Determination of arterial blood pressure

Abstract

Various exemplary embodiments relate to a regimen for non-invasively determining continuous arterial blood pressure waveforms or systolic and diastolic pressures. In one embodiment, an optical plethysmography, PPG, signal may be acquired from a PPG sensor, and an optical plethysmography, PPG, signal may be acquired from the PPG sensor. A flux signal may be acquired from a flux sensor. The arterial blood pressure waveform or the systolic and diastolic pressures may be determined based at least in part on the PPG signal and the flux signal.

Description

TECHNICAL FIELD

[0001] Various exemplary embodiments relate generally to the field of determining arterial blood pressure. BACKGROUND

[0002] Various methods can be used to measure blood pressure directly, such as inserting a catheter with a pressure sensor into the pulmonary artery or through an inflatable cuff. The catheter-based measurement method is an invasive method and is typically used in critical care. The inflatable cuff method is widely used in clinical practice, but is an inconvenient and uncomfortable measurement method. A major drawback of both methods is that they do not provide convenient and regular around-the-clock blood pressure measurements.

[0003] An indirect blood pressure measurement method can be performed using a photoplethysmography (PPG) sensor. The PPG sensor measures the relative blood volume changes of a peripheral vascular bed. The principle of PPG is based on the absorption of light by blood at the measurement site. When the blood volume changes due to the pumping of blood by the heart through the arteries, the absorption of light by blood changes, and thus the light intensity is modulated by the cardiovascular system. While PPG sensors, in combination with data-driven computational models including machine learning and deep learning models, can enable indirect measurement of blood pressure, there are still problems with the accuracy and reliability of the measurements. SUMMARY

[0004] The summary introduces some concepts briefly, which will be described further in the detailed description. The summary is not intended to identify key or essential features of the claimed subject matter, nor is it intended to be used to limit the scope of the claimed subject matter. The purpose of the present invention is to enable continuous and non-invasive accurate measurement of arterial blood pressure waveforms or systolic and diastolic blood pressure while still maintaining the ease of measuring blood pressure. Other implementations are apparent from the dependent claims, the description, and the accompanying drawings.

[0005] According to a first aspect, a method for non-invasively determining a continuous arterial blood pressure waveform or a diastolic and systolic blood pressure. The method comprises: obtaining a photoplethysmography (PPG) signal from a PPG sensor; obtaining a flux signal from a flux sensor; determining the arterial blood pressure waveform or the diastolic and systolic blood pressure based at least in part on the PPG signal and the flux signal. For example, this can enable an accurate scheme for continuous and non-invasive real-time measurement of blood pressure waveforms or diastolic and systolic blood pressure.

[0006] According to one implementation of the first aspect, the method includes: determining an index of the vascular cross-sectional area based on the PPG signal; determining a blood flow index based on the flux signal; determining arterial compliance based on the flux signal and the PPG signal; and determining the arterial blood pressure waveform or the diastolic and systolic blood pressure based at least in part on the index of the vascular cross-sectional area, the blood flow index, and the arterial compliance.

[0007] According to one implementation of the first aspect, the method includes: estimating indices of vascular cross-sectional area, tissue oxygenation, arterial oxygenation, blood and hemoglobin content based on the PPG signal; determining indices of blood flow, blood velocity, and mobile red blood cell concentration based on the flux signal; determining arterial compliance based on the flux signal and the PPG signal; and determining the arterial blood pressure waveform or the systolic and diastolic blood pressure based at least in part on cardiovascular and hemodynamic indices obtained partially from the PPG signal and the flux signal. For example, this can enable an accurate scheme for continuous and non-invasive real-time measurement of blood pressure waveforms or systolic and diastolic blood pressure.

[0008] According to one implementation of the first aspect, the method further includes: determining a first pulse wave analysis (PWA) feature based on the PPG signal; determining a second PWA feature based on the flux signal; and determining the arterial blood pressure waveform or the systolic and diastolic blood pressure based at least in part on the first and second PWA features. For example, this can enable an accurate scheme for real-time measurement of continuous blood pressure or measurement of systolic and diastolic blood pressure per cardiac cycle.

[0009] According to one implementation of the first aspect, the method further includes: applying at least one trained machine learning model when determining the arterial blood pressure waveform or the systolic and diastolic blood pressure, said at least one trained machine learning model being trained at least in part based on the PPG signal, the flux signal, and blood pressure reference measurements. For example, this can provide an efficient method for determining continuous arterial blood pressure or systolic and diastolic blood pressure for each cardiac cycle.

[0010] According to one implementation of the first aspect, the at least one trained machine learning model comprises a physics-informed neural network (PINN) based on a mathematical model of blood flow in the human vascular system. For example, this can enable an efficient method for determining continuous arterial blood pressure, or systolic and diastolic blood pressure, for each cardiac cycle.

[0011] According to one implementation of the first aspect, the flux sensor includes a laser Doppler flowmeter (LDF) sensor, a dynamic light scattering (DLS) sensor, a speckle plethysmography (SPG) sensor, a diffusion correlation spectroscopy (DCS) sensor, or a laser speckle contrast imaging (LCSI) sensor. For example, this can enable an accurate protocol for real-time measurement of continuous blood pressure or for measuring systolic and diastolic blood pressure per cardiac cycle.

[0012] According to a second aspect, a device for noninvasively determining a continuous arterial blood pressure waveform or systolic and diastolic blood pressure. The device is used to: acquire a PPG signal from a photoplethysmography (PPG) sensor; acquire a flux signal from a flux sensor; and determine the arterial blood pressure waveform or the systolic and diastolic blood pressure based at least in part on the PPG signal and the flux signal.

[0013] According to one implementation of the second aspect, the device is used to: determine an index of the vascular cross-sectional area based on the PPG signal; determine a blood flow index based on the flux signal; determine arterial compliance based on the flux signal and the PPG signal; and determine the arterial blood pressure waveform or the systolic and diastolic blood pressure based at least in part on the index of the vascular cross-sectional area, the blood flow index, and the arterial compliance.

[0014] According to one implementation of the second aspect, the device is used to: estimate indices of vascular cross-sectional area, tissue oxygenation, arterial oxygenation, blood and hemoglobin content based on the PPG signal; determine indices of blood flow, blood velocity and mobile red blood cell concentration based on the flux signal; determine arterial compliance based on the flux signal and the PPG signal; and determine the arterial blood pressure waveform based at least in part on cardiovascular and hemodynamic indices obtained in part from the PPG signal and the flux signal.

[0015] According to one implementation of the second aspect, the device is used to: determine a first pulse wave analysis (PWA) feature based on the PPG signal; determine a second PWA feature based on the flux signal; and determine the arterial blood pressure waveform or the systolic and diastolic blood pressure based at least in part on the first PWA feature and the second PWA feature.

[0016] According to one implementation of the second aspect, the device is used to: apply at least one trained machine learning model when determining the arterial blood pressure waveform or the systolic and diastolic blood pressure, said at least one trained machine learning model being trained at least in part based on the PPG signal, the flux signal, and blood pressure reference measurements.

[0017] According to one implementation of the second aspect, the at least one trained machine learning model includes a physics-informed neural network (PINN) based on a mathematical model of blood flow in the human vascular system, including but not limited to the Windkessel model and an incompressible fluid flow model based on the Navier-Stokes equations.

[0018] According to one implementation of the second aspect, the flux sensor includes a laser Doppler flowmeter (LDF) sensor, a dynamic light scattering (DLS) sensor, a speckle plethysmography (SPG) sensor, a diffuse correlation spectroscopy (DCS) sensor, a laser speckle contrast imaging (LCSI) sensor, or a laser Doppler perfusion imaging sensor (LDPI).

[0019] According to a third aspect, a wearable device is provided, including the device described in the second aspect.

[0020] According to a fourth aspect, a computer program is provided. The computer program includes program code that, when executed on a device, causes the method according to the first aspect to be performed.

[0021] According to a fifth aspect, an apparatus is provided for noninvasively determining a continuous arterial blood pressure waveform or systolic and diastolic blood pressure. The apparatus includes: a module for acquiring a PPG signal from an optical photoplethysmography (PPG) sensor; a module for acquiring a flux signal from a flux sensor; and a module for determining the arterial blood pressure waveform or the systolic and diastolic blood pressure based at least in part on the PPG signal and the flux signal.

[0022] According to a sixth aspect, an apparatus is provided for noninvasively determining a continuous arterial blood pressure waveform or systolic and diastolic blood pressure. The apparatus includes at least one processor and at least one memory storing instructions, which, when executed by the at least one processor, cause the apparatus to perform the following operations: acquire a PPG signal from a photoplethysmography (PPG) sensor; acquire a flux signal from a flux sensor; and determine the arterial blood pressure waveform or the systolic and diastolic blood pressure based at least in part on the PPG signal and the flux signal. Attached Figure Description

[0023] The accompanying drawings are provided to further illustrate exemplary embodiments and form part of this specification. These drawings depict exemplary embodiments and, together with the specification, help to explain them. In the drawings:

[0024] Figure 1 An example of a device for implementing one or more embodiments is shown;

[0025] Figure 2 An exemplary embodiment is shown, illustrating a machine learning model for estimating at least one parameter from a flux signal;

[0026] Figure 3 Example features are shown, calculated from the AC portion of the flux signal and its first derivative, according to an exemplary embodiment.

[0027] Figure 4 An example of time-domain characteristics calculated from the time difference between the flux and the PPG signal is shown;

[0028] Figure 5 An example of amplitude-based features calculated by combining flux and PPG signal characteristics is shown;

[0029] Figure 6 An exemplary embodiment is shown, illustrating a machine learning model for calculating at least one parameter from a PPG signal;

[0030] Figure 7 An exemplary embodiment is shown, illustrating a machine learning model for calculating arterial compliance from flux signals and PPG signals;

[0031] Figure 8 An exemplary embodiment is shown, illustrating a machine learning model for calculating blood pressure from PPG signals and flux signals, and PWA features calculated from PPG signals and flux signals and their derivatives.

[0032] Figure 9 Possible embodiments of the system architecture provided by exemplary models are illustrated;

[0033] Figure 10 Another possible embodiment of the system architecture provided by the exemplary embodiments is shown;

[0034] Figure 11A The possible component layouts provided by the exemplary embodiments are shown;

[0035] Figure 11B Another possible component layout provided by an exemplary embodiment is shown;

[0036] Figure 12 An exemplary embodiment is shown, illustrating a method for noninvasively determining continuous arterial blood pressure waveforms or systolic and diastolic blood pressure.

[0037] In the accompanying drawings, the same reference numerals are used to denote the same parts. Detailed Implementation

[0038] Reference will now be made in detail to exemplary embodiments, examples of which are illustrated in the accompanying drawings. The detailed description provided below, in conjunction with the accompanying drawings, is intended as a description of the present example and is not intended to merely represent a form in which the present example can be constructed or used. The description illustrates the functionality of the example and the sequence of steps for constructing and operating the example. However, the same or equivalent functionality and sequence can be achieved through different embodiments.

[0039] Photoplethysmography (PPG) sensors can be used for blood pressure measurement. PPG sensors measure relative changes in blood volume in the peripheral vascular bed. The principle of PPG is based on the absorption of light by red blood cells in the blood. When blood volume changes due to the heart pumping blood through the arteries, the absorption of light by red blood cells changes, thus modulating the light intensity. This modulation of light intensity is then detected by a photodetector.

[0040] PPG signals can be divided into two parts: the non-pulsatile DC portion and the pulsatile AC portion. The DC portion occurs because light passes through tissue without changing its volume or density. The AC portion occurs because light passes through tissue and changes its volume or density. The volume change effect is particularly pronounced in the dilation of blood vessels (arteries and arterioles) as the pulse wave propagates. The AC portion is typically generated due to changes in blood volume within the tissue caused by the heart's pumping action. In addition, motion causes changes in tissue density and volume, resulting in motion artifacts in the AC signal.

[0041] According to the modified Beer-Lambert law, changes in PPG signal are related to changes in red blood cell concentration.

[0042] (1)

[0043] in, These are the optical path length, red blood cell concentration (Cg), and photodetector current, respectively. Red blood cell concentration is related to blood volume. Therefore, changes in the PPG signal are related to changes in blood volume. Changes in current... Divide by DC current The perfusion index (PI) is defined as follows:

[0044] (2)

[0045] If we assume that the vessel length remains constant, then changes in blood volume are related to changes in the vessel cross-section. This means that PI (pigmentation index) is related to changes in vessel cross-section. On the other hand, changes in blood vessel cross-section are non-linearly correlated with blood pressure.

[0046] (3)

[0047] in, The elasticity of the arterial wall is described by p, where p is blood pressure. On the other hand, the Bramwell-Hill equation gives the arterial compliance C as...

[0048] (4)

[0049] Wherein, PWV is pulse wave velocity. It's the density of blood. Also, I know...

[0050] (5)

[0051] By combining equations (4) and (5), we can write:

[0052] (6)

[0053] By combining equations (1) and (6), we can write:

[0054] (7)

[0055] Since arterial compliance and arterial cross-sectional area A are difficult to measure, and their relationship with pressure is non-linear, machine learning models are typically used to estimate them. These machine learning models attempt to estimate missing parameters from features computed from PI signals.

[0056] Mathematical models of blood flow in the vascular system are based on the Navier-Stokes equations and / or the Windkessel model of fluid flow. While mathematical models of blood flow on the three-dimensional geometry of the human vascular system support the assessment of blood flow and hemodynamic properties in complex geometries, 1D, 0D, and 1D-0D models demonstrate acceptable performance in real-time measurements of blood flow and cardiovascular and hemodynamic properties. For example, the following equations used to model an artery as a one-dimensional elastic tube can be written as follows:

[0057] (8)

[0058] and

[0059] (9)

[0060] In equations (8) and (9), A is the cross-sectional area of ​​the artery. It refers to blood flow, K is a parameter characterizing the elasticity of the arterial wall, and F is the coefficient of friction. This refers to blood density. Assuming blood density is constant, and the coefficient of friction can be estimated, for example, by the following equation:

[0061] (10)

[0062] in, ξ is the kinematic viscosity of blood, and its value ξ = 9 is commonly used for one-dimensional blood flow. Equations (8) and (9) show that, in addition to changes in cross-sectional area, changes in blood flow and elasticity of the arterial wall can also be measured.

[0063] The embodiments discussed in more detail below provide a scheme for noninvasively determining continuous arterial blood pressure waveforms or minimum (systolic) and maximum (diastolic) blood pressure. In this scheme, the PPG signal can be acquired from a PPG sensor. The flux signal can be acquired from a flux sensor. The arterial blood pressure waveform, or systolic and diastolic blood pressure, can be determined at least in part based on the PPG signal and the flux signal.

[0064] Figure 1Examples of devices 100 for implementing one or more embodiments are shown. For example, device 100 may include a mobile device, a smart device, a smartwatch, a bracelet, a chest strap, a wristband, an audio or wearable device, or any device generally used to implement any of the functions described herein. Device 100 may include at least one processor 102. At least one processor 102 may include one or more of various processing devices, such as a coprocessor, microprocessor, controller, digital signal processor (DSP), processing circuitry with or without an accompanying DSP, or various other processing devices including integrated circuits, such as application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), microcontroller units (MCUs), hardware accelerators, dedicated computer chips, and so on.

[0065] Device 100 may also include at least one memory 104. For example, memory 104 may be used to store computer program code, such as operating system software and application software. Memory 104 may include one or more volatile memory devices, one or more non-volatile memory devices, and / or combinations thereof. For example, memory 104 may be embodied as a magnetic storage device (e.g., hard disk drive, magnetic tape, etc.), an optical storage device, or a semiconductor memory (e.g., mask ROM, programmable ROM (PROM), erasable PROM (EPROM), flash ROM, random access memory (RAM), etc.).

[0066] Device 100 may also include a communication interface 108 for enabling device 100 to send and / or receive information. The communication interface may also be used to provide wireless radio connectivity, such as 3GPP mobile broadband connectivity (e.g., 3G, 4G, 5G, 6G, or next-generation), wireless local area network (WLAN) connectivity, such as that standardized by the IEEE 802.11 series or the Wi-Fi Alliance, or short-range wireless network connectivity, such as Bluetooth connectivity. Therefore, communication interface 108 may include one or more antennas to enable over-the-air transmission and / or reception of radio frequency signals.

[0067] The device 100 may also include a user interface 110, such as a display, for displaying information, such as blood pressure measurements.

[0068] When device 100 is used to implement certain functions, one or more components of the device (e.g., at least one processor 102 and / or at least one memory 104) may be used to implement those functions. Furthermore, when at least one processor 102 is used to implement certain functions, those functions may be implemented using program code 106, such as that included in at least one memory 104.

[0069] The functions described herein may be performed at least in part by one or more computer program product components (e.g., software components). According to an embodiment, device 100 includes a processor or processor circuitry, such as a microcontroller, configured by program code 106 when executed to perform embodiments of the operations and functions described herein. Alternatively or additionally, the functions described herein may be performed at least in part by one or more hardware logic components. Exemplary types of hardware logic components that may be used, without limitation, include field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip (SOCs), complex programmable logic devices (CPLDs), and graphics processing units (GPUs), etc.

[0070] Device 100 may be used to perform one or more methods described herein or may include modules for performing one or more methods described herein. In one example, the module includes at least one processor 102 and at least one memory 104 including program code 106 for causing device 100 to perform one or more methods when executed by at least one processor 102.

[0071] Although device 100 is shown as a single device, it should be understood that, where applicable, the functionality of device 100 may be distributed among multiple devices, such as among components of a transmitter, receiver, or transceiver.

[0072] Figure 2A machine learning model for estimating at least one parameter from a flux signal is shown. Equations (4) and (5) indicate that, to deterministically solve for blood pressure within an artery, the absolute value of blood flow can be measured. In an exemplary embodiment, the same equations can be applied to an arterial system, with flow rate then replaced by a blood flow index. Figure 2 In the illustrated scheme, the first moment of the power spectral density (flux) can be calculated from the flux signal. For example, the flux sensor can be a laser Doppler flowmeter (LDF) sensor, a dynamic light scattering (DLS) sensor, a speckle plethysmography (SPG) sensor, a diffusion correlation spectroscopy (DSC) sensor, a laser speckle contrast imaging (LCSI) sensor, or a laser Doppler perfusion imaging (LDPI) sensor.

[0073] Flux can then be used to estimate the blood flow index, since the flux signal has been shown to be related to blood flow. However, this correlation can be non-linear. Figure 2 One approach is illustrated where a machine learning (ML) model is applied to learn how flux signals map to blood flow indices. In various embodiments, different machine learning models can be used to extract blood flow indices from flux signals. In some embodiments, to optimize the learning of the ML model, a Physics Informed Neural Network (PINN) can be used in conjunction with constraint equations derived from theoretical modeling of blood flow in a vascular tree.

[0074] Figure 2 An example is shown where PINN learns from input data. 200 Estimated Blood Flow Index The input data includes blood pressure p, time constant τ, and arterial compliance C, comprising a flux signal and pulse waveforms derived from the flux signal and features calculated from measured continuous blood pressure references. Parameters These are one or more parameters derived from features calculated from flux, and / or one or more parameters derived from blood flow, blood velocity, and moving RBC concentration calculated from speckle contrast signals. Neural network 202 maps the input data to control equation variables. Reference 204 illustrates the components of the control equation. Figure 2In the example shown, governing equation 206, i.e., the governing partial differential equation, is based on the bielement Windkessel model:

[0075] (12)

[0076] In this case, the loss function is a combination of the residuals of the governing equation (GE):

[0077] (13)

[0078] Error between PINN's estimated pressure and the measured reference arterial blood pressure:

[0079] (14)

[0080] Give the loss function 208

[0081] (15)

[0082] The GE portion of the loss function can be adjusted based on the selected blood flow control equation. For example, the Windkessel model can have several variants: two-parameter, three-parameter, four-parameter, and multi-partition models can be used for the control equation. In some embodiments, equations (8) and (9) can be used as the GE loss function for PINN, but other inputs such as the PPG signal can subsequently be used to include an estimate of the exponential of the vascular cross-sectional area A required for blood pressure measurement.

[0083] In addition to the flux signal, the input data 200 of the ML model may also include features calculated from the waveform of the flux signal and its derivatives. Examples of such features are as follows: Figure 3 As shown. Features include, but are not limited to, the duration between peaks and troughs, the amplitude difference between peaks and troughs, the area under different parts of the curve, the higher-order derivatives of the signal, and the ratios of the above features.

[0084] In addition, in some embodiments, one or more arterial blood pressure points from reference measurements, such as those measured using a blood pressure cuff, may be provided as input to the ML model.

[0085] In addition to individual flux signal characteristics, in some embodiments, it can be as follows: Figure 4 and Figure 5 The diagram shows additional features of the combined flux signal waveform and the PPG signal waveform. Figure 4 An example is shown of the time-domain characteristics calculated from the time difference between the flux and the PPG signal; in Figure 4 For illustrative purposes, the flux and PPG signal amplitudes have been scaled. Figure 5An example of an amplitude-based feature fn calculated by combining flux and PPG signal characteristics is shown. Figure 5 In this context, flux signals and PPG signals are represented in arbitrary units.

[0086] Figure 6 An exemplary embodiment is shown, illustrating a machine learning model for calculating at least one parameter from PPG measurements.

[0087] Figure 6 A scheme for estimating hemodynamic and cardiovascular characteristics from tissue volume using a PPG sensor and a flux sensor is illustrated. In some embodiments, the same penetration depth can be achieved using PPG measurements at the same wavelength as the LED used in the laser of the flux sensor. Furthermore, in some embodiments, the location of the source and detector and the source-detector separation may be similar for both PPG and flux measurements.

[0088] Machine learning can be used to determine the correlation between PPG signals and the exponent of blood vessel cross-sectional area. In various embodiments, different machine learning models can be used to estimate the exponent of blood vessel cross-sectional area using PPG signals. In some embodiments, to optimize the learning of the ML model, a Physics Informed Neural Network (PINN) can be used in conjunction with constraint equations derived from theoretical modeling of blood pressure.

[0089] Figure 6 An example is shown where PINN learns from input data. 600 Estimate at least one parameter, such as the exponent of the blood vessel cross-sectional area. The input data includes PPG signals, pulse waveforms from the PPG signals, and characteristics calculated from measured continuous blood pressure reference values. In some embodiments, parameters... It is one or more parameters derived from characteristics calculated from the PPG signal (including its derivative), and / or one or more parameters derived from tissue oxygenation, arterial oxygenation, blood and hemoglobin content calculated from the PPG signal.

[0090] In some embodiments, multiple PPG wavelengths can be used to provide more input signals to the PINN. The neural network 602 maps the input data 600 to the control equation variables. In some embodiments, one or more reference arterial blood pressure points (e.g., from a blood pressure cuff) may also be given as input to the ML model.

[0091] Reference 604 shows the components of the governing equations. Figure 6 In the example shown, the governing equation (GE) 606 is

[0092] (16)

[0093] (17)

[0094] The loss function (15) 608 is based on the control equation and training data. For training the neural network, reference blood pressure can be measured.

[0095] Figure 7 An exemplary embodiment is shown, illustrating a machine learning model for calculating arterial compliance from flux signals and PPG signals.

[0096] Pulse wave velocity (PWV) can be expressed as:

[0097] (18)

[0098] in, D represents blood density, and D represents arterial dilatation. Arterial dilatation is the ability of an artery to expand due to internal pressure. Arterial dilatation can be expressed as...

[0099] (19)

[0100] Where A is the arterial cross-sectional area and P is the arterial blood pressure. The combination of equations (18) and (19) provides:

[0101] (20)

[0102] Wave intensity analysis (WIA) can be used to represent forward-propagating blood pressure waves as...

[0103] (twenty one)

[0104] The established water hammer equation provides another equation for forward-propagating blood pressure waves:

[0105] (twenty two)

[0106] Where U is the blood velocity. Replacing the blood velocity with flow rate and arterial cross-sectional area, we can obtain...

[0107] (twenty three)

[0108] The combination of equations (21) and (22) provides:

[0109] (twenty four)

[0110] It is now possible to solve PWV as follows:

[0111] (25)

[0112] Since the above only considers forward (away from the heart) propagating waves, the above equation is only valid at the beginning of the cardiac cycle, and only when there are only forward propagating waves.

[0113] The Bramwell-Hill equation gives the arterial compliance C as...

[0114] (26)

[0115] The combination of equations (25) and (26) provides:

[0116] (27)

[0117] This can be approximated by using the difference between the variables:

[0118] (28)

[0119] Equation (23) shows that arterial compliance C can be expressed as the blood flow at the start of the cardiac cycle after the previous diastolic cycle. It is a function of the arterial cross-sectional area A. Assume blood density... It is constant. Although the equation can be applied to a single artery, it can be extended to the arterial system by replacing the cross-sectional area with the cross-sectional area exponent and the flow rate with the blood flow exponent.

[0120] The preceding text describes how to calculate an estimate of the blood flow index from the flux signal and its pulse wave analysis (PWA) characteristics (including its derivative). In some embodiments, the blood flow index derived above can be used. As input to the ML model, or to enable the ML system to relearn how to obtain the blood flow index from the input signal. In some embodiments, in addition to the flux signal, the PPG signal and features calculated from the PPG signal (including its derivative) may also be used as input, allowing the ML model to learn the relationship between the exponent of the vessel cross-sectional area and the PPG signal. In some embodiments, multiple wavelength PPG signals with different light source-detector intervals may also be used.

[0121] In some embodiments, different machine learning models can be used to derive arterial compliance from flux and PPG signals. In some embodiments, PINN can be used in conjunction with constraint equations derived from theoretical modeling of blood flow in the vascular tree.

[0122] Figure 7 An example is shown where PINN learns from input data. The 700 algorithm estimates arterial compliance (C), blood flow (Q), and arterial cross-sectional area (A) using input data including flux signals, features calculated from the pulse waveform of the flux signals, features calculated from the derivative of the flux, PPG signals, features calculated from the pulse waveform of the PPG signals, and features calculated from the derivative of the PPG signals. Parameters These are one or more parameters derived from characteristics calculated from flux, and / or one or more parameters derived from indices of blood flow, blood velocity, and moving RBC concentration calculated from speckle contrast signals, and / or one or more parameters derived from characteristics calculated from PPG signals, and / or one or more parameters derived from indices of arterial and tissue oxygenation, blood and hemoglobin content calculated from PPG signals and their derivatives.

[0123] For the PINN training process, arterial compliance is implicitly measured through closely related parameters. Arterial compliance can be defined as...

[0124] (29)

[0125] However, only when p is approximately equal to This definition is only valid when the reference data is available. In this example, equations (27) and (29) can also be used to constrain PINN during training. In some embodiments, one or more arterial blood pressure calibration points can also be given as input to the ML model.

[0126] The neural network 702 maps the input data 700 to the variables of the control equation. Reference 704 illustrates the components of the control equation. Figure 7 In the example shown, governing equation (GE) 706 is

[0127] (30)

[0128] (31)

[0129] The loss function (15) 708 is based on the control equation and training data.

[0130] Figure 8 An exemplary embodiment is shown, illustrating a machine learning model for calculating blood pressure from PPG signals and flux signals, and PWA features calculated from PPG signals and flux signals and their derivatives.

[0131] In some embodiments, a combination of some or all of the ML models described above can be used to compute the variables used in equation (7) for blood pressure. In some embodiments, the equations derived above can also be followed and the machine learning model extended to predict blood pressure from flux and PPG signals and waveform features computed from these signals. Various machine learning models can be used to compute blood pressure. Figure 8 An example of how to calculate arterial blood pressure from flux and PPG signals using PINN is shown.

[0132] The ML model uses PPG and flux signals, along with their PWA features, as inputs. Reference blood pressure is used along with control equations to train the neural network. In this example, equations (7), (13), (27), and (29) are used as control equations to constrain the behavior of the neural network. In some embodiments, one or more arterial blood pressure calibration points may also be given as inputs to the ML model.

[0133] One advantage of PINN is that, in addition to guiding the neural network on what the correct value is at each reference point, PINN can also incorporate the temporal and spatial evolution of cardiovascular and hemodynamic variables, which are typically described by partial differential equations.

[0134] Figure 8 An example is shown where PINN learns from input data. The 800 learns to estimate blood pressure, and the input data includes a flux signal, features calculated from the pulse waveform of the flux signal, features calculated from the derivative of the flux, a PPG signal, features calculated from the pulse waveform of the PPG signal, and features calculated from the derivative of the PPG signal. Parameters These are one or more parameters derived from features calculated from flux, and / or one or more parameters derived from indices of blood flow, blood velocity, and moving RBC concentration calculated from speckle contrast signals, and / or one or more parameters derived from features calculated from PPG signals, and / or one or more parameters derived from indices of arterial and tissue oxygenation, blood, and hemoglobin content calculated from PPG signals. The neural network 802 maps the input data 800 to the variables of the governing equation. Reference 804 illustrates the components of the governing equation. Figure 8 In the example shown, governing equation (GE) 806 is

[0135] (32)

[0136] (32)

[0137] (33)

[0138] (34)

[0139] The loss function (15) 808 is based on the control equation and training data.

[0140] When using devices (such as smartwatches, fitness trackers, or other wearable devices including hearing devices) to acquire PPG and flux signals, combinations can be applied. Figure 2 , Figure 6 , Figure 7 and Figure 8 The discussion focuses on trained machine learning models. Using these models, arterial blood pressure can be calculated or estimated based on PPG and flux signals (including their derivatives).

[0141] Figure 9 and Figure 10 Possible embodiments of the system architecture for implementing the above scheme are described. The PPG sensor consists of an amplified analog-to-digital converted signal to drive an LED. One or more photodetectors detect the received signal transmitted back from the tissue. The photodetector signal can be amplified and then converted from an analog signal to a digital signal for processing in a microcontroller unit (MCU) or digital signal processing (DSP) unit.

[0142] The signals from the flux sensor consist of analog-to-digital converted signals, which are amplified to drive a vertical-cavity surface-emitting laser (VCSEL) or one or more other miniaturized laser sources. One or more photodetectors detect the received signals transmitted from the tissue. The photodetector signals can also be amplified and then converted from analog to digital signals for processing in an MCU or DSP unit.

[0143] In some embodiments, the system may also include electrodes and analog circuitry to measure electrocardiogram (ECG) signals. The ECG signals can be used to calculate pulse wave volume (PWV), which provides information about the elastic properties of the arterial wall.

[0144] In some embodiments, the system may have two PPG sensors to measure PWV from the pulse waveform transmission time between the wrist and fingers.

[0145] In some embodiments, the system may include a blood pressure cuff and the electronic and mechanical components required for cuff operation. The blood pressure cuff may be used to provide single-point calibration values ​​for systolic and diastolic blood pressure.

[0146] Figure 9 orFigure 10 The system shown can be included in devices such as smartwatches, wristbands, or other wearable devices.

[0147] Figure 11A The illustration shows a possible component layout provided by an exemplary embodiment. It should be noted that... Figure 11A The layout shown is only one possible layout, and other layouts can also be applied.

[0148] The layout includes two photodetectors 1100, an LED 1102, a VCSEL 1104, and a light barrier 1106. The distances from the VCSEL 1104 and the LED 1102 to the photodetectors 1100 are the same, and the wavelengths of the VCSEL 1104 and the LED 1102 are similar, allowing light from both sources to penetrate to the same depth and thus through the same tissue volume.

[0149] A light barrier 1106 surrounding the light sources 1102 and 1104 prevents direct light leakage from the light sources 1102 and 1104 to the photodetector 1100. The photodetector is preferably placed symmetrically on both sides of the light sources 1102 and 1104 to reduce signal noise.

[0150] In an exemplary embodiment, calibration information from the cuff blood pressure sensor can be provided to improve blood pressure determination from the PPG signal and flux signal.

[0151] Figure 11B Another possible component layout provided by an exemplary embodiment is shown. Figure 11B The layout shown is Figure 11A The difference in the layout shown is that, instead of using two photodetectors, four photodetectors are symmetrically placed around the light sources 1102 and 1104. It should be noted that... Figure 11B The layout shown is only one possible layout, and other layouts can also be applied.

[0152] In an exemplary embodiment, calibration information from the cuff blood pressure sensor can be provided to improve blood pressure determination from the PPG signal and flux signal.

[0153] Figure 12 An example of a method for noninvasively determining continuous arterial blood pressure waveforms or systolic and diastolic blood pressure is shown. For example, the method can be implemented by device 100 or an application performed by device 100.

[0154] At 1200, the method may include acquiring an optical volumetric (PPG) signal from a PPG sensor.

[0155] At 1202, the method may include acquiring a flux signal from a flux sensor.

[0156] At 1204, the method may include determining arterial blood pressure waveforms or systolic and diastolic pressures based at least in part on PPG signals and flux signals.

[0157] In an exemplary embodiment, calibration information from the cuff blood pressure sensor can be provided to improve blood pressure determination from the PPG signal and flux signal.

[0158] The device can be used to perform or cause to perform any aspect of one or more methods described herein. Furthermore, the embedded program or embedded program product may include instructions for causing the device to perform any aspect of one or more methods described herein when executed. Additionally, the device may include modules for performing any aspect of one or more methods described herein. According to an exemplary embodiment, the module may include at least one processor and a memory including program code, the at least one processor, and the program code for causing the device to perform any aspect of one or more methods when executed by the at least one processor.

[0159] Any ranges or device values ​​given herein may be extended or modified without losing the desired effect. Furthermore, unless expressly prohibited, any embodiment may be combined with another embodiment.

[0160] Although the subject matter has been described in structural and / or action-specific language, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or actions described above. Rather, the specific features and actions described above are disclosed as examples of implementing the claims, and other equivalent features and actions are intended to be included within the scope of the claims.

[0161] It should be understood that the advantages and benefits described above may relate to one embodiment or multiple embodiments. The embodiments are not limited to solving any or all of the described problems, nor are they limited to embodiments having any or all of the described advantages and benefits. It should also be understood that reference to "one" item may refer to one or more of these items.

[0162] The steps or operations described herein can be performed in any suitable order, or simultaneously where appropriate. Furthermore, individual boxes can be removed from any method without departing from the scope of the subject matter described herein. Aspects of any exemplary embodiments described above can be combined with aspects of any other exemplary embodiments described to form further exemplary embodiments without affecting the desired effect.

[0163] The term “comprising” is used herein to mean including identified methods, blocks or elements, but such blocks or elements do not include an exclusive list, and methods or apparatus may include additional blocks or elements.

[0164] While a topic may be referred to as the "first" or "second" topic, this does not necessarily indicate any order or importance of the topics. Rather, these attributes can be used simply to distinguish between subjects.

[0165] It should be understood that the above description is given by way of example only, and various modifications can be made by those skilled in the art. The foregoing specification, examples, and data provide a complete description of the structure and use of exemplary embodiments. Although various embodiments have been described above with some degree of specificity or with reference to one or more individual examples, those skilled in the art can make various changes to the disclosed embodiments without departing from the scope of this specification.

Claims

1. A method for non-invasively determining a continuous arterial blood pressure waveform or systolic and diastolic pressure, characterized in that, The method comprises: acquiring a photoplethysmography (PPG) signal from a PPG sensor; acquiring a flux signal from a flux sensor; determining the arterial blood pressure waveform or the systolic and diastolic pressures based at least in part on the PPG signal and the flux signal.

2. The method of claim 1, wherein, Further comprising: determining an index of vascular cross-sectional area based on the PPG signal; determining an index of blood flow based on the flux signal; determining arterial compliance based on the flux signal and the PPG signal; determining the arterial blood pressure waveform or the systolic and diastolic pressures based at least in part on the index of vascular cross-sectional area, the index of blood flow, and the arterial compliance.

3. The method of claim 1, wherein, Further comprising: determining an index of vascular cross-sectional area, tissue oxygenation, arterial oxygenation, blood, and hemoglobin content based on the PPG signal; determining an index of blood flow, blood velocity, and moving red blood cell concentration based on the flux signal; determining arterial compliance based on the flux signal and the PPG signal; determining the arterial blood pressure waveform or the systolic and diastolic pressures based at least in part on cardiovascular and hemodynamic indices acquired in part from the PPG signal and the flux signal.

4. The method according to claim 2 or 3, characterized in that, Further comprising: determining first pulse wave analysis (PWA) features based on the PPG signal; determining second PWA features based on the flux signal; determining the arterial blood pressure waveform or the systolic and diastolic pressures based at least in part on the first PWA features and the second PWA features.

5. The method according to any one of claims 1 to 4, characterized in that, Further comprising: applying at least one trained machine learning model in determining the arterial blood pressure waveform or the systolic and diastolic pressures, the at least one trained machine learning model trained based at least in part on the PPG signal, the flux signal, and blood pressure reference measurements.

6. The method of claim 5, wherein, The at least one trained machine learning model comprises a physics-informed neural network (PINN) based on a mathematical model of blood flow in a human vasculature.

7. The method according to any one of claims 1 to 6, characterized in that, The flux sensor comprises a laser Doppler flowmeter (LDF) sensor, a dynamic light scattering (DLS) sensor, a speckle plethysmography (SPG) sensor, a diffusion correlation spectroscopy (DCS) sensor, a laser speckle contrast imaging (LCSI) sensor, or a laser Doppler perfusion imaging sensor.

8. An apparatus for non-invasively determining a continuous arterial blood pressure waveform or systolic and diastolic pressures, characterized by, The device is for: acquiring a photoplethysmography (PPG) signal from a PPG sensor; acquiring a flux signal from a flux sensor; determine the arterial blood pressure waveform or the systolic and diastolic pressures based at least in part on the PPG signal and the flux signal.

9. The apparatus of claim 8, wherein, Also for: determining an index of a blood vessel cross-sectional area based on the PPG signal; determining an index of blood flow based on the flux signal; determining an arterial compliance based on the flux signal and the PPG signal; determine the arterial blood pressure waveform or the systolic and diastolic pressures based at least in part on the index of the blood vessel cross-sectional area, the index of the blood flow, and the arterial compliance.

10. The apparatus of claim 8, wherein, Also for: estimating an index of a blood vessel cross-sectional area, tissue oxygenation, arterial oxygenation, blood, and hemoglobin content based on the PPG signal; determining an index of blood flow, blood velocity, and moving red blood cell concentration based on the flux signal; determining an arterial compliance based on the flux signal and the PPG signal; determine the arterial blood pressure waveform or the systolic and diastolic pressures based on cardiovascular and hemodynamic indices obtained at least in part from the PPG signal and the flux signal.

11. The apparatus of claim 9 or 10, wherein, Also for: determining a first pulse wave analysis (PWA) feature based on the PPG signal; determining a second PWA feature based on the flux signal; determine the arterial blood pressure waveform or the systolic and diastolic pressures based at least in part on the first PWA feature and the second PWA feature.

12. The apparatus of any one of claims 8 to 11, wherein, Also for: applying at least one trained machine learning model in determining the arterial blood pressure waveform or the systolic and diastolic pressures, the at least one trained machine learning model trained at least in part based on the PPG signal, the flux signal, and blood pressure reference measurements.

13. The apparatus of claim 12, wherein, the at least one trained machine learning model comprises a physics-informed neural network (PINN) based on a mathematical model of blood flow in a human vascular system.

14. The apparatus of any one of claims 8 to 13, wherein, the flux sensor comprises a laser Doppler flowmeter (LDF) sensor, a dynamic light scattering (DLS) sensor, a speckle plethysmography (SPG) sensor, a diffusion correlation spectroscopy (DCS) sensor, a laser speckle contrast image (LCSI) sensor, or a laser Doppler perfusion imaging (LDPI) sensor.

15. The apparatus of any one of claims 8 to 14, wherein, the device comprises the PPG sensor and the flux sensor.

16. A wearable device, comprising: comprising the device according to any one of claims 8 to 15.

17. A computer program, characterized in that, comprising program code for causing execution of the method according to any one of claims 1 to 7 when the computer program is executed on a computer.

Images