Blood pressure monitoring method, apparatus, device, and storage medium

Abstract

The application provides a blood pressure monitoring method, device, equipment and storage medium. The method comprises the following steps: collecting a PPG signal of a user; processing the PPG signal by using a blood pressure waveform estimation model to obtain a target blood pressure waveform; determining a blood pressure value of the user according to the target blood pressure waveform; the blood pressure waveform estimation model comprises an embedding layer, an encoding layer, a decoding layer and a generation layer; the embedding layer increases the dimension of a PPG signal segment and a blood pressure waveform segment output by a previous time model; the encoding layer extracts features of the PPG signal segment; the decoding layer obtains high-dimensional feature distribution information of the blood pressure waveform segment corresponding to the PPG signal segment according to the output of the encoding layer; the generation layer reduces the dimension of the feature information and maps the feature information to a blood pressure waveform space to obtain a corresponding blood pressure waveform segment; and the blood pressure waveform segments are spliced to obtain the target blood pressure waveform. The method uses the blood pressure estimation model to obtain the target blood pressure waveform by collecting the PPG signal, reduces the pressure measurement error, and realizes continuous blood pressure monitoring.

Description

Technical Field

[0001] This invention relates to the field of medical monitoring technology, and in particular to a blood pressure monitoring method, device, equipment, and storage medium. Background Technology

[0002] In recent years, hypertension has become the most common chronic cardiovascular disease, and blood pressure monitoring is an important means of hypertension prevention and treatment. Currently, the most commonly used blood pressure measurement device is the cuff-type electronic blood pressure monitor.

[0003] Cuff-type electronic blood pressure monitors use the oscilloscope measurement method, which is easily affected by various factors during use, often leading to measurement errors. These factors include the physical condition of the person being measured, improper operation by the user, external interference, and equipment malfunction. Summary of the Invention

[0004] This invention provides a blood pressure monitoring method, device, equipment, and storage medium to address the shortcomings of existing blood pressure monitoring technologies that are prone to measurement errors. It can accurately estimate the user's blood pressure waveform in real time and reduce blood pressure measurement errors.

[0005] This invention provides a blood pressure monitoring method, comprising:

[0006] Collect the user's photoplethysmography (PPG) signal;

[0007] The PPG signal is processed using a blood pressure waveform estimation model to obtain the target blood pressure waveform; the PPG signal includes multiple PPG signal segments;

[0008] Determine the user's blood pressure value based on the target blood pressure waveform;

[0009] The blood pressure waveform estimation model includes an embedding layer, an encoding layer, a decoding layer, and a generation layer. For each PPG signal segment, the embedding layer increases the dimensionality of the PPG signal segment and the blood pressure waveform segment output by the blood pressure waveform estimation model at the previous moment. The encoding layer extracts features from the increased-dimensional PPG signal segment to obtain PPG features. The decoding layer matches the PPG features with the increased-dimensional blood pressure waveform segment to obtain the high-dimensional feature distribution information of the blood pressure waveform segment corresponding to the PPG signal segment. The generation layer reduces the dimensionality of the high-dimensional feature distribution information of the blood pressure waveform segment corresponding to the PPG signal segment and maps the reduced-dimensional feature information to the blood pressure waveform space to obtain the corresponding blood pressure waveform segment. The target blood pressure waveform is obtained based on the blood pressure waveform segment.

[0010] According to a blood pressure monitoring method provided by the present invention, the embedding layer is further used to superimpose position encoding information on the dimensionality-upgraded PPG signal segment and the blood pressure waveform segment, respectively. The dimensionality-upgraded PPG signal segment and the blood pressure waveform segment each include multiple sub-segments, each sub-segment being a 512-dimensional vector. The position encoding information of each vector element includes PE(pos,2i) and PE(pos,2i+1), wherein PE(pos,2i) and PE(pos,2i+1) are represented by the following formula:

[0011] PE(pos,2i)=sin(pos / 10000 2i / 512 );

[0012] PE(pos,2i+1)=cos(pos / 10000 2i / 512 );

[0013] Where pos represents the absolute position of a vector element, and i represents the dimension.

[0014] According to a blood pressure monitoring method provided by the present invention, the generation layer is further used to splice the blood pressure waveform segment with the blood pressure waveform segment output by the blood pressure waveform estimation model at the previous moment to obtain the target blood pressure waveform.

[0015] According to a blood pressure monitoring method provided by the present invention, the user's blood pressure is monitored based on a target blood pressure waveform, including:

[0016] The user's health status is determined based on their blood pressure readings.

[0017] According to a blood pressure monitoring method provided by the present invention, determining a user's blood pressure value based on a target blood pressure waveform includes:

[0018] Determine the first-order difference symbol corresponding to each sampling point in the target blood pressure waveform; the target blood pressure waveform includes n sampling points, where n is an integer greater than 1;

[0019] Based on the first-order difference sign corresponding to each sampling point, determine the difference preservation information corresponding to each sampling point;

[0020] Based on the differential hold information corresponding to each sampling point, determine the peak and trough information corresponding to the target blood pressure waveform;

[0021] The user's blood pressure value is determined based on the peak and trough information corresponding to the target blood pressure waveform.

[0022] According to a blood pressure monitoring method provided by the present invention, before processing the PPG signal using a blood pressure waveform estimation model to obtain the target blood pressure waveform, the method further includes:

[0023] The PPG signal is preprocessed, including at least one of the following: signal segmentation, signal filtering and denoising, abnormal data processing, and signal normalization.

[0024] According to a blood pressure monitoring method provided by the present invention, before upscaling the blood pressure waveform segment output by the blood pressure waveform estimation model at the previous moment, the method further includes:

[0025] Normalize the signal of the blood pressure waveform segment.

[0026] According to a blood pressure monitoring method provided by the present invention, the signal filtering and denoising is wavelet transform denoising, and after performing signal filtering and denoising on the PPG signal, the method further includes:

[0027] The DC component of the PPG signal before denoising is superimposed on the denoised PPG signal.

[0028] According to a blood pressure monitoring method provided by the present invention, after the PPG signal is segmented into multiple PPG signal segments, abnormal data processing includes at least one of the following:

[0029] If there is an upper cutoff and / or lower cutoff in the PPG signal segment, delete the PPG signal segment.

[0030] Peak detection is performed on PPG signal segments. If outliers are found in multiple detected peaks, the PPG signal segments are deleted.

[0031] The present invention also provides a blood pressure monitoring device, comprising:

[0032] The acquisition module is used to acquire the user's photoplethysmography (PPG) signal;

[0033] The processing module is used to process the PPG signal using a blood pressure waveform estimation model to obtain the target blood pressure waveform; the PPG signal includes multiple PPG signal segments;

[0034] The processing module is also used to determine the user's blood pressure value based on the target blood pressure waveform;

[0035] The blood pressure waveform estimation model includes an embedding layer, an encoding layer, a decoding layer, and a generation layer. For each PPG signal segment, the embedding layer increases the dimensionality of the PPG signal segment and the blood pressure waveform segment output by the blood pressure waveform estimation model at the previous moment. The encoding layer extracts features from the increased-dimensional PPG signal segment to obtain PPG features. The decoding layer matches the PPG features with the increased-dimensional blood pressure waveform segment to obtain the high-dimensional feature distribution information of the blood pressure waveform segment corresponding to the PPG signal segment. The generation layer reduces the dimensionality of the high-dimensional feature distribution information of the blood pressure waveform segment corresponding to the PPG signal segment and maps the reduced-dimensional feature information to the blood pressure waveform space to obtain the corresponding blood pressure waveform segment. The target blood pressure waveform is obtained based on the blood pressure waveform segment.

[0036] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the blood pressure monitoring method as described above.

[0037] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the blood pressure monitoring method as described above.

[0038] This invention provides a blood pressure monitoring method, device, equipment, and storage medium. It acquires a user's PPG signal using a photoplethysmography (PPG) sensor, and obtains a target blood pressure waveform using a trained blood pressure estimation model. Specifically, an embedding layer increases the dimensionality of the PPG signal segment and the blood pressure waveform segment output by the previous time-phase blood pressure waveform estimation model. An encoding layer extracts features from the increased-dimensional PPG signal segment to obtain PPG features. A decoding layer matches the PPG features with the increased-dimensional blood pressure waveform segment to obtain the high-dimensional feature distribution information of the blood pressure waveform segment corresponding to the PPG signal segment. Finally, a generation layer reduces the dimensionality of the high-dimensional feature distribution of the blood pressure waveform segment corresponding to the PPG signal segment and maps the reduced-dimensional feature information to the blood pressure waveform space to obtain the corresponding blood pressure waveform segment. This corresponding blood pressure waveform segment is then concatenated with the blood pressure waveform segment obtained at the previous time-phase to finally obtain the target blood pressure waveform. Since the target blood pressure waveform obtained through the above blood pressure estimation model is relatively accurate, calculating the user's blood pressure value from the target blood pressure waveform can reduce measurement errors and improve the accuracy of blood pressure measurement. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0040] Figure 1 This is a schematic flowchart of a blood pressure monitoring method provided by the present invention;

[0041] Figure 2 This is a network structure diagram of the blood pressure waveform estimation model provided by the present invention;

[0042] Figure 3 This is a schematic diagram of the blood pressure waveform estimation model training process provided by the present invention;

[0043] Figure 4 This is a flowchart of the data preprocessing method provided by the present invention;

[0044] Figure 5 This is a schematic diagram of the peak detection algorithm provided by the present invention;

[0045] Figure 6 This is a schematic diagram of the blood pressure monitoring device provided by the present invention;

[0046] Figure 7 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0047] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0048] First, some of the terms and application scenarios involved in the embodiments of the present invention will be introduced.

[0049] In the embodiments of the present invention, the cuff-type electronic blood pressure monitor has drawbacks in daily use, such as cumbersome testing procedures, inconvenient carrying of the measuring device, and unfriendly design for certain groups of people. Furthermore, in the field of continuous blood pressure monitoring, the inflation and deflation process of the cuff during use can cause pressure on the patient, and the need to inflate and deflate the cuff every half hour can affect the patient's daily life during the day and sleep at night.

[0050] In embodiments of the present invention, photoplethysmograph (PPG) signals are photoelectric signals that reflect changes in blood volume within blood vessels. The pulsating changes in blood volume can reflect many important cardiovascular information, such as cardiac function, blood flow, peripheral blood vessels, and microcirculation. When blood flows through microvessels such as arterioles, capillaries, and venules in peripheral blood vessels, the blood volume in these microvessels also exhibits pulsating changes under cardiac pulsation. This pulsation of blood volume can be recorded by a photoelectric sensor, i.e., photoplethysmographing. The basic principle is as follows: a probe consisting of a light source and a photodetector contacts the skin; the emitted light beam is reflected, absorbed, and scattered by the skin tissue and blood. The photodetector receives the light signal and converts it into an electrical signal, thereby reflecting the pulsation of the volumetric pulse wave. The signal obtained using this method is called a photoplethysmograph (PPG) signal. Because PPG sensors are small, inexpensive, and have a simple acquisition method, they are widely used in wearable devices.

[0051] The following is combined with Figures 1-7 The technical solutions of the embodiments of the present invention will be described in detail with reference to specific examples. The following specific examples can be combined with each other, and the same or similar concepts or processes may not be described again in some examples.

[0052] Figure 1 This is a schematic flowchart of a blood pressure monitoring method provided by the present invention. Figure 1 As shown, the method provided in this embodiment includes:

[0053] Step 101: Acquire the user's photoplethysmography (PPG) signal;

[0054] Specifically, for example, a PPG sensor is used to collect the user's PPG signal. Assuming the PPG signal is relatively long, preprocessing can be performed for easier subsequent processing. For instance, the PPG signal can be segmented into multiple segments, each containing 1280 sampling points and a length of 10.24 seconds. Optionally, the PPG signal can be filtered before segmentation. If the recording time of the PPG signal is less than a preset duration (e.g., 60 seconds), that segment is discarded; PPG signals with a recording time greater than or equal to the preset duration are then segmented.

[0055] Because PPG sensors are convenient and compact, they can reduce the wearing pressure on users and improve the user experience.

[0056] Step 102: Process the PPG signal using a blood pressure waveform estimation model to obtain the target blood pressure waveform; the PPG signal includes: multiple PPG signal segments;

[0057] Specifically, the blood pressure waveform estimation model includes an embedding layer, an encoding layer, a decoding layer, and a generation layer. For each PPG signal segment, the embedding layer is used to increase the dimensionality of the PPG signal segment and the blood pressure waveform segment output by the blood pressure waveform estimation model at the previous time step. The encoding layer is used to extract features from the dimensionality-increased PPG signal segment to obtain PPG features. The decoding layer is used to match the PPG features with the dimensionality-increased blood pressure waveform segment to obtain the high-dimensional feature distribution information of the blood pressure waveform segment corresponding to the PPG signal segment. The generation layer reduces the dimensionality of the high-dimensional feature distribution information of the blood pressure waveform segment corresponding to the PPG signal segment and maps the dimensionality-increased feature information to the blood pressure waveform space to obtain the corresponding blood pressure waveform segment. The corresponding blood pressure waveform segment is then concatenated with the blood pressure waveform segment obtained at the previous time step to finally obtain the target blood pressure waveform.

[0058] The PPG signal comprises multiple PPG signal segments. The PPG signal is the raw signal acquired by the PPG sensor, typically recorded for a long time, such as tens of minutes. The PPG signal segment is the basic unit of the PPG signal processed in the blood pressure estimation model. The raw PPG signal acquired by the sensor is segmented and processed. First, the sensor-acquired PPG signal is segmented, resulting in multiple PPG signal segments. For example, each signal segment contains 1280 sampling points and has a length of 10.24 seconds. Then, each PPG signal segment is input into the blood pressure waveform estimation model. To facilitate signal processing, the model further segments each PPG signal segment at the embedding layer, resulting in multiple PPG signal segments. Each PPG signal segment is the basic unit of the model's processing. For example, each PPG signal segment contains 64 sampling points, corresponding to a 64-dimensional vector, which can be scaled up to 256 or 512 dimensions.

[0059] Furthermore, the blood pressure waveform estimation model is trained based on PPG signal sample data and arterial blood pressure (ABP) waveform sample data. For a detailed flowchart of the blood pressure waveform estimation model training method, please refer to [link to flowchart]. Figure 2 For example, the sample data collected for training the blood pressure estimation model totaled 12,000 samples, namely the PPG signals and their synchronous arterial blood pressure waveforms (ABP signals) of 12,000 subjects. The sample data was extracted from the MIMIC (Medical Information Mart for Intensive Care) II public database, which is a free and open public resource intensive care unit research database with medical and health data and records of more than 40,000 patients.

[0060] Furthermore, multiple PPG signal segments are continuously input into the blood pressure waveform estimation model. For each PPG signal segment, the dimensionality of this signal segment and the blood pressure waveform segment output by the blood pressure waveform estimation model at the previous time step is increased through an embedding layer. Assuming that the signal segment is the first input segment, the blood pressure waveform segment output by the blood pressure waveform estimation model at the previous time step can be a preset blood pressure waveform segment, such as a blank segment (all zeros segment) with the same length as the signal segment. For example, if the signal segment contains 64 sampling points, it corresponds to a 64-dimensional vector, which can be increased to 256 or 512 dimensions. After dimensionality increase, the feature distribution of the signal segment becomes more concentrated, which facilitates subsequent feature extraction. The upscaled signal segment is input into the encoding layer, which extracts features from the PPG signal segment to obtain PPG features. Then, the decoding layer matches the PPG features with the upscaled blood pressure waveform segment to obtain the high-dimensional feature distribution information of the blood pressure waveform segment corresponding to the PPG signal segment. Finally, the generation layer reduces the dimensionality of the high-dimensional feature distribution information of the blood pressure waveform segments corresponding to multiple PPG signal segments and maps it to the blood pressure waveform space to obtain the corresponding blood pressure waveform segment. The corresponding blood pressure waveform segment is then concatenated with the blood pressure waveform segment obtained at the previous moment to finally obtain the target blood pressure waveform.

[0061] Step 103: Determine the user's blood pressure value based on the target blood pressure waveform.

[0062] Specifically, the method for calculating the user's blood pressure value based on the target blood pressure waveform is as follows: the peak and trough of the target blood pressure waveform are detected by a peak detection algorithm. The average value of the peak blood pressure is taken as the user's systolic blood pressure (SBP) value, and the average value of the trough blood pressure is taken as the user's diastolic blood pressure (DBP) value.

[0063] In the foregoing embodiment, the user's photoplethysmography (PPG) signal is collected by a PPG sensor, and the target blood pressure waveform is obtained using the trained blood pressure estimation model. Specifically, the PPG signal segment and the blood pressure waveform segment output by the blood pressure waveform estimation model at the previous moment are dimension-increased through an embedding layer, and an encoding layer performs feature extraction on the dimension-increased PPG signal segment to obtain PPG features. Then, a decoding layer performs matching according to the PPG features and the dimension-increased blood pressure waveform segment to obtain high-dimensional feature distribution information of the blood pressure waveform segment corresponding to the PPG signal segment. Finally, a generation layer reduces the dimension of the high-dimensional feature distribution information of the blood pressure waveform segment corresponding to the PPG signal segment, maps the dimension-reduced feature information to a blood pressure waveform space to obtain a corresponding blood pressure waveform segment, and splices the corresponding blood pressure waveform segment with the blood pressure waveform segment obtained at the previous moment to finally obtain the target blood pressure waveform. Since the target blood pressure waveform obtained by the above blood pressure estimation model is relatively accurate, calculating the user's blood pressure value from the target blood pressure waveform can reduce the pressure measurement error and improve the accuracy of blood pressure measurement.

[0064] The blood pressure monitoring method provided by the present invention further comprises:

[0065] Determining the user's health status according to the user's blood pressure value.

[0066] Specifically, the user's blood pressure values include a systolic blood pressure (SBP) value and a diastolic blood pressure (DBP) value. The user's health status can be determined according to the user's SBP value and DBP value, for example, the user's hypertension status can be judged. The judgment method is as follows:

[0067] When SBP<130 and DBP<85, it is normal blood pressure;

[0068] When 130<SBP<140 or 85<DBP<90, it is high-normal blood pressure;

[0069] When 140<SBP<160 or 90<DBP<100, it is grade 1 hypertension;

[0070] When 160<SBP or 100<DBP, it is grade 2 hypertension.

[0071] In the foregoing embodiment, the user's target blood pressure waveform is estimated by the blood pressure estimation model, the user's blood pressure value is calculated from the target blood pressure waveform, and then the user's hypertension status is judged. Since the estimated target blood pressure waveform is dynamic and continuous, dynamic and continuous monitoring of the user's blood pressure data can be achieved.

[0072] Figure 2 is a network structure diagram of the blood pressure waveform estimation model provided by the present invention.

[0073] as shown in Figure 2As shown, the blood pressure waveform estimation model network structure provided in this embodiment includes:

[0074] The model consists of four main modules: embedding layer, encoding layer, decoding layer, and generation layer. The embedding layer includes a fully connected layer, which upscales the PPG and ABP signals. The upscaled data can be embedded into a suitable high-dimensional feature space compared to the original data, making the feature distribution of the PPG signal sequence more concentrated, facilitating subsequent PPG feature extraction and selection, and better controlling overfitting to ensure model generalization performance. Furthermore, the embedding layer also overlays positional encoding information onto the upscaled PPG and ABP signals. The upscaled PPG and ABP signals each consist of multiple data segments, each a 512-dimensional vector. The positional encoding information for each vector element includes PE(pos,2i) and PE(pos,2i+1), which are represented by the following formula:

[0075] PE(pos,2i)=sin(pos / 10000 2i / 512 );

[0076] PE(pos,2i+1)=cos(pos / 10000 2i / 512 );

[0077] Where pos represents the absolute position of a vector element, and i represents the dimension.

[0078] Exemplarily, in the above embodiments,

[0079] For example, to add position encoding to the upscaled PPG signal segment, the specific steps are as follows:

[0080] For example, each PPG signal segment after dimensionality upgrade is a 21×512 matrix. Each input vector element is encoded as follows:

[0081] PE(pos,2i)=sin(pos / 10000 2i / 512 );

[0082] PE(pos,2i+1)=cos(pos / 10000 2i / 512 );

[0083] Where pos represents the absolute position of a vector element, with values ​​from 0 to 20; 2i represents an even position in the dimension with values ​​from 0 to 512; and 2i+1 represents an odd position in the dimension with values ​​from 0 to 512.

[0084] For example, we let

[0085] C i =1 / 10000 2i / 512 ,

[0086] Then the PE value at position t is:

[0087]

[0088] Furthermore, the PE value at each position is a 1×512 matrix, and the PE size at each position is fixed. By adding positional encoding information to the upscaled PPG signal segments, the model can obtain the relative positional information of each signal segment. This facilitates the model in distinguishing the temporal order of the input signal segments during the self-attention mechanism calculation, improving the accuracy of the model results and the efficiency of model processing.

[0089] The role of the coding layer is to extract and select features from multiple PPG signals. Through the self-attention mechanism, it captures the correlation between PPG signal features and focuses the computation on the features with the most information.

[0090] The function of the decoding layer is to match the PPG features with the ABP signal to obtain the high-dimensional feature distribution information of the blood pressure waveform segment corresponding to the PPG signal segment.

[0091] The role of the generation layer is to reduce the dimensionality of the high-dimensional feature distribution information of the blood pressure waveform segment corresponding to each PPG signal segment, and map the dimensionality-reduced feature information to the blood pressure waveform space to obtain the corresponding blood pressure waveform segment. Then, the corresponding blood pressure waveform segment is spliced ​​with the blood pressure waveform segment obtained at the previous moment to finally obtain the target blood pressure waveform.

[0092] Furthermore, the blood pressure waveform estimation model described above uses the Transformer model as the backbone network. It effectively acquires global information through the self-attention mechanism, solves the problem of long-distance dependence, and effectively avoids the problem of receptive field size when using one-dimensional convolution and the problem of information attenuation when multiple convolutional layers are stacked.

[0093] The blood pressure waveform estimation model in this invention is obtained through training. Figure 3 This is a schematic diagram of the blood pressure waveform estimation model training process provided by the present invention.

[0094] like Figure 3 As shown, the model training process method provided in this embodiment includes:

[0095] Step 301: Read the model training sample data and preprocess the sample data;

[0096] Specifically, the training sample data used for the blood pressure waveform estimation model is read, and the training sample data is preprocessed. The preprocessed training sample data consists of multiple data segments, for example, each data segment contains 1280 sampling points and has a length of 10.24s.

[0097] Furthermore, the training sample data used for the blood pressure waveform estimation model in this invention are the PPG signals of the subjects and their synchronous arterial blood pressure waveforms (ABP signals), for example, the PPG signals and their synchronous arterial blood pressure waveforms (ABP signals) of a total of 12,000 subjects.

[0098] Furthermore, the filtered data segments are randomly shuffled and divided into training and test sets in a 4:1 ratio. The sample data in the training set is used for model training, and the sample data in the test set is used to test the trained model and evaluate its performance.

[0099] Furthermore, the data preprocessing operations performed on the PPG signal sample data include signal segmentation, signal filtering and denoising, outlier handling, and signal normalization. The data preprocessing operations performed on the synchronized arterial blood pressure waveform (ABP) signal data include signal segmentation, outlier handling, and signal normalization. For detailed diagrams of the specific data preprocessing methods, please refer to [link to diagram]. Figure 4 .

[0100] Step 302: Input the preprocessed sample data into the blood pressure estimation model for model training.

[0101] Specifically, the preprocessed training sample data consists of multiple data segments, for example, each data segment contains 1280 sampling points and has a length of 10.24 seconds. A blank segment of a preset length is added to the beginning of each PPG signal sample data segment as a starting segment, for example, a blank segment of 64 sampling points is added as a starting segment. Subsequently, each PPG signal sample data segment is divided into preset data segments, for example, divided into 64 data segments with 21 sampling points. Each PPG signal segment is input into the embedding layer of the blood pressure estimation model. The embedding layer includes fully connected layers. Each PPG signal segment is updimensionalized through 21 fully connected layers with shared weights. Each updimensionalized PPG signal segment is a 512-dimensional vector.

[0102] The embedding layer is also used to superimpose position encoding information onto the upscaled PPG signal segments. Each data segment is a 512-dimensional vector, and the position encoding information for each vector element includes PE(pos,2i) and PE(pos,2i+1), which are represented by the following formula:

[0103] PE(pos,2i)=sin(pos / 100002i / 512 );

[0104] PE(pos,2i+1)=cos(pos / 10000 2i / 512 );

[0105] Where pos represents the absolute position of a vector element, and i represents the dimension.

[0106] PPG signal fragments that have been upgraded in dimension and superimposed with location-encoded information are passed through the encoding layer of a blood pressure estimation model (e.g., an encoder, such as a Transformer Encoder). Figure 2 As shown), the coding layer has an internal dimension of 512, the hidden layer has a dimension of 2048, the number of multi-head attention is 8, and it contains 6 decoders. It is stacked six times in total, and then the output memory of the coding layer is obtained. The output memory of the coding layer is a PPG signal segment with extracted features, which is a 21×512 matrix.

[0107] Each ABP sample data segment is up-dimensioned through 21 fully connected layers with shared weights. Each up-dimensioned data segment is also a 512-dimensional vector. Then, positional encoding information is superimposed on the up-dimensioned ABP signal segments. The positional encoding method is the same as that for PPG signals, which will not be elaborated here. Simultaneously, a 21-dimensional lower triangular matrix with all elements equal to 1 is generated for use in the decoding layer (e.g., a Transformer Decoder). Figure 2 The mask shown above is used to input the upscaled ABP segment and the PPG signal segment with extracted features from the output of the coding layer into the decoding layer. The decoding layer has an internal dimension of 512 and a hidden layer dimension of 2048. The number of multi-head attention is 8, and it is stacked six times in total. Based on the blood pressure waveform ABP segment input to the decoder, the decoding layer outputs the high-dimensional feature distribution information of the blood pressure waveform segment at the next input time. The output of the decoding layer is a matrix of the number of input segments × 512.

[0108] Specifically, in the above steps, the decoding layer adds masking information to the ABP sample data fragments, as follows:

[0109] "Mask" refers to masking, and "code" is the numerical value in our vector, generally containing only 1s and 0s. It is represented as a vector. In this embodiment, during model training, a lower triangular matrix with all elements being 1s and a dimension of 21 is generated. This matrix serves as the masking information to cover or replace information from future moments in the ABP segment, enabling the model training process to match the ABP segment with the PPG signal segment.

[0110] The output data of the decoding layer is input into the generation layer of the blood pressure estimation model. The generation layer reduces the dimensionality of the ABP sample data segments through 21 fully connected layers with shared weights. Each data segment after dimensionality reduction is a 64-dimensional vector. Furthermore, the generation layer concatenates the output results in sequence to obtain, for example, the target blood pressure waveform with 1280 sampling points, thus completing the forward propagation of the model.

[0111] Step 303: Adjust and optimize the model parameters until the model converges.

[0112] Specifically, for example, the loss function MAE is used to measure the predictive ability of the blood pressure estimation model, and the error between the target blood pressure waveform and the reference blood pressure waveform is calculated; the gradient is backpropagated through the Adaptive Moment Estimation (Adam) optimizer, for example, the learning rate adopts the Warmup mechanism, the learning rate gradually increases in the first 10 epochs, and then slowly decreases after 10 epochs, thereby updating the network parameters, and repeating step 302 until the network converges.

[0113] Furthermore, the loss function is used to estimate the degree of inconsistency between the model's predictions and the true values. In other words, it can be interpreted as the difference between the predicted values ​​obtained from our model and the true values. It is a non-negative real-valued function, usually represented by L(Y,f(x)). The smaller the loss function, the better the robustness of the model.

[0114] The loss function MAE (Mean Absolute Error) used in this invention is the loss function for regression models. MAE is the sum of the absolute differences between the target and predictor variables. The formula for Mean Absolute Error is as follows:

[0115]

[0116] in It is the predicted result, and y i The difference between the two values ​​is called the residual.

[0117] Furthermore, the significance of using a warmup learning rate mechanism lies in the fact that in the initial stage of model training, the model is unfamiliar with the sample data and needs to learn slowly with a small learning rate, continuously correcting the weight distribution. After the model has been trained for a certain period, the learned distribution has become largely fixed. Continuing to use a large learning rate at this point might disrupt the stability of the model's weight distribution. Therefore, in the process of optimizing the model's parameters, the learning rate gradually increases in the first 10 epochs and then slowly decreases after 10 epochs, thereby updating the network parameters. Using the Adam optimizer, gradient descent is fast.

[0118] Furthermore, once the model training is complete, the test set from the aforementioned sample data can be input into the model to test the performance of the blood pressure estimation model, continuously adjusting and optimizing the model parameters until the model converges.

[0119] Figure 4 This is a schematic diagram of the data preprocessing method provided by the present invention.

[0120] like Figure 4 As shown, the method provided in this embodiment includes:

[0121] Data preprocessing operations include: signal segmentation, signal filtering and denoising, outlier handling, and signal normalization. Among these:

[0122] Step 401: Signal segmentation. The signal segmentation operation is performed on the acquired PPG signal and ABP signal. Each signal segment contains, for example, 1280 sampling points and has a length of 10.24s.

[0123] Step 402: Signal filtering and denoising. Signal filtering and denoising is performed on the segmented PPG signal. The signal filtering and denoising operation is as follows: wavelet transform denoising is performed on the segmented PPG signal. The wavelet transform decomposition level is 10. Daubechies 8 (db8) is used as the mother wavelet. Since the sampling frequency of the PPG signal is 125Hz, the decomposition coefficients of the 0-0.24Hz and 15.625-62.5Hz frequency bands are set to 0 to filter out the baseline drift and high-frequency noise of the PPG signal, respectively.

[0124] Step 403: Abnormal data processing. Abnormal data processing operations are performed on PPG signal segmentation and ABP signal segmentation. The abnormal data processing operation is to judge the segments after filtering and delete abnormal data that does not meet the requirements after signal quality assessment.

[0125] Specifically, for PPG signal segments, if an upper cutoff and / or lower cutoff occurs in a PPG signal segment, the PPG signal segment is deleted; peak detection is performed on the PPG signal segment, and if outliers are found in multiple detected peaks, the PPG signal segment is deleted.

[0126] For ABP signal segmentation, if the systolic blood pressure (SBP) is less than 80 mmHg or greater than 180 mmHg, and the diastolic blood pressure (DBP) is less than 50 mmHg or greater than 100 mmHg, the ABP signal segment is deleted; if outliers are found in multiple detected peaks of the ABP signal segment, the ABP signal segment is deleted.

[0127] Step 404: Signal normalization. Signal normalization is performed on PPG signal segments and ABP signal segments. The specific operation of signal normalization is as follows: normalize the PPG signal segments and ABP signal segments to map the data to the interval [0, 1].

[0128] Specifically, normalization is a data standardization method that removes unit restrictions from data, transforming it into dimensionless pure numerical values. This facilitates comparison and weighting of indicators with different units or magnitudes, avoiding large prediction errors caused by significant differences in the magnitudes of input and output data. Since the dimensions of the PPG signal and the ABP signal differ significantly, standardization of the original indicator data is necessary to ensure the reliability of the results; this is known as signal normalization.

[0129] In this embodiment of the invention, the PPG signal and ABP signal are preprocessed before being input into the blood pressure waveform estimation model. This results in better feature extraction of the PPG signal, making it easier to match blood pressure waveform segments more accurately and thus making the blood pressure waveform estimation model more accurate in estimating the blood pressure waveform.

[0130] Figure 5 This is a schematic diagram of the peak detection algorithm.

[0131] like Figure 5 As shown, the method provided in this embodiment includes:

[0132] Determine the first-order difference symbol corresponding to each sampling point in the target blood pressure waveform; the target blood pressure waveform includes n sampling points, where n is an integer greater than 1;

[0133] Based on the first-order difference sign corresponding to each sampling point, determine the difference preservation information corresponding to each sampling point;

[0134] Based on the differential hold information corresponding to each sampling point, determine the peak and trough information corresponding to the target blood pressure waveform;

[0135] The user's blood pressure value is determined based on the peak and trough information corresponding to the target blood pressure waveform.

[0136] The method includes the following steps:

[0137] Step 501: Determine the first-order difference symbol corresponding to each sampling point in the target blood pressure waveform; the target blood pressure waveform includes n sampling points, where n is an integer greater than 1;

[0138] Specifically, the target blood pressure waveform includes n sampling points, where n is an integer greater than 1, and the target blood pressure waveform data is as follows:

[0139] X = [x1, x2, ..., x n ].

[0140] The first-order difference sign corresponding to each sampling point in the target blood pressure waveform is calculated using the following formula:

[0141] diff i =sign(x) i+1 -x i ),

[0142] The sign() function is used to retrieve the sign of an object.

[0143] Step 502: Determine the difference preservation information corresponding to each sampling point based on the first-order difference sign corresponding to each sampling point;

[0144] Specifically, the first-order difference symbol has three possibilities: -1, 0, and 1. If the current first-order difference symbol is 0, then the symbol at this time maintains the first-order difference symbol of the previous sampling point. The general formula for the first-order difference symbol preservation algorithm is:

[0145]

[0146] Step 503: Determine the peak and trough information corresponding to the target blood pressure waveform based on the differential hold information corresponding to each sampling point;

[0147] Specifically, the formula for calculating the first-order sign-preserving difference is as follows:

[0148] Diff2 i =Diff i+1 -Diff i .

[0149] Among them, Diff2 i =2 represents the location of the trough, Diff2 i =-2 indicates the location of the wave crest.

[0150] Furthermore, to determine the peak and trough information corresponding to the target blood pressure waveform, it is necessary to remove interference from secondary peaks within the cardiac cycle, such as dicrotic waves.

[0151] Step 504: Determine the user's blood pressure value based on the peak and trough information corresponding to the target blood pressure waveform.

[0152] Specifically, the average of peak blood pressure is used as the systolic blood pressure (SBP) value to be measured, and the average of trough blood pressure is used as the diastolic blood pressure (DBP) value to be measured.

[0153] The blood pressure monitoring device provided by the present invention is described below. The blood pressure monitoring device described below can be referred to in correspondence with the blood pressure monitoring method described above.

[0154] Figure 6 This is a schematic diagram of the blood pressure monitoring device provided by the present invention.

[0155] like Figure 6 As shown, the blood pressure monitoring device includes:

[0156] Acquisition module 610 is used to acquire the user's photoplethysmography (PPG) signal;

[0157] Processing module 620 is used to process the PPG signal using a blood pressure waveform estimation model to obtain a target blood pressure waveform; the PPG signal includes multiple PPG signal segments;

[0158] The processing module is also used to determine the user's blood pressure value based on the target blood pressure waveform;

[0159] The blood pressure waveform estimation model includes an embedding layer, an encoding layer, a decoding layer, and a generation layer. For each PPG signal segment, the embedding layer increases the dimensionality of the PPG signal segment and the blood pressure waveform segment output by the blood pressure waveform estimation model at the previous moment. The encoding layer extracts features from the increased-dimensional PPG signal segment to obtain PPG features. The decoding layer matches the PPG features with the increased-dimensional blood pressure waveform segment to obtain high-dimensional feature distribution information of the blood pressure waveform segment corresponding to the PPG signal segment. The generation layer reduces the dimensionality of the high-dimensional feature distribution information of the blood pressure waveform segment corresponding to the PPG signal segment and maps the reduced-dimensional feature information to the blood pressure waveform space to obtain the corresponding blood pressure waveform segment. The target blood pressure waveform is then obtained based on the blood pressure waveform segment.

[0160] Optionally, the embedding layer is further used to superimpose position encoding information onto the upscaled PPG signal segment and the blood pressure waveform segment, respectively. The upscaled PPG signal segment and the blood pressure waveform segment each include multiple sub-segments, each sub-segment being a 512-dimensional vector. The position encoding information for each vector element includes PE(pos,2i) and PE(pos,2i+1), where PE(pos,2i) and PE(pos,2i+1) are represented by the following formula:

[0161] PE(pos,2i)=sin(pos / 10000 2i / 512 );

[0162] PE(pos,2i+1)=cos(pos / 10000 2i / 512 );

[0163] Where pos represents the absolute position of a vector element, and i represents the dimension.

[0164] Optionally,

[0165] The generation layer is also used to splice the blood pressure waveform segment with the blood pressure waveform segment output by the blood pressure waveform estimation model at the previous moment to obtain the target blood pressure waveform.

[0166] Optionally, the processing module is specifically used for:

[0167] The user's health status is determined based on the user's blood pressure value.

[0168] Optionally, the processing module is specifically used for:

[0169] Determine the first-order difference symbol corresponding to each sampling point of the target blood pressure waveform; the target blood pressure waveform includes n sampling points, where n is an integer greater than 1;

[0170] Based on the first-order difference sign corresponding to each sampling point, determine the difference preservation information corresponding to each sampling point;

[0171] Based on the differential hold information corresponding to each sampling point, determine the peak and trough information corresponding to the target blood pressure waveform;

[0172] The user's blood pressure value is determined based on the peak and trough information corresponding to the target blood pressure waveform.

[0173] Optionally, the processing module is specifically used for:

[0174] Before processing the PPG signal using the blood pressure waveform estimation model to obtain the target blood pressure waveform, the method is further used for:

[0175] The PPG signal is preprocessed, and the preprocessing includes at least one of the following: signal segmentation, signal filtering and noise reduction, abnormal data processing, and signal normalization.

[0176] Optionally, the processing module is further configured to:

[0177] Before increasing the dimensionality of the blood pressure waveform segment output by the blood pressure waveform estimation model at the previous moment, the blood pressure waveform segment is normalized.

[0178] Optionally, the processing module is specifically used for:

[0179] The signal filtering and denoising is wavelet transform denoising. After the PPG signal is filtered and denoised, the DC component of the PPG signal before denoising is superimposed on the denoised PPG signal.

[0180] Optionally,

[0181] After the PPG signal is segmented into multiple PPG signal segments, the abnormal data processing includes at least one of the following:

[0182] If an upper cutoff and / or lower cutoff occurs in the PPG signal segment, the PPG signal segment is deleted.

[0183] Peak detection is performed on the PPG signal segments. If outliers are found in multiple detected peaks, the PPG signal segments are deleted.

[0184] Figure 7 This is a schematic diagram of the structure of the electronic device provided by the present invention.

[0185] Figure 7 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 7 As shown, the electronic device may include: a processor 710, a communication interface 720, a memory 730, and a communication bus 740, wherein the processor 710, the communication interface 720, and the memory 730 communicate with each other via the communication bus 740. The processor 710 can call logical instructions in the memory 730 to execute the blood pressure monitoring method, which includes:

[0186] Collect the user's photoplethysmography (PPG) signal;

[0187] The PPG signal is processed using a blood pressure waveform estimation model to obtain a target blood pressure waveform; the PPG signal includes multiple PPG signal segments.

[0188] The user's blood pressure value is determined based on the target blood pressure waveform;

[0189] The blood pressure waveform estimation model includes an embedding layer, an encoding layer, a decoding layer, and a generation layer. For each PPG signal segment, the embedding layer increases the dimensionality of the PPG signal segment and the blood pressure waveform segment output by the blood pressure waveform estimation model at the previous moment. The encoding layer extracts features from the increased-dimensional PPG signal segment to obtain PPG features. The decoding layer matches the PPG features with the increased-dimensional blood pressure waveform segment to obtain high-dimensional feature distribution information of the blood pressure waveform segment corresponding to the PPG signal segment. The generation layer reduces the dimensionality of the high-dimensional feature distribution information of the blood pressure waveform segment corresponding to the PPG signal segment and maps the reduced-dimensional feature information to the blood pressure waveform space to obtain the corresponding blood pressure waveform segment. The target blood pressure waveform is then obtained based on the blood pressure waveform segment.

[0190] Furthermore, the logical instructions in the memory 730 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0191] On the other hand, the present invention also provides a computer program product, the computer program product comprising a computer program that can be stored on a non-transitory computer-readable storage medium, wherein when the computer program is executed by a processor, the computer is able to execute the blood pressure monitoring method provided by the various methods, the method comprising:

[0192] Collect the user's photoplethysmography (PPG) signal;

[0193] The PPG signal is processed using a blood pressure waveform estimation model to obtain a target blood pressure waveform; the PPG signal includes multiple PPG signal segments.

[0194] The user's blood pressure value is determined based on the target blood pressure waveform;

[0195] The blood pressure waveform estimation model includes an embedding layer, an encoding layer, a decoding layer, and a generation layer. For each PPG signal segment, the embedding layer increases the dimensionality of the PPG signal segment and the blood pressure waveform segment output by the blood pressure waveform estimation model at the previous moment. The encoding layer extracts features from the increased-dimensional PPG signal segment to obtain PPG features. The decoding layer matches the PPG features with the increased-dimensional blood pressure waveform segment to obtain high-dimensional feature distribution information of the blood pressure waveform segment corresponding to the PPG signal segment. The generation layer reduces the dimensionality of the high-dimensional feature distribution information of the blood pressure waveform segment corresponding to the PPG signal segment and maps the reduced-dimensional feature information to the blood pressure waveform space to obtain the corresponding blood pressure waveform segment. The target blood pressure waveform is then obtained based on the blood pressure waveform segment.

[0196] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the blood pressure monitoring method provided by the methods, the method comprising:

[0197] Collect the user's photoplethysmography (PPG) signal;

[0198] The PPG signal is processed using a blood pressure waveform estimation model to obtain a target blood pressure waveform; the PPG signal includes multiple PPG signal segments.

[0199] The user's blood pressure value is determined based on the target blood pressure waveform;

[0200] The blood pressure waveform estimation model includes an embedding layer, an encoding layer, a decoding layer, and a generation layer. For each PPG signal segment, the embedding layer increases the dimensionality of the PPG signal segment and the blood pressure waveform segment output by the blood pressure waveform estimation model at the previous moment. The encoding layer extracts features from the increased-dimensional PPG signal segment to obtain PPG features. The decoding layer matches the PPG features with the increased-dimensional blood pressure waveform segment to obtain high-dimensional feature distribution information of the blood pressure waveform segment corresponding to the PPG signal segment. The generation layer reduces the dimensionality of the high-dimensional feature distribution information of the blood pressure waveform segment corresponding to the PPG signal segment and maps the reduced-dimensional feature information to the blood pressure waveform space to obtain the corresponding blood pressure waveform segment. The target blood pressure waveform is then obtained based on the blood pressure waveform segment.

[0201] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0202] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the technical solution, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments.

[0203] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for monitoring blood pressure, characterized in that, include: Collect the user's photoplethysmography (PPG) signal; The PPG signal is processed using a blood pressure waveform estimation model to obtain the target blood pressure waveform; The PPG signal includes: multiple PPG signal segments; The user's blood pressure value is determined based on the target blood pressure waveform. This determination includes: determining the first-order difference sign corresponding to each sampling point in the target blood pressure waveform; the target blood pressure waveform includes n sampling points, where n is an integer greater than 1; determining the difference preservation information corresponding to each sampling point based on the first-order difference sign; determining the peak and trough information corresponding to the target blood pressure waveform based on the difference preservation information; and determining the user's blood pressure value based on the peak and trough information corresponding to the target blood pressure waveform. The blood pressure waveform estimation model includes an embedding layer, an encoding layer, a decoding layer, and a generation layer. For each PPG signal segment, the embedding layer increases the dimensionality of the PPG signal segment and the blood pressure waveform segment output by the blood pressure waveform estimation model at the previous moment. The encoding layer extracts features from the increased-dimensional PPG signal segment to obtain PPG features. The decoding layer matches the PPG features with the increased-dimensional blood pressure waveform segment to obtain high-dimensional feature distribution information of the blood pressure waveform segment corresponding to the PPG signal segment. The generation layer reduces the dimensionality of the high-dimensional feature distribution information of the blood pressure waveform segment corresponding to the PPG signal segment and maps the reduced-dimensional feature information to the blood pressure waveform space to obtain the corresponding blood pressure waveform segment. Based on the blood pressure waveform segment, the target blood pressure waveform is obtained. The generation layer also concatenates the blood pressure waveform segment with the blood pressure waveform segment output by the blood pressure waveform estimation model at the previous moment to obtain the target blood pressure waveform.

2. The blood pressure monitoring method according to claim 1, characterized in that, The embedding layer is also used to superimpose position encoding information onto the upscaled PPG signal segment and the blood pressure waveform segment, respectively. The upscaled PPG signal segment and the blood pressure waveform segment each include multiple sub-segments, each sub-segment being a 512-dimensional vector. The position encoding information for each vector element includes... and in and This can be expressed by the following formula: ； ； Where pos represents the absolute position of the vector element. i Indicates dimension.

3. The blood pressure monitoring method according to claim 1 or 2, characterized in that, Before processing the PPG signal using a blood pressure waveform estimation model to obtain the target blood pressure waveform, the method further includes: The PPG signal is preprocessed, and the preprocessing includes at least one of the following: signal segmentation, signal filtering and noise reduction, abnormal data processing, and signal normalization.

4. The blood pressure monitoring method according to claim 1 or 2, characterized in that, Before increasing the dimensionality of the blood pressure waveform segment output by the blood pressure waveform estimation model from the previous moment, the following steps are also included: The blood pressure waveform segment is normalized.

5. The blood pressure monitoring method according to claim 3, characterized in that, The signal filtering and denoising is wavelet transform denoising, and after performing signal filtering and denoising on the PPG signal, it further includes: The DC component of the PPG signal before denoising is superimposed on the denoised PPG signal.

6. The blood pressure monitoring method according to claim 3, characterized in that, The PPG signal is segmented to obtain multiple PPG signal segments, and the abnormal data processing includes at least one of the following: If an upper cutoff and / or lower cutoff occurs in the PPG signal segment, the PPG signal segment is deleted. Peak detection is performed on the PPG signal segments. If outliers are found in multiple detected peaks, the PPG signal segments are deleted.

7. A blood pressure monitoring device, characterized in that, include: The acquisition module is used to acquire the user's photoplethysmography (PPG) signal; The processing module is used to process the PPG signal using a blood pressure waveform estimation model to obtain a target blood pressure waveform; the PPG signal includes multiple PPG signal segments; The processing module is further configured to determine the user's blood pressure value based on the target blood pressure waveform; the processing module is further configured to determine the first-order difference sign corresponding to each sampling point in the target blood pressure waveform; the target blood pressure waveform includes n sampling points, where n is an integer greater than 1; determine the difference preservation information corresponding to each sampling point based on the first-order difference sign corresponding to each sampling point; determine the peak and trough information corresponding to the target blood pressure waveform based on the difference preservation information corresponding to each sampling point; and determine the user's blood pressure value based on the peak and trough information corresponding to the target blood pressure waveform. The blood pressure waveform estimation model includes an embedding layer, an encoding layer, a decoding layer, and a generation layer. For each PPG signal segment, the embedding layer increases the dimensionality of the PPG signal segment and the blood pressure waveform segment output by the blood pressure waveform estimation model at the previous moment. The encoding layer extracts features from the increased-dimensional PPG signal segment to obtain PPG features. The decoding layer matches the PPG features with the increased-dimensional blood pressure waveform segment to obtain high-dimensional feature distribution information of the blood pressure waveform segment corresponding to the PPG signal segment. The generation layer reduces the dimensionality of the high-dimensional feature distribution information of the blood pressure waveform segment corresponding to the PPG signal segment and maps the reduced-dimensional feature information to the blood pressure waveform space to obtain the corresponding blood pressure waveform segment. Based on the blood pressure waveform segment, the target blood pressure waveform is obtained. The generation layer also concatenates the blood pressure waveform segment with the blood pressure waveform segment output by the blood pressure waveform estimation model at the previous moment to obtain the target blood pressure waveform.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the blood pressure monitoring method as described in any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the blood pressure monitoring method as described in any one of claims 1 to 6.

Images