A physiological semantic consistent cuffless blood pressure model contrast learning pre-training method

Abstract

The application provides a physiological semantic consistent cuffless blood pressure model contrast learning pre-training method, the blood pressure prediction model construction method first adopts a label-free PPG signal to perform unsupervised training on the model to improve the generalization ability of the model, then adopts a labeled PPG signal to perform supervised training on the model to train the prediction performance of the model, and finally adopts a small amount of labeled PPG signal of a target object to perform individual training on the model to enable the model to have the performance of predicting blood pressure data of the target object, wherein in the unsupervised training process, data enhancement is realized based on key physiological knowledge of blood pressure to ensure the quality and practicability of the enhanced data, and the diversity of the features extracted by the model is improved through random feature masking and feature resampling. Compared with the existing blood pressure prediction method, the application reduces the demand amount of labeled data in the training process, and at the same time, data enhancement and feature reconstruction based on key physiological information of blood pressure are performed to improve the generalization ability of the blood pressure prediction model.

Description

TECHNICAL FIELD

[0001] The present application relates to the fields of medical health, physiological measurement, deep learning, data enhancement, and the like, in particular to a physiological semantic consistent cuffless blood pressure model contrast learning pre-training method. BACKGROUND

[0002] According to statistics of the World Health Organization, about 1.65 million people die of cardiovascular diseases every year, among which hypertension is one of the most important factors. Therefore, accurate and continuous blood pressure prediction is crucial for the early detection, prevention and treatment of hypertension and related cardiovascular diseases. Traditional blood pressure measurement techniques are invasive or based on inflatable cuff measurement methods. Invasive blood pressure measurement is invasive and even causes infection. The extremely high cost also limits its popularization and use. The blood pressure measurement device based on inflatable cuff has poor user experience due to the inflation of the cuff, making it unsuitable for chronic disease management and community health screening.

[0003] In the past few decades, scholars have tried to develop a non-invasive continuous blood pressure prediction technology based on physiological signals. Among them, photoplethysmography (PPG) is a simple and economical optical method for detecting changes in peripheral circulation system blood volume. In recent years, due to the advantages of deep learning such as large model capacity and good robustness to noisy signals, researchers have applied deep learning technology to predict blood pressure. The existing blood pressure prediction method based on deep model requires a large amount of data, but repeated cuff inflation is very time-consuming and uncomfortable, so it is very laborious to obtain the blood pressure label of PPG signal. On the other hand, the current popular wearable devices such as Huawei bracelet, Xiaomi Watch and Apple Watch have a large amount of unlabelled PPG signal data. These wearable devices have a wider range of coverage of daily activities and population statistics, and their potential is yet to be developed. Contrast learning is a promising self-supervised paradigm that can use unlabelled data for representation learning. Contrast learning has made significant progress in computer vision. However, compared with the methods of computer vision tasks, the time series enhancement method must consider the semantic uncertainty that occurs during the data enhancement process. Therefore, there are still many problems to be solved in directly applying contrast learning to the blood pressure prediction task.

[0004] In summary, in the current research of blood pressure prediction model, the model training excessively depends on the feature distribution of a specific data set and the blood pressure prediction model trained on the limited labeled data basis has insufficient generalization ability, thereby limiting the application ability of the model on new data or different patient groups. In addition, although contrast learning can improve the generalization ability of the model, the existing contrast learning faces challenges in semantic consistency when enhancing physiological time series data. Traditional data enhancement techniques are not suitable for time series data, which can destroy the time dependence in physiological signals and ignore the inherent complexity of physiological signals. Although current time series data enhancement methods mainly focus on the time and frequency domains of signals, they can also destroy the time dependence in physiological signals. Therefore, how to improve the generalization ability of the model and how to ensure the semantic consistency between the physiological time series data after data enhancement and the original data are crucial.

[0005] It should be noted that the background art is only used to introduce the relevant information of the present application, so as to help understand the technical solutions of the present application, but does not mean that the relevant information must be the prior art. In the absence of evidence that the relevant information has been disclosed before the filing date of the present application, the relevant information should not be regarded as prior art. SUMMARY

[0006] Therefore, the purpose of the present application is to overcome the defects of the prior art, and to provide a physiological semantic consistent cuffless blood pressure model contrast learning pre-training method. The purpose of the present application is achieved through the following technical solutions.

[0007] According to a first aspect of the present application, a blood pressure prediction model construction method is provided, the blood pressure prediction model being used to predict the corresponding systolic pressure and diastolic pressure of a target object according to the photoplethysmogram of the target object, the method comprising: step S1, acquiring a first training set containing a plurality of unlabeled photoplethysmograms, a second training set containing a plurality of labeled photoplethysmograms, and a third training set containing a plurality of labeled photoplethysmograms of the target object, wherein the label of each photoplethysmogram is the corresponding systolic pressure and diastolic pressure thereof; step S2, constructing an initial model, the initial model comprising a feature encoding module, a feature reconstruction module, a projection module, and a prediction module, the feature encoding module being used to encode the photoplethysmogram into a feature representation, the feature reconstruction module being used to resample the feature representation to obtain a new feature representation of the photoplethysmogram, the projection module being used to convert the new feature representation of the photoplethysmogram into a corresponding low-dimensional feature vector thereof, and the prediction module being used to predict the corresponding systolic pressure and diastolic pressure of the photoplethysmogram based on the low-dimensional feature vector of the photoplethysmogram; step S3, performing data augmentation on the photoplethysmograms in the first training set based on blood pressure key physiological information to obtain a plurality of positive sample pairs, each positive sample pair containing an original photoplethysmogram and a data-augmented photoplethysmogram thereof, training the initial model in a contrast learning manner based on all the positive sample pairs, and adjusting the parameters of the feature encoding module, the feature reconstruction module, and the projection module based on a preset first loss function; step S4, performing supervised training on the model trained in step S3 based on the second training set, adjusting the model parameters based on a preset second loss function to obtain an initial blood pressure prediction model; and step S5.

[0008] According to the first aspect of the present application, the feature encoding module is based on a ConvNeXt network architecture and consists of four consecutive convolutional neural networks, wherein the number of convolutional blocks of the four consecutive convolutional neural networks is configured as (2, 2, 6, 2) in turn, and the number of channels of the four consecutive convolutional neural networks is configured as (64, 128, 256, 512) in turn.

[0009] According to the first aspect of the present application, the feature reconstruction module is configured to: perform feature reconstruction on the feature representation encoded by the encoding module in a manner of random feature masking and feature resampling in turn to obtain a new feature representation, and dynamically adjust the weight of the feature resampling based on a self-attention mechanism.

[0010] According to the first aspect of the present application, the projection module is a fully connected layer containing a preset number of neurons.

[0011] According to the first aspect of the present application, the prediction module consists of two fully connected layers and uses a rectified linear unit as an activation function.

[0012] According to the first aspect of the present invention, data enhancement is performed on each photoelectric volumetric pulse wave in the first training set based on key physiological information of blood pressure in the following manner: all key signal sampling points containing key physiological information of blood pressure in the photoelectric volumetric pulse wave are obtained, and the key signal sampling points include signal sampling points corresponding to the start of all systolic pulses, signal sampling points corresponding to the maximum peak value during the ejection period of all systolic periods, signal sampling points corresponding to all transitions from systolic period to diastolic period, signal sampling points corresponding to the lowest pressure in all arterial systems, and all morphological feature points of the accelerated photoelectric volumetric pulse wave corresponding to the photoelectric volumetric pulse wave signal, and signal sampling points corresponding to the diastolic peak, wherein the accelerated photoelectric volumetric pulse wave is the second-order derivative of the photoelectric volumetric pulse wave. Based on the acquired signal sampling points, data enhancement is performed on the photoplethysmogram using any of the following methods: Method 1, which is a data enhancement operation of adding noise to the amplitudes of other signal sampling points in the photoplethysmogram except the key signal sampling point and its adjacent signal sampling points, wherein the noise is a value randomly generated by a pre-configured normal distribution probability density function; Method 2, which is a data enhancement operation of adjusting the amplitudes corresponding to the signal sampling points in part of the sampling interval of the photoplethysmogram except the key signal sampling point and its adjacent signal sampling points to 0 and keeping the amplitudes of other signal sampling points unchanged; Method 3, which is a data enhancement operation of rolling the photoplethysmogram by multiple signal sampling points, wherein the number of signal sampling points of the rolling displacement is determined by the following method:

[0013]

[0014] Wherein, T represents the number of signal sampling points of rolling displacement, M represents the total number of signal sampling points corresponding to the start of cardiac systolic pulse, and O t represents the index of the signal sampling point corresponding to the start of the t-th systolic pulse, O t-1 Represents the index of the signal sampling point corresponding to the start of the t-1th cardiac systolic pulse; Method 4, which is to divide the photoplethysmography wave into multiple wavelets and perform a data enhancement operation on all the wavelets in a random arrangement, wherein each wavelet is the waveform between the signal sampling points corresponding to the start of two adjacent cardiac systolic pulses; Method 5, which is to perform data enhancement on the photoplethysmography wave using the operations in Method 1 and Method 4 in sequence.

[0015] According to the first aspect of the present invention, the preset first loss function is configured as:

[0016]

[0017] in, represents the contrastive learning loss, z i 、z jrepresents the feature vector corresponding to the original photoplethysmogram and its corresponding data-augmented photoplethysmogram in a positive sample pair, respectively, sim(·) is a function to calculate the similarity of two feature vectors, N represents the number of positive sample pairs processed simultaneously in one training process, exp(·) represents the exponential function, τ represents a hyperparameter, z k represents the k-th feature vector in the 2N feature vectors corresponding to the N positive sample pairs processed simultaneously in one training process, f(k,i) is an indicator function, when k≠i, the value of f(k,i) is 1, when k=i, the value of f(k,i) is 0.

[0018] According to a first aspect of the present application, the preset second loss function is configured to:

[0019]

[0020] wherein l MAE represents the mean absolute error loss, n represents the number of photoplethysmograms in the third training set, represents the predicted systolic pressure and predicted diastolic pressure corresponding to the i-th photoplethysmogram in the third training set, represents the true systolic pressure and true diastolic pressure corresponding to the i-th photoplethysmogram in the third training set.

[0021] According to a second aspect of the present application, a blood pressure prediction method is provided, which is used to predict the systolic pressure and diastolic pressure corresponding to a target photoplethysmogram based on the target photoplethysmogram, the method comprising: obtaining the target photoplethysmogram; and predicting the systolic pressure and diastolic pressure of the target object based on the target photoplethysmogram using a target blood pressure prediction model constructed by any of the methods according to the first aspect of the present application.

[0022] According to a third aspect of the present application, a blood pressure prediction system is provided, the system comprising: a feature encoding module for encoding a photoplethysmogram into a feature representation; a feature reconstruction module for resampling the feature representation to obtain a new feature representation of the photoplethysmogram; a projection module for converting the new feature representation of the photoplethysmogram into its corresponding low-dimensional feature vector; and a prediction module for predicting the systolic pressure and diastolic pressure corresponding to the photoplethysmogram based on the low-dimensional feature vector of the photoplethysmogram.

[0023] According to a fourth aspect of the present application, an electronic device is provided, comprising: one or more processors; and a memory, wherein the memory is configured to store executable instructions; and the one or more processors are configured to implement the steps of the method according to any of the first aspect of the present application by executing the executable instructions.

[0024] Compared with the prior art, the present application has the advantages that:

[0025] The present application proposes a blood pressure prediction model construction method and a blood pressure prediction method by combining contrast learning and data enhancement technology. The method first uses an unlabelled PPG signal for unsupervised training of the model to improve the generalization ability of the model, then uses a labelled PPG signal for supervised training of the model to train the prediction performance of the model, and finally uses a small amount of labelled PPG signal of the target object for individual training of the model to make the model have the performance of predicting the blood pressure data of the target object. In the unsupervised training process, data enhancement is realized based on the key physiological knowledge of blood pressure to ensure the quality and practicability of the enhanced data, and the diversity of the features extracted by the model is improved through random feature masking and feature resampling. BRIEF DESCRIPTION OF DRAWINGS

[0026] The embodiments of the present application are further described below with reference to the accompanying drawings, in which:

[0027] Figure 1 The blood pressure prediction model construction method according to the embodiment of the present application is shown in the flowchart;

[0028] Figure 2 The key signal sampling point diagram according to the embodiment of the present application is shown;

[0029] Figure 3 The data enhancement contrast diagram according to the embodiment of the present application is shown;

[0030] Figure 4 The blood pressure prediction model construction method according to the embodiment of the present application is shown;

[0031] Figure 5 The contrast experiment result diagram according to the embodiment of the present application is shown;

[0032] Figure 6 The ablation experiment result diagram according to the embodiment of the present application is shown. DETAILED DESCRIPTION

[0033] In order to make the purpose, technical scheme and advantages of the present application clearer, the present application is further described in detail below through specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0034] As mentioned in the background section, in the current research of blood pressure prediction model, the model training excessively relies on the feature distribution of a specific data set and the blood pressure prediction model trained on the basis of limited labeled data has insufficient generalization ability, thereby limiting the application ability of the model on new data or different patient groups. In addition, although contrast learning can improve the generalization ability of the model, the existing contrast learning faces challenges in semantic consistency when enhancing physiological time series data. Traditional data enhancement techniques are not suitable for time series data, which will destroy the time dependence in physiological signals and ignore the inherent complexity of physiological signals. Current time series data enhancement methods mainly focus on the time domain and frequency domain of signals, which will also destroy the time dependence in physiological signals.

[0035] Before the embodiments of the present application are specifically introduced, some terms used therein are explained as follows:

[0036] Photoplethysmography (PPG): Photoplethysmography is a non-invasive biometric technique that measures the change in blood volume in the skin vessels to obtain a pulse wave signal. PPG technology uses LED light sources and detectors to capture the color change of the skin caused by the heart cycle through transmission or reflection, thereby recording the pulsation state of the blood vessels and measuring the pulse wave.

[0037] In order to better understand the present application, the inventive concept of the present application is first introduced. In the scheme of the present application, first, the model is unsupervised trained by using the label-free PPG signal to improve the generalization ability of the model, then the model is supervised trained by using the labeled PPG signal to train the prediction performance of the model, finally the model is personalized trained by using a small amount of labeled PPG signal of the target object to make the model have the performance of predicting the blood pressure data of the target object, wherein in the unsupervised training process, data augmentation is realized based on the key physiological knowledge of blood pressure to ensure the quality and practicality of the augmented data, and random feature masking and feature resampling are used to improve the diversity of the features extracted by the model.

[0038] In summary, the present application proposes a blood pressure prediction model construction method, which is used to predict the corresponding systolic pressure and diastolic pressure of a target object according to the photoplethysmogram of the target object, and the steps of the method are as follows: Figure 1As shown, it comprises: step S1, obtaining a first training set containing a plurality of untagged photoplethysmogram, a second training set containing a plurality of tagged photoplethysmogram, and a third training set containing a plurality of tagged photoplethysmogram of a target object, wherein the label of each photoplethysmogram is its corresponding systolic pressure and diastolic pressure; step S2, constructing an initial model, the initial model comprising a feature encoding module, a feature reconstruction module, a projection module, and a prediction module, the feature encoding module being used to encode the photoplethysmogram into a feature representation, the feature reconstruction module being used to resample the feature representation to obtain a new feature representation of the photoplethysmogram, the projection module being used to convert the new feature representation of the photoplethysmogram into its corresponding low-dimensional feature vector, and the prediction module being used to predict the systolic pressure and diastolic pressure corresponding to the photoplethysmogram based on the low-dimensional feature vector of the photoplethysmogram; step S3, performing data augmentation on the photoplethysmogram in the first training set based on the key physiological information of blood pressure to obtain a plurality of positive sample pairs, each positive sample pair containing an original photoplethysmogram and its data-augmented photoplethysmogram, training the initial model in a contrast learning manner based on all the positive sample pairs, and adjusting the parameters of the feature encoding module, the feature reconstruction module, and the projection module based on a preset first loss function; step S4, performing supervised training on the model trained in step S3 based on the second training set, adjusting the model parameters based on a preset second loss function to obtain an initial blood pressure prediction model; and step S5, performing personalized supervised training on the initial blood pressure prediction model based on the third training set, and adjusting the model parameters based on a preset second loss function to obtain a target blood pressure prediction model.

[0039] In order to better understand the present application, each step will be described in detail below in combination with specific embodiments.

[0040] According to an embodiment of the present application, in the step S1, the photoplethysmogram in the first training set and the second training set is the photoplethysmogram of a non-target object. The present application trains the initial model using untagged photoplethysmogram, which can effectively tap the value of a large amount of unlabeled PPG data collected by the current wearable device, thereby fully releasing the potential of these data.

[0041] According to an embodiment of the present application, the feature encoding module is configured to comprise four consecutive convolutional neural networks based on the ConvNeXt network architecture, wherein the number of convolutional blocks of the four consecutive convolutional neural networks is configured as (2, 2, 6, 2) in turn, and the number of channels of the four consecutive convolutional neural networks is configured as (64, 128, 256, 512) in turn.

[0042] ConvNeXt is a variant of Convolut ional Neural Network (CNN) that improves and optimizes the traditional CNN architecture. ConvNeXt improves the performance and efficiency of image processing and pattern recognition tasks. The key features of ConvNeXt include: ConvNeXt Block: It is the basic building block of ConvNeXt, each block consists of depthwise separable convolution, LayerNorm, GELU activation function and pointwise convolution. These components work together to extract and transform features efficiently. Neural network channel number: ConvNeXt carefully configures the number of channels at different stages to adapt to different scales of feature representation and computational requirements. For example, ConvNeXt-T has a channel number configuration of (96, 192, 384, 768), while ConvNeXt-B has a configuration of (128, 256, 512, 1024). These configurations reflect the model's adjustment of feature depth and width at different stages.

[0043] According to one embodiment of the present application, the feature reconstruction module sequentially masks and resamples the feature representation encoded by the encoding module. By randomly removing features, the model obtains different feature combinations in each training process, increasing feature diversity. Then, the features are resampled, and a self-attention mechanism is used to dynamically assign different weights to each resampled feature to optimize the weighting process of the features.

[0044] According to one embodiment of the present application, the projection module is configured as a fully connected layer containing a preset number of neurons. Preferably, the projection module of the present application is configured as a fully connected layer (also known as a dense layer or linear layer) containing 256 neurons. The main function of the projection module is to map the feature representation output by the feature reconstruction module to a low-dimensional feature vector. During the unsupervised training phase in step S3, the projection module assists the encoding module and the feature reconstruction module to ensure that the model learns more robust feature representations. Specifically, on the one hand, it helps to reduce overfitting and may improve the generalization ability of the model, and on the other hand, it helps the model to focus on learning the most important information, thereby enhancing the robustness of the encoding module.

[0045] According to one embodiment of the present application, the prediction module is composed of two fully connected layers and uses a rectified linear unit as an activation function for supervised training and prediction, preferably, the number of neurons of the two fully connected layers is 64 and 2 respectively. The prediction module receives label information during the training process and optimizes the model according to the label information, and predicts the corresponding diastolic pressure and systolic pressure during the prediction process according to the feature vector output by the projection module. The activation function of the prediction module is used to increase nonlinearity and help the model learn complex feature relationships. Specifically, the first layer of the prediction module (full connection containing 64 neurons) further converts the output of the projection head, and the second layer of the full connection (containing 2 neurons) generates the final prediction result, wherein the 2 neurons of the second layer correspond to the output of the binary classification task (i.e. systolic pressure and diastolic pressure) of the present application.

[0046] According to one embodiment of the present application, in the step S3, the key signal sampling points containing key physiological knowledge in the photoplethysmogram are extracted based on the blood pressure key physiological information, and the data enhancement operation of these signal sampling points and their adjacent signal sampling points is constrained during the data enhancement process, avoiding the destruction of the time dependence in the physiological signal by the data enhancement operation to retain the key features in the photoplethysmogram. In order to more clearly illustrate the distribution of the selected key sampling points in the photoplethysmogram, the following will be described in detail in combination with the drawings, as shown in Figure 2 The distribution of the selected key signal sampling points (O, S, N, D, a, b, c, d, e, f) in the photoplethysmogram is shown, wherein O is the signal sampling point corresponding to the starting point of a cardiac cycle, i.e. the beginning of the systolic pulse; S is the signal sampling point corresponding to the maximum peak during the ejection period of the systolic period; N is the signal sampling point corresponding to the transition from the systolic period to the diastolic period; D is the signal sampling point corresponding to the lowest pressure in the arterial system; a, b, c, d, e, f are signal sampling points selected from the corresponding accelerated photoplethysmogram signal of the photoplethysmogram, which are related to blood pressure information. These signal sampling points include additional hemodynamic information and correspond to the signal sampling points with significant acceleration changes in the cardiac cycle of the photoplethysmogram, wherein the accelerated photoplethysmogram signal is the second derivative of the photoplethysmogram, a, b, c, d are signal sampling points corresponding to four morphological feature points in the systolic period; e, f are signal sampling points corresponding to two peak points in the diastolic period. Each photoplethysmogram is subjected to data enhancement in any one of modes 1, 2, 3, 4, and 5 based on all its key signal sampling points. Each mode will be described in detail below.

[0047] Mode 1, noise based on semantic consistency

[0048] add noise on the amplitude of the signal sampling points other than the key signal sampling points and the signal sampling points adjacent thereto, the noise being configured as a value randomly generated by a normal distribution probability density function. In order to more clearly illustrate the data enhancement operation in mode 1, mode 1 will be described in detail below in combination with examples.

[0049] According to an embodiment of the present application, mode 1 realizes data enhancement through the following steps: step 1, initializing an array m[j] = 1, where j ∈ [0, N], N is the total number of sampling points of the photoplethysmogram; step 2, if j ∈ [i FP -a,i FP + a], then m[j] is set to 0, where FP ∈ {O, N, S, D, a, b, c, d, e, f}, a is a preset value, the value of a determines the range of the signal sampling points adjacent to the key signal sampling points, when a takes the value of 3, it means that m[j] is set to 0 when j ∈ [i FP -3,i FP + 3]; step 3, traversing each signal sampling point of the photoplethysmogram, calculating s'[i] = n(0, σ) * m[i] + s[i], where s'[i] is the updated amplitude of the i-th signal sampling point, s[i] is the original amplitude of the i-th signal sampling point, and n(0, σ) is a pre-configured normal distribution probability density function with a distribution mean of 0 and a standard deviation of σ.

[0050] According to an embodiment of the present application, in the operation of steps 1-3 of mode 1, when calculating the updated amplitudes of the key signal sampling points and the signal sampling points adjacent thereto, the value of m[i] is 0, the amplitudes of these signal sampling points remain unchanged, while the m[i] corresponding to other signal sampling points is 1, and noise is added to the original amplitudes. When mode 1 is used to perform data enhancement on the photoplethysmogram, the amplitude and trend of the key sampling points are preserved, and noise is added to other sampling points, so as to increase data diversity while avoiding feature loss caused by data enhancement.

[0051] Mode 2, mask based on semantic consistency

[0052] randomly intercepting part of the waveform of the photoplethysmogram, and adjusting the amplitudes of the signal sampling points in the intercepted part of the waveform to 0, except for the key signal sampling points and the signal sampling points adjacent thereto. In order to more clearly illustrate the data enhancement operation in mode 2, mode 2 will be described in detail below in combination with examples.

[0053] According to one embodiment of the present application, the mode 2 realizes data enhancement through the following steps: step 1, initializing an array m[j] = 1, where j ∈ [0, N]; step 2, generating a random number t in the interval [0, N-b]; step 3, setting the value of m[j] to 0 for j ∈ [t, t+150], and setting m[j] to 1 if j ∈ [iFP-a, iFP+a], where FP ∈ {O, N, S, D, a, b, c, d, e, f}, b is a preset value, and the value of b determines the range of signal sampling points adjacent to the key signal sampling points; and step 4, traversing each signal sampling point of the photoplethysmogram, and updating the amplitude of each signal sampling point to m[i]*s[i], where s[i] is the original amplitude of the i th signal sampling point of the photoplethysmogram.

[0054] According to one embodiment of the present application, in the operations in steps 1-4 of the mode 2, a part of the sampling interval in the photoplethysmogram is randomly intercepted, the key feature information in the interval is retained, and other features in the interval are masked to realize data enhancement. The mode 2 adopts the feature masking mode to perform data enhancement, and retains the features of the signal sampling points containing key physiological knowledge in the feature masking process to solve the problem of loss of key features of the photoplethysmogram caused by feature masking.

[0055] Mode 3: Time shift based on semantic consistency

[0056] The unlabeled photoplethysmogram is shifted by a plurality of signal sampling points. In order to more clearly illustrate the data enhancement operation in mode 3, mode 3 will be described in detail below in conjunction with an example.

[0057] According to one embodiment of the present application, the mode 3 realizes data enhancement through the following steps: step 1, determining the number of signal sampling points corresponding to the start of each cardiac systole pulse, and the index of the signal sampling point corresponding to the start of each cardiac systole pulse; step 2, configuring the number of signal sampling points for the rolling shift as follows: where T represents the number of signal sampling points for the rolling shift, M represents the total number of signal sampling points corresponding to the start of the cardiac systole pulse, O t represents the index of the signal sampling point corresponding to the start of the t th cardiac systole pulse, O t-1 represents the index of the signal sampling point corresponding to the start of the t-1 th cardiac systole pulse; and step 3, rolling shifting the original photoplethysmogram by T signal sampling points.

[0058] According to one embodiment of the present application, in the operation in steps 1-3 of the mode 3, the number of signal sampling points contained in each cardiac cycle is obtained based on each cardiac systole pulse start corresponding signal sampling point and the average value is taken as the distance of the rolling displacement, and the mode 3 can better maintain the consistency and periodicity of the physiological signal based on the rolling displacement of the data of the average cardiac cycle of the photoplethysmogram, thereby improving the generalization ability and prediction accuracy of the model.

[0059] Mode 4, replacement based on semantic consistency

[0060] The unlabeled photoplethysmogram is divided into multiple sub-waves based on all the signal sampling points corresponding to the start of the cardiac systole pulse, and all the sub-waves are randomly rearranged, wherein each sub-wave is the photoplethysmogram between the signal sampling points corresponding to the start of two adjacent cardiac systole pulses. In order to more clearly illustrate the data enhancement operation in mode 4, mode 4 will be described in detail below in conjunction with an example.

[0061] According to one embodiment of the present application, the mode 4 achieves data enhancement by the following steps: step 1, dividing the photoplethysmogram into multiple sub-waves based on all the signal sampling points corresponding to the start of the cardiac systole pulse, wherein each sub-wave is the waveform between the signal sampling points corresponding to the start of two adjacent cardiac systole pulses of the photoplethysmogram; step 2, randomly arranging all the sub-waves to obtain the enhanced photoplethysmogram.

[0062] According to one embodiment of the present application, in steps 1-3 of the mode 3, the photoplethysmogram is divided into multiple sub-waves corresponding to the cardiac cycle and then rearranged, which can maintain the timing consistency of the heartbeat event and more realistically simulate physiological changes, thereby improving the recognition ability of the model for the periodic characteristics of the cardiac cycle.

[0063] Mode 5, replacement based on semantic consistency and noise addition

[0064] The mode 1 and the mode 4 are sequentially used for data enhancement operation of the unlabeled photoplethysmogram.

[0065] In order to more clearly illustrate the difference between the five data enhancement modes proposed by the present application and the classical data enhancement mode, the inventors processed the same segment of photoplethysmogram by using the five data enhancement modes proposed by the present application and the traditional data enhancement mode respectively, and the results are as follows: Figure 3As shown, the results in the analysis chart can be obtained, the signal after data enhancement based on the semantic consistent noise and the semantic consistent mask can ensure that the amplitude of the feature point does not change, the signal after data enhancement based on the semantic consistent time shift and the semantic consistent replacement can maintain the time sequence consistent performance of the heartbeat event and more truly simulate the physiological changes, so that the data enhancement method proposed in the application can ensure that the semantic information of the signal after data enhancement is consistent with the original signal.

[0066] According to an embodiment of the present application, the preset first loss function is used to calculate the contrast loss between the two feature vectors corresponding to the two photoplethysmogram pairs in each positive sample pair, and the preset first loss function is configured to:

[0067]

[0068] wherein, represents the contrast learning loss, z i , z j , respectively, represent the feature vectors corresponding to the original unlabeled photoplethysmogram and the data-enhanced unlabeled photoplethysmogram in a positive sample pair, sim(·) is a function for calculating the similarity between two feature vectors, N represents the number of positive sample pairs processed simultaneously in one training process, exp(·) represents an exponential function, τ represents a hyperparameter, preferably, in the present application, τ is configured to be 0.1, z k represents the kth feature vector in all feature vectors corresponding to the positive sample pairs processed simultaneously in one training process, f(k, i) is an indicator function, when k≠i, the value of f(k, i) is 1, and when k=i, the value of f(k, i) is 0.

[0069] According to an embodiment of the present application, the preset second loss function is configured to:

[0070]

[0071] wherein, l MAE represents the mean absolute error loss, n represents the number of labeled photoplethysmograms used to train the blood pressure prediction model, represents the predicted systolic pressure and diastolic pressure corresponding to the ith labeled photoplethysmogram, represents the label of the ith labeled photoplethysmogram.

[0072] According to an embodiment of the present application, the blood pressure prediction model construction method can be divided into three stages, as Figure 4As shown in the figure, the three stages shown in the figure include: a first stage, unsupervised pre-training using a large amount of unlabeled data; a second stage, supervised pre-training using a small amount of labeled data; and a third stage, personalized training using labeled data of a target object. Among them, the first stage is the unsupervised pre-training stage, which is the basis of the blood pressure prediction model construction method, including four basic components: a semantic consistency-based data enhancement library, a deep model-based encoder, a semantic consistency-based feature reconstruction, and a contrastive loss. The semantic consistency-based data enhancement library generates positive sample pairs using physiological knowledge related to blood pressure in the PPG signal, ensuring semantic consistency during data enhancement. The encoder consists of a series of convolutional neural network layers that capture spatial and temporal features from the PPG signal and output high-level semantics. In order to enhance the diversity of the encoder output, the semantic consistency-based feature reconstruction is performed by random masking and self-attention mechanism for feature resampling and reweighting, respectively. The output of the feature reconstruction is input into the projection head, which maps the reconstructed features into the feature space and calculates the loss based on the contrastive learning loss function. The second stage is the supervised pre-training stage, which uses a small amount of labeled data to fine-tune the model to obtain a supervised pre-training model consistent with other contrastive learning method frameworks. Due to the significant distribution difference between the data used for supervised pre-training and the personalized data collected on the wearable device, the third stage, the personalized supervised training stage, we follow the previous blood pressure prediction work settings, and further fine-tune the model for each individual using their small amount of personalized PPG signal to improve the accuracy of the model.

[0073] According to one embodiment of the present application, a blood pressure prediction method is proposed for predicting the systolic and diastolic blood pressure of a target object based on its photoplethysmogram. The method comprises: obtaining the photoplethysmogram of the target object; and using the target blood pressure prediction model constructed by the blood pressure prediction model construction method to predict the systolic and diastolic blood pressure of the target object based on the photoplethysmogram of the target object.

[0074] According to one embodiment of the present application, a blood pressure prediction system is proposed, characterized in that the system comprises: a feature encoding module for encoding the photoplethysmogram into a feature representation; a feature reconstruction module for resampling the feature representation to obtain a new feature representation of the photoplethysmogram; a projection module for converting the new feature representation of the photoplethysmogram into its corresponding low-dimensional feature vector; and a prediction module for predicting the systolic and diastolic blood pressure corresponding to the photoplethysmogram based on the low-dimensional feature vector of the photoplethysmogram.

[0075] In order to verify the beneficial effects of the present application, the inventors designed a comparative experiment to evaluate the experimental results of the comparative experiment designed by the inventors, which will be described in detail below.

[0076] According to one embodiment of the present application, comparative experiments compare the blood pressure prediction model construction method (PhysCL) of the present application with the SOTA (state-of-the-art) contrast learning baseline method (such as SimCLR, VICREG, TS-TCC and TF-C) to prove the beneficial effects of the present application. The comparative experiments use the data of 50 subjects in the MIMIC III dataset for pre-training, that is, unsupervised training and supervised training of the initial model to obtain an initial blood pressure prediction model, and then use other datasets for personalized supervised training of the initial blood pressure prediction model. The methods tested in the comparative experiments are allocated and datasets in the following manner: 88% of the data is used for unsupervised pre-training of unlabeled data, 2% of the data is used for supervised pre-training of labeled data, and the remaining 10% of the data is used for validation in the pre-training phase. For each individual, 5, 10, and 25 samples are used to personalize the initial blood pressure prediction model to improve the performance of the method. The size of the validation set sval idat ion for each individual is set to 1024 (MIMIC III), 1024 (MIMIC IV), and 360 (UQVS), respectively. The personalized model is evaluated according to the remaining data of the individual, and the mean absolute error between individuals is recorded.

[0077] Comparing PhysCL with SOTA (state-of-the-art) contrast learning baseline methods (such as SimCLR, VICREG, TS-TCC and TF-C), the results of the comparison of PhysCL with the classic contrast learning baseline method are shown in Table 1 and Table 2, Table 1 shows the systolic blood pressure prediction accuracy of each method, and Table 2 shows the diastolic blood pressure prediction accuracy of each method. The symbols used in Table 1 and Table 2 are explained as follows: labelFor the percentage of labeled data in supervised pre-training, "5", "10", "25" are the number of personalized labeled data, "MIMIC III→MIMIC III" represents pre-training the initial model on the MIMIC III dataset and performing personalized supervised training on the MIMIC III dataset; "MIMIC III→MIMIC IV" represents pre-training the model on the MIMIC III dataset and performing personalized supervised training on the MIMIC IV dataset; "MIMIC III→UQVS" represents pre-training the model on the MIMIC III dataset and performing personalized supervised training on the UQVS dataset. By analyzing the data in Table 1, Table 2, the prediction accuracy of blood pressure MAE of PhysCL can reach the minimum under 9 different experimental settings of 3 datasets compared with other contrast learning methods. Among them, using 2% labeled data from MIMIC III for supervised pre-training and using 25 samples from MIMIC III for each individual for personalization, the prediction accuracy of PhysCL in systolic blood pressure MAE and diastolic blood pressure MAE reaches 7.600 mmHg and 3.940 mmHg, respectively.

[0078] Table 1

[0079]

[0080] Table 2

[0081]

[0082]

[0083] The inventors compared the influence of the number of training samples used in the personalized supervised training phase on the prediction accuracy of the model through experiments, and obtained the results as shown in Figure 5 When performing personalized supervised training on the MIMIC III and MIMIC IV datasets, PhysCL can achieve SOTA prediction accuracy using 5 and 10 samples for training. When performing personalized supervised training on the UQVS dataset, PhysCL can achieve SOTA prediction accuracy using 5 samples for training. Although BPCRNN can achieve SOTA performance by training the model using 25 training samples, in the real world, it is challenging and complex to collect labeled samples from individuals. Therefore, PhysCL can achieve better performance with fewer personalized samples, which is more meaningful for practical applications.

[0084] In order to demonstrate the impact of the data enhancement and feature reconstruction based on key blood pressure information proposed in the present invention on the constructed blood pressure prediction model, the inventors conducted ablation experiments to verify the prediction accuracy of the blood pressure model constructed using SCAB (data enhancement based on key blood pressure information), SCFR (feature reconstruction), and the use of both SCAB and SCFR, as well as the use of neither SCAB nor SCFR. The experimental results are shown in Figure 2. Figure 6 As shown, the results show that when the model is pre-trained using 2% labeled data, data augmentation based on key information of blood pressure helps improve the experimental results in all experimental settings, and feature reconstruction helps improve the experimental results in almost all experiments.

[0085] It should be noted that although the above describes the various steps in a specific order, it does not mean that the steps must be performed in the above specific order. In fact, some of these steps can be executed concurrently or even in a different order as long as the required functions can be achieved.

[0086] The present invention may be a system, a method and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for causing a processor to implement various aspects of the present invention.

[0087] Computer-readable storage media can be a tangible device that holds and stores the instructions used by an instruction execution device. Computer-readable storage media can, for example, include, but are not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination thereof. More specific examples (non-exhaustive list) of computer-readable storage media include: 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), a static random access memory (SRAM), a portable compact disc read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, a punch card or a raised structure in a groove on which instructions are stored, for example, and any suitable combination thereof.

[0088] While various embodiments of the present invention have been described above, the above descriptions are intended to be illustrative, non-exhaustive, and not limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is selected to best explain the principles of the embodiments, their practical applications, or technological improvements in the marketplace, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A method for constructing a blood pressure prediction model, wherein the blood pressure prediction model is used to predict the corresponding systolic and diastolic blood pressure of a target subject based on the photoplethysmography of the target subject, characterized in that: The method comprises: Step S1: Acquire a first training set comprising a plurality of unlabeled photoplethysmograms, a second training set comprising a plurality of labeled photoplethysmograms, and a third training set comprising a plurality of labeled photoplethysmograms of a target subject, wherein the label of each photoplethysmogram is its corresponding systolic and diastolic pressures; Step S2: construct an initial model, which includes a feature encoding module, a feature reconstruction module, a projection module, and a prediction module, wherein: The feature encoding module is used to encode the photoplethysmography into a feature representation, the feature encoding module is based on a ConvNeXt network architecture and is composed of four consecutive convolutional neural networks, the number of convolutional blocks of the four consecutive convolutional neural networks is configured as (2, 2, 6, 2), and the number of channels of the four consecutive convolutional neural networks is configured as (64, 128, 256, 512). The feature reconstruction module is used to resample the feature representation to obtain a new feature representation of the photoplethysmography. The feature reconstruction module sequentially uses random feature masking and feature resampling to reconstruct the feature representation encoded by the encoding module to obtain a new feature representation, and dynamically adjusts the weight of the feature resampling based on the self-attention mechanism; The projection module is used to convert the new feature representation of the photoplethysmogram into its corresponding low-dimensional feature vector, and the prediction module is used to predict the systolic pressure and diastolic pressure corresponding to the photoplethysmogram based on the low-dimensional feature vector of the photoplethysmogram; Step S3: Based on the key physiological information of blood pressure, data enhancement is performed on the photoplethysmogram in the first training set in a preset manner to obtain multiple positive sample pairs, each positive sample pair including the original photoplethysmogram and its data-enhanced photoplethysmogram. An initial model is trained based on all positive sample pairs using a contrastive learning method, and parameters of a feature encoding module, a feature reconstruction module, and a projection module are adjusted based on a preset first loss function, wherein the preset first loss function is a contrastive learning loss function, and the preset method is: Acquire all key signal sampling points in the photoplethysmogram containing key physiological information of blood pressure, the key signal sampling points including signal sampling points corresponding to the start of all systolic pulses, signal sampling points corresponding to the maximum peak value during all systolic ejection periods, signal sampling points corresponding to all transitions from systole to diastole, signal sampling points corresponding to the lowest pressure in all arterial systems, all morphological feature points of the accelerated photoplethysmogram corresponding to the photoplethysmogram signal, and signal sampling points corresponding to the diastolic peak, wherein the accelerated photoplethysmogram is the second-order derivative of the photoplethysmogram; Based on the acquired signal sampling points, the photoplethysmography data is enhanced using any of the following methods: Method 1: a data augmentation operation that adds noise to the amplitudes of signal sampling points other than the key signal sampling point and its adjacent signal sampling points in the photoplethysmography, wherein the noise is a value randomly generated by a preconfigured normal distribution probability density function; Method 2 is a data enhancement operation that adjusts the amplitudes corresponding to the signal sampling points other than the key signal sampling point and its adjacent signal sampling points within a portion of the sampling interval of the photoplethysmogram to 0, while keeping the amplitudes of other signal sampling points unchanged; Method 3 is a data enhancement operation of rolling the photoplethysmography to multiple signal sampling points, wherein the number of signal sampling points of the rolling displacement is determined by the following method: in, The number of signal sampling points representing the rolling displacement, Indicates the total number of signal sampling points corresponding to the start of the systolic pulse, Indicates the The index of the signal sampling point corresponding to the start of the systolic pulse, Indicates the The index of the signal sampling point corresponding to the start of the systolic pulse; Method 4: Divide the photoplethysmography wave into multiple wavelets and perform data augmentation operations on all the wavelets in a random arrangement, wherein each wavelet is the waveform between the signal sampling points corresponding to the start of two adjacent systolic pulses; Method 5, which is to perform data enhancement on the photoplethysmography using the operations in Method 1 and Method 4 in sequence; Step S4: performing supervised training on the model trained in step S3 based on the second training set and adjusting the model parameters based on a preset second loss function to obtain an initial blood pressure prediction model; Step S5: performing personalized supervised training on the initial blood pressure prediction model based on the third training set and adjusting the model parameters based on a preset second loss function to obtain a target blood pressure prediction model.

2. The method according to claim 1, characterized in that The projection module is a fully connected layer comprising a preset number of neurons.

3. The method according to claim 1, characterized in that The prediction module consists of two fully connected layers and uses rectified linear units as activation functions.

4. The method according to claim 1, wherein The preset first loss function is configured as: in, represents the contrastive learning loss, 、 represents the original photoplethysmogram in a positive sample pair and the corresponding feature vector of the photoplethysmogram after data enhancement, is a function that calculates the similarity of two feature vectors. Indicates the number of positive sample pairs processed simultaneously during a training process, represents the exponential function, represents the hyperparameter, Indicates that the The positive sample pairs correspond to The first feature vector feature vectors, is the indicator function, when hour, The value of is 1, when hour, The value of is 0.

5. The method according to claim 1, wherein The preset second loss function is configured as: in, represents the mean absolute error loss, represents the number of photoplethysmography in the third training set, represents the third training set The predicted systolic and diastolic blood pressures corresponding to the photoplethysmogram. Indicates the third training set The true systolic and diastolic pressures corresponding to each photoplethysmogram.

6. A blood pressure prediction method for predicting the corresponding systolic and diastolic blood pressure of a target subject based on the photoplethysmography of the target subject, characterized in that: The method comprises: acquiring a photoplethysmogram of a target object; The target blood pressure prediction model constructed by the method according to any one of claims 1 to 5 predicts the systolic and diastolic blood pressures of the target object based on the photoplethysmogram of the target object.

7. A blood pressure prediction system constructed based on the method according to any one of claims 1 to 6, characterized in that: The system comprises: a feature encoding module for encoding the photoplethysmography into a feature representation; a feature reconstruction module for resampling the feature representation to obtain a new feature representation of the photoplethysmogram; A projection module, configured to convert the new feature representation of the photoplethysmogram into its corresponding low-dimensional feature vector; The prediction module is used to predict the systolic pressure and diastolic pressure corresponding to the photoplethysmography based on the low-dimensional feature vector of the photoplethysmography.

8. An electronic device, characterized in that: include: one or more processors; as well as a memory, wherein the memory is used to store executable instructions; The one or more processors are configured to implement the steps of the method of any one of claims 1 to 6 by executing the executable instructions.

Images