Sleeveless individual blood pressure estimation method based on age grouping fine tuning and two-stage integration

Abstract

The invention discloses a cuff-free individual blood pressure estimation method based on age grouping fine tuning and two-stage integration. The method comprises the following steps of data collection and preprocessing; pre-training the model; finely adjusting the pre-training model; performing two-stage integrated learning to obtain an individual blood pressure estimation model; and calculating a final blood pressure prediction result based on the model. According to the cuff-free individual blood pressure estimation method based on age grouping fine tuning and two-stage integration, the blood pressure estimation performance of a traditional fine tuning normal form can be improved. The age grouping fine tuning emphasizes the matching age grouping fine tuning model to have better accuracy in cuff-free blood pressure estimation; the two-stage integrated learning unit comprises feature-level integration and result-level integration, the feature-level integration fuses encoder features and regression device weights of a plurality of fine tuning models based on age grouping fine tuning to construct an optimal blood pressure model, and the result-level integration can significantly reduce the uncertainty of selection of a plurality of blood pressure models and significantly improve the final blood pressure estimation precision.

Description

Technical Field

[0001] The present invention relates to the technical field of personalized blood pressure estimation, and particularly relates to a cuffless individual blood pressure estimation method with age-group fine-tuning and two-stage integration. Background Art

[0002] With the development of sensor and artificial intelligence technologies, many scholars have conducted extensive research on cuffless blood pressure estimation based on electrocardiogram (ECG) signals and photoplethysmogram (PPG) signals obtained by wearable devices, and used physiological models or neural network methods. For example, some scholars have proposed to extract physiological indexes such as pulse wave conduction time, pulse wave propagation velocity, PPG intensity ratio, and pulse wave amplitude half-width that can reflect arterial blood pressure changes from ECG and PPG signals, establish a physiological mechanism model, and then estimate continuous arterial blood pressure. Some other scholars use the feature automatic extraction characteristics of deep learning technology to automatically extract feature information from ECG or PPG signals, and use networks such as support vector machines or long short-term memory to train deep models to estimate arterial blood pressure.

[0003] According to the invention patent with the publication number: CN117958778A and the publication date: May 3, 2024, a cuffless blood pressure estimation method and system based on unsupervised pulse wave representation learning are disclosed. First, the PPG signal to be measured is obtained; the signal is input into the constructed blood pressure monitoring model to obtain a blood pressure waveform; the construction method of the model includes: obtaining a sample data set, and performing data preprocessing and data augmentation to obtain a pre-training data set and a downstream fine-tuning data set; constructing an unsupervised pulse wave representation learning module and a downstream fine-tuning blood pressure estimation module; training the unsupervised pulse wave representation learning module with the unlabeled pre-training data set, and migrating the parameters of the first encoder in the unsupervised pulse wave representation learning module to the second encoder in the downstream fine-tuning blood pressure estimation module; training the downstream fine-tuning blood pressure estimation module with the labeled downstream fine-tuning data set. Its main technical effect is: it does not need to rely on blood pressure labels and can learn high-value blood pressure-related physiological representations on a large amount of unlabeled PPG signal data sets.

[0004] Another invention patent with the publication number: CN109512410A and the publication date: March 26, 2019 discloses a cuffless continuous blood pressure measurement method for multi-physiological signal feature fusion, including: 1. Deriving a correlation model between arterial blood pressure and the characteristic quantities of the photoplethysmogram pulse signal (PPG) and electrocardiogram signal (ECG) of the human body from the Moens-Korteweg pulse wave conduction model and the arterial baroreflex (ABR) model; 2. According to the model, through a calibration experiment in comparison with a traditional cuff-type sphygmomanometer, determining the individual differential parameters in the model; 3. Using the calibrated model, the arterial blood pressure estimation value can be obtained by measuring PPG and ECG. Its main technical effect is that the continuous blood pressure value of the patient can be obtained without a traditional cuff-type sphygmomanometer.

[0005] In the existing personalized blood pressure estimation methods, due to the lack of introduction of physiological differences in the scheme design, the blood pressure model lacks an understanding of physiological differences, and thus the blood pressure estimation is not accurate enough in some cases. For this reason, age-grouped pre-training is adopted in the pre-training stage to implicitly introduce physiological differences into the deep learning blood pressure model. However, the blood pressure models for different age groups usually reflect the typical vascular characteristics of most people in that age group. However, the actual age of an individual is not always consistent with the vascular age. This physiological difference brings natural uncertainty to the selection of blood pressure models for different age groups, and it is necessary to effectively integrate the advantages of multiple age-grouped blood pressure models to improve the accuracy of the final blood pressure estimation. For this reason, an age-grouped fine-tuning and two-stage integration cuffless individual blood pressure estimation method is proposed, aiming to solve the problems that the existing personalized blood pressure estimation methods lack an understanding of physiological differences and reduce the selection uncertainty of blood pressure models for different age groups, and further improve the blood pressure estimation accuracy based on the traditional transfer learning fine-tuning paradigm. Summary of the Invention

[0006] The purpose of the present invention is to provide an age-grouped fine-tuning and two-stage integration cuffless individual blood pressure estimation method, aiming to solve the problems that the existing personalized blood pressure estimation methods lack an understanding of physiological differences and reduce the selection uncertainty of blood pressure models for multiple age groups after introducing the important physiological factor of age, and further improve the blood pressure estimation accuracy based on the traditional transfer learning fine-tuning paradigm.

[0007] In order to achieve the above purpose, the present invention provides the following technical solutions:

[0008] An age-grouped fine-tuning and two-stage integration cuffless individual blood pressure estimation method, including the following steps:

[0009] Data collection and preprocessing;

[0010] Collect the fingertip pulse wave data and the waveform data of each beat arterial blood pressure of the target patient from the public database. At the same time, collect the phenotypic data such as the gender and age of the target patient. Perform data preprocessing on the fingertip pulse wave data and the waveform data of each beat arterial blood pressure respectively, and output a dataset of pulse wave data signal segments with a fixed duration, fixed frequency, and fixed number of data points, as well as their corresponding diastolic and systolic blood pressure values, and gender and age information, finally forming a preliminary pre-training dataset;

[0011] Collect the pulse wave data of the measured object, and at the same time collect the blood pressure value of the measured object. Perform data preprocessing on the pulse wave data, and finally output a fine-tuning dataset with the same format as the pre-training dataset. A small number of segments in the fine-tuning dataset are the training samples of the fine-tuning dataset, and the remaining segments are the test samples of the fine-tuning dataset;

[0012] The data preprocessing includes:

[0013] Use a time window to divide the pulse wave data with a fixed sampling frequency and the waveform data of each beat arterial blood pressure into several paired segments with a fixed duration;

[0014] Use an autocorrelation filter to eliminate the paired segments with unclear periodicity and damage in several paired segments, so as to screen out high-quality paired segments. The mathematical expression of the autocorrelation filter is:

[0015]

[0016] Among them, R(τ) represents the autocorrelation coefficient at the offset time τ, ξ(t) is the original signal, ξ(t - τ) is the offset signal, and N is the number of overlapping signal points;

[0017] Perform sliding smoothing, filtering, downsampling, and normalization processing on the pulse wave data in the high-quality paired segments. For the waveform data of each beat arterial blood pressure, calculate the representative diastolic and systolic blood pressure values of each segment through the peak-valley extraction algorithm and mean processing, where the mean of the peaks is the diastolic blood pressure and the mean of the valleys is the systolic blood pressure;

[0018] Finally, output the pulse wave signal slices of each object with a fixed number of pieces, fixed duration, fixed frequency, and fixed number of data points, as well as their corresponding diastolic and systolic blood pressure values, and phenotypic information, including gender and age information

[0019] Model pre-training;

[0020] The dataset of pulse wave data signal segments with blood pressure labels in the pre-trained dataset to be prepared is divided into n - 1 specific age groups and 1 mixed age group according to age ranges. Then, a part of the target sample data is extracted from each age group to form multiple pre-trained datasets with the same number of target sample data, which are input into a blood pressure model based on deep learning for supervised pre-training, and finally n pre-trained models for age groups are output.

[0021] Fine-tuning of the pre-trained model;

[0022] Using the training samples in the fine-tuning dataset of the object to be measured, all parameters of the n pre-trained models are updated by full-parameter fine-tuning, and n - 1 fine-tuned blood pressure models for specific age groups and 1 fine-tuned blood pressure model for the mixed age group are output.

[0023] The specific full-parameter fine-tuning is as follows:

[0024]

[0025] Among them, f BP (x; θ) represents the pre-trained blood pressure model defined by the parameter θ pre-trained where x i is the i-th training sample in the fine-tuning dataset, y i is the blood pressure label of this training sample, λ is the weight of the regularization term, R(θ) is the regularization function, and Λ is the loss function.

[0026] Two-stage ensemble learning includes a feature-level ensemble learning unit and a result-level ensemble learning unit. An individual blood pressure estimation model is obtained based on two-stage ensemble learning, and the final blood pressure prediction result is calculated and output based on the individual blood pressure estimation model.

[0027] Feature-level ensemble learning unit;

[0028] The feature integration learning model includes an integration encoder and an integration regressor. Let the encoders of the fine-tuned blood pressure models for n - 1 specific age groups be the integration encoder.

[0029] The training samples of the fine-tuning dataset are respectively input into the fine-tuned blood pressure models for n - 1 specific age groups.

[0030] Features are extracted from the encoders of the n - 1 fine-tuned blood pressure models and normalized to obtain the normalized features

[0031] For the normalized features weighted summation is performed to obtain the output feature F of the integration encoder ens , and F ens is normalized to obtain the final expression form of the integration encoder

[0032] The features generated by the integrated encoder are specifically as follows:

[0033]

[0034] Among them, represents the features of each specific age-grouped blood pressure model after normalization, and ω i represents the weight of the features of each age-grouped blood pressure model in the feature fusion process;

[0035] The parameter matrix of the integrated regressor is obtained by weighted averaging the regressor parameter matrices of n - 1 specific age-grouped fine-tuned blood pressure models

[0036] The parameters of the integrated regressor are specifically as follows:

[0037]

[0038] Among them, is the parameter matrix of the regressors of each age-grouped blood pressure model, and ω i represents the weight of the features of each age-grouped blood pressure model in the feature fusion process;

[0039] Concatenate with the phenotypic information in the fine-tuning dataset to obtain comprehensive features;

[0040] Use to map the comprehensive features to generate the predicted blood pressure values, and the predicted blood pressure values include the predicted systolic blood pressure and the predicted diastolic blood pressure;

[0041] Use the training samples in the fine-tuning dataset to train the feature integrated learning model to generate the final feature integrated learning model M ens ;

[0042] Result-level integrated learning unit;

[0043] The specific calculation steps of the result-level integrated learning unit are as follows:

[0044] Let the prior probabilities P(M ens ) of the n - 1 specific age-grouped fine-tuned blood pressure models, 1 mixed age-grouped fine-tuned blood pressure model, and the feature integrated learning model M i be the same;

[0045] The prior probability P(M i ) is:

[0046]

[0047] Among them, K represents the number of blood pressure models;

[0048] Calculate the marginal likelihood function of model M using the negative absolute error exponential form with variance i Given the training samples in the fine-tuning dataset as the observed data D train Under the condition of the marginal likelihood P(D train |M i )

[0049] The specific form of the marginal likelihood function is as follows:

[0050]

[0051] where is the prediction result of model M i for the nth training sample in the training set, represents the true label of this training sample, and σ is set empirically;

[0052] For the fine-tuned blood pressure models for n - 1 specific age groups, 1 fine-tuned blood pressure model for the mixed age group, and the feature ensemble learning model, calculate the posterior probability P(M i |D train )

[0053] After obtaining the posterior probabilities of the fine-tuned blood pressure models for n - 1 specific age groups, 1 fine-tuned blood pressure model for the mixed age group, and the feature ensemble learning model, an individual blood pressure estimation model is obtained;

[0054] The specific form of the individual blood pressure estimation model is the fine-tuned blood pressure models for n - 1 specific age groups, 1 fine-tuned blood pressure model for the mixed age group, the feature ensemble learning model, and the calculated posterior probability;

[0055] Weighted sum the posterior probability with the estimated blood pressure values output by the above blood pressure models on the test samples in the fine-tuning dataset to obtain the pulse wave sample x of a certain measured object test Finally, the blood pressure prediction result obtained under the two-stage ensemble learning

[0056] In the above technical solution, a cuffless individual blood pressure estimation method with age-group fine-tuning and two-stage ensemble learning provided by the present invention has the following beneficial effects:

[0057] It can improve the estimation performance of diastolic blood pressure and systolic blood pressure in the traditional transfer learning fine-tuning paradigm. Compared with the model obtained by the traditional fine-tuning method, the fine-tuning model that emphasizes matching age groups in age-group fine-tuning has better accuracy in cuffless blood pressure estimation. The two-stage ensemble learning unit includes feature-level ensemble and result-level ensemble. The former fuses the weights of multiple fine-tuning models based on age-group fine-tuning to construct the optimal blood pressure model; the latter can significantly reduce the uncertainty of model selection and significantly improve the final blood pressure estimation accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] To more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the accompanying drawings required in the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments recorded in the present invention. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings.

[0059] Figure 1 Schematic flowchart of the method provided by the embodiment of the present invention;

[0060] Figure 2 Schematic detailed flowchart of the method provided by the embodiment of the present invention;

[0061] Figure 3 Schematic flowchart of the data preprocessing provided by the embodiment of the present invention;

[0062] Figure 4 Schematic flowchart of the model pre-training steps provided by the embodiment of the present invention;

[0063] Figure 5 Schematic flowchart of the feature-level ensemble learning unit provided by the embodiment of the present invention;

[0064] Figure 6 Schematic flowchart of the result-level ensemble learning unit provided by the embodiment of the present invention;

[0065] Figure 7 Schematic flowchart of the blood pressure estimation of test samples provided by the embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0066] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the protection scope of the present invention.

[0067] As Figures 1-7 shown, a cuffless individual blood pressure estimation method based on age-group fine-tuning and two-stage ensemble includes the following steps:

[0068] Data collection and preprocessing;

[0069] Collect the fingertip pulse wave data and the waveform data of each arterial blood pressure beat of the target patient from the public database. At the same time, collect the phenotypic data such as the gender and age of the target patient. Respectively preprocess the fingertip pulse wave data and the waveform data of each arterial blood pressure beat, and output a dataset of pulse wave data signal segments with a fixed duration, fixed frequency, and fixed number of data points, as well as their corresponding diastolic blood pressure and systolic blood pressure values, and gender and age information, and finally form a preliminary pre-training dataset;

[0070] Collect the pulse wave data of the measured object, and at the same time collect the blood pressure value of the measured object. Preprocess the pulse wave data, and finally output a fine-tuning dataset in the same format as the pre-training dataset. A small number of segments in the fine-tuning dataset are the training samples of the fine-tuning dataset, and the remaining segments are the test samples of the fine-tuning dataset;

[0071] In the embodiment provided by the present invention, the number of training samples in the fine-tuning dataset is 30 - 60, and the number of test samples is at least 30;

[0072] As the optimal embodiment provided by the present invention, the number of training samples in the fine-tuning dataset is 50, and the number of test samples is 50, so as to obtain relatively accurate blood pressure estimation data while ensuring the optimal performance of the model.

[0073] In the embodiment provided by the present invention, the public dataset is the publicly available clinical database Medical Information Mart for Intensive Care Ⅲ (MIMIC-III). In this embodiment, the waveform database and the clinical database therein are used. The waveform database records the waveform data and digital records of ICU patients, while the clinical database contains relevant data such as the phenotypic information, disease diagnosis, and drug treatment of patients. In this embodiment, only the waveform data and phenotypic data are adopted.

[0074] When collecting data from the measured object, collect the pulse wave data of the distal phalanx or middle phalanx of the finger of the measured object, and at the same time use a gold-standard electronic sphygmomanometer to record the blood pressure value of the measured object;

[0075] The data preprocessing includes:

[0076] Use a time window to divide the pulse wave data and the waveform data of each arterial blood pressure beat with a fixed sampling frequency into a number of paired segments with a fixed duration;

[0077] Use an autocorrelation filter to eliminate the paired segments with unclear periodicity and damage in the number of paired segments, so as to screen out high-quality paired segments. The mathematical expression of the autocorrelation filter is:

[0078]

[0079] Among them, R(τ) represents the autocorrelation coefficient at the offset time τ, ξ(t) is the original signal, ξ(t - τ) is the offset signal, and N is the number of overlapping signal points;

[0080] Perform sliding smoothing, filtering, downsampling, and normalization on the pulse wave data in high-quality paired segments. For each beat arterial blood pressure waveform data, calculate the representative diastolic and systolic blood pressure values of each segment through the peak-valley extraction algorithm and mean processing, where the mean of the peaks is the diastolic blood pressure and the mean of the valleys is the systolic blood pressure;

[0081] Finally, output the pulse wave signal segments of each object with a fixed number of strips, fixed duration, fixed frequency, and fixed number of data points, as well as their corresponding diastolic and systolic blood pressure values, and phenotypic information, including gender and age information;

[0082] In the embodiment provided by the present invention, the fingertip pulse wave data, the beat arterial blood pressure waveform data, and their key phenotypic data such as gender and age of 1215 patients in the MIMIC-III database are used.

[0083] Use a time window to divide the pulse wave data and the beat arterial blood pressure waveform data with a sampling frequency of 125 Hz into several paired segments with a duration of 8 seconds;

[0084] In the final data preprocessing stage, 4500 8-second, 40-Hz, and 320-data-point pulse wave signal paired segments of each object and their corresponding diastolic and systolic blood pressure values are retained.

[0085] Model pre-training;

[0086] According to age groups, the pulse wave data signal segment dataset with blood pressure labels in the prepared pre-training dataset is divided into n - 1 specific age groups and 1 mixed age group, and a part of the target sample data is extracted from each age group to form multiple pre-training datasets with the same number of target sample data, which are input into the blood pressure model based on deep learning for supervised pre-training, and finally n pre-training models for age groups are output;

[0087] As an embodiment provided by the present invention, the pulse wave dataset is divided into five age - groupings according to age, namely n - 1 specific age - groupings and 1 mixed - age grouping. The n - 1 specific age - groupings are specifically four specific age - groupings, which specifically include a young group, a young - middle - aged group, a middle - old - aged group, and an old - aged group. It should be noted that the division of specific age - groupings can be adjusted according to different needs. As an embodiment provided by the present invention, among them, the age span of the young group is 18 - 29 years old, the age span of the young - middle - aged group is 30 - 49 years old, the age span of the middle - old - aged group is 50 - 69 years old, the age span of the old - aged group is 70 - 89 years old, and the age span of the mixed - age group is 18 - 89 years old. It should be noted that the division span of the actual age - groups can be flexibly adjusted, not limited to the age - span values given in the above - mentioned embodiment; if there are elderly people over 89 years old in the public database, the upper age limit of the mixed - age grouping can be further increased.

[0088] As a specific embodiment provided by the present invention, a certain number of object - patient data are respectively extracted from the pulse wave datasets of the first four specific age - groupings of the object patients as the pre - training dataset, and the data of the measured object is used as the fine - tuning dataset. The pre - training dataset of the mixed - age group is taken from the pre - training datasets of these four specific age - groupings. By extracting data of some subjects to form the pre - training dataset, the pre - training dataset is input into a blood - pressure model based on deep learning for pre - training;

[0089] In this embodiment, the number of object - patient data extracted is 250.

[0090] In the embodiment provided by the present invention, the blood - pressure model based on deep learning specifically includes a single - waveform feature extraction unit, a multi - waveform feature extraction unit, and a blood - pressure regression unit that fuses phenotypic information, which can capture the local features of a single pulse - wave waveform while effectively depicting the global dependence relationship between multiple pulse - wave waveforms;

[0091] Fine - tuning of the pre - trained model;

[0092] Using the training samples in the fine - tuning dataset of the measured object, all parameters of the n pre - trained models are updated by full - parameter fine - tuning, and n - 1 fine - tuning blood - pressure models for specific age - groupings and 1 fine - tuning blood - pressure model for the mixed - age grouping are output;

[0093] Specifically, all parameters of the pre - trained model are updated by full - parameter fine - tuning;

[0094] Full - parameter fine - tuning is specifically expressed as:

[0095]

[0096] Among them, f BP (x; θ) represents a pre-trained blood pressure model defined by parameters θ pre-trained , where x i is the i-th training sample of the fine-tuning data set, y i is the blood pressure label of this training sample, λ is the weight of the regularization term, R(θ) is the regularization function, and Λ is the loss function;

[0097] In the pre-training stage, individuals in the pre-training data set were grouped according to age. Therefore, for individuals in the fine-tuning data set, the pre-trained model matching their age usually provides the best initial parameters. Fine-tuning based on these pre-trained models can significantly improve the accuracy of blood pressure estimation.

[0098] Based on the above-mentioned transfer learning fine-tuning paradigm guided by age grouping, we provided four fine-tuning blood pressure models for specific age groups and one fine-tuning blood pressure model for mixed age groups for each subject; in the prior art, usually a model is randomly selected or directly the model with the best performance on the training set is selected, but for the methods adopted in the prior art, there are limitations: randomly selecting a model may not find the optimal model, thus affecting the accuracy of blood pressure estimation; while only looking at the performance on the training set is likely to lead to overfitting, especially in the case of few samples, the model often cannot maintain the same excellent performance on the test set, ultimately resulting in affecting the accuracy of blood pressure estimation.

[0099] In the embodiments provided by the present invention, it is proposed to use ensemble learning to solve the uncertainty in the selection of blood pressure models based on deep learning, and by integrating the advantages of multiple models, the stability and accuracy of blood pressure estimation are improved;

[0100] Two-stage ensemble learning;

[0101] Two-stage ensemble learning includes a feature-level ensemble learning unit and a result-level ensemble learning unit. An individual blood pressure estimation model is obtained based on two-stage ensemble learning, and the final blood pressure prediction result is calculated and output based on the individual blood pressure estimation model;

[0102] Feature-level ensemble learning unit;

[0103] The feature ensemble learning unit includes an ensemble encoder and an ensemble regressor;

[0104] Let the encoders of n - 1 fine-tuning blood pressure models for specific age groups be the ensemble encoder;

[0105] Input the training samples of the fine-tuning data set into n - 1 fine-tuning blood pressure models for specific age groups respectively;

[0106] Extract features from the encoders of n - 1 fine-tuning blood pressure models And normalize it to obtain the normalized features

[0107] In this embodiment, the training samples x of the fine-tuning dataset are respectively input into the fine-tuning blood pressure models of four specific age groups, namely the youth group, the young and middle-aged group, the middle-aged and elderly group, and the elderly group;

[0108] It should be noted that in this embodiment, a small number of training samples are used for fine-tuning, and the training samples x of the fine-tuning dataset used are 30 - 60.

[0109] As the optimal embodiment provided by the present invention, the training sample x of the fine-tuning dataset used is 50.

[0110] In this embodiment, features are extracted from the encoders of the fine-tuning blood pressure models of four specific age groups, that is, features are extracted from the youth group, the young and middle-aged group, the middle-aged and elderly group, and the elderly group respectively

[0111] From the encoders of the fine-tuning blood pressure models of four specific age groups:

[0112] Features are extracted from Encoderi (i ∈ I, I = {main, aux1, aux2, aux3})

[0113] To further improve the accuracy of the feature integration learning model in the blood pressure estimation task, according to the age group to which the measured object belongs, a suitable model is selected as the main model and the auxiliary model;

[0114] Specifically, let the fine-tuning blood pressure models of the youth group, the young and middle-aged group, the middle-aged and elderly group, and the elderly group be M1, M2, M3, and M4 respectively, and the age group of the measured object be a ∈ {1, 2, 3, 4}, where a = 1 represents the youth group, a = 2 represents the young and middle-aged group, a = 3 represents the middle-aged and elderly group, and a = 4 represents the elderly group. In this way, the main model M of the measured object can be represented main and the auxiliary model M aux1 , M aux2 , M aux3 's relationship:

[0115]

[0116] As a specific embodiment provided by the present invention, if the measured object belongs to the young and middle-aged group, then the fine-tuning blood pressure model of this group is regarded as M main , while the fine-tuning blood pressure models of the youth group, the middle-aged and elderly group, and the elderly group are respectively regarded as M aux1 , M aux2 and M aux3 .

[0117] In the integrated encoder, the fine-tuning training data of the object to be measured is simultaneously input into M main 、M aux1 、M aux2 and M aux3 encoders, which will respectively generate the main feature, auxiliary feature 1, auxiliary feature 2 and auxiliary feature 3. The main feature, auxiliary feature 1, auxiliary feature 2 and auxiliary feature 3 are represented as where i ∈ {main, aux1, aux2, aux3}, and the age group corresponding to main is the age group where the object to be measured is located;

[0118] For the normalized features weighted summation is performed to obtain the output feature F ens ;

[0119] The features generated by the integrated encoder are specifically:

[0120]

[0121] where, represents the features of each specific age group blood pressure model after normalization, and ω i represents the weight of each feature age group blood pressure model feature in the feature fusion process. This weight is related to the actual age of the object to be measured. According to experience, ω i is set to ω main = 0.7, ω aux1 = 0.1, ω aux2 = 0.1, ω aux3 = 0.1;

[0122] For F ens normalization is performed to obtain the final expression form of the integrated encoder

[0123]

[0124] The regression parameter matrices of n - 1 specific age group fine-tuning blood pressure models are weighted averaged to obtain the regression parameter matrix of the integrated regressor

[0125] As an embodiment provided by the present invention, the regression parameter matrices of four specific age group fine-tuning blood pressure are weighted averaged according to the weights in the integrated encoder to ensure that the integrated regressor parameters of the feature integration learning model can be aligned with the integrated encoder, which is expressed as:

[0126]

[0127] The integrated regressor parameters are specifically:

[0128]

[0129] Among them, is the parameter matrix of the blood pressure model regressor for each age group, ω i represents the weight of each feature of the blood pressure model for the characteristic age group in the feature fusion process. According to experience, ω i is set to ω main = 0.7, ω aux1 = 0.1, ω aux2 = 0.1, ω aux3 = 0.1;

[0130] Concatenate the features generated by the integrated encoder with the phenotypic information x plus in the fine-tuning dataset to obtain comprehensive features;

[0131] Use to map the comprehensive features to generate the output y output ;

[0132] The output y output is the estimated blood pressure value, and the estimated blood pressure value includes the estimated systolic blood pressure and the estimated diastolic blood pressure, which is expressed as:

[0133]

[0134] Use a small number of training samples in the fine-tuning dataset to train the feature ensemble learning model, and generate the final feature ensemble learning model M ens of this subject;

[0135] As an embodiment provided by the present invention, the model is trained using the training samples of 30 - 60 measured objects;

[0136] In this embodiment, the model is trained using the training samples of 50 measured objects, and the final feature ensemble learning model M ens of this measured object is generated.

[0137] Result-level ensemble learning unit;

[0138] As an embodiment provided by the present invention, the result-level ensemble learning unit is specifically an improved Bayesian model averaging method.

[0139] The calculation of the posterior probability distribution depends on the marginal likelihood function, which comprehensively considers the explanatory ability of all possible models for the observed data, thereby providing an important basis for model selection and parameter inference. For this reason, the present invention proposes a reasonable marginal likelihood function, which can directly affect the posterior probability distribution of the model, and thus determines the weight of each model in the overall prediction.

[0140] In the prior art, Bayesian models often use numerical approximation methods such as Markov Chain Monte Carlo to estimate the marginal likelihood. However, for deep learning models, their ultra-high-dimensional parameter space makes it extremely time-consuming to calculate the marginal likelihood integral using traditional Bayesian model averaging methods, so it is not applicable.

[0141] The specific calculation steps of the result-level ensemble learning unit are as follows:

[0142] Let the prior probabilities of the fine-tuned blood pressure models for n - 1 specific age groups, 1 fine-tuned blood pressure model for the mixed age group, and the feature ensemble learning model M ens be the same;

[0143] In this embodiment, the prior probabilities P(M i ) of the above three fine-tuned blood pressure models for specific age groups, one fine-tuned blood pressure model for the mixed age group, and the above feature ensemble learning model are set to be the same;

[0144] Specifically, the prior probability P(M i ) is:

[0145]

[0146] where K represents the number of blood pressure models;

[0147] This application proposes a marginal likelihood function in the form of a negative absolute error exponent with variance, aiming to more quickly and intuitively measure the stability and accuracy of each model in fitting actual data. By introducing variance to adjust the volatility of the absolute error, the assumption that the absolute error follows a normal distribution is avoided, so that the marginal likelihood function can adapt to a wider range of error distributions.

[0148] The marginal likelihood function provided by this application fully considers the important evaluation index in the blood pressure estimation task - the mean absolute error, to ensure that the marginal likelihood function is highly consistent with the task objective of the model.

[0149] Calculate the marginal likelihood P(D i |M train ) of model M train under the condition that the training samples in the given fine-tuning dataset are used as the observed data D i ;

[0150] The marginal likelihood function is specifically:

[0151]

[0152] where is the prediction result of model M i for the nth training sample in the training set, denotes the true label of the training sample, where σ is an empirically set value and is set to 5 according to experience;

[0153] The calculation of the posterior probability distribution depends on the marginal likelihood function, which comprehensively considers the explanatory power of all possible models for the observed data, thus providing an important basis for model selection and parameter inference. The marginal likelihood function can directly affect the posterior probability distribution of the model, thereby determining the weight of each model in the overall prediction;

[0154] For the fine-tuned blood pressure models for n - 1 specific age groups, 1 fine-tuned blood pressure model for the mixed age group, and the feature ensemble learning model, the posterior probability P(M i |D train ) is calculated by combining their prior probabilities and marginal likelihood probabilities;

[0155] In this embodiment, for the fine-tuned blood pressure models for four specific age groups, 1 fine-tuned blood pressure model for the mixed age group, and the feature ensemble learning model M ens the posterior probability is calculated by combining their prior probabilities and marginal likelihood probabilities;

[0156] Specifically, multiply the marginal likelihood probability P(D train |M i ) corresponding to each blood pressure model by its prior probability, then normalize the multiplication results for each blood pressure model, and finally output the posterior probability P(M i |D train ), which is expressed as:

[0157]

[0158] After obtaining the posterior probabilities of the fine-tuned blood pressure models for n - 1 specific age groups, 1 fine-tuned blood pressure model for the mixed age group, and the feature ensemble learning model, an individual blood pressure estimation model is obtained;

[0159] The individual blood pressure estimation model is specifically the fine-tuned blood pressure models for n - 1 specific age groups, 1 fine-tuned blood pressure model for the mixed age group, the feature ensemble learning model, and the calculated posterior probability;

[0160] The posterior probability is weighted and summed with the estimated blood pressure values output by the above blood pressure models on the test samples of the fine-tuning dataset to obtain the pulse wave sample x of a certain subject to be measured test Finally, the blood pressure prediction result obtained under the two-stage ensemble learning is expressed as:

[0161]

[0162] where, M i (x test) represents the prediction result of the i-th model for a certain test sample x test of the prediction result.

[0163] As Figure 7 shown, after adopting the method provided by the present invention, after pre-integrating the fine-tuned blood pressure model and the feature ensemble learning model corresponding to the test sample and obtaining the posterior probability of the model, when inputting the test sample of the new object to be measured, the weighted sum of a group of outputs generated by the new pulse wave test sample can be directly calculated through the posterior probability, that is, the final estimated diastolic blood pressure and systolic blood pressure of the new input test sample of the object to be measured can be directly obtained.

[0164] Those skilled in the art in the following fields should understand that the embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0165] The present invention is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram can be implemented by computer program instructions, and the combination of the processes and / or blocks in the flowchart and / or block diagram can also be implemented. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0166] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including instruction means, and the instruction means implements the functions specified in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0167] These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide means for implementing the functions specified in Figure 1One or more processes and / or boxes Figure 1 Steps of functions specified in one box or multiple boxes.

[0168] In the present invention, specific embodiments are used to expound the principles and implementation manners of the present invention. The description of the above embodiments is only used to help understand the method and its core idea of the present invention; at the same time, for those of ordinary skill in the art, according to the idea of the present invention, there will be changes in the specific implementation manners and application scopes. To sum up, the content of this specification should not be construed as a limitation to the present invention.

[0169] The embodiments of the present application also provide a specific implementation manner of an electronic device that can implement all the steps in the method in the above embodiments. The electronic device specifically includes the following:

[0170] A processor, a memory, a communication interface, and a bus;

[0171] Among them, the processor, the memory, and the communication interface complete communication with each other through the bus;

[0172] The processor is used to call the computer program in the memory. When the processor executes the computer program, all the steps in the method in the above embodiments are implemented. For example, when the processor executes the computer program, the following steps are implemented:

[0173] Data collection and preprocessing;

[0174] Collect fingertip pulse wave data and waveform data of each cardiac cycle arterial blood pressure of the target patient from a public database. At the same time, collect phenotypic data such as the gender and age of the target patient. Respectively preprocess the fingertip pulse wave data and waveform data of each cardiac cycle arterial blood pressure, and output a dataset of pulse wave data signal segments with a fixed duration, a fixed frequency, and a fixed number of data points, as well as their corresponding diastolic blood pressure and systolic blood pressure values, and gender and age information, and finally form a preliminary pre-training dataset;

[0175] Collect the pulse wave data of the measured object, and at the same time collect the blood pressure value of the measured object. Preprocess the pulse wave data, and finally output a fine-tuning dataset in the same format as the pre-training dataset. A small number of segments in the fine-tuning dataset are training samples of the fine-tuning dataset, and the remaining segments are test samples of the fine-tuning dataset;

[0176] Model pre-training;

[0177] The dataset of pulse wave data signal segments with blood pressure labels in the pre-trained dataset to be prepared is divided into n-1 specific age groups and 1 mixed age group according to age groups, and a part of the target sample data is extracted from each age group to form multiple pre-trained datasets with the same number of target sample data, which are input into the blood pressure model based on deep learning for supervised pre-training, and finally n pre-trained models for age groups are output;

[0178] Fine-tuning of the pre-trained model;

[0179] Using the training samples in the fine-tuning dataset of the measured object, all parameters of the n pre-trained models are updated by full-parameter fine-tuning, and n-1 fine-tuned blood pressure models for specific age groups and 1 fine-tuned blood pressure model for the mixed age group are output;

[0180] Two-stage ensemble learning;

[0181] The two-stage ensemble learning includes a feature-level ensemble learning unit and a result-level ensemble learning unit. An individual blood pressure estimation model is obtained based on the two-stage ensemble learning, and the final blood pressure prediction result is calculated and output based on the individual blood pressure estimation model.

[0182] An embodiment of the present application also provides a computer-readable storage medium capable of implementing all steps in the method in the above embodiment. A computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, all steps of the method in the above embodiment are implemented. For example, when the processor executes the computer program, the following steps are implemented:

[0183] Data collection and preprocessing;

[0184] Collect the fingertip pulse wave data and the waveform data of each cardiac cycle blood pressure of the object patient from the public database. At the same time, collect the phenotypic data such as the gender and age of the object patient. The fingertip pulse wave data and the waveform data of each cardiac cycle blood pressure are respectively preprocessed to output a dataset of pulse wave data signal segments with a fixed duration, a fixed frequency, and a fixed number of data points, as well as their corresponding diastolic and systolic blood pressure values, and gender and age information, and finally a pre-trained dataset to be prepared is formed;

[0185] Collect the pulse wave data of the measured object, and at the same time collect the blood pressure value of the measured object. The pulse wave data is preprocessed, and finally a fine-tuning dataset in the same format as the pre-trained dataset is output. A small number of segments in the fine-tuning dataset are the training samples of the fine-tuning dataset, and the remaining segments are the test samples of the fine-tuning dataset;

[0186] Model pre-training;

[0187] The dataset of pulse wave data signal segments with blood pressure labels in the pre-trained dataset to be prepared is divided into n - 1 specific age groups and 1 mixed age group according to age ranges. Then, part of the target sample data is extracted from each age group to form multiple pre-trained datasets with the same number of target sample data, which are input into a deep learning-based blood pressure model for supervised pre-training, and finally, pre-trained models for n age groups are output.

[0188] Fine-tuning of the pre-trained model;

[0189] Using the training samples in the fine-tuning dataset of the object to be measured, all parameters of the n pre-trained models are updated by full-parameter fine-tuning, and n - 1 fine-tuned blood pressure models for specific age groups and 1 fine-tuned blood pressure model for the mixed age group are output.

[0190] Two-stage ensemble learning;

[0191] Two-stage ensemble learning includes a feature-level ensemble learning unit and a result-level ensemble learning unit. An individual blood pressure estimation model is obtained based on two-stage ensemble learning, and the final blood pressure prediction result is calculated and output based on the individual blood pressure estimation model.

[0192] The various embodiments in this specification are described in a progressive manner. For the same or similar parts among the various embodiments, reference can be made to each other, and the key points of each embodiment are the differences from other embodiments. In particular, for the hardware + program type embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the corresponding descriptions in the method embodiments. Although the method operation steps as described in the embodiments of this specification are provided, based on conventional or non-creative means, there may be more or fewer operation steps. The order of steps listed in the embodiments is only one of the ways of the execution order of numerous steps and does not represent the only execution order. When the actual device or terminal product is executed, it can be executed in the order shown in the embodiments or the drawings or executed in parallel (for example, in an environment of parallel processors or multi-threaded processing, or even in a distributed data processing environment). The term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, product or device comprising a series of elements not only includes those elements but also includes other elements not explicitly listed, or further includes elements inherent to such process, method, product or device. Without further limitation, it does not exclude the existence of additional identical or equivalent elements in the process, method, product or device comprising the said elements. For the convenience of description, the above device is described by dividing it into various modules according to functions. Of course, when implementing the embodiments of this specification, the functions of each module can be implemented in the same or multiple software and / or hardware, or the modules implementing the same function can be realized by a combination of multiple sub-modules or sub-units, etc. The device embodiments described above are merely illustrative. For example, the division of the said units is only a logical function division, and there may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling or direct coupling or communication connection to each other can be through some interfaces, and the indirect coupling or communication connection of the device or unit can be in electrical, mechanical or other forms. The present invention is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of the present invention. It should be understood that each flow and / or block in the flowchart and / or block diagram, and the combination of flows and / or blocks in the flowchart and / or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate a device for implementing the functions specified in one or more flows and / or blocks or multiple flows and / or blocks. Figure 1 in one or more flows and / or blocks Figure 1 or multiple blocks

[0193] Those skilled in the art should understand that the embodiments of this specification can be provided as methods, systems, or computer program products. Therefore, the embodiments of this specification can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the embodiments of this specification can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code. Each embodiment in this specification is described in a progressive manner. For the same or similar parts among the embodiments, reference can be made to each other. The key point of each embodiment is to illustrate the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can refer to the partial description of the method embodiment. In the description of this specification, the description of reference terms such as "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the embodiments of this specification.

[0194] In this specification, the schematic expressions of the above terms do not necessarily refer to the same embodiment or example. In addition, without contradiction, those skilled in the art can combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples. The above is only the embodiments of the embodiments of this specification and is not used to limit the embodiments of this specification. For those skilled in the art, various changes and modifications can be made to the embodiments of this specification. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the embodiments of this specification shall be included within the scope of the claims of the embodiments of this specification.

Claims

1. A cuffless individual blood pressure estimation method with age group fine-tuning and two-stage integration, characterized in that, It includes the following steps: Data collection and preprocessing; Collect the fingertip pulse wave data and the waveform data of each beat arterial blood pressure of the target patient from the public database. At the same time, collect the phenotypic data such as the gender and age of the target patient. Perform data preprocessing on the fingertip pulse wave data and the waveform data of each beat arterial blood pressure respectively, and output the pulse wave data signal segment dataset with a fixed duration, fixed frequency, and fixed number of data points, as well as the corresponding diastolic blood pressure and systolic blood pressure values, and gender and age information, and finally form a preliminary pre-training dataset; Collect the pulse wave data of the measured object, and at the same time collect the blood pressure value of the measured object. Perform data preprocessing on the pulse wave data, and finally output a fine-tuning dataset with the same format as the pre-training dataset. A small number of segments in the fine-tuning dataset are the training samples of the fine-tuning dataset, and the remaining segments are the test samples of the fine-tuning dataset; Model pre-training; According to the age group, divide the pulse wave data signal segment dataset with blood pressure labels in the preliminary pre-training dataset into n - 1 specific age groups and 1 mixed age group, and extract part of the target sample data from each age group to form multiple pre-training datasets with the same number of target sample data, and input them into the blood pressure model based on deep learning for supervised pre-training, and finally output the pre-training models of n age groups; Fine-tuning of the pre-trained model; Use the training samples in the fine-tuning dataset of the measured object, and update all the parameters of the n pre-trained models by full-parameter fine-tuning, and output the fine-tuning blood pressure models of n - 1 specific age groups and 1 mixed age group; Two-stage ensemble learning; The two-stage ensemble learning includes a feature-level ensemble learning unit and a result-level ensemble learning unit. An individual blood pressure estimation model is obtained based on the two-stage ensemble learning, and the final blood pressure prediction result is calculated and output based on the individual blood pressure estimation model.

2. According to the method for estimating the cuffless individual blood pressure with age group fine-tuning and two-stage ensemble learning as described in claim 1, the feature integration learning unit includes an integration encoder and an integration regressor, and the specific calculation steps are as follows: Let the encoders of the fine-tuning blood pressure models of n - 1 specific age groups be the integration encoder; Input the training samples of the fine-tuning dataset into the fine-tuning blood pressure models of n - 1 specific age groups respectively; Extract features from the encoders of n-1 fine-tuned blood pressure models And normalize them to obtain normalized features For the normalized features perform weighted summation to obtain the output feature F of the integrated encoder ens , for F ens normalize to obtain the final expression form of the integrated encoder The parameter matrix of the integrated regressor is obtained by weighted averaging the regressor parameter matrices of the fine-tuned blood pressure models for n-1 specific age groups Combine with the phenotypic information in the fine-tuning dataset to obtain comprehensive features; Use Generate an output estimated blood pressure value by mapping comprehensive features, where the estimated blood pressure value includes an estimated systolic blood pressure and an estimated diastolic blood pressure; Train the feature ensemble learning model using the training samples in the fine-tuning dataset to generate the final feature ensemble learning model M of the subject ens .

3. According to the method for estimating the cuffless individual blood pressure with age group fine-tuning and two-stage ensemble learning as described in claim 1, the specific calculation steps of the result-level ensemble learning unit are as follows: Let the fine-tuned blood pressure models for n - 1 specific age groups, the fine-tuned blood pressure model for one mixed age group, and the feature ensemble learning model M ens have the same prior probability P(M i ); Calculate the marginal likelihood function of model M using the negative absolute error exponential form with variance i Given the training samples in the fine-tuning dataset as the observed data D train Under the condition of the marginal likelihood P(D train |M i ); Fine-tuned blood pressure models for n-1 specific age groups, a fine-tuned blood pressure model for a mixed age group, and a feature ensemble learning model calculate the posterior probability P(M i |D train ) by combining their prior probabilities and marginal likelihood probabilities; After obtaining the posterior probabilities of the fine-tuning blood pressure models of n - 1 specific age groups, 1 fine-tuning blood pressure model of the mixed age group, and the feature integration learning model, obtain the individual blood pressure estimation model; The individual blood pressure estimation model is specifically the fine-tuning blood pressure models of n - 1 specific age groups, 1 fine-tuning blood pressure model of the mixed age group, the feature integration learning model, and the calculated posterior probabilities; The posterior probability is weighted and summed with the estimated blood pressure value output by the above blood pressure model on the test samples of the fine-tuning dataset to obtain the pulse wave sample x of a certain subject to be measured. test The blood pressure prediction result finally obtained under the two-stage ensemble learning.

4. A cuffless individual blood pressure estimation method with age group fine-tuning and two-stage integration according to claim 1, characterized in that, The full-parameter fine-tuning is specifically: Among them, f BP (x; θ) represents a pre-trained blood pressure model defined by the parameter θ pre-trained , x i is the i-th training sample of the fine-tuning data set, and y i is the blood pressure label of this training sample. λ is the weight of the regularization term, R(θ) is the regularization function, and Λ is the loss function.

5. The cuffless individual blood pressure estimation method with age group fine-tuning and two-stage integration according to claim 1, characterized in that, The features generated by the integration encoder are specifically: Among them, represents the features of each specific age-grouped blood pressure model after normalization, and ω i represents the weight of the features of each age-grouped blood pressure model in the feature fusion process.

6. The cuffless individual blood pressure estimation method with age grouping fine-tuning and two-stage integration according to claim 2, characterized in that, The parameters of the integration regressor are specifically: Among them, is the parameter matrix of the blood pressure model regressor for each age group, and ω i represents the weight of each feature of the blood pressure model for the characteristic age group in the feature fusion process.

7. An individual blood pressure estimation method without a cuff with age group fine-tuning and two-stage integration according to claim 3, characterized in that, The marginal likelihood function is specifically: Among them, is the prediction result of the model M i for the nth training sample in the training set, represents the true label of this training sample, and σ is set empirically.

8. The cuffless individual blood pressure estimation method with age group fine-tuning and two-stage integration according to claim 3, wherein The prior probability P(M i ) is as follows: Where K represents the number of blood pressure models.

9. The cuffless individual blood pressure estimation method with age group fine-tuning and two-stage integration according to claim 3, wherein, The calculation of the posterior probability is specifically: Multiply the marginal likelihood probability P(D train |M i ) corresponding to each blood pressure model by its prior probability, then normalize the multiplication results for each blood pressure model, and finally output the posterior probability, expressed as:

10. A cuffless individual blood pressure estimation method with age group fine-tuning and two-stage integration according to claim 1, characterized in that The data preprocessing includes: Using a time window, the pulse wave data with a fixed sampling frequency and the waveform data of each beat arterial blood pressure are segmented into several paired segments of a fixed duration; Using an autocorrelation filter to eliminate the paired segments with unclear periodicity and damage in the several paired segments, so as to screen and obtain high-quality paired segments. The mathematical expression of the autocorrelation filter is: where R(τ) represents the autocorrelation coefficient at an offset time of τ, ξ(t) is the original signal, ξ(t - τ) is the offset signal, and N is the number of overlapping signal points; Performing sliding smoothing, filtering, downsampling, and normalization processing on the pulse wave data in the high-quality paired segments, and performing peak-valley extraction algorithm and mean processing on the waveform data of each beat arterial blood pressure to calculate the representative diastolic and systolic blood pressure values of each segment. Among them, the mean of the peaks is the diastolic blood pressure, and the mean of the valleys is the systolic blood pressure; Finally, output the pulse wave signal segments of each object with a fixed number of strips, a fixed duration, a fixed frequency, and a fixed number of data points, as well as their corresponding diastolic and systolic blood pressure values, and phenotypic information, including gender and age information.

Images