WO2025175923A1 - Method and apparatus for blood pressure estimation, electronic device, and storage medium - Google Patents

Abstract

Provided are a method and apparatus for blood pressure estimation, an electronic device, and a storage medium. The method comprises: acquiring a target feature according to at least one of a photoplethysmography (PPG) signal, an inertial measurement unit (IMU) signal, and an electrocardiogram (ECG) signal (101); and performing blood pressure estimation according to the target feature to acquire a blood pressure estimation result (102).

Description

Blood pressure estimation method and device, electronic device, and storage medium

This application claims priority to the Chinese patent application filed on February 23, 2024, with application number 202410207006.X, and invention name “Blood Pressure Estimation Method and Device, Electronic Device, Storage Medium”, the entire contents of which are incorporated by reference into this application.

Technical Field

The embodiments of the present application relate to the technical field of blood pressure analysis, and are related to but not limited to a blood pressure estimation method and device, electronic equipment, and storage medium.

Background Art

The traditional blood pressure measurement solution is based on a pressurized sensor and uses blood vessel occlusion to measure blood pressure.

Since traditional blood pressure measurement methods are manual operations with obvious errors, the blood pressure estimation results obtained are not very accurate.

Summary of the Invention

In a first aspect, the blood pressure estimation method provided in an embodiment of the present application is applied to an electronic device, including:

acquiring a target feature based on at least one of a photoplethysmography (PPG) signal, an inertial measurement unit (IMU) signal, and an electrocardiogram (ECG) signal, wherein the target feature includes at least one of a PPG signal feature, an IMU signal feature, and an ECG signal feature;

Blood pressure estimation is performed according to the target feature to obtain a blood pressure estimation result; wherein,

The PPG signal feature includes at least one of the following: a pulse waveform analysis PWA feature, a zero-crossing feature, a nonlinear dynamics feature, a pulse transit time PTT feature, a pulse arrival time PAT feature, a first heart rate feature, and a first neural network feature; and/or,

The IMU signal feature includes at least one of the following: a second heart rate feature, a relative stroke volume feature, a heart rhythm signal BCG high frequency feature, an IMU multi-axis feature, and a second neural network feature; and/or,

The ECG signal features include third neural network features.

In a second aspect, the blood pressure estimation device provided in an embodiment of the present application is applied to an electronic device, including:

a feature acquisition module, configured to acquire a target feature based on at least one of a photoplethysmography (PPG) signal, an inertial measurement unit (IMU) signal, and an electrocardiogram (ECG) signal, wherein the target feature includes at least one of a PPG signal feature, an IMU signal feature, and an ECG signal feature; wherein the PPG signal feature includes at least one of: a pulse waveform analysis (PWA) feature, a zero-crossing feature, a nonlinear dynamics feature, a pulse transit time (PTT) feature, a pulse arrival time (PAT) feature, a first heart rate feature, and a first neural network feature; and/or the IMU signal feature includes at least one of: a second heart rate feature, a relative stroke volume feature, a heart rhythm signal (BCG) high-frequency feature, an IMU multi-axis feature, and a second neural network feature; and/or the ECG signal feature includes a third neural network feature;

The result acquisition module is used to estimate the blood pressure according to the target characteristics and obtain the blood pressure estimation result.

In a third aspect, the blood pressure estimation device provided by the embodiment of the present application includes: at least one sensor selected from a photoplethysmography (PPG) sensor, an inertial measurement unit (IMU) sensor, and an electrocardiogram (ECG) sensor, and a processor, wherein:

The PPG sensor is configured to collect PPG signals;

The IMU sensor is configured to collect IMU signals;

The ECG sensor is configured to collect ECG signals;

The processor is configured to obtain a target feature based on at least one of the PPG signal, the IMU signal, and the ECG signal, wherein the target feature includes at least one of a PPG signal feature, an IMU signal feature, and an ECG signal feature; and perform blood pressure estimation based on the target feature to obtain a blood pressure estimation result;

The PPG signal feature includes at least one of the following: a pulse waveform analysis PWA feature, a zero-crossing feature, a nonlinear dynamics feature, a pulse transit time PTT feature, a pulse arrival time PAT feature, a first heart rate feature, and a first neural network feature; and/or,

The IMU signal feature includes at least one of the following: a second heart rate feature, a relative stroke volume feature, a heart rhythm signal BCG high frequency feature, an IMU multi-axis feature, and a second neural network feature; and/or,

The ECG signal features include third neural network features.

In a fourth aspect, the blood pressure estimation device provided in an embodiment of the present application includes a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and when the processor executes the program, the steps of the blood pressure estimation method provided in the first aspect of the embodiment of the present application are implemented.

In a fifth aspect, the computer-readable storage medium provided in an embodiment of the present application stores a computer program thereon, which, when executed by a processor, implements the steps of the blood pressure estimation method provided in the first aspect of the embodiment of the present application.

In a sixth aspect, the computer program product provided in an embodiment of the present application includes a computer program, which, when executed by a processor, implements the steps in the blood pressure estimation method provided in the first aspect of the embodiment of the present application.

BRIEF DESCRIPTION OF THE DRAWINGS

The drawings herein are incorporated into and constitute a part of the specification. These drawings illustrate embodiments consistent with the present application and, together with the specification, are used to illustrate the technical solutions of the present application.

FIG1 is a schematic diagram of the structure of a blood pressure estimation system provided in an embodiment of the present application;

FIG2 is a schematic diagram of an implementation flow of a blood pressure estimation method provided in an embodiment of the present application;

FIG3 is a flow chart of a method for obtaining zero-crossing features according to an embodiment of the present application;

FIG4 is a schematic diagram of a scenario of a zero-crossing feature extraction method provided in an embodiment of the present application;

FIG5 is a schematic diagram of a flow chart of a method for acquiring nonlinear dynamic characteristics provided in an embodiment of the present application;

FIG6 is a flow chart of a method for obtaining features of a neural network autoencoder provided in an embodiment of the present application;

FIG7 is a flowchart of a method for acquiring PWA features according to an embodiment of the present application;

FIG8 is a diagram showing an application example of a nonlinear dynamic feature provided by an embodiment of the present application;

FIG9 is a flow chart of a method for extracting a second heart rate feature according to an embodiment of the present application;

FIG10 is a flow chart of a method for obtaining relative stroke volume according to an embodiment of the present application;

FIG11 is a flow chart of a method for obtaining BCG high-frequency features according to an embodiment of the present application;

FIG12 is a waveform diagram of a BCG signal provided in an embodiment of the present application;

FIG13 is a schematic structural diagram of a blood pressure estimation device provided in an embodiment of the present application;

FIG14 is a schematic structural diagram of an electronic device provided in an embodiment of the present application.

DETAILED DESCRIPTION

To make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the specific technical solutions of the present application will be further described in detail below in conjunction with the drawings in the embodiments of the present application. The following embodiments are used to illustrate the present application but are not intended to limit the scope of the present application.

Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this application pertains. The terms used herein are for the purpose of describing the embodiments of this application only and are not intended to limit this application.

In the following description, references to "some embodiments" describe a subset of all possible embodiments. However, it should be understood that "some embodiments" may be the same subset or different subsets of all possible embodiments, and may be combined with each other without conflict. The term "at least one" in this application may also be understood to mean "one or more," where "a plurality" may include "two or more."

It should be pointed out that the terms "first\second\third" involved in the embodiments of the present application are used to distinguish similar or different objects, and do not represent a specific ordering of the objects. It can be understood that "first\second\third" can be interchanged with a specific order or sequence where permitted, so that the embodiments of the present application described here can be implemented in an order other than that illustrated or described here.

Traditional blood pressure measurement methods rely on pressurized sensors that use vascular occlusion to measure blood pressure. Due to significant human error, the resulting blood pressure estimates are less accurate.

In view of this, an embodiment of the present application provides a blood pressure estimation method that can obtain target features based on at least one of a photoplethysmography (PPG) signal, an inertial measurement unit (IMU) signal, and an electrocardiogram (ECG) signal. The target features include at least one of the PPG signal features, the IMU signal features, and the ECG signal features. Blood pressure is estimated based on the target features to obtain a blood pressure estimation result. This method improves the accuracy of blood pressure estimation by extracting target features from at least one signal collected by a cuffless detection sensor.

FIG1 is a schematic diagram of the structure of a blood pressure estimation system provided in an embodiment of the present application. As shown in FIG1 , the blood pressure estimation system may include a blood pressure estimation device 11 and a terminal device 12 , and the terminal device 12 may establish a wireless connection with the blood pressure estimation device 11 .

Among them, the terminal device 12 can be: a mobile phone, tablet computer, TV or audio device that can interact with the blood pressure estimation device 11 and has data calculation functions. The embodiment of this application does not limit the type of terminal device 12.

The blood pressure estimation device 11 may be a device such as a smart watch or smart bracelet embedded with a sensor, which can collect signals through the sensor with or without contact with the object being measured. The embodiment of the present application does not limit the blood pressure estimation device 11. After collecting the signal through the sensor, the blood pressure estimation device 11 can send the signal to the terminal device 12, so that the terminal device 12 performs signal processing and/or performs blood pressure estimation through algorithm calculation. Exemplarily, the sensor embedded in the blood pressure estimation device 11 may include any one, any two, or three of a photoelectric volumetric pulse wave (PPG) sensor, an inertial measurement unit (IMU) sensor, and an electrocardiogram (ECG) sensor. The present application does not limit the type of sensor embedded in the blood pressure estimation device 11.

In some embodiments, when the terminal device 12 is performing blood pressure measurement or obtaining the result of the blood pressure measurement, the measurement information and/or the measurement result may be displayed to the user so that the user can understand the situation of the blood pressure measurement.

It can be understood that the blood pressure estimation method provided in the embodiment of the present application can be applied to the terminal device 12 in the above-mentioned blood pressure measurement system, that is, the terminal device 12 performs blood pressure estimation based on the signal sent by the blood pressure estimation device 11 to obtain a blood pressure estimation result.

In some embodiments, the blood pressure measurement system may include only the blood pressure estimation device 11. After collecting signals through sensors, the blood pressure estimation device 11 may perform signal processing using its own processor and/or perform algorithmic calculations to estimate blood pressure. The blood pressure estimation device 11 may also display blood pressure measurement process and result information to the user, thereby allowing the user to understand the blood pressure measurement status.

It can be understood that when the blood pressure measurement system only includes the blood pressure estimation device 11, the blood pressure estimation method provided in the embodiment of the present application can also be applied to the blood pressure estimation device 11, that is, the blood pressure estimation device 11 performs blood pressure estimation based on the signal collected by itself to obtain a blood pressure estimation result.

The technical solutions in the embodiments of the present application will be described below in conjunction with the drawings in the embodiments of the present application.

FIG2 is a schematic diagram of an implementation flow of a blood pressure estimation method provided in an embodiment of the present application. The blood pressure estimation method can be applied to electronic devices, which can be various types of devices with information processing capabilities during implementation. For example, the electronic device can be the terminal device 12 shown in FIG1 , or the blood pressure estimation device 11 shown in FIG1 . As shown in FIG2 , the method can include the following steps 101 to 102:

Step 101: Acquire target features based on at least one of a photoplethysmography (PPG) signal, an inertial measurement unit (IMU) signal, and an electrocardiogram (ECG) signal, wherein the target features include at least one of a PPG signal feature, an IMU signal feature, and an ECG signal feature. The PPG signal features include at least one of: a pulse waveform analysis (PWA) feature, a zero-crossing feature, a nonlinear dynamics feature, a pulse transit time (PTT) feature, a pulse arrival time (PAT) feature, a first heart rate feature, and a first neural network feature; and/or the IMU signal features include at least one of: a second heart rate feature, a relative stroke volume feature, a heart rhythm signal (BCG) high-frequency feature, an IMU multi-axis feature, and a second neural network feature; and/or the ECG signal features include a third neural network feature.

It should be noted that after acquiring at least one of the PPG signal, the IMU signal, and the ECG signal, the electronic device can obtain a target feature based on the acquired signal. The target feature can be at least one of the PPG signal feature, the IMU signal feature, and the ECG signal feature. That is, the target feature can include only the PPG signal feature, or only the IMU signal feature, or only the ECG feature, or any two of the three features, or all three features, and this embodiment of the application is not limited to this.

In some embodiments, the PPG signal feature may include at least one of the following: a pulse waveform analysis PWA feature, a zero-crossing feature, a nonlinear dynamics feature, a pulse transit time PTT feature, a pulse arrival time PAT feature, a first heart rate feature, and a first neural network feature.

Exemplarily, the PWA feature may be extracted based on waveform feature analysis of the PPG signal. The zero-crossing feature may be extracted based on the derivative signal of the standardized PPG signal. The nonlinear dynamic feature may be extracted based on PPG signals corresponding to multiple heartbeat cycles in the PPG signal. The PTT feature and the PAT feature may be extracted based on signal peaks of the PPG signal. The first heart rate feature may be extracted by segmenting the waveform features of the PPG signal and matching the positions of corresponding feature points in each segment. The first neural network feature may be obtained by extracting features from the PPG signal based on a preset neural network.

In some embodiments, the IMU signal feature may include at least one of the following: a second heart rate feature, a relative stroke volume feature, a heart rhythm signal BCG high frequency feature, an IMU multi-axis feature, and a second neural network feature.

Exemplarily, the relative stroke volume feature can be extracted based on the amplitude of the BCG signal extracted from the IMU signal when the object under measurement is in a stable measurement state, the BCG high-frequency feature can be extracted based on the waveform morphology characteristics of the BCG signal extracted from the IMU signal, the IMU multi-axis feature can be extracted based on data of acceleration and attitude angle in three directions, and the second neural network feature can be obtained by extracting features from the IMU signal based on a preset neural network.

In some embodiments, the ECG signal features may include third neural network features.

Exemplarily, the third neural network feature may be obtained by extracting features from the ECG signal based on a preset neural network.

It is understandable that the electronic device may obtain at least one of the photoplethysmography (PPG) signal, the inertial measurement unit (IMU) signal, and the electrocardiogram (ECG) signal by receiving signals sent by other devices or by collecting signals itself, and this embodiment of the present application does not limit this.

Step 102: Estimating blood pressure based on the target characteristics to obtain a blood pressure estimation result.

It should be noted that there are many ways to estimate blood pressure based on target characteristics. For example, a trained blood pressure estimation model can be input for calculation, or calculation can be performed through a preset algorithm formula. The embodiment of the present application does not limit the method of estimating blood pressure based on the target characteristics and obtaining the blood pressure estimation result.

The blood pressure estimation method provided in an embodiment of the present application obtains target features based on at least one of a photoplethysmography (PPG) signal, an inertial measurement unit (IMU) signal, and an electrocardiogram (ECG) signal. The target features include at least one of the PPG signal features, the IMU signal features, and the ECG signal features. A blood pressure estimation result is obtained based on the target features. Because the signals collected by the cuffless detection sensor can be continuous or multimodal, this solution uses target features extracted from at least one signal collected by the cuffless detection sensor to perform blood pressure estimation. Therefore, the resulting blood pressure estimation result is more accurate than the detection results of the prior art, thereby improving the accuracy of blood pressure estimation.

In some embodiments, before obtaining the target feature based on at least one of the photoplethysmography (PPG) signal, the inertial measurement unit (IMU) signal, and the electrocardiogram (ECG) signal, the blood pressure estimation method may further include: collecting at least one of the photoplethysmography (PPG) signal, the inertial measurement unit (IMU) signal, and the electrocardiogram (ECG) signal.

It should be noted that the above-mentioned signals can be collected by electronic devices, or collected by other devices and then input into electronic devices. The embodiments of the present application do not limit the collection method and transmission method.

In an embodiment of the present application, obtaining target features based on at least one of a photoplethysmogram (PPG) signal, an inertial measurement unit (IMU) signal, and an electrocardiogram (ECG) signal may include a variety of situations: obtaining target features based on a PPG signal alone; obtaining target features based on an IMU signal alone; obtaining target features based on an ECG signal alone; obtaining target features based on a PPG signal and an IMU signal; obtaining target features based on a PPG signal and an ECG signal; obtaining target features based on an IMU signal and an ECG signal; or obtaining target features based on a PPG signal, an IMU signal, and an ECG signal. The embodiments of the present application do not limit the combination of the three signals and the type of combination of the target features obtained. This will be explained below with some examples.

In some embodiments, obtaining target features based on at least one of a photoplethysmography (PPG) signal, an inertial measurement unit (IMU) signal, and an electrocardiogram (ECG) signal may include: obtaining target features based on the PPG signal, wherein the target features include at least one of a PPG signal feature, an IMU signal feature, and an ECG signal feature.

Among them, the PPG signal feature may include at least one of the following: pulse waveform analysis PWA feature, zero crossing feature, nonlinear dynamic feature, pulse transit time PTT feature, pulse arrival time PAT feature, first heart rate feature and first neural network feature.

Exemplarily, the PWA feature can be extracted based on waveform feature analysis of the PPG signal, the zero-crossing feature can be extracted based on the derivative signal of the standardized PPG signal, the nonlinear dynamic feature can be extracted based on the PPG signals corresponding to multiple heartbeat cycles in the PPG signal, the PTT feature or the PAT feature can be extracted based on the signal peak of the PPG signal, and the first neural network feature can be obtained by performing feature extraction on the PPG signal based on a preset neural network.

It can be understood that when obtaining target features based on PPG signals, the target features may include at least one of pulse wave waveform analysis PWA features, zero crossing features, nonlinear dynamic features, pulse transfer time PTT features, pulse arrival time PAT features, first heart rate features and first neural network features.

In some embodiments, the blood pressure estimation method further includes at least one of the following steps: extracting the PWA feature based on waveform feature analysis of the PPG signal; extracting the zero-crossing feature based on the derivative signal of the standardized PPG signal; extracting the nonlinear dynamic feature based on the PPG signal corresponding to multiple heartbeat cycles in the PPG signal; extracting the PTT feature or the PAT feature based on the signal peak of the PPG signal; segmenting the waveform feature of the PPG signal, and matching the positions of corresponding feature points in each segment to extract the first heart rate feature; and extracting the first neural network feature from the PPG signal based on a preset neural network.

It is understandable that the above steps may be interrelated or independent, and the embodiments of the present application do not limit this.

In some embodiments, the blood pressure estimation method may include extracting the PWA feature based on waveform feature analysis of the PPG signal.

It should be noted that the method for extracting PWA features based on the waveform features of the PPG signal may include multiple steps such as signal preprocessing, heartbeat detection, feature extraction, and feature analysis. Among them, the PWA features may include at least one of the following features: Peak time: the time when the peak of the heartbeat waveform occurs, which is related to the heart contraction time. Peak amplitude: the peak amplitude of the heartbeat waveform, which is related to the strength of the heart contraction. Dicrotic wave: in the heartbeat waveform, the second smaller peak following the main peak, which is related to the diastolic function of the heart. Waveform width: the width of the heartbeat waveform, which is related to the duration of the heart contraction. Waveform symmetry: the symmetry of the heartbeat waveform, which is related to the elasticity of the blood vessels. This application does not limit the method for extracting PWA features and the types of features included in the PWA features.

In some embodiments, the blood pressure estimation method may include extracting the zero-crossing feature based on a derivative signal of the normalized PPG signal.

It's important to note that when extracting PWA features from PPG signals, the peaks of the SDPPG waveform (the second-order derivative of the PPG signal sequence, or the first-order difference sequence of the PPG signal sequence) are susceptible to noise, and extracting these waveform features is particularly difficult for low peak heights. Young people with good vascular elasticity will have distinct peaks in their SDPPG waveforms. However, for older people whose vascular elasticity deteriorates due to aging, the SDPPG waveform may lack noticeable fluctuations, and/or the waveform after noise reduction filtering may lack continuous downstream peaks. The disappearance of these peaks can lead to the loss of corresponding features. Therefore, other features can be extracted from the PPG signal to supplement the PWA features.

FIG3 is a flow chart of a method for obtaining a zero-crossing feature according to an embodiment of the present application. As shown in FIG3 , extracting the zero-crossing feature based on the derivative signal of the standardized PPG signal may include:

Step 201: Obtain a standardized target order derivative signal according to the PPG signal, wherein the standardized target order derivative signal includes a first-order derivative signal, a second-order derivative signal, and a higher-order derivative signal.

Step 202: Draw a plurality of contour lines at preset intervals on the waveform of the normalized target order derivative signal to obtain horizontal intersection points between the plurality of contour lines and the waveform of the normalized target order derivative signal.

Step 203: The number of the horizontal crossing points, the time length in the waveform of the normalized target order derivative signal, and the proportion of the time length in each heartbeat cycle in the PPG signal are used as the zero-crossing features.

It should be noted that high-order mean filtering and differential filtering can be used to enhance the PPG harmonic frequency information and make the characteristic peaks in the waveform more prominent.

FIG4 is a schematic diagram of a scenario for extracting zero-crossing features provided by an embodiment of the present application. As shown in FIG4 , based on the standardized first-order, second-order, and higher-order derivatives of the signal, assuming that the maximum amplitude of the waveform is 100%, multiple contour lines are drawn at arbitrary intervals (intervals can be selected as 5%, 10%, 20%, etc.) in the waveform. As shown in FIG4 , the horizontal intersections of the multiple contour lines and the waveform are obtained, and the number of intersections, the duration of the intersections within the waveform, and the proportion of the duration within the cycle are used as features for blood pressure estimation.

As shown in Figure 4, through the efficient zero-crossing feature extraction method, two intersection points x1 and x2 are obtained at the interval of -30%. Six horizontal intersection points are obtained at the interval of 0%, with the time length between the x3 and x4 intersection points being 0.21s, the time length between the x5 and x6 intersection points being 0.08s, and the time length between the x7 and x8 intersection points being 0.21s.

In some embodiments, the blood pressure estimation method may include extracting the nonlinear dynamic feature based on PPG signals corresponding to multiple heartbeat cycles in the PPG signal.

It is understandable that the related technical solutions do not contain targeted analysis tools based on nonlinear dynamics, and lack analysis of the periodic characteristics of PPG signals. The embodiments of the present application can reconstruct chaotic attractors based on pulse wave PPG and/or ballistocardiograph (BCG).

FIG5 is a flow chart of a method for acquiring nonlinear dynamic features provided by an embodiment of the present application. As shown in FIG5 , the method for extracting nonlinear dynamic features based on the PPG signals corresponding to multiple heartbeat cycles in the PPG signal may include:

Step 301: Performing a high-order expansion on the PPG signal based on calculation of an embedding dimension and a delay time to obtain an expanded signal, wherein the calculation of the embedding dimension includes a Takkens embedding theorem method, and the calculation of the delay time includes at least one of a mutual information method and an autocorrelation method.

Step 302: quantizing the expanded signal to obtain the nonlinear dynamic characteristics, where the nonlinear dynamic characteristics include at least one of a Lyapunov exponent and a correlation dimension.

It should be noted that by reconstructing the chaotic attractor of the PPG signal, the high-dimensional nonlinear dynamic characteristics of the signal can be effectively expanded. This includes constructing the embedding dimension and delay time, allowing the signal to be maximally expanded in the optimal dimensional space. For feature quantification, classical calculation parameters such as the correlation dimension and Lyapunov exponent can be used to quantify nonlinear characteristics.

Exemplarily, the embedding dimension and delay time can also be used as the input dimensions of the preset neural network for extracting nonlinear dynamic features, and the chaotic attractor expansion can be used as the input layer of the neural network for analysis. For example, the learning target is the relative change of blood pressure in multiple measurements, and the nonlinearly expanded signal attractor signal (including but not limited to PPG single input, PPG multi-channel input, BCG single-channel input, BCG multi-axis input and PPG and BCG mixed input) is used as the input layer, and the neural network with the relative change of blood pressure as the target is trained.

In some embodiments, the blood pressure estimation method may include extracting the PTT feature or the PAT feature based on a signal peak of the PPG signal.

Exemplary methods for extracting the PTT feature based on the peak value of the PPG signal can include extracting the PTT feature by collecting PPG signals at multiple wavelengths, performing signal preprocessing, identifying feature points, calculating time differences, and analyzing the results. Signal acquisition: PPG signals can be collected simultaneously from two different locations (e.g., the wrist and finger). This can be achieved by placing photosensors at the corresponding locations. Signal preprocessing: The collected PPG signals are preprocessed, including operations such as denoising, filtering, and smoothing, to improve signal quality and accuracy. Feature point identification: Feature points are identified in the PPG signal, such as peaks or valleys in the waveform. These feature points typically correspond to the arrival time of the pulse wave. Time difference calculation: The time difference between feature points in the PPG signals from two different locations, i.e., the pulse wave transit time (PTT), is calculated. This can be achieved by measuring the time interval between the two feature points. Result analysis: Analysis is performed based on the calculated PTT value.

In some embodiments, the time difference between PPG signals of different wavelengths collected at the same body part can also be regarded as the pulse transit time (PTT_MW) of blood vessels under the skin within a short period of time.

Exemplarily, obtaining the PTT feature based on the PPG signal may include: extracting the peak value within the beat cycle of each PPG signal among the multiple PPG signals, and calculating the time difference between the peak values between each two of the PPG signals to obtain a plurality of time differences of pulse transit times; performing correlation analysis on the plurality of time differences of pulse transit times and the actually measured blood pressure values to obtain a multi-wavelength pulse transit time feature, wherein the multi-wavelength pulse transit time feature is used to indicate the optimal wavelength combination corresponding to the multi-wavelength pulse transit time.

It should be noted that the time difference between MW PPG signals (i.e., PTT_MW) may also be related to arterial wall characteristics and can therefore be used to track blood pressure. The acquisition system is used to simultaneously collect PPG signals generated by blue, green, yellow, and infrared light, extract the peak values within the beat cycle of the four PPG signals, and calculate the difference between the peak values of each pair at different wavelengths (i.e., PTT_MW, the time difference characteristic between multiple PPG sensors). The correlation between PTT_MW and blood pressure between light of different wavelengths is analyzed to determine the optimal wavelength combination for PTT_MW. PTT_MW is also a special form of PTT, so the PTT_MW method can also address the challenges of existing PTT-based blood pressure measurement methods, such as the establishment and calibration of the PTT-BP model.

Exemplarily, the method for extracting the PAT feature based on the peak value of the PPG signal can also be implemented through the steps of signal acquisition, signal preprocessing, R-wave identification, feature point identification, time calculation, and result calculation. Signal acquisition: A photoelectric sensor can be used to collect PPG signals from specific body parts (such as fingers or wrists). Ensure close contact between the sensor and the skin to obtain high-quality signals. Signal preprocessing: The collected PPG signal is preprocessed, including operations such as denoising, filtering, and smoothing, to reduce interference and artifacts in the signal. R-wave identification: The R wave (i.e., the pulse wave caused by the QRS complex in the electrocardiogram) is identified in the PPG signal. The R wave is a distinct feature in the PPG signal and typically corresponds to the onset of cardiac contraction. Feature point identification: Feature points corresponding to the R wave are identified in the PPG signal, such as the peak or trough of the pulse wave. These feature points indicate the time when the pulse wave arrives at a specific body part. Time calculation: The time interval from the R wave to the feature point, i.e., the pulse arrival time (PAT), is calculated. This time interval can be measured using a timestamp or timer. Result analysis: Analysis is performed based on the calculated PAT value. PAT is related to cardiovascular health status and can be used to assess physiological parameters such as arterial stiffness and blood pressure.

In some embodiments, the blood pressure estimation method may include segmenting the waveform features of the PPG signal and matching the positions of corresponding feature points in each segment to extract the first heart rate feature.

In one exemplary method for extracting instantaneous heart rate from a PPG signal, heart rate extraction is accomplished by segmenting the PPG waveform into segments and matching the corresponding feature points within each segment. In another example, a machine learning algorithm is trained based on a gold standard (e.g., the ECG R peak) or by adding weak labels (e.g., strategy-based instantaneous peak search). One example employs a pre-defined one-dimensional convolutional neural network (CNN) for extracting the first heart rate feature, employing multiple network layers, multiple fully connected layers, and a self-attention mechanism to output the probability of the corresponding feature peak location.

In some embodiments, the blood pressure estimation method may include extracting the first neural network feature from the PPG signal based on a preset neural network.

It should be noted that the features of the PPG signal can be extracted through a preset neural network.

Among them, the first neural network feature may include at least one of a neural network autoencoder feature and a discriminant neural network feature, the neural network autoencoder feature is extracted based on an autoencoder of a convolutional neural network, and the discriminant neural network feature is extracted based on a discriminant neural network model.

For example, a convolutional neural network-based autoencoder can first compress and then restore the collected PPG signal, optimizing the network using a loss function such as reconstruction loss. Overall, the "information bottleneck" principle is utilized to extract key information from the PPG signal, resulting in a final encoding as a salient feature. A discriminant-based neural network model is trained and validated using the PPG signal, and features obtained during model training are extracted as first neural network features. The present embodiment does not limit the convolutional neural network-based autoencoder or the method for obtaining the first neural network features using a discriminant-based neural network model.

FIG6 is a flow chart of a method for obtaining neural network autoencoder features provided in an embodiment of the present application. As shown in FIG6 , the neural network autoencoder features are extracted by an autoencoder based on a convolutional neural network, and the autoencoder based on the convolutional neural network includes an encoder and a decoder. The convolutional neural network autoencoder extracts the neural network autoencoder features, which may include:

Step 401: Map the PPG signal to a latent space encoding through the encoder to obtain the latent variable.

Step 402: The latent variable is decoded by the decoder and mapped back to the signal space of the PPG signal to obtain the neural network autoencoder feature.

It should be noted that the neural network autoencoder features are features obtained through signal compression and salient feature recognition based on machine learning. The PPG signal is compressed through a network with an auto-encoder structure. The network is divided into two parts: an encoder and a decoder. The encoder maps the preprocessed PPG signal to a code in the latent space, and the decoder decodes the code and maps it back to the PPG signal space. The encoder and decoder networks here can be basic networks such as multi-layer perceptrons (MLPs), convolutional neural networks (CNNs), recurrent neural networks (RNNs), and transformers. The network details are designed to first compress the signal and then restore it. The network is optimized using loss functions such as reconstruction loss. Overall, the "information bottleneck" principle is used to extract key information from the PPG signal, and the final code is obtained as a salient feature.

In some embodiments, the first neural network feature is a discriminant neural network feature extracted by a discriminant-based neural network model, the discriminant-based neural network model is a network for time series data modeling, the discriminant-based neural network model includes a fully connected layer, and obtaining the discriminant neural network feature based on the collected PPG signal may include: when the discriminant-based neural network model is trained by the PPG signal, extracting the tensor before the fully connected layer of the discriminant-based neural network model as the discriminant neural network feature.

It should be noted that a discriminative neural network is first constructed. This neural network can be a deep residual neural network (ResNet), a deep learning time series classification (InceptionTime), an attention-based sequence model (Transformer), or any other network or variant thereof suitable for time series data modeling. Taking ResNet as an example, the input is set to a single channel with a length of 1000. The kernel size of the first convolutional layer is 15 and the number of channels is 16. The kernel size of subsequent layers is 3. Four subsequent stages are followed, each with four basic blocks. Each basic block consists of two convolutional layers, an activation layer, and a MaxPooling layer with a cross-layer connection. The number of channels doubles with each basic block. After the last basic block, an average pooling layer is used to pool the length of each channel in the tensor to 1. Then, a two-layer fully connected layer and an input gate sigmoid layer are used to map the tensor to the predicted result, outputting the probability of positive. The model is trained and validated using PPG signals from a subset of individuals, and a trained neural network is obtained upon convergence. The tensor before the fully connected layer is also extracted as a deep learning feature and combined with the PWA feature.

In some application scenarios, the PPG signal measured by a single-wavelength light source is mixed with pulsation components of different blood vessel types, which may lead to inaccurate physiological measurement results. Therefore, the embodiment of the present application proposes a multi-wavelength (MW) photoplethysmography (PPG) method, which analyzes the correlation delay and waveform differences of multiple PPG sensors.

FIG7 is a flow chart of a method for acquiring PWA features according to an embodiment of the present application. As shown in FIG7 , the PPG signal includes multiple PPG signals of different wavelengths, and the analysis and extraction of the PWA features based on the waveform characteristics of the PPG signal may include:

Step 501: performing calculations to remove capillary pulsation interference based on at least two PPG signals of different wavelengths among the multiple PPG signals to obtain a signal after removal.

Step 502: extracting arterial blood pulsation from the removed signal to obtain an arterial blood pulsation waveform.

Step 503: Extract features from the arterial blood pulsation waveform to obtain the PWA features.

It should be noted that the multi-wavelength photoplethysmography (MW PPG) signal contains blood pulsation information of different blood vessels at different skin depths. Blue light and green light can only reach shallow capillaries, yellow light can further reach the arterioles in the dermis, and longer-wavelength red light and infrared light can penetrate the skin to reach the arteries in the subcutaneous tissue. The MW PPG system requires two to three light sources of different wavelengths to remove the interference of capillary pulsation contained in the short-wavelength PPG from the long-wavelength PPG signal, leaving behind a pure waveform of arterial blood pulsation. For example, arterial blood pulsation is extracted by using a multi-wavelength multi-layer light-skin interaction model derived from the modified Beer-Lambert law and a quasi-analytical self-calibration algorithm. By extracting PWA features from the arterial vascular pulsation waveform or combining it with ECG calculations, an accurate PTT can be obtained as input to the blood pressure model to estimate the blood pressure value.

It can be understood that the above embodiments may be interrelated or exist independently.

The above content describes feature extraction methods based on different aspects of PPG signals, including: extracting PWA features, zero-crossing features, nonlinear dynamic features, pulse transit time (PTT) features, pulse arrival time (PAT) features, first heart rate features, and first neural network features. It also includes extracting PTT features and PAT features based on PPG signals of multiple wavelengths. The combination of features from different aspects can be used to model the direction and measurement of blood pressure changes.

In some embodiments, performing blood pressure estimation based on the target feature to obtain the blood pressure estimation result may include: inputting the PPG signal feature into a preset model, such as a trained blood pressure estimator, to obtain the blood pressure estimation result.

For example, the trained blood pressure estimator is obtained by merging the collected PPG signal features to obtain a merged feature, then obtaining the mean feature corresponding to the merged feature of a preset time length in the merged feature, and training the initial blood pressure estimator through the mean feature.

Among them, blood pressure estimation can include blood pressure trend estimation, hypertension grade estimation, or hypertension stratification estimation, among others. Among them, the grading of hypertension can include categorizing blood pressure into grades 1, 2, and 3, with a cutoff of 140/90 mmHg. Hypertension risk stratification can include four levels: low risk, moderate risk, high risk, and very high risk. The embodiments of this application do not limit the content of blood pressure estimation.

As can be understood, because PPG signal features include at least one of the following: PWA feature, zero-crossing feature, nonlinear dynamic feature, pulse transit time (PTT), characteristic pulse arrival time (PAT), first heart rate feature, and first neural network feature, some or all of these PPG signal features can be input into a trained blood pressure estimator to obtain a blood pressure estimation result. In actual product applications, the input PPG signal features can be considered based on the training status of the estimator, the validity of the features extracted from the collected signal, and other factors, and can even be dynamically adjusted.

The following describes an exemplary application of some embodiments of the present application in a practical application scenario.

Example 1: Blood pressure stratification based on the zero-crossing feature of the PPG signal. The steps are as follows:

1. Collect PPG signals, such as a 1-minute 100Hz signal with a length of 6000;

2. Filter the signal to obtain a filtered signal;

3. Use the peak-finding algorithm to split the signal into multiple signals and obtain the signals corresponding to all heartbeat cycles;

4. Extract features from the signal corresponding to each heartbeat cycle:

The filtered PPG signal is derivatived to obtain its first-, second-, and higher-order derivatives, all of which exhibit periodic fluctuations. The zero-crossing points of each derivative are determined, and each periodic signal is segmented using these zero-crossing points. The waveform is then normalized based on the maximum amplitude of each periodic signal, with the maximum amplitude being 100%. Contour lines are drawn on the normalized waveform at arbitrary intervals (intervals can be set to any value not exceeding 100%, such as 5% or 10%). The horizontal intersections of the contour lines with the waveform, x1, x2, x3, x4, etc., are obtained. The number of intersections, the duration of these intersections within the waveform (|x2-x1|, |x4-x3|, etc.), and the proportion of these durations within the period (|x2-x1|/T, |x4-x3|/T, etc.) are used as features for blood pressure estimation.

5. Combine the extracted features and take the average of the features between multiple heartbeat cycles in one minute to obtain the features of this minute.

6. Take the blood pressure layer labels corresponding to the signal and train a blood pressure estimator, such as the extreme gradient boosting algorithm XGBoost.

7. Then you can apply this classifier and perform the above filtering and extraction operations on any signal data of the user to obtain the corresponding features. Input them into the classifier to obtain the blood pressure stratification results.

Example 2: Blood pressure stratification based on the PWA feature and zero-crossing feature of the PPG signal. The steps are as follows:

1. Collect PPG signals, such as a 1-minute 100Hz signal with a length of 6000;

2. Filter the signal to obtain a filtered signal;

3. Use the peak-finding algorithm to split the signal into multiple signals and obtain the signals corresponding to all heartbeat cycles;

4. Extract features from the signal corresponding to each heartbeat cycle:

a) PWA characteristics;

b) Derivatives of the filtered PPG signal are calculated to obtain the first-order, second-order, and higher-order derivatives of the PPG signal, all of which exhibit periodic fluctuations. The zero-crossing points of each derivative are found, and each periodic signal is segmented using these zero-crossing points. The waveform is then normalized based on the maximum amplitude of each periodic signal, with the maximum amplitude being 100%. Contour lines are drawn on the normalized waveform at arbitrary intervals (intervals can be set to any value not exceeding 100%, such as 5% or 10%). The horizontal intersection points x1, x2, x3, x4, etc. of the contour lines with the waveform are obtained. The number of intersection points, the duration of these intersections within the waveform (|x2-x1|, |x4-x3|, etc.), and the proportion of these durations within the period (|x2-x1|/T, |x4-x3|/T, etc.) are used as features for blood pressure estimation.

5. Combine the extracted features and take the average of the features between multiple heartbeat cycles in one minute to obtain the features of this minute.

6. Take the blood pressure layer labels corresponding to the signal and train a blood pressure estimator, such as the extreme gradient boosting algorithm XGBoost.

7. Then you can apply this classifier and perform the above filtering and extraction operations on any signal data of the user to obtain the corresponding features. Input them into the classifier to obtain the blood pressure stratification results.

Example 3: Blood pressure stratification based on the PWA feature, zero-crossing feature, and nonlinear dynamics feature of the PPG signal. The steps are as follows:

1. Collect PPG signals, such as a 1-minute 100Hz signal with a length of 6000;

2. Filter the signal to obtain a filtered signal;

3. Use the peak-finding algorithm to split the signal into multiple signals and obtain the signals corresponding to all heartbeat cycles;

4. Extract features from the signal corresponding to each heartbeat cycle:

a) PWA characteristics;

b) Zero-crossing characteristics;

c) Nonlinear dynamic characteristics. Based on the calculation of embedding dimension and delay time, the signal is expanded to a higher order, such as [y(t), y(t-τ), y(t-2τ), …, y(t-2nτ)]. The calculation of embedding dimension includes but is not limited to Takens Embedding Theorem. Effective delay time selection includes but is not limited to mutual information and autocorrelation. Quantifying the nonlinear characteristics of the system brings about effective correlation with the dynamic changes of blood pressure. For example, the Lyapunov index characterizes the average exponential divergence rate of adjacent trajectories in phase space and identifies the characteristics of the chaotic motion of the system. The correlation dimension describes the measure of the spatial dimension occupied by random points. Under different blood pressure values or hemodynamic states, the nonlinear dynamic characteristics of the system change accordingly.

Figure 8 illustrates an example application of nonlinear dynamic features provided by an embodiment of the present application. As shown in Figure 8, two sets of PPG sensor signals are shown, along with an example of an attractor reconstruction with an embedding dimension of 2. The attractor reconstruction demonstrates the periodicity of the PPG signals collected over multiple cycles, enabling calculations for stratified blood pressure estimation.

5. Combine the extracted features and take the average of the features between multiple heartbeat cycles in one minute to obtain the features of this minute.

6. Take the blood pressure layer labels corresponding to the signal and train a blood pressure estimator, such as the extreme gradient boosting algorithm XGBoost.

7. Then you can apply this classifier and perform the above filtering and extraction operations on any signal data of the user to obtain the corresponding features. Input them into the classifier to obtain the blood pressure stratification results.

Example 4: Blood pressure stratification based on multiple PPG signal features. The steps are as follows:

1. Collect the PPG signal of an individual;

2. Filter the PPG signal to obtain a filtered signal;

3. Use a peak-finding algorithm to segment the signal into multiple heartbeat cycles and filter out the bands where peaks and troughs cannot be correctly identified.

4. Extract the following features for each heartbeat cycle:

a) PWA characteristics;

b) Zero-crossing features. Typical PWA features are extracted based on the waveform morphology of the PPG waveform and its derivatives, which are susceptible to noise, especially from multi-order derivative waveforms. Therefore, extracting these PWA features from these waveforms is particularly difficult. Furthermore, for individuals with poor vascular elasticity, the lack of significant peak fluctuations in multi-order derivative waveforms can also result in the loss of corresponding PWA feature extraction. The method proposed in the patent can reliably extract features even when peaks are missing from the waveform, improving both feature accuracy and usability.

c) Nonlinear dynamic characteristics; among them, it is difficult to efficiently extract periodic information from one-dimensional PPG signals, while this feature can characterize the periodic correlation of the signal in detail.

d) Neural network self-encoding features. Among them, the design cost of manual features in feature engineering is high and limited to known prior knowledge, while data-driven neural network-based features can automatically learn and extract features from data.

5. Take the median of each dimension feature for each signal segment to obtain a set of features;

6. Feed the features into the classification model and calculate the blood pressure stratification results.

Example 5: Blood pressure stratification based on multiple PPG signals of different wavelengths. The steps are as follows:

1. Collect multiple PPG signals of different wavelengths, such as a 100Hz signal of 1 minute per sensor, with a length of 6000. These multiple PPG signals can be collected by multiple sensors, each emitting different light sources with different wavelengths. Therefore, the signals from these sensors have differences in blood vessel depth. Since it is a photoplethysmogram of multiple wavelength signals, it is denoted as MW_PPG.

2. Filter the signal to obtain a filtered signal;

3. Use the peak-finding algorithm to split the signal into multiple signals and obtain the signals corresponding to all heartbeat cycles;

4. For each heartbeat cycle, arterial blood pulsation is extracted using a multi-wavelength, multi-layer light-skin interaction model derived from the modified Beer-Lambert law and a quasi-analytical self-calibration algorithm. PWA features are then extracted from the arterial pulsation waveform. Assuming a simultaneous acquisition system for PPG signals generated by blue (B), green (G), yellow (Y), and infrared (IR) light, the peaks of the four wavelength PPG signals within the beat cycle are extracted. The time difference between the peaks of each pair of PPG signals at different wavelengths is then calculated, defined as the PTT_MW. Six PTT_MWs are extracted from the time difference between various combinations of two-wavelength PPG pairs: IR-Y PTT_MW, IR-G PTT_MW, IR-BPTT_MW, Y-G PTT_MW, Y-B PTT_MW, and G-B PTT_MW. Correlations between the PTT_MWs of each pair of wavelengths and the actual blood pressure values are analyzed to determine the optimal PTT_MW wavelength combination and obtain the PTT_MW features.

5. Take the average of the features between multiple heartbeat cycles in one minute to obtain the features of this minute.

6. Take the blood pressure layer labels corresponding to the signal and train a blood pressure estimator, such as the extreme gradient boosting algorithm XGBoost.

7. Then you can apply this classifier and perform the above filtering and extraction operations on any signal data of the user to obtain the corresponding features. Input them into the classifier to obtain the blood pressure stratification results.

Example 6: Blood pressure stratification based on multiple PPG signal features of multiple PPG signals at multiple wavelengths. The steps are as follows:

1. Collect multi-dimensional PPG signals from an individual;

2. Filter the PPG signal to obtain a filtered signal;

3. Use a peak-finding algorithm to segment the signal into multiple heartbeat cycles and filter out the bands where peaks and troughs cannot be correctly identified.

4. Extract the following features for each heartbeat cycle:

a) PWA characteristics;

b) Zero-crossing features. Typical PWA features are extracted based on the waveform morphology of the PPG waveform and its derivatives, which are susceptible to noise, especially from multi-order derivative waveforms. Therefore, extracting these PWA features from them is particularly difficult. Furthermore, for people with poor vascular elasticity, the lack of significant peak fluctuations in multi-order derivative waveforms can also lead to the loss of corresponding PWA feature extraction. The method proposed in the patent can reliably extract features even when peaks are missing from the waveform, improving both feature accuracy and usability.

c) Nonlinear dynamic characteristics, where it is difficult to efficiently extract periodic information from one-dimensional PPG signals, while this feature can characterize the periodic correlation of the signal in detail;

d) the time difference feature PTT_MW between multiple PPG signals, where the PPG signal measured by a single wavelength light source has mixed pulsation components of different blood vessel types, which may lead to inaccurate physiological measurement results;

e) Neural network self-encoding features. In feature engineering, the design cost of manual features is high and limited to known prior knowledge, while data-driven neural network-based features can automatically learn and extract features from data;

5. Take the median of each dimension feature for each signal segment to obtain a set of features;

6. Feed the features into the classification model and calculate the blood pressure stratification results.

Example 7: Blood pressure stratification based on the neural network autoencoder features and PWA features of the PPG signal. The steps are as follows:

1. Collect single/multiple PPG signals. For example, multiple PPG signals can be collected by multiple sensors, where each sensor collects a 100 Hz signal for 1 minute with a length of 6000. Multiple sensors emit different light sources with different wavelengths, so the signals of these sensors have differences in blood vessel depth.

2. Filter and denoise the PPG signal, using a peak-finding algorithm to find the starting point of each heartbeat cycle. If there are 10 consecutive heartbeat cycles, extract a 10-second sample from the start of the first heartbeat cycle of the 10-heartbeat cycle. If there are no 10 consecutive heartbeat cycles or the extraction fails, delete the sample, resulting in a series of clean 10-second signal data of equal length.

3. Build an autoencoder, consisting of an encoder and a decoder. This can be any network capable of processing time series data. For example, build an autoencoder based on a convolutional neural network: the encoder consists of three convolutional layers. The first layer has an input tensor of length 1000, 1 channel, and 15 convolution kernels. The second and third layers have convolution kernels of 3. Between each layer, there is a max pooling layer and an activation layer (ReLU) to compress information. After the signal passes through the encoder, the latent variable is obtained. The decoder consists of three deconvolution layers, with an upsampling layer and an activation layer in between to map the latent variable back to the original space. Use reconstruction loss and the adaptive moment estimation Adam optimizer to optimize the model parameters. This is a basic example; different variations can be obtained using various deep learning techniques.

4. Use the PPG signals of a portion of individuals to train and verify the model, and converge to obtain a trained autoencoder based on a convolutional neural network.

5. For the remaining individual signals, the encoder is used to extract the corresponding latent variable. Each value of this latent variable is a deep learning feature. These signals are also extracted using the PWA feature extraction method to obtain PWA features. The two are combined to form a comprehensive feature. The blood pressure stratification labels of these signals are used to train a blood pressure estimator, such as the extreme gradient boosting algorithm XGBoost.

6. This classifier can then be applied to any PPG signal data of the user, performing the above filtering and extraction operations to obtain deep learning features and PWA features, which are then input into the classifier to obtain blood pressure stratification results.

Example 8: Blood pressure stratification based on the discriminant neural network features and PWA features of the PPG signal. The steps are as follows:

1. Collect PPG signals (can also be IMU/ECG, etc.), such as a 1-minute 100Hz signal with a length of 6000;

2. Filter and denoise the signal, using a peak-finding algorithm to find the starting point of each heartbeat cycle. If there are 10 consecutive heartbeat cycles, extract a 10-second sample from the start of the first heartbeat cycle of the 10-heartbeat cycle. If there are no 10 consecutive heartbeat cycles or the extraction fails, delete the sample, resulting in a series of clean 10-second signal data of equal length.

3. Build a discriminative neural network. This neural network can be ResNet, InceptionTime, Transformer, or any other network that can be used for time series data modeling, or any of their variants. Taking ResNet as an example, the input is set to a single channel with a length of 1000. The kernel size of the first convolutional layer is 15, with 16 channels. Subsequent kernel sizes are all 3. Four subsequent stages follow, each with four basic blocks. Each basic block contains two convolutional layers, an activation layer, and a MaxPooling layer, which includes a cross-layer connection. The number of channels doubles with each basic block. After the last basic block, an average pooling layer is used to pool the length of each channel of the tensor to 1. The tensor is then mapped to the predicted result through a two-layer fully connected layer and an input gate Sigmoid layer, outputting the probability of positive results.

4. Use the signals of some individuals to train and verify the discriminant neural network, and converge to obtain a trained discriminant neural network.

5. At the same time, the tensor before the fully connected layer of the discriminant neural network can also be extracted as deep learning features, combined with manual features such as PWA as sample features, and then a blood pressure estimator can be trained, such as the extreme gradient boosting algorithm XGBoost.

6. This classifier can then be applied to any PPG signal data of the user, performing the above filtering and extraction operations to obtain deep learning features and PWA features, which are then input into the classifier to obtain blood pressure stratification results.

It can be understood that the steps in the above examples can be swapped in order if there is no conflict.

The technical solution provided in the embodiments of the present application can extract various types of blood pressure-related features based on the PPG signals extracted from a single sensor/multiple sensors, and use them individually or organically integrate them to significantly improve the accuracy of blood pressure estimation.

It is understandable that the above examples are for illustration only, and the various features extracted from the PPG signal can be applied individually or in combination according to actual needs. The accuracy of the blood pressure estimation results may vary depending on the application method.

In some embodiments, obtaining target features based on at least one of a photoplethysmography (PPG) signal, an inertial measurement unit (IMU) signal, and an electrocardiogram (ECG) signal may include: obtaining target features based on the IMU signal, wherein the target features include at least one of a PPG signal feature, an IMU signal feature, and an ECG signal feature.

Among them, the IMU signal characteristics include at least one of the following: a second heart rate characteristic, a relative stroke volume characteristic, a heart rhythm signal BCG high-frequency characteristic, an IMU multi-axis characteristic and a second neural network characteristic.

Exemplarily, the second heart rate feature can be extracted by segmenting the waveform of the BCG signal extracted from the IMU signal and matching the positions of the corresponding time domain feature points in each segment. The relative stroke volume feature can be extracted based on the amplitude of the BCG signal extracted from the IMU signal when the object under measurement is in a stable measurement state. The BCG high-frequency feature can be extracted based on the waveform morphology characteristics of the BCG signal extracted from the IMU signal. The IMU multi-axis feature can be extracted based on the acceleration and attitude angle data in three directions. The second neural network feature can be obtained by extracting features from the IMU signal based on a preset neural network.

It can be understood that when obtaining target features based on IMU signals, the target features may include at least one of a second heart rate feature, a relative stroke volume feature, a heart rhythm signal BCG high-frequency feature, an IMU multi-axis feature, and a second neural network feature.

In some embodiments, the blood pressure estimation method also includes at least one of the following steps: segmenting the waveform of the BCG signal extracted from the IMU signal, and matching the positions of the corresponding time domain feature points in each segment to extract the second heart rate feature; when the object being measured is in a stable measurement state, extracting the relative stroke volume feature based on the amplitude of the BCG signal extracted from the IMU signal; extracting the BCG high-frequency feature based on the waveform morphology characteristics of the BCG signal extracted from the IMU signal; extracting the IMU multi-axis feature based on the acceleration and attitude angle data in three directions; and extracting the second neural network feature from the IMU signal based on a preset neural network.

It is understandable that the above steps may be interrelated or independent, and the embodiments of the present application do not limit this.

It should be noted that based on the collected IMU signals, BCG-related information and features can be extracted, including but not limited to instantaneous heart rate (i.e., second heart rate feature), relative stroke volume, BCG-based high-frequency features, machine learning-based signal compression and significant feature recognition, and IMU multi-axis features.

In some embodiments, the blood pressure estimation method includes segmenting the waveform of the BCG signal extracted from the IMU signal, and matching the positions of corresponding time domain feature points in each segment to extract the second heart rate feature.

FIG9 is a flow chart of a method for extracting a second heart rate feature according to an embodiment of the present application. As shown in FIG9 , extracting the second heart rate feature based on the BGC signal extracted from the IMU signal may include:

Step 601: Calculate a frequency spectrum according to a target gyroscope signal, wherein the target gyroscope signal is obtained by filtering the collected gyroscope signal.

Step 602: Determine the second heart rate feature according to the peak frequency of the frequency spectrum.

It should be noted that BCG-related information and features, including but not limited to instantaneous heart rate, are extracted based on IMU signals. In one example, instantaneous heart rate is obtained by segmenting the IMU signal waveform and matching the corresponding time domain feature points in each segment. The time interval between adjacent J peaks of the BCG signal is the heart rate cycle. In another example, the gyroscope signal collected by the IMU sensor is denoised, the signal envelope is extracted, and filtering is performed. The spectrum of the filtered signal is calculated. The peak frequency of the spectrum is the estimated heart rate frequency, and this frequency multiplied by 60 is the heart rate.

In some embodiments, the blood pressure estimation method includes extracting the relative stroke volume feature based on the amplitude of the BCG signal extracted from the IMU signal when the subject is in a stable measurement state.

FIG10 is a flow chart of a method for obtaining relative stroke volume according to an embodiment of the present application. As shown in FIG10 , obtaining relative stroke volume characteristics from BCG signals extracted from IMU signals may include:

Step 701: Filtering the BCG signal extracted from the IMU signal to obtain a filtered BCG signal;

Step 702: Segment the filtered BCG signal and extract peak amplitude information to obtain the relative stroke volume.

It's important to note that when the subject is relatively still and the measurement state is stable (e.g., measurement angle, location, etc.), the BCG signal amplitude extracted from the IMU signal effectively represents relative changes in stroke volume. In sleep scenarios, when the subject wears the watch while sleeping, information from the accelerometer and angular velocity meter can be used to quantify the user's orientation in real time, and relative stroke volume is calculated based on the multi-axis amplitudes. When orientation information changes, breakpoints are set as new relative stroke volume intervals, and segments with similar orientation information are considered for fusion. Incorporating relative stroke volume information as blood pressure feature calibration information can reliably compensate for changes in blood flow caused by changes in body position. For example, in traditional pressurized blood pressure measurement methods, the user must maintain the pressurized upper arm at the same level as the heart. However, PPG-based solutions do not have measurement constraints. Incorporating IMU orientation information and relative stroke volume information can effectively compensate for measurement discrepancies.

In some embodiments, the blood pressure estimation method includes extracting the BCG high-frequency features based on the waveform morphology characteristics of the BCG signal extracted from the IMU signal.

FIG11 is a flow chart of a method for obtaining BCG high-frequency features according to an embodiment of the present application. As shown in FIG11 , obtaining BCG high-frequency features based on BCG signals extracted from IMU signals may include:

Step 801: Extract the amplitude, time interval, area, slope and energy characteristics of the peaks and troughs of each heartbeat cycle of the BCG signal based on the maximum peak of each heartbeat cycle in the BCG signal extracted from the IMU signal, the trough and peak closest to the maximum peak, and the trough and peak closest to the maximum peak.

Step 802: Determine the BCG high-frequency features according to the amplitude, time interval, area, slope and energy characteristics of the peaks and troughs of each heartbeat cycle of the BCG signal.

It should be noted that the BCG-based high-frequency features are based on the BCG waveform morphology. The J-peak is determined by identifying the maximum peak in each BCG cycle signal. The closest trough and peak before the J-peak are defined as I and H, respectively. The closest trough and peak after the J-peak are defined as K and L, respectively. The BCG high-frequency feature set is derived by extracting the amplitude, time interval, area, slope, and energy characteristics of the peaks and troughs and combining these features.

Figure 12 is a schematic diagram of a BCG signal waveform provided by an embodiment of the present application. As shown in Figure 12, the maximum peak value of the BCG signal in the positive y-axis direction is the J peak. Before the J peak, the closest trough and peak are defined as the I peak and H peak, respectively. After the J peak, the closest trough and peak are defined as the K peak and L peak, respectively.

In some embodiments, the blood pressure estimation method includes extracting the IMU multi-axis features based on acceleration and attitude angle data in three directions.

Exemplarily, obtaining IMU multi-axis features based on IMU signals may include: obtaining speed change information and posture change information in the IMU signal; and determining at least one of the speed change information and the posture change information as the IMU multi-axis features.

It's important to note that analysis of the signal correlation between multiple IMU axes reveals that the raw IMU data exhibits linear acceleration, and this linear acceleration exists in all three coordinate axes. Velocity can be calculated by integrating the linear acceleration along the three defined coordinate axes, but in practice, the IMU exhibits significant noise, making this integral approach impractical. Therefore, the absolute value of the linear acceleration for each set of data can be calculated and entered into a matrix. The absolute value calculation is as follows: |a| = 1/3√(a_x^2 + a_y^2 + a_z^2), where ax represents the acceleration along the x-axis, ay represents the acceleration along the y-axis, and az represents the acceleration along the z-axis. To some extent, the absolute value of the linear acceleration |a| can represent the magnitude of the velocity change.

Raw IMU data also includes Euler angles and quaternions. These represent the IMU's pose. If each quaternion represents a pose in space, the arc cosine of the dot product of two quaternions represents the angle between the two quaternion poses. The angle between the quaternion of each data set and the previous data set is calculated and then populated into a matrix. The time between each data set is approximately one second, which is equivalent to calculating the IMU position in one second. The angle change in pose is calculated as follows: Δangle = arccos(q^pre · q^now) × 180/π, where Δangle represents the angle difference, q^pre represents the angle between the previous data set, and q^now represents the angle between the current data set. Motion information such as velocity change and pose angle change is used as IMU multi-axis features. Features extracted from other sensors, such as PPG and ECG, are integrated into the blood pressure prediction model to provide quantifiable calibration information for the user's blood pressure prediction model. This establishes a correlation between blood pressure and motion, improves the accuracy of cuffless blood pressure measurement, and verifies the feasibility of calibrating continuous cuffless blood pressure measurement based on motion information.

In some embodiments, the blood pressure estimation method includes extracting the second neural network feature from the IMU signal based on a preset neural network.

It should be noted that the IMU signal can be feature extracted through a preset neural network.

Among them, the second neural network feature may include at least one of a neural network autoencoder feature and a discriminant neural network feature, the neural network autoencoder feature is extracted based on an autoencoder of a convolutional neural network, and the discriminant neural network feature is extracted based on a discriminant neural network model.

The manner in which the second neural network features are extracted from the IMU signal based on a preset neural network is similar to the aforementioned manner in which the features of the PPG signal are extracted based on a preset neural network. Reference may be made to the description of the manner in which the features of the PPG signal are extracted based on a preset neural network in the aforementioned embodiment, and no further details will be given here.

It can be understood that the above embodiments may be interrelated or exist independently.

The above content describes the feature extraction methods based on different aspects of IMU signals, including: extracting the second heart rate feature, the relative stroke volume feature, the BCG high-frequency feature of the heart rhythm signal, the IMU multi-axis feature and the second neural network feature. The combination of features from different aspects can be used to model the direction and measurement of blood pressure changes.

In some embodiments, performing blood pressure estimation based on the target feature to obtain the blood pressure estimation result may include: inputting the IMU signal feature into a preset model, such as a trained blood pressure estimator, to obtain the blood pressure estimation result.

It can be understood that the above embodiments may be interrelated or exist independently.

For example, the trained blood pressure estimator is obtained by merging the collected IMU signal features to obtain a merged feature, then obtaining the mean feature corresponding to the merged feature of a preset time length in the merged feature, and training the initial blood pressure estimator through the mean feature.

The following describes an exemplary application of some embodiments of the present application in a practical application scenario.

Example 9: Blood pressure stratification based on BCG high-frequency features of IMU signals. The steps are as follows:

1. Collect IMU signals, such as a 1-minute 100Hz signal with a length of 6000;

2. Filter the signal to obtain a filtered signal;

3. Use the peak-finding algorithm to split the signal into multiple signals and obtain the signals of all heartbeats;

4. Feature extraction for the signal corresponding to each heartbeat cycle: Based on the BCG waveform morphology, the J peak is determined by identifying the maximum peak in each BCG heartbeat signal. The trough and peak closest to the J peak are defined as I and H, respectively. The trough and peak closest to the J peak are defined as K and L, respectively. The amplitude, time interval, area, slope, and energy characteristics of the peaks and troughs are extracted, and these features are combined to calculate the high-frequency features of the BCG.

5. Combine the above features as the corresponding features of each sample;

6. Take the blood pressure stratification labels corresponding to the signal and train a blood pressure classifier, such as the extreme gradient boosting algorithm XGBoost.

7. Then you can apply this classifier and perform the above filtering and extraction operations on any signal data of the user to obtain the corresponding features. Input them into the classifier to obtain the blood pressure stratification results.

Example 10: Blood pressure stratification based on the discriminant neural network features of the IMU signal and the high-frequency features of the BCG signal. The steps are as follows:

1. Collect IMU signals, such as a 1-minute 100Hz signal with a length of 6000;

2. Filter and denoise the signal, using a peak-finding algorithm to find the starting point of each heartbeat cycle. If there are 10 consecutive heartbeat cycles, extract a 10-second sample from the start of the first heartbeat cycle of the 10-heartbeat cycle. If there are no 10 consecutive heartbeat cycles or the extraction fails, delete the sample, resulting in a series of clean 10-second signal data of equal length.

3. Build a discriminative neural network. This neural network can be ResNet, InceptionTime, Transformer, or any other network that can be used for time series data modeling, or any of their variants. Taking ResNet as an example, the input is set to a single channel with a length of 1000. The kernel size of the first convolutional layer is 15, with 16 channels. Subsequent kernel sizes are all 3. Four subsequent stages follow, each with four basic blocks. Each basic block contains two convolutional layers, an activation layer, and a MaxPooling layer, which includes a cross-layer connection. The number of channels doubles with each basic block. After the last basic block, an average pooling layer is used to pool the length of each channel of the tensor to 1. The tensor is then mapped to the predicted result through a two-layer fully connected layer and an input gate Sigmoid layer, outputting the probability of positive results.

4. Use the signals of some individuals to train and verify the discriminant neural network, and converge to obtain a trained discriminant neural network.

5. At the same time, the tensor before the fully connected layer of the discriminant neural network can also be extracted as a deep learning feature, combined with the high-frequency features of the BCG signal as sample features, and then a blood pressure estimator can be trained, such as the extreme gradient boosting algorithm XGBoost.

6. This classifier can then be applied to any IMU signal data of the user, performing the above filtering and extraction operations to obtain deep learning features and BCG high-frequency features, which can be input into the classifier to obtain blood pressure stratification results.

It can be understood that the steps in the above examples can be swapped in order if there is no conflict.

The technical solution provided in the embodiment of the present application can extract various types of features related to blood pressure based on the extracted IMU signal, which can be used alone or organically integrated to significantly improve the accuracy of blood pressure estimation.

It is understandable that the above examples are for illustration only. The various features extracted from the IMU signal can be applied individually or in combination according to actual needs. The accuracy of the blood pressure estimation results may vary depending on the application method.

In some embodiments, obtaining target features based on at least one of a photoplethysmography (PPG) signal, an inertial measurement unit (IMU) signal, and an electrocardiogram (ECG) signal may include: obtaining target features based on the ECG signal, wherein the target features include at least one of a PPG signal feature, an IMU signal feature, and an ECG signal feature.

The ECG signal feature may include a third neural network feature.

Exemplarily, the third neural network feature may be obtained by extracting features from the ECG signal based on a preset neural network.

In some embodiments, the blood pressure estimation method further includes extracting the third neural network feature from the ECG signal based on a preset neural network.

It should be noted that the IMU signal can be feature extracted through a preset neural network.

Among them, the third neural network feature may include at least one of a neural network autoencoder feature and a discriminant neural network feature, the neural network autoencoder feature is extracted based on an autoencoder of a convolutional neural network, and the discriminant neural network feature is extracted based on a discriminant neural network model.

The manner in which the third neural network features are extracted from the ECG signal based on the preset neural network is similar to the aforementioned manner in which the features of the PPG signal are extracted based on the preset neural network. Reference may be made to the description of the manner in which the features of the PPG signal are extracted based on the preset neural network in the aforementioned embodiment, and no further details will be given here.

It can be understood that the above embodiments may be interrelated or exist independently.

The above content describes the feature extraction method based on ECG signals. The combination of features from different aspects can be used to model the direction and measurement of blood pressure changes.

In some embodiments, performing blood pressure estimation based on the target feature to obtain the blood pressure estimation result may include: inputting the ECU signal feature into a preset model, such as a trained blood pressure estimator, to obtain the blood pressure estimation result.

It can be understood that the above embodiments may be interrelated or exist independently.

For example, the trained blood pressure estimator is obtained by merging the collected ECG signal features to obtain a merged feature, then obtaining the mean feature corresponding to the merged feature of a preset time length in the merged feature, and training the initial blood pressure estimator through the mean feature.

The following describes an exemplary application of some embodiments of the present application in a practical application scenario.

Example 11: Blood pressure stratification based on the discriminant neural network features of ECG signals. The steps are as follows:

1. Collect ECG signals, such as a 1-minute 100Hz signal with a length of 6000;

2. Filter and denoise the signal, and use a peak-finding algorithm to find the starting point of each heartbeat cycle.

3. Build a discriminative neural network. This neural network can be ResNet, InceptionTime, Transformer, or any other network that can be used for time series data modeling, or any of their variants. Taking ResNet as an example, the input is set to a single channel with a length of 1000. The kernel size of the first convolutional layer is 15, with 16 channels. Subsequent kernel sizes are all 3. Four subsequent stages follow, each with four basic blocks. Each basic block contains two convolutional layers, an activation layer, and a MaxPooling layer, which includes a cross-layer connection. The number of channels doubles with each basic block. After the last basic block, an average pooling layer is used to pool the length of each channel of the tensor to 1. The tensor is then mapped to the predicted result through a two-layer fully connected layer and an input gate Sigmoid layer, outputting the probability of positive results.

4. Use the signals of some individuals to train and verify the discriminant neural network, and converge to obtain a trained discriminant neural network.

5. At the same time, the tensor before the fully connected layer of the discriminant neural network can be extracted as a deep learning feature, and the deep learning feature can be used as a sample feature to train a blood pressure estimator, such as the extreme gradient boosting algorithm XGBoost.

6. Then you can apply this classifier and perform the above filtering and extraction operations on any ECG signal data of the user to obtain deep learning features. Input them into the classifier to obtain blood pressure stratification results.

Example 12: Blood pressure stratification based on the feature set of ECG signals. The steps are as follows:

1. Collect ECG signals, such as a 1-minute 100Hz signal with a length of 6000;

2. Filter and denoise the signal to obtain a filtered signal;

3. Use the peak-finding algorithm to find the starting point of each heartbeat cycle.

4. Feature extraction: Based on the waveform characteristics of the ECG, identify the key waveforms such as the P wave, QRS complex, T wave of each ECG heartbeat and the corresponding starting points, ending points, and extreme points of the waveform. Use different key points to construct a series of feature sets based on angles such as area, amplitude, time interval, slope, and energy.

5. Combine the above features as the corresponding features for each sample.

6. Take the blood pressure stratification labels corresponding to the signal and train a blood pressure classifier, such as the extreme gradient boosting algorithm XGBoost.

7. Then you can apply this classifier and perform the above filtering and extraction operations on any signal data of the user to obtain the corresponding features. Input them into the classifier to obtain the blood pressure stratification results.

It can be understood that the steps in the above examples can be swapped in order if there is no conflict.

The technical solution provided in the embodiment of the present application can extract various types of features related to blood pressure based on the extracted ECG signal, which can be used alone or organically integrated to significantly improve the accuracy of blood pressure estimation.

It is understandable that the above examples are for illustration only, and the various features extracted from the ECG signal can be applied individually or in combination according to actual needs. The accuracy of the blood pressure estimation results may vary depending on the application method.

In some embodiments, the PTT feature can also be obtained using a PPG signal and an IMU signal. Specifically, extracting the PTT feature based on the signal peak of the PPG signal can include determining the pulse transit time (PTT) feature based on the time difference between the signal peak position of the PPG signal and the J peak or K peak of the BCG signal extracted from the IMU signal, where the J peak is the maximum peak of the BCG signal in a first direction, and the K peak is the second-largest peak of the BCG signal in a second direction, the first and second directions being opposite.

It should be noted that the correlation analysis between multiple sensors, such as the signal delay between PPG and IMU, can be used to derive PAT information. For example, the time difference between the position of the PPG feature point and the position of the BCG feature point (such as the J peak or K peak) can be recorded as the PTT parameter. Among them, PTT is an active or background measurement parameter. In the background measurement scenario, an example is the sleep scenario: during sleep, continuous recording of PTT parameter information based on PPG and IMU sensors, combined with other background characteristic parameters, can more reliably quantify blood pressure trends.

In some embodiments, obtaining target features based on at least one of a photoplethysmography (PPG) signal, an inertial measurement unit (IMU) signal, and an electrocardiogram (ECG) signal may include: obtaining target features based on the PPG signal and the IMU signal, wherein the target features include at least one of a PPG signal feature, an IMU signal feature, and an ECG signal feature.

The target feature may include a pulse transit time (PTT) feature.

The following describes an exemplary application of some embodiments of the present application in a practical application scenario.

Example 13: Blood pressure stratification based on the characteristics of PPG and IMU signals. The steps are as follows:

1. Collect PPG and IMU signals for 1 minute;

2. Filter the signal to obtain a filtered signal;

3. Use the peak-finding algorithm to split the signal into multiple signals and obtain the signal of all heartbeat cycles;

4. Extract the following features:

a) Extract PWA features from the PPG signal of each heart cycle;

b) Real-time heart rate: After denoising the gyro signal collected by the IMU sensor, extract the signal envelope, filter it, and calculate the spectrum of the filtered signal. The peak frequency of the spectrum is the estimated heart rate frequency, and this frequency is multiplied by 60 to obtain the heart rate.

c) Relative Stroke Volume: The BCG signal amplitude extracted by the IMU effectively represents relative changes in stroke volume. The BCG signal amplitude and stroke volume are correlated. Filtering the collected BCG signal improves the signal-to-noise ratio (SNR) and facilitates subsequent feature point extraction. The BCG signal is segmented, and the positions of the H, I, J, and K peaks are determined based on the local maximum of the signal and the relative positions of each peak. The peak amplitudes are then extracted to represent relative changes in stroke volume. In sleep scenarios, when the subject wears the watch while sleeping, the user's orientation is quantified in real time based on information from the accelerometer and angular velocity meter. When the orientation changes, a breakpoint is set in the collected data to represent a new relative stroke volume interval. Segments with similar orientation information are fused, and the relative stroke volume for different orientations is calculated based on the multi-axis amplitudes. Incorporating relative stroke volume information as blood pressure feature calibration information reliably compensates for changes in blood flow caused by changes in body position. For example, in traditional pressurized blood pressure measurement methods, users need to keep the pressurized upper arm and the heart at the same level; the PPG-based technical solution does not yet have measurement constraints. Adding IMU orientation information and relative stroke volume information can effectively compensate for measurement differences.

d) Based on BCG high-frequency features: Based on the BCG waveform morphology, the J-peak is determined by identifying the maximum peak in each BCG cycle signal. The closest trough and peak before the J-peak are defined as I and H, respectively. The closest trough and peak after the J-peak are defined as K and L, respectively. The amplitude, time interval, area, slope, and energy characteristics of the peaks and troughs of the BCG cycle signal are extracted and combined to form the BCG high-frequency feature set.

e) IMU multi-axis characteristics: The original IMU data has linear acceleration, and the linear acceleration exists in the three directions of the coordinate axis. The velocity can be solved by integrating the linear acceleration in the three directions of the defined coordinate axis, but in actual work, the noise of the IMU is large, so the integral solution is not feasible. Therefore, we calculate the absolute value of the linear acceleration of each set of data and fill it into the matrix. The absolute value calculation process is as follows: |a| = 1/3√(a_x^2 + a_y^2 + a_z^2), where ax represents the acceleration on the x-axis, ay represents the acceleration on the y-axis, and az represents the acceleration on the z-axis. To a certain extent, the absolute value of the linear acceleration |a| can represent the magnitude of the velocity change.

At the same time, the raw IMU data also includes Euler angles and quaternions. Euler angles and quaternions represent the posture of the IMU. If each quaternion represents a posture in space, the arc cosine of the dot product of two quaternions represents the angle between the two quaternion postures. We calculate the angle between the quaternion of each set of data and the previous set of data, and then fill it into the matrix. The time between each set of data is about 1 second, which is equivalent to calculating the IMU position in one second. The calculation process of the angle change value of the posture is as follows: Δangle = arccos(q^pre·q^now)×180/π, where Δangle represents the angle difference, q^pre represents the angle of the previous set of data, and q^now represents the angle of the current set of data. Motion information such as speed change and posture angle change is used as the multi-axis feature of the IMU.

a) Optional: Neural network autoencoder features or discriminant neural network features can be added.

5. Combine the above features as the corresponding features of each sample;

6. Take the blood pressure layer labels corresponding to the signal and train a blood pressure estimator, such as the extreme gradient boosting algorithm XGBoost.

7. Then you can apply this classifier and perform the above filtering and extraction operations on any signal data of the user to obtain the corresponding features. Input them into the classifier to obtain the blood pressure stratification results.

It can be understood that the steps in the above examples can be swapped in order if there is no conflict.

The technical solution provided in the embodiment of the present application can extract various types of features related to blood pressure based on the extracted PPG signals and IMU signals, which can be used alone or organically integrated to significantly improve the accuracy of blood pressure estimation.

It is understandable that the above examples are for illustrative purposes only. The various features extracted from the PPG signal and the IMU signal can be applied individually or in combination according to actual needs. The accuracy of the blood pressure estimation results may vary depending on the application method.

In some embodiments, the PAT feature can also be obtained from a PPG signal and an ECG signal. Specifically, extracting the PAT feature based on the signal peak of the PPG signal may include determining the PAT feature based on the time difference between the signal peak position of the PPG signal and the R-peak of the ECG signal, where the R-peak is the highest peak point of the ECG signal.

It should be noted that PTT information can be derived from correlation analysis between multiple sensors, such as the signal delay between PPG and ECG. In one example, a user performs an active point-to-point measurement: the system collects PPG and ECG signals in the point-to-point state, and records the time difference between the characteristic point position of the PPG (such as the position of maximum acceleration on the rising edge) and the R peak of the ECG as a PAT parameter.

In some embodiments, obtaining target features based on at least one of a photoplethysmography (PPG) signal, an inertial measurement unit (IMU) signal, and an electrocardiogram (ECG) signal may include: obtaining target features based on the PPG signal and the ECG signal, wherein the target features include at least one of a PPG signal feature, an IMU signal feature, and an ECG signal feature.

The target feature may include a pulse transit time (PAT) feature.

The following describes an exemplary application of the embodiments of the present application in a practical application scenario.

Example 14: Blood pressure stratification based on the characteristics of PPG and ECG signals. The steps are as follows:

1. Collect PPG and ECG signals for 1 minute;

2. Filter the signal to obtain a filtered signal;

3. Use the peak-finding algorithm to split the signal into multiple signals and obtain the signal of all heartbeat cycles;

4. Extract the following features:

a) Extract PWA features from the PPG signal of each heart cycle;

b) PAT/PTT features: The time difference between the PPG feature point position (e.g., the position of maximum acceleration on the rising edge) and the ECG R peak is recorded as the PAT parameter. (Similarly, the PTT feature can be obtained using the BCG signal from an IMU sensor. The time difference between the PPG feature point position and the BCG feature point position (e.g., the J peak or K peak) is recorded as the PTT parameter.)

c) Optional, neural network autoencoder features or discriminant neural network features can be added;

d) Optional, can be combined with other IMU features.

5. Combine the above features as the corresponding features of each sample;

6. Take the blood pressure layer labels corresponding to the signal and train a blood pressure estimator, such as the extreme gradient boosting algorithm XGBoost.

7. Then you can apply this classifier and perform the above filtering and extraction operations on any signal data of the user to obtain the corresponding features. Input them into the classifier to obtain the blood pressure stratification results.

It can be understood that the steps in the above examples can be swapped in order if there is no conflict.

The technical solution provided in the embodiment of the present application can extract various types of features related to blood pressure based on the extracted PPG signals and ECG signals, which can be used alone or organically integrated to significantly improve the accuracy of blood pressure estimation.

It is understandable that the above examples are for illustration only. The various features extracted from the PPG signal and ECG signal can be applied individually or in combination according to actual needs. The accuracy of the blood pressure estimation results may vary depending on the application method.

In some embodiments, obtaining target features based on at least one of a photoplethysmography (PPG) signal, an inertial measurement unit (IMU) signal, and an electrocardiogram (ECG) signal may include: obtaining target features based on the IMU signal and the ECG signal, wherein the target features include at least one of a PPG signal feature, an IMU signal feature, and an ECG signal feature.

The following describes an exemplary application of the embodiments of the present application in a practical application scenario.

Example 15: Blood pressure stratification based on the features of IMU and ECG signals. The steps are as follows:

1. Collect IMU and ECG signals for 1 minute;

2. Filter the signal to obtain a filtered signal;

3. Use the peak-finding algorithm to split the signal into multiple signals and obtain the signal of all heartbeat cycles;

4. Extract the following features:

a) Based on the BCG waveform morphology, the J peak is determined by identifying the maximum peak in each BCG heartbeat signal. The nearest trough and peak before the J peak are defined as I and H, respectively. The nearest trough and peak after the J peak are defined as K and L, respectively. The amplitude, time interval, area, slope, and energy features of the peaks and troughs are extracted and combined to obtain the BCG high-frequency feature set.

b) Based on the waveform characteristics of ECG, identify the key waveforms such as the P wave, QRS complex, T wave of each ECG heartbeat and the corresponding starting points, ending points and extreme points of the waveforms, and use different key points to construct a series of feature sets based on angles such as area, amplitude, time interval, slope, and energy.

c) Based on the key points of the BCG heartbeat and the corresponding ECG heartbeat points, the different time intervals between the key points of the two signals are calculated as features.

5. Combine the above features as the corresponding features of each sample;

6. Take the blood pressure layer labels corresponding to the signal and train a blood pressure estimator, such as the extreme gradient boosting algorithm XGBoost.

7. Then you can apply this classifier and perform the above filtering and extraction operations on any signal data of the user to obtain the corresponding features. Input them into the classifier to obtain the blood pressure stratification results.

It can be understood that the steps in the above examples can be swapped in order if there is no conflict.

The technical solution provided in the embodiment of the present application can extract various types of features related to blood pressure based on the extracted IMU signals and ECG signals, and use them individually or organically integrate them to significantly improve the accuracy of blood pressure estimation.

It is understandable that the above examples are for illustrative purposes only. The various features extracted from the IMU signal and ECG signal can be applied individually or in combination according to actual needs. The accuracy of the blood pressure estimation results may vary depending on the application method.

In some embodiments, obtaining target features based on at least one of a photoplethysmography (PPG) signal, an inertial measurement unit (IMU) signal, and an electrocardiogram (ECG) signal may include: obtaining target features based on the PPG signal, the IMU signal, and the ECG signal, wherein the target features include at least one of a PPG signal feature, an IMU signal feature, and an ECG signal feature.

It should be noted that the multi-sensor based machine learning fusion strategy includes mapping the signal data of multiple sensors into the same latent space through machine learning to efficiently fuse the information between different sensors. Multimodal learning methods can be used for further optimization, such as mutual information constraints. In one example, PPG signals, ECG signals, and IMU signals are sent as three inputs to a neural network model. The model has three network branches to compress the three input signals respectively, such as three convolutional neural networks (CNNs) with different parameters or a sequence model Transformer network branch based on an attention mechanism. The three branches are then sent to a backbone network for information fusion to obtain fused features in the same latent space, namely, neural network multimodal features. The target results are then output through a head network. The target results vary depending on the downstream task, such as population classification or blood pressure regression. The model is optimized through the corresponding objective function and constraints on information fusion.

The following describes an exemplary application of the embodiments of the present application in a practical application scenario.

Example 16: Blood pressure stratification based on the features of PPG, IMU, and ECG signals. The steps are as follows:

1. Collect PPG signals, IMU signals, and ECG signals of an individual;

2. Filter the PPG signal, IMU signal, and ECG signal separately;

3. Use the peak-finding algorithm to filter out the bands that cannot be correctly peak-finded. For the remaining bands with good quality, capture 10 seconds of waveform data from the starting point of the most recent cycle.

4. The three waveform data were fed into a neural network to calculate multimodal features. Manual feature engineering is expensive and limited to known prior knowledge, whereas data-driven neural network-based features can automatically learn and extract features from the data. Multimodal neural networks can further integrate information from different sensors.

5. Then use a classifier (such as a multi-layer perceptron) to calculate the blood pressure stratification results.

Example 17: Blood pressure stratification based on the features of PPG, IMU, and ECG signals. The steps are as follows:

1. Collect PPG signals, IMU signals, and ECG signals of an individual;

2. Filter the PPG signal, IMU signal, and ECG signal separately;

3. Use the peak-finding algorithm to filter out the bands that cannot be correctly peak-finded. For the remaining bands with good quality, capture 10 seconds of waveform data from the starting point of the most recent cycle.

4. The three waveform data were fed into a neural network to calculate multimodal features. Manual feature engineering is expensive and limited to known prior knowledge, whereas data-driven neural network-based features can automatically learn and extract features from the data. Multimodal neural networks can further integrate information from different sensors.

5. Extract PWA features from the filtered PPG signal, then merge them with multimodal features and use the classifier to obtain blood pressure stratification results.

Example 18: Blood pressure stratification based on the features of PPG, IMU, and ECG signals. The steps are as follows:

1. Collect PPG signals, IMU signals, and ECG signals of an individual;

2. Filter the PPG signal, IMU signal, and ECG signal separately;

3. Use the peak-finding algorithm to filter out the bands that cannot be correctly peak-finded. For the remaining bands with good quality, capture 10 seconds of waveform data from the starting point of the most recent cycle.

4. Each signal is subjected to a multidimensional expansion with time delays, resulting in a nonlinear dynamical representation called a time-delay embedding. While one-dimensional PPG signals are difficult to efficiently extract periodic information from, this representation can characterize the signal's periodic correlations in detail. Furthermore, the multidimensional embedding provides a high-dimensional representation of the time series signal at a specific delay time.

5. This representation is fed into a neural network, and after the classifier, blood pressure stratification results are obtained. Compared to the one-dimensional representation of the original time series signal, the high-dimensional representation of time delay embedding can more significantly characterize the cyclical changes and trends of the signal, thereby improving the performance of subsequent classification and regression models.

The technical solution provided in the embodiments of the present application can extract various types of blood pressure-related features from signals collected by multiple sensors and organically integrate them, thereby significantly improving the accuracy of blood pressure estimation.

It can be understood that various target features in the examples of the present application can be used for blood pressure estimation, that is, in addition to being used for blood pressure stratification estimation, they can also be used for other blood pressure estimations, as well as for other combinations.

In some embodiments, the blood pressure estimation based on the target feature to obtain the blood pressure estimation result may include: inputting the target feature into a trained blood pressure estimator to obtain the blood pressure estimation result; wherein, the trained blood pressure estimator merges the target features to obtain a merged feature, and then obtains the mean feature corresponding to the merged feature of a preset time length in the merged feature, and trains the initial blood pressure estimator through the mean feature.

The above describes different aspects of feature extraction, such as those based on PPG signals, IMU signals, ECG signals, and multi-sensor correlations. A single feature or a combination of features can be used to model the direction and magnitude of blood pressure changes.

Based on this technical solution, various types of blood pressure-related features can be extracted from multiple sensors and organically integrated. The multi-sensor information complements each other, which significantly improves the accuracy of blood pressure estimation.

It should be understood that, although the various steps in the above flow chart are shown in sequence as indicated by the arrows, these steps are not necessarily performed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be performed in other orders. Moreover, at least a portion of the steps in the flow chart may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily performed at the same time, but can be performed at different times. The execution order of these sub-steps or stages is not necessarily to be performed in sequence, but can be performed in turn or alternately with other steps or at least a portion of the sub-steps or stages of other steps.

Based on the foregoing embodiments, an embodiment of the present application provides a blood pressure estimation device, and the modules included in the device and the units included in each module can be implemented by a processor; of course, they can also be implemented by a specific logic circuit; in the implementation process, the processor can be a central processing unit (CPU), a microprocessor (MPU), a digital signal processor (DSP) or a field programmable gate array (FPGA), etc.

FIG13 is a schematic diagram of the structure of a blood pressure estimation device provided in an embodiment of the present application. As shown in FIG13 , the device 900 includes a feature acquisition module 901 and a result acquisition module 902, wherein:

Feature acquisition module 901 is configured to acquire target features based on at least one of a photoplethysmography (PPG) signal, an inertial measurement unit (IMU) signal, and an electrocardiogram (ECG) signal, wherein the target features include at least one of a PPG signal feature, an IMU signal feature, and an ECG signal feature; wherein the PPG signal features include at least one of: a pulse waveform analysis (PWA) feature, a zero-crossing feature, a nonlinear dynamics feature, a pulse transit time (PTT) feature, a pulse arrival time (PAT) feature, a first heart rate feature, and a first neural network feature; and/or the IMU signal features include at least one of: a second heart rate feature, a relative stroke volume feature, a heart rhythm signal (BCG) high-frequency feature, an IMU multi-axis feature, and a second neural network feature; and/or the ECG signal features include a third neural network feature;

The result acquisition module 902 is configured to perform blood pressure estimation based on the target feature to obtain a blood pressure estimation result.

In some embodiments, the apparatus further includes a first execution module configured to perform at least one of the following steps: extracting the PWA feature based on waveform feature analysis of the PPG signal;

Extracting the zero-crossing feature according to the derivative signal of the standardized PPG signal;

Extracting the nonlinear dynamic features according to the PPG signals corresponding to a plurality of heartbeat cycles in the PPG signal;

extracting the PTT feature or the PAT feature based on a signal peak of the PPG signal;

Segmenting the waveform features of the PPG signal, and matching the positions of corresponding feature points in each segment to extract the first heart rate feature;

The first neural network feature is extracted from the PPG signal based on a preset neural network.

In some embodiments, the PPG signal includes a plurality of PPG signals of different wavelengths, and the first execution module is specifically configured to: perform a calculation to remove capillary pulsation interference based on at least two PPG signals of different wavelengths among the plurality of PPG signals to obtain a post-capillary pulsation interference removal signal;

extracting arterial blood pulsation from the removed signal to obtain an arterial blood pulsation waveform;

Features are extracted from the arterial blood pulsation waveform to obtain the PWA features.

In some embodiments, the PPG signal includes multiple PPG signals of different wavelengths, and the first execution module is specifically configured to: extract a peak value within a beat cycle of each of the multiple PPG signals, and calculate a time difference between any two of the peak values in the PPG signals to obtain a plurality of time differences of pulse transit times;

The time differences of the multiple pulse transit times are correlated with the actually measured blood pressure values to obtain a multi-wavelength pulse transit time feature, which is used to indicate the optimal wavelength combination corresponding to the multi-wavelength pulse transit time.

In some embodiments, the first execution module is specifically used to determine the pulse transit time PTT feature based on the time difference between the signal peak position of the PPG signal and the J peak or K peak of the BCG signal extracted by the IMU signal, wherein the J peak is the maximum peak of the BCG signal in the first direction, and the K peak is the second largest peak of the BCG signal in the second direction, and the first direction and the second direction are opposite.

In some embodiments, the first execution module is specifically configured to determine a PAT feature based on a time difference between a signal peak position of the PPG signal and an R peak of the ECG signal, where the R peak is the highest peak point of the ECG signal.

In some embodiments, the apparatus further includes a second execution module configured to perform at least one of the following steps: segmenting the waveform of the BCG signal extracted from the IMU signal, and matching the positions of corresponding time domain feature points in each segment to extract the second heart rate feature;

When the measured object is in a stable measurement state, extracting the relative stroke volume feature according to the amplitude of the BCG signal extracted from the IMU signal;

Extracting the BCG high-frequency features based on the waveform morphology characteristics of the BCG signal extracted from the IMU signal;

Extracting the IMU multi-axis features based on the acceleration and attitude angle data in three directions;

The second neural network feature is extracted from the IMU signal based on a preset neural network.

In some embodiments, the device further includes a third execution module, and the second execution module is used to execute feature extraction of the ECG signal based on a preset neural network to extract the third neural network feature.

In some embodiments, the result acquisition module is specifically configured to: input the target feature into a trained blood pressure estimator to obtain the blood pressure estimation result;

Among them, the trained blood pressure estimator is obtained by merging the target features to obtain a merged feature, then obtaining the mean feature corresponding to the merged feature of a preset time length in the merged feature, and training the initial blood pressure estimator through the mean feature.

In the embodiment of the present application, blood pressure estimation can be performed by using target features collected by the cuffless detection sensor to obtain a blood pressure estimation result, thereby improving the accuracy of blood pressure estimation.

The description of the above device embodiment is similar to the description of the above method embodiment and has similar beneficial effects as the method embodiment. For technical details not disclosed in the device embodiment of this application, please refer to the description of the method embodiment of this application for understanding.

It should be noted that the division of modules in the blood pressure estimation device shown in FIG13 in the embodiment of the present application is schematic and is merely a logical functional division. In actual implementation, other division methods may be used. Furthermore, the functional units in the various embodiments of the present application may be integrated into a single processing unit, or may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units. They may also be implemented in the form of a combination of software and hardware.

It should be noted that in the embodiments of the present application, if the above-mentioned method is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiments of the present application, or the part that contributes to the relevant technology, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling an electronic device to execute all or part of the methods described in each embodiment of the present application. The aforementioned storage media include various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROMs), magnetic disks or optical disks. In this way, the embodiments of the present application are not limited to any specific combination of hardware and software.

Figure 14 is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. As shown in Figure 14, the electronic device 100 may include a processor 110, a memory 120, a wireless communication module 130, a sensor module 140, a camera 150, a USB interface 160, a display screen 170, etc.

The processor 110 may include one or more processing units. For example, the processor 110 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present application, such as one or more microprocessors (digital signal processors, DSPs) or one or more field programmable gate arrays (FPGAs). The different processing units may be independent devices or integrated into one or more processors.

The memory 120 can be used to store computer executable program code, which includes instructions. The internal memory 120 may include a program storage area and a data storage area. The program storage area may store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc. The data storage area may store data created during the use of the electronic device 100 (such as audio data, video data, etc.), etc. In addition, the memory 120 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, a universal flash storage (UFS), etc. The processor 110 executes various functional applications and data processing of the electronic device 100 by running instructions stored in the memory 120 and/or instructions stored in a memory provided in the processor.

The wireless communication module 130 can provide wireless communication solutions for the electronic device 100, including WLAN, such as Wi-Fi networks, Bluetooth, NFC, IR, and the like. The wireless communication module 130 can be one or more devices that integrate at least one communication processing module. In some embodiments of the present application, the electronic device 100 can establish a wireless communication connection with other electronic devices through the wireless communication module 130.

The sensor module 140 may include a photoplethysmography sensor, a gyroscope sensor, an electrocardiogram sensor, an air pressure sensor, a magnetic sensor, an acceleration sensor, a distance sensor, a proximity light sensor, etc. The sensor module may be used to collect at least one of a PPG signal, an IMU signal, and an ECG signal.

The camera 150 is used to capture still images or videos. The object generates an optical image through the lens and projects it onto the photosensitive element. The photosensitive element can be a charge coupled device (CCD) or a complementary metal oxide semiconductor (CMOS) phototransistor. The photosensitive element converts the optical signal into an electrical signal, and then passes the electrical signal to the ISP to be converted into a digital image signal. The ISP outputs the digital image signal to the DSP for processing. The DSP converts the digital image signal into an image signal in a standard RGB, YUV or other format. In some embodiments, the electronic device 100 may include 1 or N cameras 150, where N is a positive integer greater than 1.

USB interface 160 is an interface that complies with USB standards and may be a Mini USB interface, a Micro USB interface, a USB Type-C interface, or the like. USB interface 160 can be used to connect to other electronic devices. In some other embodiments, electronic device 100 can also be connected to an external camera via USB interface 160 for image capture.

Display screen 170 is used to display images, videos, and the like. Display screen 170 includes a display panel. The display panel can be a liquid crystal display (LCD), an organic light-emitting diode (OLED), an active-matrix organic light-emitting diode (AMOLED), a flexible light-emitting diode (FLED), a MiniLED, a MicroLED, a Micro-oLED, or a quantum dot light-emitting diode (QLED). In some embodiments, electronic device 100 may include one or N display screens 170, where N is a positive integer greater than one.

It should be understood that the structure illustrated in the embodiments of the present invention does not constitute a specific limitation on the electronic device 100. In other embodiments of the present application, the electronic device 100 may include more or fewer components than shown, or may combine or separate certain components, or arrange the components differently. The illustrated components may be implemented in hardware, software, or a combination of software and hardware.

An embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the blood pressure estimation method provided in the above embodiment are implemented.

The above-mentioned computer-readable storage medium may adopt any combination of one or more computer-readable media. The computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus or device, or any combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM) or flash memory, an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus or device.

A computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, which carries computer-readable program code. Such a propagated data signal may take a variety of forms, including, but not limited to, electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transport a program for use by or in conjunction with an instruction execution system, apparatus, or device.

The program code contained on the computer-readable medium can be transmitted using any appropriate medium, including but not limited to wireless, wire, optical cable, radio frequency (RF), etc., or any suitable combination of the above.

Computer program code for performing the operations of this specification may be written in one or more programming languages, or a combination thereof, including object-oriented programming languages such as Java, Smalltalk, C++, and conventional procedural programming languages such as "C" or similar programming languages. The program code may execute entirely on the user's computer, partially on the user's computer, as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on the remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., through the Internet using an Internet service provider).

An embodiment of the present application provides a computer program product comprising instructions, which, when executed on a computer, enables the computer to execute the steps of the blood pressure estimation method provided in the above method embodiment.

The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in accordance with the embodiments of the present invention is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via a wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) method. The computer-readable storage medium may be any available medium that a computer can store or a data storage device such as a server or data center that includes one or more available media. The available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a magnetic tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive (SSD)).

It should be noted that the descriptions of the above storage medium, program product, and device embodiments are similar to the descriptions of the above method embodiments and have similar beneficial effects as the method embodiments. For technical details not disclosed in the storage medium, program product, and device embodiments of this application, please refer to the description of the method embodiments of this application for understanding.

It should be understood that "one embodiment" or "an embodiment" or "some embodiments" mentioned throughout the specification means that the specific features, structures or characteristics related to the embodiment are included in at least one embodiment of the present application. Therefore, "in one embodiment" or "in an embodiment" or "in some embodiments" appearing throughout the specification do not necessarily refer to the same embodiment. In addition, these specific features, structures or characteristics can be combined in one or more embodiments in any suitable manner. It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application. The above-mentioned serial numbers of the embodiments of the present application are for description only and do not represent the advantages and disadvantages of the embodiments. The above description of the various embodiments tends to emphasize the differences between the various embodiments. The same or similar aspects can be referenced to each other. For the sake of brevity, they will not be repeated here.

The term "and/or" in this article is only a description of the association relationship between associated objects, indicating that there can be three relationships. For example, object A and/or object B can mean: object A exists alone, object A and object B exist at the same time, and object B exists alone.

It should be noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or apparatus comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or apparatus comprising the element.

In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The embodiments described above are merely illustrative. For example, the division of the modules is merely a logical function division. In actual implementation, there may be other division methods, such as: multiple modules or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of devices or modules can be electrical, mechanical or other forms.

The modules described above as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical modules; they may be located in one place or distributed across multiple network units; some or all of the modules may be selected according to actual needs to achieve the purpose of this embodiment.

In addition, all functional modules in the embodiments of the present application can be integrated into one processing unit, or each module can be a separate unit, or two or more modules can be integrated into one unit; the above-mentioned integrated modules can be implemented in the form of hardware or in the form of hardware plus software functional units.

Those skilled in the art will understand that all or part of the steps of implementing the above-mentioned method embodiment can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above-mentioned method embodiment; and the aforementioned storage medium includes: mobile storage devices, read-only memories (ROM), magnetic disks or optical disks, and other media that can store program codes.

Alternatively, if the above-mentioned integrated unit of the present application is implemented in the form of a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present application, or the part that contributes to the relevant technology, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a number of instructions for enabling an electronic device to execute all or part of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, ROMs, magnetic disks or optical disks.

The methods disclosed in the several method embodiments provided in this application can be arbitrarily combined, if they do not conflict, to obtain new method embodiments. The features disclosed in the several product embodiments provided in this application can be arbitrarily combined, if they do not conflict, to obtain new product embodiments. The features disclosed in the several method or device embodiments provided in this application can be arbitrarily combined, if they do not conflict, to obtain new method embodiments or device embodiments.

The above is merely an embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims (21)

A blood pressure estimation method, characterized by being applied to an electronic device, comprising: acquiring a target feature based on at least one of a photoplethysmography (PPG) signal, an inertial measurement unit (IMU) signal, and an electrocardiogram (ECG) signal, wherein the target feature includes at least one of a PPG signal feature, an IMU signal feature, and an ECG signal feature; Blood pressure estimation is performed according to the target feature to obtain a blood pressure estimation result; wherein, The PPG signal feature includes at least one of the following: a pulse waveform analysis PWA feature, a zero-crossing feature, a nonlinear dynamics feature, a pulse transit time PTT feature, a pulse arrival time PAT feature, a first heart rate feature, and a first neural network feature; and/or, The IMU signal feature includes at least one of the following: a second heart rate feature, a relative stroke volume feature, a heart rhythm signal BCG high frequency feature, an IMU multi-axis feature, and a second neural network feature; and/or, The ECG signal features include third neural network features. The method according to claim 1, further comprising at least one of the following steps: Extracting the PWA feature based on waveform feature analysis of the PPG signal; Extracting the zero-crossing feature according to the derivative signal of the standardized PPG signal; Extracting the nonlinear dynamic features according to the PPG signals corresponding to a plurality of heartbeat cycles in the PPG signal; extracting the PTT feature or the PAT feature based on a signal peak of the PPG signal; Segmenting the waveform features of the PPG signal, and matching the positions of corresponding feature points in each segment to extract the first heart rate feature; The first neural network feature is extracted from the PPG signal based on a preset neural network. The method according to claim 2, wherein the PPG signal includes multiple PPG signals of different wavelengths, and extracting the PWA feature based on waveform feature analysis of the PPG signal comprises: performing calculations to remove capillary pulsation interference based on at least two PPG signals of different wavelengths among the multiple PPG signals to obtain a signal after removal; extracting arterial blood pulsation from the removed signal to obtain an arterial blood pulsation waveform; Features are extracted from the arterial blood pulsation waveform to obtain the PWA features. The method according to claim 2, wherein extracting the nonlinear dynamic feature based on the PPG signals corresponding to multiple heartbeat cycles in the PPG signal comprises: performing a high-order expansion on the PPG signal based on calculation of an embedding dimension and a delay time to obtain an expanded signal, wherein the calculation of the embedding dimension includes a Takkens embedding theorem method, and the calculation of the delay time includes at least one of a mutual information method and an autocorrelation method; The expanded signal is quantized to obtain the nonlinear dynamic characteristics, where the nonlinear dynamic characteristics include at least one of a Lyapunov exponent and a correlation dimension. The method according to claim 2, wherein the PPG signal includes a plurality of PPG signals of different wavelengths, and extracting the PTT feature based on a signal peak of the PPG signal comprises: Extracting a peak value within a beat cycle of each of the multiple PPG signals, and calculating a time difference between the peak values of each of the PPG signals to obtain a plurality of time differences of pulse transit times; The time differences of the multiple pulse transit times are correlated with the actually measured blood pressure values to obtain a multi-wavelength pulse transit time feature, which is used to indicate the optimal wavelength combination corresponding to the multi-wavelength pulse transit time. The method according to claim 2, wherein extracting the PTT feature based on a signal peak of the PPG signal comprises: The pulse transit time (PTT) feature is determined based on the time difference between the signal peak position of the PPG signal and the J peak or K peak of the BCG signal extracted from the IMU signal, where the J peak is the maximum peak of the BCG signal in the first direction, and the K peak is the second largest peak of the BCG signal in the second direction, and the first direction and the second direction are opposite. The method according to claim 2, characterized in that extracting the PAT feature based on the signal peak of the PPG signal comprises: The PAT feature is determined based on the time difference between the signal peak position of the PPG signal and the R peak of the ECG signal, where the R peak is the highest peak point of the ECG signal. The method according to claim 1 is characterized in that the first neural network feature includes at least one of a neural network autoencoder feature and a discriminant neural network feature, the neural network autoencoder feature is extracted based on an autoencoder of a convolutional neural network, and the discriminant neural network feature is extracted based on a discriminant neural network model. The method according to claim 8, wherein the convolutional neural network-based autoencoder includes an encoder and a decoder, and extracting neural network autoencoder features based on the convolutional neural network autoencoder includes: Mapping the PPG signal to an encoding of a latent space by the encoder to obtain a latent variable; The latent variable is decoded by the decoder and mapped back to the signal space of the PPG signal to obtain the neural network autoencoder feature. The method according to claim 1, further comprising at least one of the following steps: Segmenting the waveform of the BCG signal extracted from the IMU signal, and matching the positions of corresponding time domain feature points in each segment to extract the second heart rate feature; When the measured object is in a stable measurement state, extracting the relative stroke volume feature according to the amplitude of the BCG signal extracted from the IMU signal; Extracting the BCG high-frequency features based on the waveform morphology characteristics of the BCG signal extracted from the IMU signal; Extracting the IMU multi-axis features based on the acceleration and attitude angle data in three directions; The second neural network feature is extracted from the IMU signal based on a preset neural network. The method according to claim 10, wherein extracting the relative stroke volume feature based on the amplitude of the BCG signal extracted from the IMU signal comprises: Filtering the BCG signal extracted from the IMU signal to obtain a filtered BCG signal; The filtered BCG signal is segmented and peak amplitude information is extracted to obtain the relative stroke volume. The method according to claim 10, wherein extracting the BCG high-frequency features based on the waveform morphology of the BCG signal extracted from the IMU signal comprises: Extracting the amplitude, time interval, area, slope, and energy characteristics of the peaks and troughs of each cardiac cycle of the BCG signal based on the maximum peak of each cardiac cycle in the BCG signal extracted from the IMU signal, the trough and peak closest to the maximum peak, and the trough and peak closest to the maximum peak; The BCG high-frequency feature is determined based on the amplitude, time interval, area, slope and energy characteristics of the peaks and troughs of each heartbeat cycle of the BCG signal. The method according to claim 1, further comprising: The third neural network feature is extracted from the ECG signal based on a preset neural network. The method according to claim 1, wherein estimating blood pressure based on the target feature to obtain a blood pressure estimation result comprises: Inputting the target feature into a trained blood pressure estimator to obtain the blood pressure estimation result; Among them, the trained blood pressure estimator is obtained by merging the target features to obtain a merged feature, then obtaining the mean feature corresponding to the merged feature of a preset time length in the merged feature, and training the initial blood pressure estimator through the mean feature. A blood pressure estimation device, characterized in that it is applied to an electronic device, comprising: a feature acquisition module, configured to acquire a target feature based on at least one of a photoplethysmography (PPG) signal, an inertial measurement unit (IMU) signal, and an electrocardiogram (ECG) signal, wherein the target feature includes at least one of a PPG signal feature, an IMU signal feature, and an ECG signal feature; wherein the PPG signal feature includes at least one of: a pulse waveform analysis (PWA) feature, a zero-crossing feature, a nonlinear dynamics feature, a pulse transit time (PTT) feature, a pulse arrival time (PAT) feature, a first heart rate feature, and a first neural network feature; and/or the IMU signal feature includes at least one of: a second heart rate feature, a relative stroke volume feature, a heart rhythm signal (BCG) high-frequency feature, an IMU multi-axis feature, and a second neural network feature; and/or the ECG signal feature includes a third neural network feature; The result acquisition module is used to estimate the blood pressure according to the target characteristics and obtain the blood pressure estimation result. The device according to claim 15, further comprising a first execution module, configured to perform at least one of the following steps: extracting the PWA feature based on waveform feature analysis of the PPG signal; Extracting the zero-crossing feature according to the derivative signal of the standardized PPG signal; Extracting the nonlinear dynamic features according to the PPG signals corresponding to a plurality of heartbeat cycles in the PPG signal; extracting the PTT feature or the PAT feature based on a signal peak of the PPG signal; Segmenting the waveform features of the PPG signal, and matching the positions of corresponding feature points in each segment to extract the first heart rate feature; The first neural network feature is extracted from the PPG signal based on a preset neural network. The device according to claim 15, wherein the PPG signal comprises a plurality of PPG signals of different wavelengths, and the first execution module is specifically configured to: perform a calculation to remove capillary pulsation interference based on at least two PPG signals of different wavelengths among the plurality of PPG signals to obtain a post-capillary pulsation interference removal signal; extracting arterial blood pulsation from the removed signal to obtain an arterial blood pulsation waveform; Features are extracted from the arterial blood pulsation waveform to obtain the PWA features. A blood pressure estimation device, characterized by comprising: at least one sensor selected from a photoplethysmography (PPG) sensor, an inertial measurement unit (IMU) sensor, and an electrocardiogram (ECG) sensor, and a processor, wherein: The PPG sensor is configured to collect PPG signals; The IMU sensor is configured to collect IMU signals; The ECG sensor is configured to collect ECG signals; The processor is configured to obtain a target feature based on at least one of the PPG signal, the IMU signal, and the ECG signal, wherein the target feature includes at least one of a PPG signal feature, an IMU signal feature, and an ECG signal feature; and perform blood pressure estimation based on the target feature to obtain a blood pressure estimation result; The PPG signal feature includes at least one of the following: a pulse waveform analysis PWA feature, a zero-crossing feature, a nonlinear dynamics feature, a pulse transit time PTT feature, a pulse arrival time PAT feature, a first heart rate feature, and a first neural network feature; and/or, The IMU signal feature includes at least one of the following: a second heart rate feature, a relative stroke volume feature, a heart rhythm signal BCG high frequency feature, an IMU multi-axis feature, and a second neural network feature; and/or, The ECG signal features include third neural network features. A blood pressure estimation device comprises a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and wherein the processor implements the steps of the blood pressure estimation method according to any one of claims 1 to 14 when executing the program. A computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the steps in the blood pressure estimation method according to any one of claims 1 to 14 are implemented. A computer program product, comprising a computer program, characterized in that when the computer program is executed by a processor, the steps in the blood pressure estimation method according to any one of claims 1 to 14 are implemented.