIGS. 8A and 8B can be implemented as computer- implemented methods, or using computer-readable media. For example, in the system 200c shown in FIG. 2C, the methods (and computer- implemented methods) of training and deploying machine learning models described herein can be implemented using the computing device 250 and / or the portable electronic device 230.

[0212] An example system 200c can include a wearable device 210 including one or more physiological sensors 220. The physiological sensors 220 can include any of the physiological sensors described herein, including, for example, PPG sensors. Additional non-limiting examples of physiological sensors that can be used in implementations of the present disclosure include optical sensors, electrocardiogram (ECG) electrodes and sensors, temperature sensors, SpO2 (oxygen saturation) sensors, impedance cardiography (ICG) electrodes, pressure sensors, accelerometers and / or gyroscopes, ultrasonic transducers, skin conductance sensors, phonocardiogram sensors, ballistocardiography sensors, piezoelectric / piezoresistive sensors, ultrasound sensors, arterial tonometry sensors, and / or capacitive respiration sensors.

[0213] In the example system shown in FIG. 2C, the physiological sensors 220 include an optica] sensor 222, temperature sensor 224, and pressure sensor 226. The example system 200c is configured measure a PPG signal by the optical sensor 222 and output a calibrated blood pressure value based on both the temperature sensor 224 and pressuresensor 226. However, it should be understood that in some implementations the physiological sensors 220 could include only the optical sensor and temperature sensor 224, or only the optical sensor and pressure sensor 226. Alternatively or additionally, it should be understood that any number or combination of the physiological sensors described herein could be included in the physiological sensors 220.

[0214] FIG. 2D illustrates an example of the wearable device 210 deployed on a skin surface of a subject. The example vascular bed is beneath the skin and includes an artery. Optionally, the vascular bed can include any vascular tissue (veins, capillaries, etc.) In the example shown in FIG. 2D, the optical sensor 222 is a PPG sensor as described herein that illuminates the skin, vascular bed, and artery and measures reflected light over time (a PPG signal) and can thereby detect that the artery is changing in size as the heart pumps blood through the artery. As shown in FIG. 2D, if the wearable device 210 and / or wristband compress the skin, the vascular bed and / or artery can also be compressed, which can compress the artery and change how the artery reacts to changes in blood pressure. Likewise, as shown in FIG. 2D, the wearable device 210 and / or wristband can cover a skin surface preventing heat from escaping the skin, while also causing the skin to be heated due to heat generated by the device (e.g., the heat generated by the physiological sensors 220). The heating of the vascular bed and / or artery can also change how the artery responds to changes in blood pressure, affecting the PPG signal received by the optical sensor.

[0215] The example system can further include a computing device 250 operably coupled to the wearable device 210. It should be understood that the computing device 250 and wearable device 210 can be coupled using any combination of wireless and / or wired connections, including Bluetooth, Wi-Fi, cellular / LTE, etc. Additionally, it should be understood that the computing device can be part of the wearable device in some implementations of the present disclosure. Alternatively or additionally, the computing device can be separate from the wearable device (e.g., a server). Optionally, the computing device 250 can be a cloud server.

[0216] The computing device 250 can include any of the elements shown and described with reference to the computing device illustrated in FIG. 4.

[0217] The computing device 250 can include at least one processor and memory (not shown), where the includes computer-executable instructions that, cause the processor to implement any of the methods described herein (e.g., the methods 800, 820) described with reference to FIGS. 8 A and 8B).

[0218] For example, the system 200d can be configured to receive a plurality of physiological signals measured by the plurality of physiological sensors 220.

[0219] The system 200d can further be configured to input the plurality of physiological signals into a trained machine learning model 255 and output, using the trained machine learning model 255, a physiological measurement. Optionally, the computing device and / or wearable device can include a display 260. The display 260 can be configured to receive the physiological measurement and display the physiological measurement to the user. The trained machine learning model 255 can be trained and / or deployed using the methods described in FIGS. 8 A and 8B.

[0220] In some implementations, the system 200d can further include a portable electronic device 230 operably coupled to the wearable device 210, where the portable electronic device 230 is configured to: receive the plurality of physiological signals from the plurality of physiological sensors 220; and transmit the plurality of physiological signals to the computing device 250. Optionally, the computing device 250 can transmit the physiological measurements to the portable electronic device 230. In some implementations of the present disclosure, the trained machine learning model 255 can be included in the portable electronic device 230.

[0221] In some implementations, the portable electronic device 230 can include a display 270, and the portable electronic device 230 can be configured to display the physiological measurement on a graphical user interface (GUI).

[0222] Example Implementations: Classifying Waveforms by trained machine learning models

[0223] In some implementations, implementations of the present disclosure include methods and / or computer implemented methods of training machine learning models to select an algorithm for converting a physiological waveform (e.g., PPG signal) to a physiological measurement (e.g., blood pressure). The example implementations further include methods of using the trained machine learning models to estimate blood pressure based on PPG signals.

[0224] FIG. 1 A illustrates an example of using a trained machine learning model in inference mode to select an algorithm from a set of algorithms. With reference to FIG.1 A, a method 100 for determining a physiological measurement from a physiological sensor is shown. Optionally the physiological measurement can include a measurement of blood pressure.

[0225] At step 110 the method can include receiving a physiological signal from the physiological sensor. Optionally, the physiological sensor is a photoplethysmography (PPG) sensor, but any of the physiological sensors described herein can be used, and / or any of the physiological sensors described herein can be used in combination (as described with reference to FIG. IB).

[0226] At step 120, the method can include inputting the physiological signal into a trained machine learning model.

[0227] Non- limiting examples of machine learning models include recurrent neural networks (RNN) and long short term memory networks (LSTM). Alternatively or additionally, the trained machine learning model can include one or more classifiers, as described herein.

[0228] At step 130 the method can include outputting, using the trained machine learning model, an algorithm from a set of algorithms. For example, the trained machine learning model can include classifiers that are configured (e.g., trained according to the methods disclosed herein) to select an algorithm from the set of algorithms configured to estimate or determine a physiological measurement.

[0229] The set of algorithms can optionally include algorithms configured to estimate or determine physiological measurements for patients with different health states. For example, patients can have different heart conditions (e.g., aortic valve stenosis) and / or different levels of arterial stiffness. Patients with different health states can have the same blood pressure even when their PPG waveforms are different.

[0230] In an example implementation, the algorithms include a set of algorithms that estimate blood pressure based on PPG signals, and the machine learning classifier is configured to select an algorithm from the set of algorithms by classifying the patient’ s PPG signal and selecting the corresponding algorithm from the set of algorithms.

[0231] In some implementations, the method 100 can further include determining a physiological measurement based on the algorithm and the physiological signal.

[0232] In some implementations, the method 100 can further include inputting a patient health state and / or activity profile into the trained machine learning model as an alternative or addition to the physiological signal.

[0233] In some implementations, the method 100 can further include inputting environmental data into the trained machine learning model.

[0234] With reference to FIG. IB, implementations of the present disclosure include methods of determining a physiological measure from a plurality of physiologicalsensors. The plurality of physiological sensors can include any of the sensors described herein, including those described with reference to FIG. 1 A and FIGS. 2A-2D.

[0235] An example method 150 is shown in FIG. IB. At step 160 the method can include receiving a plurality of physiological signals from a plurality of physiological sensors.

[0236] At step 170 the method can include inputting the plurality of physiological signals into a trained machine learning model. Optionally, the plurality of signals can include a PPG signal (i.e., a signal acquired from a PPG sensor). Alternatively or additionally, the plurality of physiological signals can correspond to signals and / or combinations of signals acquired from any of the sensors described herein, (e.g., with reference to FIG. 1A and FIG. 2A). For example, an ECG sensor can output an ECG signal, which can be included in the plurality of physiological signals. As another example, an ICG sensor can output an ICG signal, which can be included as an alternative to, or in addition to, the ECG signal and / or PPG signal (and so on).

[0237] At step 180 the method can include outputting, using the trained machine learning model, a physiological measurement. A non-limiting example of a physiological measurement is blood pressure.

[0238] Optionally, the method 150 can further include inputting each of the plurality of physiological signals into a plurality of corresponding sensor-specific trained machine learning models. As a non-limiting example, if the plurality of signals include an ICG signal, an ECG signal, and PPG signal, the method can input the ICG signal into an ICG- specific trained machine learning model, the ECG signal into an ECG-specific trained machine learning model, and the PPG signal into a PPG-specific trained machine learning model. It should be understood that any number of signals can be included in the plurality of signals, so that any number of sensor-specific machine learning models can be included in implementations of the present disclosure.

[0239] In some implementations of the present disclosure, the trained machine learning models can include one or more recurrent neural networks. Alternatively or additionally, the trained machine learning models can include one or more long short term memory networks.

[0240] In some implementations, the method 150 can further include inputting a patient health state, and / or environmental data into any or all of the plurality of trained machine learning models. The patient health state and environmental data can include any of the information described with reference to the patient health state and environmental data of FIG. 1A.

[0241] Example Implementations: Estimating blood pressure using optical data, calibration data, and / or selected algorithms.

[0242] Implementations of the present disclosure include systems and methods to train machine learning models to estimate physiological measurements of a subject (e.g., blood pressure) based on sensor data (e.g., PPG or optical sensor data). The machine learning models can be further calibrated based on the subject’s (e.g., a user of a wearable device) attributes (e.g., age, health conditions, other demographic information, etc.) as well as environmental conditions (e.g., temperature, time of year, altitude, etc.). These systems can benefit patients by providing efficient and non-invasive monitoring of blood pressure and other physiological measurements.

[0243] With reference to FIG. 6 A, a method 600 for training a machine learning model according to implementations of the present disclosure is shown.

[0244] At step 602, the method includes receiving a plurality of PPG signals.

[0245] At step 604 the method includes receiving a plurality of corresponding groundtruth blood pressure measurements.

[0246] At step 606 the method includes receiving a plurality of corresponding selected blood pressure algorithms wherein the plurality of corresponding selected blood pressure algorithms each provide an estimate of blood pressure based on a respective PPG signal and wherein a corresponding selected blood pressure algorithm for each respective PPG signal provides an estimate of blood pressure that is closest to the corresponding groundtruth blood pressure measurement of for the respective PPG signal.

[0247] As used herein, the “corresponding selected blood pressure algorithm” is the blood pressure algorithm that provides the estimate of blood pressure that is closest to the ground truth blood pressure. Variations between different patients, and variation between the same patient over time, means that a single algorithm may not produce the most accurate blood pressure for all patients, or the same patient at all times. Training a machine learning module to select algorithms based on the PPG input can therefore increase the accuracy of blood pressure estimation by selecting different algorithms over time for the same patient, and / or different algorithms for different patients.

[0248] In some implementations, each PPG signal can correspond to one of the groundtruth blood pressure measurements, so that the machine learning model can be trained using a dataset that includes one or more: ground-truth blood pressure, an algorithm that estimates blood pressure using a PPG signal, and the PPG signal that is used to estimate the blood pressure using the algorithm. Thus, the systems and methods described herein can compare the blood pressure estimates produced by different algorithms to a groundtruth blood pressure in order to determine which algorithm produces the most accurate blood pressure estimates for different types of input PPG signals. As described throughout the present disclosure, machine learning models can be trained to select the most accurate blood pressure estimation algorithm for a new input PPG signal, where the ground truth blood pressure for the new input PPG signal is unknown.

[0249] At step 608 the method includes creating a training set comprising the plurality of PPG signals and the respective selected blood pressure algorithm for the plurality of PPG signals.

[0250] At step 610 the method includes training a machine learning model using the training set, wherein the trained machine learning model is configured to identify a corresponding selected a blood pressure algorithm for a patient based on a test PPG signal measured from the patient.

[0251] As non-limiting examples, the trained machine learning model can be a recurrent neural network or a long short term memory (LSTM) model in implementations of the present disclosure, or any of the other machine learning models described herein.

[0252] In some implementations, the method can further include receiving sets of calibration data values, where each of the sets of calibration data values correspond to at least one PPG signal from the plurality of PPG signals, and where the training set comprises the sets of calibration data values. Optionally, the calibration data values can include one or more patient health states and / or one or more activity profiles. The patient health states can include physiological parameters of the patient.

[0253] With reference to FIG. 6B, a method 620 of estimating blood pressure using trained machine learning models is shown. The method can optionally be implemented using the system 200a shown in FIG. 2A. In some implementations of the present disclosure, the model is a model trained according to the methods described with reference to FIG. 6A, FIG. 6C and / or FIG. 6D.

[0254] At step 622, the method includes receiving optical data from an optical sensor (e.g., the physiological sensor 220 shown in FIG. 2 A can be an optical sensor).

[0255] At step 624, the method includes receiving calibration data. Optionally the calibration data can be received from a portable electronic device 230 shown in FIG. 2A. Alternatively or additionally, the calibration data can be received from the computing device 250. In some implementations, the display 260 and / or display 270 can be configured to input calibration data used by the system to implement the methods described herein, including the method 620 of FIG. 6B.

[0256] Optionally, the calibration data values include a plurality of patient health states and / one or more activity profiles. The patient health states can include physiological parameters of the patient. Non-limiting examples of physiological parameters of the patient that can be used as patient health states in implementations of the present disclosure include age, gender, weight, ethnicity, health history, and / or any combination thereof.

[0257] In some implementations of the present disclosure, the calibration data values can include one or more environmental data values. Non-limiting examples of environmental data values can include location data, altitude, time zone changes, temperature and / or combinations thereof.

[0258] At step 626, the method includes: deploying a trained machine learning model, where the trained machine learning model is configured to output an algorithm for estimating blood pressure based on optical data. As shown and described with reference to FIGS. 2A-2D, the trained machine learning model 255 can be deployed to a computing device 250 operably coupled to a wearable device 210, and / or the trained machine learning model 255 can be deployed to a portable electronic device 230 operationally connected to the wearable device 210.

[0259] Again, as non-limiting examples, the trained machine learning model can be a recurrent neural network or a long short term memory (LSTM) model in implementations of the present disclosure.

[0260] At step 628, the method includes selecting, by the trained machine learning model, a selected algorithm from a plurality of algorithms based on the calibration data and the optical data, where the selected algorithm is the algorithm chosen to be closest to ground truth blood pressure for the optical data and calibration data.

[0261] In some implementations of the present disclosure, the plurality of algorithms each correspond to a physical state of a patient, and the method 620 can optionally further include determining a disease state of the patient based on the selected algorithm. In some implementations of the present disclosure, the algorithms from the plurality of algorithms can each correspond to one or more disease states, and / or one or more disease states can correspond to one or more algorithms from the plurality of algorithms. Therefore the disease state or states of a patient can be determined by the selected algorithm.

[0262] It should be understood that the physical state of a patient can include states other than disease states. Additional non-limiting examples of physical states include: posture,motion, and activity of the patient’s body, as well as the position and orientation of the patient’s body in space.

[0263] At step 630, the method can further include includes estimating blood pressure based on the optical data, the calibration data, and the selected algorithm.

[0264] With reference to FIG. 6C, a method 650 for training a trained machine learning model using additional calibration data and / or optical data is shown according to implementations of the present disclosure. The method 650 can be used to individualize or refine the trained machine learning model based on calibration data and / or PPG signals that correspond to a specific user. In some implementations of the present disclosure, a system (e.g., the system 200a) can be configured so that it includes a deployed trained machine learning model (e.g., the trained machine learning model 255 shown in FIG. 2A) that is trained based according to the methods described in FIG. 6A. When the system is used with an individual user, the method shown in FIG. 6C can be used to further individualize or adapt the trained machine learning model to a specific user to increase the accuracy of the physiological measurements.

[0265] At step 652, the method includes receiving optical data from an optical sensor.

[0266] At step 654, the method includes receiving calibration data.

[0267] At step 656, the method includes deploying a trained machine learning model, where the trained machine learning model is configured to output an algorithm for estimating blood pressure based on optical data. Optionally, the trained machine learning model is an LTSM model.

[0268] At step 658, the method includes selecting, by the trained machine learning model, a selected algorithm from a plurality of algorithms based on the calibration data and the optical data, wherein the selected algorithm is the algorithm chosen to estimate a respective blood pressure closest to a ground truth blood pressure for the optical data and calibration data;

[0269] At step 660, the method includes receiving second optical data from the optical sensor.

[0270] At step 662, the method includes receiving second calibration data; and

[0271] At step 664, the method includes training the trained machine learning model using the second optical data and the second calibration data.

[0272] Optionally, the calibration data and / or second calibration data can include patient health states and / one or more activity profiles. The patient health states can include physiological parameters of the patient. Non-limiting examples of physiological parameters of the patient that can be used as patient health states in implementations ofthe present disclosure include age, gender, weight, ethnicity, health history, and / or any combination thereof. Optionally, the calibration data can include any number of physiological measurements and / or physiological signals, where the physiological measurements and / or physiological signals can be acquired by any number of physiological sensors.

[0273] In some implementations of the present disclosure, the calibration data and / or second calibration data values can include a one or more environmental data values. Non- limiting examples of environmental data values can include location data, altitude, time zone changes, temperature and / or combinations thereof.

[0274] With reference to FIG. 6D, a method 670 for training a machine learning model is shown.

[0275] At step 672, the method includes receiving sensor data. The sensor data can be data collected by any or all of the sensors described herein.

[0276] At step 674, the method includes receiving a plurality of corresponding groundtruth physiological measurements. Non-limiting examples of ground-truth physiological measurements include any or all of: pulmonary capillary wedge pressure; left atrial pressure; left ventricular ejection fraction derived from ultrasound; RV volume; LV volume; blood glucose; imaging data (e.g., CT, MR, X-ray, optical imaging, etc.), and / or heart rate.

[0277] At step 676, the method includes receiving a plurality of corresponding selected physiological estimation algorithms where the plurality of corresponding selected physiological estimation algorithms each provide an estimate of a physiological measurement based on a respective sensor signal and wherein the selected physiological estimation algorithm for each respective sensor signal provides an estimate of the physiological measurement that is closest to the corresponding ground-truth physiological measurement for the respective sensor signal. The physiological estimation algorithms can be algorithms that estimate any physiological measurement. In various implementations of the present disclosure, the physiological estimation algorithm can be configured to estimate any or all of: pulmonary capillary wedge pressure; left atrial pressure; left ventricular ejection fraction derived from ultrasound; RV volume; LV volume; blood glucose; imaging data (e.g., CT, MR, X-ray, optical imaging, etc.), and / or heart rate.

[0278] Optionally the estimation algorithms can be algorithms that estimate the physiological measurements that are received at step 674. For example, the estimation algorithm can be an algorithm that relates any combination of the of the sensor datadescribed with reference to step 672 to any or all of the physiological measurements described with reference to step 674.

[0279] At step 678, the method includes creating a training set comprising the sensor data and the respective selected physiological measurement algorithm for the plurality of sensor signals.

[0280] At step 680, the method includes training a machine learning model using the training set, wherein the trained machine learning model is configured identify a corresponding selected physiological estimation algorithm for a patient based on a test sensor signal measured from the patient. The test sensor signal can be a sensor signal measured from any of the sensors described herein, for example with reference to steps 672 and 674.

[0281] Example Implementations: Estimating blood pressure using optical data and position data, estimating body position

[0282] With reference to FIG. 7A, implementations of the present disclosure include computer implemented methods of training machine learning models to estimate a proportion of a blood pressure measurement that is caused by gravity (e.g., a “gravity contribution” to the blood pressure measurement). Hydrostatic pressure is proportional to the depth of the fluid multiplied by the density of the fluid and the acceleration of gravity. Therefore a measurement of blood pressure can be increased or decreased by raising or lowering the measurement point relative to the heart, which affects the hydrostatic pressure of the blood and therefore the overall blood pressure measured.

[0283] At step 710, the computer implemented method includes accessing training PPG waveform data. The training data can include any number of PPG waveforms collected from any number of individuals.

[0284] At step 712 the computer implemented method includes accessing ground truth body position measurements. Optionally, the ground truth body position measurements can include measurements of where the PPG signal was recorded relative to the heart of the individual. Optionally, the ground truth body position measurements can include measurements of a changing body position over time. Optionally the ground truth body position measurement can include any combination of position, orientation, acceleration, and velocity for any number of body parts. For example, the ground truth body position measurement can include a height above or below the heart.

[0285] At step 714 the computer implemented method includes accessing ground truth blood pressure measurements. The ground truth blood pressure measurements can optionally include measurements of a blood pressure collected using a standardizedmeasurement. Alternatively or additionally, the ground truth blood pressure measurements can include measurements of blood pressure collected from parts of a body that are not moving. For example, if the corresponding PPG waveform is collected from a patient’s right wrist that is moving, the ground truth blood pressure can be collected from a patient’ s stationary left arm by a blood pressure cuff, for example.

[0286] At step 716, the corresponding ground truth blood pressure measurements, ground truth body position measurements, and PPG waveform data can be used to create a training data set. The training dataset can associate the PPG waveform collected from any number of individuals with the measured ground truth body position and ground truth blood pressure collected. The training dataset can therefore include both a physiological signal that blood pressure can be derived from (the PPG signal) as well as the actual blood pressure (the ground truth blood pressure) and the actual body position that the PPG signal was recorded from. Thus, the training dataset created at step 716 allows the relationships between body position and both PPG signal and blood pressure to be used to train a machine learning model to receive a PPG input and estimate both body position and / or ground truth blood pressure.

[0287] At step 718 the computer implemented method includes training a machine learning model, using the training dataset to estimate a gravity contribution in input PPG training waveforms. Optionally, the gravity contribution can represent an estimate of the amount of blood pressure and / or fraction of blood pressure that is caused by gravity. The gravity contribution can represent an “offset” from the blood pressure that would be measured at a standardized location on a patient’s body (e.g., the brachial artery of the upper arm). The trained machine learning model can therefore improve wearable devices by allowing the wearable device to determine the gravity contribution to a blood pressure estimate generated by the wearable device, and then adjusting the final blood pressure output by the wearable device by the gravity contribution to provide a blood pressure that compensates for position, orientation and motion of the wearer. This allows for wearable devices to record an estimate of blood pressure that is corrected for the motion / position of the wearer, and therefore represents a more accurate measure of the wearer’ s health throughout the day or over time.

[0288] In some implementations of the present disclosure, the respective gravity contribution includes an estimate of a column pressure caused by gravity. For example, in implementations of the present disclosure configured to estimate a body position, the column pressure can be estimated by the height of the wearable device, the estimated density of blood, and the acceleration of gravity (e.g., 9.81 m / s2).

[0289] Optionally, implementations of the present disclosure include training machine learning models based on the estimated density of blood. As described above, hydrostatic pressure is proportional to fluid density. In some implementations of the present disclosure, density of blood is considered a constant (e.g., approximately .994 g / mL).

[0290] In some implementations of the present disclosure, the training data set can include additional data corresponding to the PPG waveform data collected at step 710. Non-limiting examples of additional types of training data that can be optionally used in implementations of the present disclosure include activity profiles (e.g., representing physical activity corresponding to the PPG waveform data) and patient health states (e.g., information about the health of the individuals the PPG waveform data was measured from). Additional examples of patient health states include physiological parameters, and non-limiting examples of physiological parameters include age, gender, weight, ethnicity, or health history.

[0291] In some implementations of the present disclosure, the training data can further include environmental data values corresponding to the PPG waveform data received at step 710. Non-limiting examples of environmental data values include location data, altitude, time zone changes or temperature.

[0292] With reference to FIG. 7B, implementations of the present disclosure include computer implemented methods of training a machine learning model to estimate a body position based on a PPG waveform.

[0293] At step 720, the method includes accessing a plurality of training PPG waveform data.

[0294] At step 722, the method includes accessing a plurality of corresponding ground truth blood pressure measurements.

[0295] At step 724, the method includes accessing a plurality of physiological parameters corresponding to each of the ground truth blood pressure measurements. Non-limiting examples of physiological parameters include subject body dimensions, subject blood composition, subject blood density, subject vascular anatomy, and subject heart position.

[0296] At step 726, the method includes training a machine learning model, using the training PPG waveform data, the ground truth blood pressure measurements, and the plurality of physiological parameters, to output an estimated body position corresponding to the ground truth blood pressure measurements based on a respective training PPG waveform.

[0297] Training the machine learning model can include configuring the machine learning model to identify patterns in combinations of the plurality of physiological parameters corresponding to respective ground truth blood pressure measurements.

[0298] Optionally, the estimated body position at step 726 can be output based on determining a blood pressure change due to gravity, and estimating, based on the blood pressure change due to gravity, a height that the PPG data was collected from. Optionally, outputting the estimated body position comprises estimating a gravity contribution to the respective training PPG waveform can include calculating an estimated height dimension at which the ground truth blood pressure measurements were gathered. In some implementations of the present disclosure, the training data can include the body dimensions of the one or more subjects that the PPG data is collected.

[0299] With reference to FIG. 7C, implementations of the present disclosure further include methods of using trained machine learning models to estimate blood pressure, gravity contributions to a blood pressure measurement (e.g., a blood pressure measurement estimated by PPG) and / or a body position.

[0300] At step 730, the example method includes the example method includes receiving an input PPG signal from a PPG sensor located on a body part of a subject.

[0301] At step 732, the example method includes receiving a trained blood pressure prediction machine learning model. The trained blood pressure prediction machine learning model can include a model trained according to any of the methods described with reference to FIGS. 7A and 7B.

[0302] At step 734, the example method includes estimating, using the blood pressure prediction machine learning model, a blood pressure estimate. For example, the blood pressure estimate can include estimates of blood pressure derived using a PPG sensor and / or other sensor located on a wearable device. Optionally, the blood pressure estimate is obtained using a trained machine learning model configured to estimate a blood pressure based on a PPG signal or PPG waveform.

[0303] At step 736, the example method includes estimating, using the trained blood pressure correction model, blood pressure correction data. As described with reference to FIGS. 7A and 7B, blood pressure correction data can optionally include a proportion or amount of the measured blood pressure that is caused by gravity, motion, and / or hydrostatic pressure. As described in additional detail with reference to FIGS. 7A-7B, a wearable device can measure blood pressure continuously when the wearable device is at different positions relative to the heart, but those continuous blood pressure measurements represent the blood pressure at the point where the wearable devicemeasures, not the blood pressure that would be measured at a standard location (e.g. the brachial artery). The blood pressure correction data can represent the difference between the blood pressure estimate or measurement obtained by the wearable device and the blood pressure that would be obtained by a blood pressure cuff located at a the standard location on the body. Therefore the blood pressure correction data can be used to estimate the patient’s standardized blood pressure even when the measurements or estimates of blood pressure are taken from a location other than a standard location on the patient’s body.

[0304] At step 738, the example method includes determining, based on the blood pressure correction data and the blood pressure estimate, a corrected blood pressure. As described with respect to step 736, the blood pressure correction data can represent an amount of pressure or a proportion of pressure in a blood pressure estimate that is estimated to be caused by the position of the wearable device relative to the heart and / or relative to a standard measure of blood pressure.

[0305] Example Implementation: Estimating blood pressure including temperature and pressure calibration

[0306] Example implementations of the present disclosure further include systems, devices, and methods configured to perform temperature and pressure calibration of blood pressure estimates.

[0307] For example, any of the systems, methods, and devices described herein can be configured to calibrate blood pressure estimates performed by a wearable device to compensate for effects of skin moisture, skin temperature, and / or the pressure of the wearable device. Optionally, the systems and methods described herein can measure skin temperature, pressure exerted by the wearable device, and / or skin moisture / sweat (e.g., galvanic skin response). The measurements of temperature, pressure, and / or skin moisture / sweat can be used as additional inputs to train the machine learning models described herein to estimate blood pressure and / or select algorithms for estimating blood pressure. Alternatively or additionally, the measurements of temperature, pressure, and / or skin moisture / sweat can be used as inputs to a trained machine learning model to calibrate an estimate of blood pressure, and / or as inputs to an algorithm configured to estimate blood pressure,

[0308] With reference to FIG. 8A, a method 800 for training a machine learning model according to implementations of the present disclosure is shown.

[0309] At step 802, the method includes receiving a plurality of PPG signals.

[0310] At step 804 the method includes receiving a plurality of corresponding groundtruth blood pressure measurements.

[0311] At step 806, the method includes receiving corresponding calibration data for each of the plurality of PPG signals. The calibration data can optionally include temperature data (e.g., temperature values) and / or pressure data (e.g., pressure values). Optionally, the PPG signals of the plurality of PPG signals are collected from a vascular bed of a subject, and the calibration data is calibration data collected from the same vascular bed or tissue near the same vascular bed. For example, an optical sensor used to collect the PPG signal can illuminate an area of skin to measure a PPG signal from a vascular bed, and the temperature data and / or pressure data can be collected from the same vascular bed so that the temperature sensor measures the temperature of the skin near the vascular bed and the pressure sensor measures the pressure applied to the vascular bed.

[0312] At step 808 the method includes receiving a plurality of corresponding selected blood pressure algorithms wherein the plurality of corresponding selected blood pressure algorithms each provide an estimate of blood pressure based on a respective PPG signal and wherein a corresponding selected blood pressure algorithm for each respective PPG signal provides an estimate of blood pressure that is closest to the corresponding groundtruth blood pressure measurement of for the respective PPG signal.

[0313] As used herein, the “corresponding selected blood pressure algorithm” is the blood pressure algorithm that provides the estimate of blood pressure that is closest to the ground truth blood pressure. Variations between different patients, and variation between the same patient over time, can have the effect that a single algorithm may not produce the most accurate blood pressure for all patients, or the same patient at all times. Training a machine learning module to select algorithms based on the PPG input can therefore increase the accuracy of blood pressure estimation by selecting different algorithms over time for the same patient, and / or different algorithms for different patients.

[0314] In some implementations, each PPG signal can correspond to one of the ground-truth blood pressure measurements, so that the machine learning model can be trained using a dataset that includes one or more: ground-truth blood pressure, an algorithm that estimates blood pressure using a PPG signal, and the PPG signal that is used to estimate the blood pressure using the algorithm. Thus, the systems and methods described herein can compare the blood pressure estimates produced by different algorithms to a ground truth blood pressure in order to determine which algorithm produces the most accurate blood pressure estimates for different types of input PPGsignals. As described throughout the present disclosure, machine learning models can be trained to select the most accurate blood pressure estimation algorithm for a new input PPG signal (e.g., a PPG signal acquired using a wearable device), where the ground truth blood pressure for the new input PPG signal is unknown.

[0315] It should be understood that temperature and pressure are only non-limiting examples of calibration data values, and that in some implementations of the present disclosure, the calibration data values can include one or more environmental data values. Non- limiting examples of environmental data values can include location data, altitude, time zone changes, temperature and / or combinations thereof. In some implementations, the method can further include receiving sets of calibration data values, where each of the sets of calibration data values correspond to at least one PPG signal from the plurality of PPG signals, and where the training set comprises the sets of calibration data values. Optionally, the calibration data values can include one or more patient health states and / or one or more activity profiles. The patient health states can include physiological parameters of the patient. Non-limiting examples of physiological parameters of the patient that can be used as patient health states in implementations of the present disclosure include age, gender, weight, ethnicity, health history, and / or any combination thereof.

[0316] At step 810 the method includes creating a training set including the plurality of PPG signals and the respective selected blood pressure algorithm for the plurality of PPG signals.

[0317] At step 812 the method includes training a machine learning model using the training set, wherein the trained machine learning model is configured to identify a corresponding selected a blood pressure algorithm for a patient based on a test PPG signal measured from the patient.

[0318] As non- limiting examples, the trained machine learning model can be a recurrent neural network or a long short term memory (LSTM) model in implementations of the present disclosure.

[0319] With reference to FIG. 8B, a method 820 of estimating blood pressure using trained machine learning models is shown. The method can optionally be implemented using the system 200c shown in FIG. 2C. In some implementations of the present disclosure, the model is a model trained according to the methods described with reference to FIG. 8A.

[0320] At step 822, the method includes receiving optical data from an optical sensor (e.g., the physiological sensor 220 shown in FIG. 2C can be an optical sensor).

[0321] At step 824, the method includes receiving calibration data. Optionally the calibration data can be received from a portable electronic device 230 shown in FIG. 2C. Alternatively or additionally, the calibration data can be received from the computing device 250. In some implementations, the display 260 and / or display 270 can be configured to input calibration data used by the system 200c to implement the methods described herein, including the method 820 of FIG. 8B.

[0322] As described herein, the calibration data can optionally include temperature and / or pressure values. Optionally, as shown in FIG. 2C, the optical sensor is disposed over (and / or against) a vascular bed and the calibration data includes a temperature value of the vascular bed. Alternatively or additionally, as shown in FIGS. 2D, the calibration data can optionally include a pressure sensed by a pressure sensor disposed over the same vascular bed as the optical sensor.

[0323] In some implementations of the present disclosure the calibration data values include a plurality of patient health states and / one or more activity profiles. The patient health states can include physiological parameters of the patient. Non-limiting examples of physiological parameters of the patient that can be used as patient health states in implementations of the present disclosure include age, gender, weight, ethnicity, health history, and / or any combination thereof.

[0324] In some implementations of the present disclosure, the calibration data values can include one or more environmental data values. Non-limiting examples of environmental data values can include location data, altitude, time zone changes, temperature and / or combinations thereof.

[0325] At step 826, the method includes deploying a trained machine learning model, where the trained machine learning model is configured to output an algorithm for estimating blood pressure based on optical data. As shown and described with reference to FIG. 2C, the trained machine learning model 255 can be deployed to a computing device 250 operably coupled to a wearable device 210, and / or the trained machine learning model 255 can be deployed to a portable electronic device 230 operationally connected to the wearable device 210.

[0326] Again, as non-limiting examples, the trained machine learning model can be a recurrent neural network or a long short term memory (LSTM) model in implementations of the present disclosure.

[0327] At step 828, the method includes selecting, by the trained machine learning model, a selected algorithm from a plurality of algorithms based on the calibration data and the optical data, where the selected algorithm is the algorithm chosen to be closest toground truth blood pressure for the optical data and calibration data. As described with reference to FIG. 8A, trained machine learning model can be trained based on optical data and calibration data to select a blood pressure algorithm for determining blood pressure based on the optical data and calibration data. The trained machine learning model can optionally be trained to select the blood pressure algorithm that is most accurate (e.g., the trained machine learning model minimizes a measure of error).

[0328] In some implementations of the present disclosure, the plurality of algorithms each correspond to a physical state of a patient, and the method 820 can optionally further include determining a disease state of the patient based on the selected algorithm. In some implementations of the present disclosure, the algorithms from the plurality of algorithms can each correspond to one or more disease states, and / or one or more disease states can correspond to one or more algorithms from the plurality of algorithms. Therefore the disease state or states of a patient can be determined by the selected algorithm.

[0329] It should be understood that the physical state of a patient can include states other than disease states. Additional non-limiting examples of physical states include: posture, motion, and activity of the patient’s body, as well as the position and orientation of the patient’ s body in space.

[0330] At step 830, the method can further include includes estimating blood pressure based on the optical data, the calibration data, and the selected algorithm.

[0331] Again, while the example implementations described herein are described in the context of sensing blood pressure, it should be understood that the methods described herein can be applied to any kind of estimate of physiological measurements and / or physiological signals, waveforms, etc.

[0332] It should be appreciated that the logical operations described herein with respect to the various figures may be implemented (1) as a sequence of computer implemented acts or program modules (i.e., software) running on a computing device (e.g., the computing device described in FIG. 4), (2) as interconnected machine logic circuits or circuit modules (i.e., hardware) within the computing device and / or (3) a combination of software and hardware of the computing device. Thus, the logical operations discussed herein are not limited to any specific combination of hardware and software. The implementation is a matter of choice dependent on the performance and other requirements of the computing device. Accordingly, the logical operations described herein are referred to variously as operations, structural devices, acts, or modules. These operations, structural devices, acts and modules may be implemented in software, infirmware, in special purpose digital logic, and any combination thereof. It should also be appreciated that more or fewer operations may be performed than shown in the figures and described herein. These operations may also be performed in a different order than those described herein.

[0333] Referring to FIG. 4, an example computing device 400 upon which the methods described herein may be implemented is illustrated. It should be understood that the example computing device 400 is only one example of a suitable computing environment upon which the methods described herein may be implemented. Optionally, the computing device 400 can be a well-known computing system including, but not limited to, personal computers, servers, handheld or laptop devices, multiprocessor systems, microprocessor-based systems, network personal computers (PCs), minicomputers, mainframe computers, embedded systems, and / or distributed computing environments including a plurality of any of the above systems or devices. Distributed computing environments enable remote computing devices, which are connected to a communication network or other data transmission medium, to perform various tasks. In the distributed computing environment, the program modules, applications, and other data may be stored on local and / or remote computer storage media.

[0334] In its most basic configuration, computing device 400 typically includes at least one processing unit 406 and system memory 404. Depending on the exact configuration and type of computing device, system memory 404 may be volatile (such as random access memory (RAM)), non-volatile (such as read-only memory (ROM), flash memory, etc.), or some combination of the two. This most basic configuration is illustrated in FIG. 4 by dashed line 402. The processing unit 406 may be a standard programmable processor that performs arithmetic and logic operations necessary for operation of the computing device 400. The computing device 400 may also include a bus or other communication mechanism for communicating information among various components of the computing device 400.

[0335] Computing device 400 may have additional features / functionality. For example, computing device 400 may include additional storage such as removable storage 408 and non-removable storage 410 including, but not limited to, magnetic or optical disks or tapes. Computing device 400 may also contain network connection(s) 416 that allow the device to communicate with other devices. Computing device 400 may also have input device(s) 414 such as a keyboard, mouse, touch screen, etc. Output device(s) 412 such as a display, speakers, printer, etc. may also be included. The additional devices may be connected to the bus in order to facilitate communication of data among the componentsof the computing device 400. All these devices are well known in the art and need not be discussed at length here.

[0336] The processing unit 406 may be configured to execute program code encoded in tangible, computer-readable media. Tangible, computer-readable media refers to any media that is capable of providing data that causes the computing device 400 (i.e., a machine) to operate in a particular fashion. Various computer-readable media may be utilized to provide instructions to the processing unit 406 for execution. Example tangible, computer-readable media may include, but is not limited to, volatile media, nonvolatile media, removable media and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data. System memory 404, removable storage 408, and non-removable storage 410 are all examples of tangible, computer storage media. Example tangible, computer-readable recording media include, but are not limited to, an integrated circuit (e.g., field-programmable gate array or application-specific IC), a hard disk, an optical disk, a magneto-optical disk, a floppy disk, a magnetic tape, a holographic storage medium, a solid-state device, RAM, ROM, electrically erasable program read-only memory (EEPROM), flash memory or other memory technology, CD- ROM, digital versatile disks (DVD) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices.

[0337] In an example implementation, the processing unit 406 may execute program code stored in the system memory 404. For example, the bus may carry data to the system memory 404, from which the processing unit 406 receives and executes instructions. The data received by the system memory 404 may optionally be stored on the removable storage 408 or the non-removable storage 410 before or after execution by the processing unit 406.

[0338] It should be understood that the various techniques described herein may be implemented in connection with hardware or software or, where appropriate, with a combination thereof. Thus, the methods and apparatuses of the presently disclosed subject matter, or certain aspects or portions thereof, may take the form of program code (i.e., instructions) embodied in tangible media, such as floppy diskettes, CD-ROMs, hard drives, or any other machine-readable storage medium wherein, when the program code is loaded into and executed by a machine, such as a computing device, the machine becomes an apparatus for practicing the presently disclosed subject matter. In the case of program code execution on programmable computers, the computing device generally includes a processor, a storage medium readable by the processor (including volatile andnon-volatile memory and / or storage elements), at least one input device, and at least one output device. One or more programs may implement or utilize the processes described in connection with the presently disclosed subject matter, e.g., through the use of an application programming interface (API), reusable controls, or the like. Such programs may be implemented in a high level procedural or object-oriented programming language to communicate with a computer system. However, the program(s) can be implemented in assembly or machine language, if desired. In any case, the language may be a compiled or interpreted language and it may be combined with hardware implementations .L0339J Additional Example Implementations

[0340] Example 1: Selecting waveform templates for a patient to estimate physiological parameters

[0341] An example implementation of the present disclosure can be used to estimate a patient’ s blood pressure by using algorithms selected based on the arterial valve waveform recorded in the patient. The example implementation of the present disclosure can be used to accurately estimate blood pressures in patients with different arterial valve waveforms. As non-limiting examples, FIG. 3A illustrates an actual arterial valve waveform recorded in a “normal” subject, and FIG. 3B illustrates an arterial waveform for a subject with Aortic Valvular Stenosis. Aortic Valvular Stenosis is a very common condition that occurs with aging and with various degrees of severity.

[0342] The waveforms shown in FIG. 3A and FIG. 3B are different, and it should be understood that many different health states, including the presence or absence of disease, age, and any other demographic traits can result in a large number of different waveforms in a population. Additional examples of health states are described with reference to FIG. 1 A. Systems and methods that use a single algorithm to process the waveforms in FIG. 3A and FIG. 3B can result in inaccurate results for one or both waveforms. Additionally, a filtering algorithm that works for one of the waveforms may filter useful information from the other waveform.

[0343] The example implementation can collect data from different groups. The groups can include patients that are considered “normal,” and patients with specific disease states (e.g., aortic valvular stenosis). The groups can be further divided into different sub-groups (e.g., young and old “normal” patients), data from different groups. The data gathered from different groups can different waveforms that represent different attributes of the circulatory systems of the patients (e.g., patients with different levels of vascular distensibility).

[0344] The different waveforms for the different groups can be stored in a “library” of waveforms and groups. Optionally, the library can further store the relationship between the stored waveforms and physiological attributes of interest (e.g., blood pressure).

[0345] As another non-limiting example, the example implementation can include a library that stores waveforms that represent photoplethysmogram signals for different groups, and further store information that correlates the stored photoplethysmogram waveforms to one or more physiological attributes of interest (e.g., blood pressure).

[0346] In some implementations of the present disclosure, the library can provide a “waveform template” for a patient based on a user input. For example, the patient’s health state and demographic characteristics may be known (e.g., the patient may be a 50 year old man with Aortic Valvular Stenosis). The patient can then be assigned to a group, and the library can provide a waveform template for the patient that can estimate the patient’ s blood pressure based on a photoplethysmogram signal from the patient.

[0347] Alternatively or additionally, the library can be configured to receive a waveform from the patient, and compare the waveform to the waveform templates stored in the library. The system can determine whether the waveform is sufficiently similar to the waveform templates to “fit” one of the waveform templates. IF the waveform “fits” one of the waveform templates, then that waveform template can be provided to be used to estimate the patient’s blood pressure. If the waveform template does not “fit” an existing waveform template, a new waveform template can be created, and / or a “best fit” output can be provided to the patient (e.g., an estimate of blood pressure using a different waveform template).

[0348] Optionally, the system can include providing alerts to the user based on the waveform template provided, or whether or not the waveform from the patient “fits” a waveform template.

[0349] Example 2: Changing Waveform Templates and / or Groups

[0350] Implementations of the present disclosure can be configured to provide new and / or updated waveform templates to a patient. For example, a patient may initially receive a first waveform template when they are 40 years old and in normal health. After several years pass, the patient is older and may develop aortic valvular stenosis. Now the patient may be categorized in a different group. The system can observe the patient’s waveform over time and determine when the patient switches to a new group. Alternatively or additionally, the system can receive a user input (e.g., representing a diagnosis or treatment from a physician) that indicates that the user is in a different group, causing the system to provide a new waveform template. For example, when apatient develops aortic valvular stenosis, the system can detect the development of aortic valvular stenosis by the changes to the patient’s waveform over time. The system can alert the patient of the change. Alternatively or additionally, if the patient is treated (e.g., has surgery to correct the aortic valvular stenosis), then the patient can input that treatment into the system to select a different waveform template.

[0351] Example 3: Multivariate Analysis

[0352] In some implementations of the present disclosure, the system can correlate other available variables with a change in wave form morphology. As a non-limiting example, the system can aggregate data on the patient’s body weight and the patient’s subjective sense of shortness of breath, (e.g., given using a numerical scale). The wave form library contains a wave form template for a particular variant of heart failure. As the patient’ s waveform starts to change, the system initially may not be able to precisely match the change to any disease type. However, the system can also have data for a change in weight and increased shortness of breath. Using a multi-variate approach, the system can output an alert to possible heart failure even if the patient’s waveform is not a sufficiently close match to a waveform in the waveform library. In some implementations of the present disclosure, a 95% similarity is configured sufficiently close, and so in the present example the patient’s waveform may be less than a 95% match for all the waveform templates. It should be understood that any level of similarity can be used in various implementations of the present disclosure, and 95% is provided only as a non-limiting example.

[0353] Example 4: Research.

[0354] Implementations of the present disclosure can be configured to perform research across groups using patient data. For example, the system can be configured to search the entire patient data set and continuously look for correlations between a change in blood pressure and / or wave form template type and other factors. Non-limiting examples of correlations include determining the answer to questions like the following examples: Does a change in medication that occurred on Monday correlate with a drop in blood pressure seen on Friday? Does the patient’s sleep pattern correlate with the 24-hour blood pressure profile?

[0355] Example 5 : W earable devices

[0356] Yet another example implementation of the present disclosure includes computer implemented methods and wearable devices.

[0357] The example implementation supports wristband, finger / fingertip, and smartwatch-based data collection that can optionally be used for healthcare. The exampleimplementation includes a cloud-based algorithm-orchestration design based on an algorithm and signal repository that can hosts multiple algorithms and / or a large collection of physiological signals on cloud. The example implementation allows smart selection of different algorithms and various signal processing techniques will result into patient- specific accurate measurement of blood pressure and other physiological conditions.

[0358] The example implementation can be configured to use a variety of different algorithms and signal processing strategies to translate measurements into clinical values. Optionally, the example implementation can include a cloud-based repository of algorithms, a large collection of signals from real-life subjects, and / or an algorithm- orchestrator.

[0359] The “algorithm-orchestrator” described with reference to the present example can be configured to apply one or more algorithms as applicable to efficiently process raw data and derive clinical values out of it, based on context at a given moment. The algorithm-orchestrator can also factor in any other external parameters. Non-limiting example parameters include time of the day, recent history, body weight, exercise data, location, food & fluid intake, etc. and uses one or more algorithm as applicable.

[0360] The present example also contemplates the use of artificial intelligence and / or machine learning in the algorithm-orchestrator. One or more algorithms used by algorithm-orchestrator can perform signal comparison in near-real-time, perform various signal processing calculations on the cloud and / or provide more accurate physiological measurement to the wearable device via smartphone platform. The algorithm-orchestrator can optionally adapt to a given patient’s specific situation, including parameters such as race, gender, age, disease condition, geographic location, etc. in order to calculate patient specific physiological measurement.

[0361] The example implementation includes an algorithm & signal repository. This repository can contain many algorithms for various data processing and calculations. Optionally, one or more of these algorithms can use the signal repository underneath to perform various signal processing in near-real-time. Optionally, one or more the algorithms are ‘adaptive’ and can have ability to ‘learn’ over period of time. The module called algorithm-orchestrator can be the decision-making unit which decides which algorithm is most appropriate for given user, for given condition.

[0362] Collectively, multiple algorithms and algorithm-orchestrator can work together to pick the most suitable physiological measurement for given patient. In some implementations of the present disclosure, the system can compare an unknown patientPPG waveform and obtaining the best matches from the library that results in a probability of match table presented as a differentia] diagnosis for an unknown waveform. Alternatively or additionally, the system can starting with a known PPG pattern associated with a known disease state, and over time note changes that may better fit with a different pattern in the library (a change in waveform indicating a change in diagnosis) or the designation as “unknown waveform” for which the physician or other person with data access would be notified to provide additional analysis tools, both within the smartwatch and from external sources. FIG. 5 illustrates an example block diagram of the present example implementation.L0363 J The example implementation can further include efficient methodologies to collect physiological signal from volar wrist and archive them into ‘signal repository’.

[0364] Alternatively or additionally, the algorithm repository with a collection of smart algorithms can be capable of performing adaptive signal processing in near-real-time and / or capable of learning variations for given patient.

[0365] Implementations of the present disclosure further include computer- implemented methods to optimally select one or more algorithms from algorithm repository by algorithm-orchestrator and apply them sequentially or parallelly to process the data from single or dual-headed wearable on wrist and other parts of limbs (e.g., fingers or fingertips).

[0366] Implementations of the present disclosure further include morphology match for patient specific waveform with collection of waveforms from real human subjects. This match will provide indication for a disease state or degree of differential diagnosis to the clinician. Such parameters can help clinicians understand the disease progression for a given patient.

[0367] Optionally, the methods and algorithms described herein can be simultaneously applied to multiple sensors on the wristband, smartwatch, and / or any other location (e.g., fingers or fingertips) and can be applied to any combinations of data from multiple physiological sensors.

[0368] It should be understood that, while examples herein describe applications of PPG signals, PPG sensors, and PPG data, and methods of obtaining a blood pressure measurements / estimates, PPG and blood pressure are only non-limiting examples, and the systems and methods describe herein can be used to improve the accuracy and processing of any physiological data from any sensor, to obtain any physiological measurement based on one or more models and algorithms.

[0369] Example 6: Systems and methods for PPG acquisition, processing, and / or blood pressure estimation

[0370] With reference to FIG. 9, an additional example method of obtaining an estimated blood pressure is shown according to implementations of the present disclosure. The system 900 can optionally be implemented using the systems described with reference to FIGS. 2A-2D.

[0371] At step 902, the method includes obtaining PPG signals from one or more PPG sensors. At step 904, the method includes processing the PPG signals. Processing the PPG signals can include modulating the intensity of the light emitted by the PPG sensors (e.g., based on a feedback loop of the absorbed light). FIG. 13 provides an example method of PPG signal processing that can be used with the method of FIG. 9.

[0372] At step 910, the method includes inputting the PPG signal into an algorithm a machine learning model configured to select an algorithm for converting PPG signals into a blood pressure measurement based on the characteristic of the waveforms. At step 912, an algorithm is retrieved based on the selection at step 910 from a waveform library. The output of the selected algorithm provides an estimate of blood pressure. The waveform library can include disease and / or health status-specific algorithms for converting PPG signals to blood pressure.

[0373] At step 906, one or more additional physiological signals can be obtained using additional physiological sensors. At step 908, a data selection step can be performed on the other physiological signals. The data selection step can include filtering, signal processing, aggregating / combining data from multiple sensors, etc.

[0374] At step 920, the method can include performing a calibration step. The calibration step can include inputting the data selected at step 908 into another machine learning model configured to calibrate the output of the algorithm selected from the waveform library at step 912. Alternatively or additionally, the calibration step can include using one or more calibration algorithms (e.g., algorithms configured to calibrate a blood pressure estimate for motion or body part position). At step 930, the calibrated estimate of blood pressure is output.

[0375] With reference to FIG. 10, an example method includes receiving, at step 1002, a PPG signal and at least one of: a skin temperature, a skin tone, a body-mass-index, medical history, and / or any other information about a patient health state. During the training at step 1002, the ground truth blood pressure can optionally also be used as an input to the training of the machine learning model.

[0376] At step 1004, the method further includes using a machine learning model to select a blood pressure algorithm based at least in part on a cohort normal / hypertension, medical history, skin tone, BMI, and / or waveform template matching.

[0377] At step 1006, the method further includes performing at least one of the following operations on the signal received at step 1002: post processing, blood pressure estimation, signal phenotype, and / or trend analysis.

[0378] At step 1008, the method further includes performing subject- specific calibration to improve a blood pressure estimated by the blood pressure algorithm selected by the machine learning model, where the blood pressure algorithm is configured to estimate blood pressure based on a PPG signal.

[0379] At step 1010, the method further includes outputting the estimated blood pressure. The estimated blood pressure can include calibrations performed at step 1008 and 1006. Optionally, outputting the estimated blood pressure can include displaying the blood pressure and / or transmitting the blood pressure to a computing device.

[0380] With reference to FIG. 11, an additional example method is shown according to implementations of the present disclosure. At step 1102, the method includes training, using a computer, a convolutional neural network, and / or long short-term memory machine learning model. The training at step 1102 can be performed based on calibration data (e.g., data acquired by one or more sensors, and / or input by a user), as well as PPG signals, calibration blood pressure (e.g., a blood pressure measured using a blood pressure cuff), a measurement of skin tone, and / or a BMI measurement.

[0381] At step 1104, the method can further include receiving accelerometer, galvanic skin response, skin moisture, electrocardiogram (ECG), and / or impedance cardiography measurement.

[0382] At step 1106, the method can further include calibrating an output of the machine learning mode based on any or all of the data received at steps 1102 and / or 1104.

[0383] At step 1108, the method can further include outputting an estimated blood pressure, where the estimated blood pressure can optionally include the calibrations performed at step 1106.

[0384] Example 7: Systems and methods for PPG acquisition

[0385] Implementations of the present disclosure further include systems and methods for acquiring PPG signals. FIG. 12 illustrates a block diagram of a system for acquiring optical data (e.g., PPG data) according to implementations of the present disclosure. The system shown in FIG. 12 can optionally be used with the systems described with reference to FIGS. 2A-2D, for example.

[0386] The system can include an optical module 1202. The optical module can include any number of multicolor LEDs and any number of photodiodes (PDs).

[0387] An acquisition and processing module 1204 can be in operable communication with the optical module 1202. The acquisition and processing module 1204 can be configured to control the multicolor LED(s) and / or receive signals acquired by the one or more photodiodes. The acquisition and processing module 1204 can further be configured to perform ambient noise cancelation, data combining, and / or packetization. For example, signals from one or more photodiodes can be processed using ambient noise cancelation, combined, and packetized.

[0388] A Bluetooth low energy (BLE) module 1206 can be in operable communication with the acquisition and processing module 1204. The Bluetooth low energy module can be configured to receive data from the acquisition and processing module 1204 and transmit the data to one or more remote computing devices (not shown). For example, when the system of FIG. 12 is implemented as part of a wearable device (e.g., any of the devices described with reference to FIGS. 2A-2D) the remote computing device can be a mobile device in communication with the wearable device.

[0389] Optionally any / all of the optical module 1202, acquisition and processing module 1204, and / or BLE module 1206 can be implemented using one or more computing devices (e.g., the computing device 400 shown in FIG. 4).

[0390] The present disclosure contemplates that the optical module 1202 and / or acquisition and processing module can implement additional features configured to improve signal acquisition for PPG signals. The present disclosure contemplates that any / all of the following features can be used for PPG signal acquisition, according to implementations of the present disclosure.

[0391] Multiple Light Source / Photo Detectors

[0392] Implementations of the present disclosure can use combinations of multiple Light Sources and Photo Detectors that can improve waveform stabilization and noise cancellation.

[0393] Diverse Wavelengths

[0394] Implementations of the present disclosure include using a combination of Red, IR and Green wavelengths (and / or an additional light source of one of these if needed), which allows for functionality when acquiring PPG signals on different skin tones. Conventional devices may be ineffective when used for darker skin tones and implementations of the present disclosure can overcome the limitations of theseconventional devices by using multiple wavelengths to provide a better response across all skin tones.

[0395] Multiple PhotoDetectors

[0396] Implementations of the present disclosure can include multiple photodetectors to reduce or prevent the effects of ambient light noise and / or poor contact with skin. Implementations of the present disclosure including multiple photodetectors can also improve PPG signals by providing for differential measurements based on the different positions of the photodetectors.

[0397] Real time composite(s) of waveforms

[0398] Implementations of the present disclosure can use any number of light sources and photodetectors to generate real-time composites of waveforms. For example, if there are M light sources and N photo detectors, the system can output up to MxN number of output optical waveforms. Some of these waveforms can optionally be rejected if, for instance, the light source and photo detector are not in the same reflective / refractive path. Optionally additional or alternative criteria can be used such as a positional or motion input, and / or excess noise that can be used to reject an output optical waveform.

[0399] Optionally, the inclusion or rejection criteria of optical waveforms may or may not be static or predetermined. Optionally, a user application can customize inclusion and / or rejection of optical waveforms. The incoming stream of optical outputs can be passed through a quality check which can determine whether each optical output is suitable for further processing.

[0400] The set of optical waveforms (K <= M x N) can be combined into a single composite waveform. Optionally, multiple composite waveforms can be generated based on the needs of applications or disease conditions. Optionally, incoming optical outputs can be synchronized according to their timestamps.

[0401] Modulation of current drive / light output

[0402] Each of the light sources can have an input parameter, P, (current drive setting, for example), that determines the light energy output. In turn, this influences the light energy captured by the photo detector. Optionally, the PPG sensors described herein can modulate the input parameter P for each light source to get optimal setting for a skin tone. Therefore any number of configurations can be set for any number of light sources, and each user of the device can optionally have an ideal input parameter or set of input parameters for the light sources.

[0403] Optionally, any of the systems, methods, and devices described herein can be configured based on the skin tone of the patient / user / subject. The skin tone of thepatient / user / subject can affect the PPG waveforms received, for example. Optionally, a user can be prompted by a user interface / display to enter an skin tone measure (e.g., a metric corresponding to darkness, lightness, etc.) and the system can be configured based on the skin tone. For example, the skin tone can optionally be used to calibrate the estimates of physiological parameters described herein (e.g., blood pressure), as inputs to trained machine learning models configured to estimate physiological parameters, and / or as inputs to train machine learning models.

[0404] Alternatively or additionally, as patient demographics are entered, the parameter P can be mapped to predetermined values. Further, an individual user may perform a ‘sweep’ of P values to determine their best setting. This can optionally be performed as an automated ‘self calibration’ routine at the start of device usage and / or as-needed.

[0405] Signal Quality Indicator(s)

[0406] Alternatively or additionally, implementations of the present disclosure can perform a real time determination of signal quality. As PPG signals (raw and composite) are being collected, they are analyzed for signal quality and a metric is generated. This indicator can be used to look for optimizing / avoiding noise sources or even to find an optimal position on the wrist.

[0407] Optionally, any of the systems, methods, and devices described herein can include evaluating raw PPG signal to determine if the PPG sensor is correctly positioned on the wearer. Optionally, a machine learning model can be trained to determine whether the raw PPG signal is from a correctly positioned PPG sensor, and implementations of the present disclosure can input raw PPG signals into trained machine learning models that can output whether the PPG sensor is correctly positioned. The determination of whether the PPG sensor is correctly positioned, and / or instructions for correctly positioning the PPG sensor, can optionally be output by a display (e.g. the displays described with reference to FIG. 2A-2D). Alternatively or additionally, the system can be configured to output an alert, for example upon detecting that the PPG sensor has moved into or out of an optimal position.

[0408] Example 8: Systolic and Diastolic Blood Pressure Estimation

[0409] FIG. 13 illustrates an additional example implementation of the present disclosure configured to estimate systolic blood pressure (SBP) and / or diastolic blood pressure (DBP) based on an optical input.

[0410] At step 1 02, optical data is received by an optical module and analog frontend. The optical module and analog frontend can include the features of any of the optical / PPG sensors described herein (e.g., the optical module 1202 of FIG. 12), and canfurther include an analog front end configured to process the signals from any number of PPG sensors.

[0411] At step 1304, the method includes controlling the optical module. Controlling the optical module can optionally include any or all of: controlling a drive signal to LEDs of the optical module, adjusting the sensitivity of the photodiodes of the optical module, acquiring data from the optical module, and / or performing ambient noise cancelation on the data acquired from the optical module.

[0412] At step 1306, the method further includes combining optical data received from the optical module. As a non-limiting example, if the optical module is configured to use three wavelengths to implement a PPG sensor, the method can include combining the three wavelengths into a single PPG signal using statistical features of each of the three wavelengths.

[0413] Step 1306 can optionally further include denoising, and / or signal quality estimation. Optionally, denoising and / or signal quality (SQI) estimation can be used to control the optical module. Optionally, the method can include rejecting signals where the signal quality is below a threshold.

[0414] At step 1308, PPG features are extracted from the combined PPG signal. Feature extraction can include frequency domain, time domain, and statistical methods. Optionally, first and / or second derivative PPG waveforms and features can be estimated.

[0415] At step 1310, the method includes analyzing the features extracted at step 1308. Data, including optionally the features, can be input to a neural-network based regressor. The neural-network based regressor can be configured to output estimates of blood pressure, including, optionally, both systolic and / or diastolic blood pressure.

Claims

CLAIMSWHAT IS CLAIMED:

1. A method for determining a physiological measurement from a physiological sensor, comprising: receiving a physiological signal from the physiological sensor; inputting the physiological signal into a trained machine learning model; and outputting, using the trained machine learning model, an algorithm from a set of algorithms.

2. The method of claim 1 , further comprising determining a physiological measurement based on the algorithm and the physiological signal.

3. The method of claim 1 or claim 2, wherein the physiological sensor is a photoplethysmography (PPG) sensor.

4. The method of any one of claims 1-3, wherein the trained machine learning model is a recurrent neural network (RNN).

5. The method of any one of claims 1-4, wherein the trained machine learning model is a long short term memory (LSTM) network.

6. The method of any one of claims 1-5, wherein the physiological measurement is a measurement of blood pressure.

7. The method of any one of claims 1-6, further comprising inputting a patient health state into the trained machine learning model.

8. The method of claim 7, wherein the patient health state comprises an activity profile of a patient.

9. The method of claim 7 or claim 8, wherein the patient health state comprises physiological parameters of a patient.

10. The method of claim 9, wherein the physiological parameters of the patient comprise at least one of age, gender, weight, ethnicity, or health history.

11. The method of any one of claims 1-10, further comprising inputting an environmental data into the trained machine learning model.

12. The method of claim 11, wherein the environmental data comprises at least one of location, altitude, weather condition, time zone changes or temperature.

13. A method for determining a physiological measure from a plurality of physiological sensors, comprising: receiving a plurality of physiological signals from a plurality of physiological sensors; inputting the plurality of physiological signals into a trained machine learning model; and outputting, using the trained machine learning model, a physiological measurement.

14. The method of claim 13, wherein the plurality of physiological signals comprise a photoplethysmography (PPG) signal.

15. The method of claim 13 or claim 14, wherein the plurality of physiological sensors comprises a photoplethysmography (PPG) sensor.

16. The method of any one of claims 13-15, further comprising inputting each of the plurality of physiological signals into a plurality of corresponding sensor- specific trained machine learning models.

17. The method of any one of claims 13-16, wherein the trained machine learning model is a recurrent neural network.

18. The method of any one of claims 16-17, wherein the trained machine learning model is a long short term memory (LSTM) network.

19. The method of any one of claims 13-18, wherein the physiological measurement is a measurement of blood pressure.

20. The method of any one of claims 13-19, further comprising inputting a patient health state into the trained machine learning model.

21. The method of claim 20, wherein the patient health state comprises an activity profile of a patient.

22. The method of claim 20 or claim 21, wherein the patient health state comprises physiological parameters of a patient.

23. The method of claim 22, wherein the physiological parameters of the patient comprise at least one of age, gender, weight, ethnicity, or health history.

24. The method of any one of claims 13-23, further comprising inputting an environmental data into the trained machine learning model.

25. The method of claim 24, wherein the environmental data comprises at least one of location, altitude, weather condition, time zone changes, or temperature.

26. A system comprising: a wearable device comprising a plurality of physiological sensors; and a computing device operably coupled to the wearable device, wherein the computing device comprises at least one processor and memory, the memory having computer-executable instructions stored thereon that, when executed by the at least one processor, cause the at least one processor to: receive a plurality of physiological signals measured by the plurality of physiological sensors; input the plurality of physiological signals into a trained machine learning model; andoutputting, using the trained machine learning model, a physiological measurement.

27. The system of claim 26, further comprising a portable electronic device operably coupled to the wearable device, wherein the portable electronic device is configured to: receive the plurality of physiological signals from the plurality of physiological sensors; and transmit the plurality of physiological signals to the computing device.

28. The system of claim 27, wherein the computing device is a cloud server.

29. The system of claim 27 or claim 28, wherein the computing device is further configured to: transmit the physiological measurement to the portable electronic device.

30. The system of claim 29, wherein the portable electronic device is configured to display the physiological measurement on a graphical user interface (GUI).

31. A method comprising: receiving a plurality of PPG signals; receiving a plurality of corresponding ground- truth blood pressure measurements; receiving a plurality of corresponding selected blood pressure algorithms wherein the plurality of corresponding selected blood pressure algorithms each provide an estimate of blood pressure based on a respective PPG signal and wherein a corresponding selected blood pressure algorithm for each respective PPG signal provides an estimate of blood pressure that is closest to the corresponding ground-truth blood pressure for the respective PPG signal; creating a training set comprising the plurality of PPG signals and a respective selected blood pressure algorithm for the plurality of PPG signals; and training a machine learning model using the training set, wherein the trained machine learning model is configured to identify a corresponding selected blood pressure algorithm for a patient based on a test PPG signal measured from the patient.

32. The method of claim 31, wherein the method further comprises receiving sets of calibration data values, wherein each of the sets of calibration data values correspond toat least one PPG signal from the plurality of PPG signals, and wherein the training set comprises the sets of calibration data values.

33. The method of claim 32, wherein the calibration data values comprise a plurality of patient health states.

34. The method of claim 32, wherein the calibration data values comprise a plurality of activity profiles.

35. The method of claims 33, wherein the patient health state comprises physiological parameters of the patient.

36. The method of claim 35, wherein the physiological parameters of the patient comprise age, gender, weight, ethnicity, or health history.

37. The method of any one of claims 32-36, wherein the calibration data values further comprise a plurality of environmental data values input into the trained machine learning model.

38. The method of claim 37, wherein the environmental input comprises location data, altitude, time zone changes or temperature.

39. The method of any one of claims 31-38, wherein the trained machine learning model is a recurrent neural network.

40. The method of any one of claims 31-39, wherein the trained machine learning model is a long short term memory (LSTM) model.

41. A method comprising: receiving optical data from an optical sensor; receiving calibration data; deploying a trained machine learning model, wherein the trained machine learning model nfigured to output an algorithm for estimating blood pressure based on optical data;selecting, by the trained machine learning model, a selected algorithm from a plurality of algorithms based on the calibration data and the optical data, wherein the selected algorithm is the algorithm chosen to be closest to ground truth blood pressure for the optical data and calibration data; and estimating blood pressure based on the optical data, the calibration data, and the selected algorithm.

42. The method of claim 41, wherein the plurality of algorithms each correspond to a physical state of a patient, and wherein the method further comprises determining a disease state of the patient based on the selected algorithm.

43. The method of claim 41 or claim 42, wherein the trained machine learning model is a recurrent neural network.

44. The method of any one of claims 41-43, wherein the trained machine learning model is a long short term memory (LSTM) model.

45. The method of any one of claims 41-44, wherein the calibration data comprises a plurality of patient health states.

46. The method of claim 45, wherein the patient health state comprises physiological parameters of the patient.

47. The method of claim 46, wherein the physiological parameters of the patient comprise age, gender, weight, ethnicity, or health history.

48. The method of any one of claims 41-47, wherein the calibration data comprises a plurality of activity profiles.

49. The method of any one of claims 41-48, wherein the calibration data further comprises a plurality of environmental data values, an environmental input into the trained machine learning model.

50. The method of claim 49, wherein the environmental input comprises location data, altitude, time zone changes or temperature.

51. A method comprising: receiving optical data from an optical sensor; receiving calibration data; deploying a trained machine learning model, wherein the trained machine learning model is configured to output an algorithm for estimating blood pressure based on optical data; selecting, by the trained machine learning model, a selected algorithm from a plurality of algorithms based on the calibration data and the optical data, wherein the selected algorithm is the algorithm chosen to estimate a respective blood pressure closest to a ground truth blood pressure for the optical data and calibration data; receiving, second optical data from the optical sensor; receiving second calibration data; and training the trained machine learning model using the second optical data and the second calibration data.

52. The method of claim 51 , wherein the trained machine learning model is an LTSM model.

53. The method of claim 51 or claim 52, wherein the calibration data comprises a plurality of patient health states.

54. The method of any one of claim 53 wherein the patient health state comprises physiological parameters of the patient.

55. The method of claim 54, wherein the physiological parameters of the patient comprise age, gender, weight, ethnicity, or health history.

56. The method of any one of claims 51-55, wherein the calibration data comprises a plurality of activity profiles.

57. The method of any one of claims 51-56, wherein the calibration data further comprises a plurality of environmental data values, an environmental input into the trained machine learning model.

58. A method comprising: receiving a plurality of sensor signals; receiving a plurality of corresponding ground- truth physiological measurements; receiving a plurality of corresponding selected physiological estimation algorithms wherein the plurality of corresponding selected physiological estimation algorithms each provide an estimate of a physiological measurement based on a respective sensor signal and wherein a selected physiological estimation algorithm for each respective sensor signal provides an estimate of the physiological measurement that is closest to the corresponding ground-truth physiological measurement for the respective sensor signal; creating a training set comprising the plurality of sensor signals and a respective selected physiological measurement algorithm for the plurality of sensor signals; and training a machine learning model using the training set, wherein the trained machine learning model is configured to identify a corresponding selected physiological estimation algorithm for a patient based on a test sensor signal measured from the patient.

59. The method of claim 58, wherein the method further comprises receiving an input from a patient.

60. The method of claim 59, wherein the patient input comprises an indication of breathlessness.

61. The method of any one of claims 58-60, wherein the physiological estimation algorithm is configured to estimate at least one of: pulmonary capillary wedge pressure and left atrial pressure.

62. The method of any one of claims 58-61, wherein the physiological estimation algorithm is configured to estimate imaging data.

63. The method of any one of claims 58-62, wherein the physiological estimation algorithm is configured to estimate left ventricular ejection fraction derived from ultrasound, RV volume or LV volumes.

64. The method of any one of claims 58-63, wherein the physiological estimation algorithm is configured to estimate blood glucose.

65. The method of any one of claims 58-64, wherein the physiological estimation algorithm is configured to estimate heart rate.

66. The method of any one of claims 58-65, wherein the plurality of sensor signals comprises measurements of B-Type Natriuretic Peptide (BNP).

67. The method of any one of claims 58-66, wherein the plurality of sensor signals comprises data from at least one of: electrocardiogram (ECG) electrodes and sensors, temperature sensors, SpO2 (oxygen saturation) sensors, impedance cardiography (ICG) electrodes, accelerometers and / or gyroscopes, ultrasonic transducers, skin conductance sensors, phonocardiogram sensors, ballistocardiography sensors, piezoelectric / piezoresistive sensors, ultrasound sensors, arterial tonometry sensors, and / or capacitive respiration sensors.

68. A computer-implemented method comprising: accessing a plurality of training PPG waveform data; accessing a plurality of corresponding ground truth body position measurements; accessing a plurality of corresponding ground truth blood pressure measurements; creating a training data set comprising the plurality of training PPG waveform data, the corresponding ground truth body position measurements, and the corresponding ground truth blood pressure measurements; and training a machine learning model, using the training data set, to estimate a gravity contribution in each respective input training PPG waveform, wherein the gravity contribution corresponds to a hydrostatic pressure of blood relative to a heart.

69. The computer-implemented method of claim 68, wherein estimating the respective gravity contribution for each of the respective input training PPG waveforms comprises estimating a column pressure caused by gravity at each of the plurality of corresponding ground truth body position measurements.

70. The computer-implemented method of claim 69, wherein the corresponding ground truth body position measurements comprise respective estimates of a height dimension at which respective training PPG data were gathered.

71. The computer-implemented method of claim 69, wherein determining the respective gravity contributions for each of the plurality of PPG data comprises estimating a density of the blood.

72. The computer-implemented method of claim 71, wherein the estimated density is a constant.

73. The computer-implemented method of any one of claims 68-72, wherein the training data set further comprises a plurality of activity profiles.

74. The computer-implemented method of any one of claims 68-73, wherein the training data set further comprises a plurality of patient health states.

75. The computer-implemented of claim 74, wherein the plurality of patient health states comprise physiological parameters of the patients.

76. The computer- implemented of claim 75, wherein the physiological parameters of the patient comprise age, gender, weight, ethnicity, or health history.

77. The computer-implemented of any one of claims 68-76, wherein the training data set further comprises a plurality of environmental data values input into the trained machine learning model.

78. The computer-implemented of claim 77, wherein the environmental data values comprise location data, altitude, time zone changes or temperature.

79. A computer-implemented method comprising: accessing a plurality of training PPG waveform data; accessing a plurality of corresponding ground truth blood pressure measurements; accessing a plurality of physiological parameters corresponding to each of the ground truth blood pressure measurements; and training a machine learning model, using the training PPG waveform data, the ground truth blood pressure measurements, and the plurality of physiological parameters, to output an estimated body position corresponding to the ground truth blood pressure measurements based on a respective training PPG waveform.

80. The computer-implemented method of claim 79, wherein training the machine learning model comprises identifying patterns in combinations of the plurality of physiological parameters corresponding to respective ground truth blood pressure measurements.

81. The computer-implemented method of claim 79 or claim 80, wherein the physiological parameters comprise subject body dimensions, subject blood composition, subject blood density, subject vascular anatomy, and subject heart position.

82. The computer-implemented method of any one of claims 79-81 wherein outputting the estimated body position comprises estimating a gravity contribution to the respective training PPG waveform by calculating an estimated height dimension at which the ground truth blood pressure measurements were gathered.

83. The computer-implemented method of claim 82, wherein calculating the estimated height dimension comprises utilizing subject body dimensions at which the ground truth blood pressure measurements were gathered.

84. A computer-implemented method of estimating blood pressure, the method comprising: receiving an input PPG signal from a PPG sensor located on a body part of a subject; receiving a trained blood pressure prediction machine learning model; receiving a trained blood pressure correction model; estimating, using the blood pressure prediction machine learning model, a blood pressure estimate;estimating, using the trained blood pressure correction model, blood pressure correction data; and determining, based on the blood pressure correction data and the blood pressure estimate, a corrected blood pressure.

85. The computer-implemented method of claim 84, wherein the blood pressure correction model is configured to correct the estimated blood pressure for a position of the PPG sensor.

86. The computer-implemented method of claim 84, wherein the blood pressure correction model is configured to correct the estimated blood pressure for a hydrostatic pressure of gravity.

87. The computer-implemented method of any one of claims 84-86, wherein the trained blood pressure correction model comprises a model trained according to claims 79-83.

88. A method comprising: receiving a plurality of PPG signals; receiving a plurality of corresponding calibration data for each of the plurality of PPG signals; receiving a plurality of corresponding ground-truth blood pressure measurements for each of the plurality of PPG signals; receiving a plurality of corresponding selected blood pressure algorithms wherein the plurality of corresponding selected blood pressure algorithms each provide an estimate of blood pressure based on a respective PPG signal and wherein a corresponding selected blood pressure algorithm for each respective PPG signal provides an estimate of blood pressure that is closest to the corresponding ground-truth blood pressure for the respective PPG signal; creating a training set comprising the plurality of PPG signals and a respective selected blood pressure algorithm for the plurality of PPG signals; and training a machine learning model using the training set, wherein the trained machine learning model is configured to identify a corresponding selected blood pressure algorithm for a patient based on a test PPG signal measured from the patient.

89. The method of claim 88, wherein the corresponding calibration data comprises temperature data.

90. The method of claim 89, wherein the PPG signals are each measured from a corresponding vascular bed, and wherein the temperature data comprises pressure measured at the corresponding vascular bed of the corresponding PPG signal.

91. The method of claim 88, wherein the corresponding calibration data comprises pressure data.

92. The method of claim 91, wherein the PPG signals are each measured from a corresponding vascular bed, and wherein the pressure data comprises pressure measured at the corresponding vascular bed of the corresponding PPG signal.

93. The method of any one of claims 88-92, wherein the trained machine learning model is a recurrent neural network.

94. The method of any one of claims 88-93, wherein the trained machine learning model is a long short term memory (LSTM) model.

95. A method comprising: receiving optical data from an optical sensor; receiving calibration data; deploying a trained machine learning model, wherein the trained machine learning model is configured to output an algorithm for estimating blood pressure based on optical data; selecting, by the trained machine learning model, a selected algorithm from a plurality of algorithms based on the calibration data and the optical data, wherein the selected algorithm is the algorithm chosen to be closest to ground truth blood pressure for the optical data and calibration data; and estimating blood pressure based on the optical data, the calibration data, and the selected algorithm.

96. The method of claim 95, wherein the optical sensor is disposed over a vascular bed, and the calibration data comprises a temperature value of the vascular bed.

97. The method of claim 95 or claim 96, wherein the optical sensor is disposed over a vascular bed, and the calibration data comprises a pressure value on the vascular bed.

98. The method of any one of claims 95-97, wherein the trained machine learning model is a recurrent neural network.

99. The method of any one of claims 95-98, wherein the trained machine learning model is a long short term memory (LSTM) model.

100. A system comprising: a wearable device comprising an optical sensor and a physiological sensor; and a computing device operably coupled to the wearable device, wherein the computing device comprises at least one processor and memory, the memory having computer-executable instructions stored thereon that, when executed by the at least one processor, cause the at least one processor to: receive an optical signal from the optical sensor; receive a corresponding physiological signal measured by the physiological sensor; input the optical signal and the corresponding physiological signal into a trained machine learning model; and output, using the trained machine learning model, a physiological measurement.

101. The system of claim 100, wherein the physiological sensor comprises a temperature sensor.

102. The system of claim 100, wherein the physiological sensor comprises a pressure sensor.

103. The system of claim 100, wherein the physiological sensor comprises a pressure sensor and a temperature sensor.

104. The system of claim 102 or 103, wherein the pressure sensor comprises a piezoelectric pressure sensor.

105. The system of any one of claims 100-104, wherein the physiological sensor and the optical sensor are configured to measure the same vascular bed of a patient.

106. The system of any one of claims 100-105, further comprising a wristband configured to position the physiological sensor and the optical sensor adjacent to a single vascular bed.

107. The system of any one of claims 100-106, further comprising a portable electronic device operably coupled to the wearable device, wherein the portable electronic device is configured to: receive the corresponding physiological signal from the physiological sensor; and transmit the corresponding physiological signal to the computing device.

Images