Blood pressure prediction method and device, computer equipment, storage medium and program product

Abstract

The invention relates to a blood pressure prediction method and device, computer equipment, a storage medium and a program product. The method comprises the following steps: acquiring a pulse wave signal acquired by a monitoring watch; determining target features suitable for a plurality of trained blood pressure prediction models; based on the pulse wave signal, extracting a feature value of each target feature; through the plurality of blood pressure prediction models, determining blood pressure values predicted by the plurality of blood pressure prediction models according to the feature values of the target features suitable for the plurality of blood pressure prediction models; and determining the blood pressure value predicted by the monitoring watch based on the blood pressure values predicted by the plurality of blood pressure prediction models and the weights corresponding to the blood pressure values. The method is convenient for continuously monitoring blood pressure.

Description

Technical Field

[0001] The present application relates to the technical field of blood pressure monitoring, and in particular to a blood pressure prediction method, apparatus, computer equipment, storage medium and program product. Background Art

[0002] Hypertension is one of the common chronic diseases that affects human health. Hypertension is not only the main risk factor for cardiovascular disease, but is also closely related to a variety of serious diseases such as stroke and renal failure. However, patients with hypertension often do not show obvious clinical manifestations until target organ damage occurs, which attracts attention. Therefore, early blood pressure monitoring is of great significance. In traditional technology, blood pressure can be measured more accurately through invasive blood pressure measurement methods, specifically invasive arterial pressure measurement through vascular cannula. This method places a cannula needle in the artery to directly measure arterial pressure, but this method will cause wounds and cause pain to the subject. Improper operation may even cause infection. Therefore, non-invasive blood pressure measurement methods have emerged. The common non-invasive blood pressure measurement method is to measure through a cuff. The subject wears a cuff and obtains blood pressure information by pressurizing the cuff and combining it with a mercury sphygmomanometer or upper arm electronic blood pressure.

[0003] However, the above-mentioned traditional non-invasive blood glucose measurement method is not convenient for continuous monitoring of blood pressure. Summary of the invention

[0004] Based on this, it is necessary to provide a blood pressure prediction method, device, computer equipment, storage medium and program product that are convenient for continuous monitoring of blood pressure in response to the above technical problems.

[0005] In a first aspect, the present application provides a blood pressure prediction method, comprising:

[0006] Obtain the pulse wave signal collected by the monitoring watch;

[0007] Determine the applicable target features for each of the multiple trained blood pressure prediction models;

[0008] Based on the pulse wave signal, extracting a feature value of each target feature;

[0009] Determining the blood pressure values ​​predicted by each of the multiple blood pressure prediction models according to the characteristic values ​​of the respective applicable target characteristics by using the multiple blood pressure prediction models;

[0010] Based on the blood pressure values ​​predicted by the multiple blood pressure prediction models and their corresponding weights, the blood pressure value predicted by the monitoring watch is determined.

[0011] In a second aspect, the present application also provides a blood pressure prediction device, comprising:

[0012] An acquisition module is used to acquire the pulse wave signal collected by the monitoring watch;

[0013] A feature processing module, used to determine target features applicable to each of the multiple trained blood pressure prediction models; and extract a feature value of each target feature based on the pulse wave signal;

[0014] The blood pressure prediction module is used to determine the blood pressure values ​​predicted by each of the multiple blood pressure prediction models according to the characteristic values ​​of the respective applicable target features through the multiple blood pressure prediction models; and determine the blood pressure values ​​predicted by the monitoring watch based on the blood pressure values ​​predicted by each of the multiple blood pressure prediction models and their respective corresponding weights.

[0015] In a third aspect, the present application further provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0016] Obtain the pulse wave signal collected by the monitoring watch;

[0017] Determine the applicable target features for each of the multiple trained blood pressure prediction models;

[0018] Based on the pulse wave signal, extracting a feature value of each target feature;

[0019] Determining the blood pressure values ​​predicted by each of the multiple blood pressure prediction models according to the characteristic values ​​of the respective applicable target characteristics by using the multiple blood pressure prediction models;

[0020] Based on the blood pressure values ​​predicted by the multiple blood pressure prediction models and their corresponding weights, the blood pressure value predicted by the monitoring watch is determined.

[0021] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the following steps are implemented:

[0022] Obtain the pulse wave signal collected by the monitoring watch;

[0023] Determine the applicable target features for each of the multiple trained blood pressure prediction models;

[0024] Based on the pulse wave signal, extracting a feature value of each target feature;

[0025] Determining the blood pressure values ​​predicted by each of the multiple blood pressure prediction models according to the characteristic values ​​of the respective applicable target characteristics by using the multiple blood pressure prediction models;

[0026] Based on the blood pressure values ​​predicted by the multiple blood pressure prediction models and their corresponding weights, the blood pressure value predicted by the monitoring watch is determined.

[0027] In a fifth aspect, the present application further provides a computer program product, including a computer program, which implements the following steps when executed by a processor:

[0028] Obtain the pulse wave signal collected by the monitoring watch;

[0029] Determine the applicable target features for each of the multiple trained blood pressure prediction models;

[0030] Based on the pulse wave signal, extracting a feature value of each target feature;

[0031] Determining the blood pressure values ​​predicted by each of the multiple blood pressure prediction models according to the characteristic values ​​of the respective applicable target characteristics by using the multiple blood pressure prediction models;

[0032] Based on the blood pressure values ​​predicted by the multiple blood pressure prediction models and their corresponding weights, the blood pressure value predicted by the monitoring watch is determined.

[0033] The above-mentioned blood pressure prediction method, device, computer equipment, storage medium and program product collect pulse wave signals through a monitoring watch, extract the characteristic value of each target feature among the target features applicable to each of the multiple blood pressure prediction models based on the pulse wave signal, and each of the multiple blood pressure prediction models predicts blood pressure based on the characteristic value of the applicable target feature, and then determines the blood pressure value predicted by the monitoring watch based on the blood pressure values ​​predicted by each of the multiple blood pressure prediction models and their corresponding weights, comprehensively considers the blood pressure values ​​predicted by the different multiple blood pressure prediction models, and improves the accuracy of the blood pressure value predicted by the monitoring watch. Moreover, since the pulse wave signal is collected by the monitoring watch, it is highly portable and convenient for users to wear it for a long time. Therefore, it is convenient to continuously monitor blood pressure while ensuring a certain prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the drawings required for use in the embodiments of the present application or related technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0035] Figure 1 is a flow chart of a blood pressure prediction method in one embodiment;

[0036] Figure 2 is a flowchart of a training step in an embodiment;

[0037] Figure 3 A schematic diagram of a flow chart of a prediction step in an embodiment;

[0038] Figure 4 is a structural block diagram of a blood pressure prediction device in one embodiment;

[0039] Figure 5 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0040] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0041] In one embodiment, Figure 1 As shown, a blood pressure prediction method is provided. This embodiment uses the method applied to a computer device as an example. It can be understood that the computer device can be a monitoring watch, or a terminal or a server. The method can be applied to the monitoring watch independently, or can be implemented through the interaction between the terminal or the server and the monitoring watch. In this embodiment, the method includes the following steps:

[0042] Step 102, obtaining the pulse wave signal collected by the monitoring watch.

[0043] Among them, the monitoring watch is a smart watch with a blood pressure monitoring function. The monitoring watch can be worn on the human wrist and collects pulse wave signals based on the airbag structure. The pulse wave signal can reflect the change of the pressure value of the artery in the wrist over time, and the artery can be the radial artery. A pulse wave signal can include the change of the pressure value of the artery over time within a preset time period, and the preset time period is 30 seconds, one minute or other.

[0044] The monitoring watch may include a watch body, a watch strap and an airbag structure. The airbag structure may be arranged in the watch strap, or in a wristband that can be stacked with the watch strap, and the wristband can fit the skin when the monitoring watch is worn. The airbag structure may include a double-layer micro-airbag and a sensor. When the monitoring watch is worn, one of the double-layer micro-airbags may be close to the skin side and may be close to the skin after being inflated. A sensor for measuring pressure value may be arranged in the micro-airbag, while the other layer of micro-airbags may be away from the skin side and may improve the fit between the micro-airbag close to the skin side and the skin after being inflated.

[0045] The monitoring watch can also be provided with a posture detection module, which can be used to detect the posture of the subject wearing the monitoring watch. The posture detection module can be implemented using an inertial measurement unit. For example, the monitoring watch can detect whether the subject bends his arm. In order to improve the accuracy of the acquisition, when collecting pulse wave signals through the monitoring watch, the subject can adjust the position and tightness of the monitoring watch, and can adjust the position of the strap of the monitoring watch to cover the ulnar joint of the wrist, so as to be able to collect the pressure changes of the radial artery, and breathe steadily during the acquisition process.

[0046] Exemplarily, the computer device may obtain the pulse wave signal collected by the monitoring watch in response to a blood pressure monitoring trigger event. The blood pressure monitoring trigger event may be a manual trigger operation, for example, a trigger operation of a blood pressure monitoring trigger control in an interface of the computer device. The blood pressure monitoring trigger event may also be an automatic trigger event, for example, it may be triggered at a preset time interval, such as one day, two days or other.

[0047] Step 104, determining target features applicable to each of the multiple trained blood pressure prediction models.

[0048] The blood pressure prediction model may be a linear regression model, a support vector machine model, a random forest regression model or others. Each blood pressure prediction model may be applicable to a feature group, a feature group includes at least two target features, and different blood pressure prediction models are applicable to different feature groups.

[0049] The target feature is a feature among multiple preset features. Multiple preset features are features related to blood pressure values. Blood pressure values ​​may include systolic blood pressure (SBP) and diastolic blood pressure (DBP). Each blood pressure prediction model may include a systolic blood pressure prediction sub-model and a diastolic blood pressure prediction sub-model. The systolic blood pressure prediction sub-model and the diastolic blood pressure prediction sub-model apply the same target feature but have different model parameters.

[0050] For example, when the blood pressure prediction model is a linear regression model, in any one of the multiple blood pressure prediction models, the systolic pressure prediction sub-model of the blood pressure prediction model may be a first relationship between systolic pressure and an applicable target feature, and the diastolic pressure prediction sub-model of the blood pressure prediction model may be a second relationship between diastolic pressure and an applicable target feature. The first relationship and the second relationship have the same target feature, but the coefficients corresponding to the target feature are different.

[0051] Exemplarily, the computer device obtains target features applicable to each of the multiple blood pressure prediction models from pre-stored model information, wherein the model information may include a model identifier, a blood pressure prediction model, and a corresponding relationship between applicable target features.

[0052] In one embodiment, the target feature can be a feature among multiple preset features, including morphological features, time domain features and dynamic features; the morphological features include at least one of peak features, trough features, extreme value features and envelope features; the time domain features include heart rate; the dynamic features include at least one of change trend features and variability features.

[0053] The pulse wave signal may include multiple pulse waves, and a pulse wave is a pulse wave generated in a heartbeat cycle. The morphological feature is a feature related to the morphology of the pulse wave signal. The time domain feature is a feature related to time implicit in the pulse wave signal. The dynamic feature is used to describe the dynamic change information of different pulse waves in the pulse wave signal.

[0054] The peak feature may include a peak and a peak position. The peak is the peak value of each pulse wave in the pulse wave signal. The peak position is the time point at which each pulse wave in the pulse wave signal produces a peak value. The trough feature may include a trough and a trough position. The trough is the valley value of each pulse wave in the pulse wave signal. The trough position is the time point at which each pulse wave in the pulse wave signal produces a valley value.

[0055] The extreme value features may include extreme values, extreme value positions, extreme value differences, and others. The extreme values ​​may include the maximum and minimum values ​​in each pulse wave in the pulse wave signal. The extreme value position is the time point at which the extreme value is generated in each pulse wave in the pulse wave signal. The extreme value difference may be the difference between extreme values ​​at different positions in the same pulse wave, or the difference between extreme values ​​at different positions in two adjacent pulse waves.

[0056] The envelope feature is a feature related to the envelope formed by the pulse wave signal. The envelope feature may include envelope height, envelope mean, or others. The envelope height may be the difference between the maximum value and the minimum value in the envelope formed by the pulse wave signal. The envelope mean may be the average value of each value in the envelope formed by the pulse wave signal.

[0057] The heart rate can be calculated based on the peak positions of multiple pulse waves of the pulse wave signal. Specifically, the peak positions are clustered to identify different heartbeat cycles, and the time intervals between adjacent heartbeat cycles are determined based on the clustering results; the heart rate is calculated using the formula HR = 60 / T, where HR represents the heart rate and T represents the average time interval of the heartbeat cycle.

[0058] The change trend feature can be represented by the change trend of the peak features of the multiple primary pulse waves of the pulse wave signal. Specifically, the peak and the peak position contained in each peak feature can be used as a peak coordinate, and a linear fitting can be performed based on the peak coordinates corresponding to the multiple primary pulse waves. For example, a linear relationship is obtained by fitting, and the slope of the relationship is used as the change trend feature. The variability feature can include the standard deviation of the peaks of the multiple primary pulse waves, the coefficient of variation of the peaks of the multiple primary pulse waves, or other characteristics.

[0059] In one embodiment, the plurality of preset features may further include wrist features, and the wrist features may include wrist size, wrist bone features, artery location features or others. Among them, the wrist size may include wrist width and wrist circumference. The wrist bone features may include wrist bone protrusion dimensions, such as the protrusion distance of the ulnar styloid process and the radial styloid process compared to the straight line formed by the forearm. The artery location feature may be the location of the radial artery, which is correlated with the radial styloid process, or the location of the radial artery may be indirectly indicated by the position of the radial styloid process. The above-mentioned wrist features may be features of the wrist of the subject to which the pulse wave signal belongs, which may be obtained by manually collecting an image of the wrist of the subject to which the pulse wave signal belongs, and may also be obtained by collecting an image of the wrist of the subject to which the pulse wave signal belongs and analyzing the image.

[0060] Step 106: extracting the characteristic value of each target feature based on the pulse wave signal.

[0061] Among them, the characteristic value is the value of the target feature in the pulse wave signal.

[0062] Exemplarily, the computer device deduplicates target features applicable to each of the multiple blood pressure prediction models to obtain a feature set formed by the deduplicated target features, and extracts a feature value of each target feature in the feature set based on the pulse wave signal.

[0063] In one embodiment, the computer device may extract the characteristic value of each target feature based on the pulse wave signal according to the preconfigured algorithm corresponding to each target feature. The preconfigured algorithm is an algorithm preconfigured based on the definition of the target feature. For example, the extreme value can be obtained by taking the derivative of the pulse wave signal.

[0064] Step 108, using multiple blood pressure prediction models, according to the characteristic values ​​of the respective applicable target features, the blood pressure values ​​predicted by the multiple blood pressure prediction models are determined.

[0065] Exemplarily, the computer device may input the characteristic value of the target characteristic applicable to each blood pressure prediction model among multiple blood pressure prediction models into the blood pressure prediction model to obtain the predicted blood pressure value output by the blood pressure prediction model.

[0066] In one embodiment, the blood pressure value may include systolic pressure and diastolic pressure, and each blood pressure prediction model may include a systolic pressure prediction sub-model and a diastolic pressure prediction sub-model. The computer device may, for each of the multiple blood pressure prediction models, input the characteristic value of the target feature applicable to the blood pressure prediction model into the systolic pressure prediction sub-model and the diastolic pressure prediction sub-model included in the blood pressure prediction model to respectively obtain the systolic pressure output by the systolic pressure prediction sub-model and the diastolic pressure output by the diastolic pressure prediction sub-model.

[0067] Step 110, based on the blood pressure values ​​predicted by the multiple blood pressure prediction models and their corresponding weights, determine the blood pressure value predicted by the monitoring watch.

[0068] The weights corresponding to the multiple blood pressure prediction models can represent the influence of the blood pressure values ​​predicted by the respective models on the blood pressure values ​​predicted by the monitoring watch. The weights corresponding to the multiple blood pressure prediction models can be set according to experience or obtained through training steps.

[0069] Exemplarily, the computer device may perform a weighted average calculation based on the blood pressure values ​​predicted by each of the multiple blood pressure prediction models and the weights corresponding to each of the multiple blood pressure prediction models to obtain the blood pressure value predicted by the monitoring watch.

[0070] In the above-mentioned blood pressure prediction method, the pulse wave signal is collected by a monitoring watch, and the characteristic value of each target feature among the target features applicable to each of the multiple blood pressure prediction models is extracted based on the pulse wave signal. The multiple blood pressure prediction models each predict blood pressure based on the characteristic value of the applicable target feature, and then the blood pressure value predicted by the monitoring watch is determined based on the blood pressure values ​​predicted by the multiple blood pressure prediction models and their corresponding weights. The blood pressure values ​​predicted by the different multiple blood pressure prediction models are comprehensively considered, thereby improving the accuracy of the blood pressure value predicted by the monitoring watch. Moreover, since the pulse wave signal is collected by the monitoring watch, it is highly portable and convenient for users to wear it for a long time. Therefore, it is convenient to continuously monitor blood pressure while ensuring a certain prediction accuracy.

[0071] In an exemplary embodiment, the target feature is a feature among multiple preset features, and the training steps of training multiple blood pressure prediction models include: obtaining multiple sample pulse wave signals collected by the monitoring watch, and the actual blood pressure value corresponding to each sample pulse wave signal; the actual blood pressure value is measured by a sphygmomanometer; based on each sample pulse wave signal, extracting the characteristic values ​​of each of the multiple preset features, and obtaining the characteristic values ​​of each sample pulse signal corresponding to the multiple preset features; performing different combinations of the multiple preset features to obtain different candidate feature combinations; each candidate feature combination includes at least two preset features; based on the characteristic values ​​of each candidate feature combination in the different candidate feature combinations corresponding to the multiple sample pulse wave signals, and the actual blood pressure values ​​corresponding to the multiple sample pulse wave signals, model training is performed to obtain multiple blood pressure prediction models and the weights corresponding to each blood pressure prediction model, the multiple blood pressure prediction models correspond to at least a part of the different candidate feature combinations, and the preset features in the candidate feature combination corresponding to each blood pressure prediction model are used as applicable target features.

[0072] The sample pulse wave signal is a pulse wave signal collected as a sample from a subject during the training phase. The actual blood pressure value corresponding to the sample pulse wave signal may be a blood pressure value measured by a sphygmomanometer on the same subject in a time period adjacent to the time period of collecting the sample pulse wave signal. The sphygmomanometer may be a mercury sphygmomanometer. It is understood that a more accurate blood pressure value may be obtained by a mercury sphygmomanometer.

[0073] The plurality of preset features may include features of multiple preset types. When different candidate feature combinations are obtained by combination, the features of each candidate feature combination may be combined according to the feature containing at least two preset types to increase the feature complexity. For example, the multiple preset types may be morphological features, time domain features, and dynamic features, or may be morphological features, time domain features, dynamic features, and wrist features.

[0074] The model type of the blood pressure prediction model can be a linear regression model, a support vector machine model, a random forest regression model or others. When training the model, for each candidate feature combination, the feature values ​​of the multiple sample pulse wave signals corresponding to the preset features in the candidate feature combination can be used as input data, and the actual blood pressure values ​​corresponding to the multiple sample pulse wave signals can be used as labels of the input data. The model type of the blood pressure prediction model is trained in a training method applicable to the training method to obtain the candidate model corresponding to the candidate feature combination; multiple blood pressure prediction models are selected from the candidate models corresponding to different candidate feature combinations, and the weight corresponding to each blood pressure prediction model is determined. It can be understood that the target feature applicable to the blood pressure prediction model refers to the preset feature contained in the corresponding candidate feature combination.

[0075] Multiple blood pressure prediction models can be obtained by screening based on the evaluation scores of candidate models corresponding to different candidate feature combinations. For example, multiple blood pressure prediction models can be candidate models whose evaluation scores reach preset scores among candidate models corresponding to different candidate feature combinations, or candidate models whose evaluation scores rank high among candidate models corresponding to different candidate feature combinations. The evaluation score can be calculated based on the root mean square error (RMSE) and the coefficient of determination (R²). For example, the root mean square error and the coefficient of determination can be mapped to scores within a preset score range, so that the smaller the root mean square error, the larger the score obtained by mapping, and the larger the coefficient of determination, the larger the score obtained by mapping; and then the scores mapped by the root mean square error and the coefficient of determination are weighted averaged to obtain the evaluation score. The preset score range can be, for example, 0-100, the value of the evaluation score can also be 0-100, the preset score can be 85, 90 or others, and the preset ratio can be, for example, 20%, 10% or others. The weight corresponding to the blood pressure prediction model can be assigned according to the ranking position of the evaluation scores of the multiple blood pressure prediction models screened out, so that the higher the ranking position, the greater the weight.

[0076] In this embodiment, model training is performed based on the feature values ​​of each candidate feature combination in different candidate feature combinations corresponding to multiple sample pulse wave signals, and the actual blood pressure values ​​corresponding to each of the multiple sample pulse wave signals. Since the actual blood pressure values ​​are measured by a sphygmomanometer, this can provide a basis for obtaining an accurate blood pressure prediction model. Moreover, different candidate feature combinations are obtained by making different combinations of multiple preset features, thereby improving the diversity of multiple blood pressure prediction models obtained by subsequent training, creating conditions for subsequent monitoring watches to accurately predict blood pressure values.

[0077] In an exemplary embodiment, the step of performing model training based on the feature values ​​of each candidate feature combination in different candidate feature combinations corresponding to each of the multiple sample pulse wave signals and the actual blood pressure values ​​corresponding to each of the multiple sample pulse wave signals to obtain multiple blood pressure prediction models and the weight corresponding to each blood pressure prediction model may include:

[0078] For each candidate feature combination in different candidate feature combinations, linear regression model training is performed based on feature values ​​of preset features in the candidate feature combination corresponding to multiple sample pulse wave signals, and actual blood pressure values ​​corresponding to each of the multiple sample pulse wave signals, to obtain a candidate model corresponding to the candidate feature combination; multiple blood pressure prediction models are screened out from the candidate models corresponding to different candidate feature combinations, and the preset features in the candidate feature combination corresponding to each blood pressure prediction model are used as applicable target features; linear regression model training is performed based on sample blood pressure values ​​predicted by each blood pressure prediction model in the multiple blood pressure prediction models based on multiple sample pulse wave signals, and actual blood pressure values ​​corresponding to each of the multiple sample pulse wave signals, to obtain weights corresponding to each of the multiple blood pressure prediction models.

[0079] Among them, the blood pressure prediction model can be a linear regression model, specifically a multivariate linear regression model with applicable target features as independent variables and blood pressure as dependent variables. When performing linear regression model training to obtain a candidate model, the characteristic values ​​of the preset features in the candidate feature combination corresponding to each of the multiple sample pulse wave signals can be used as input data (for substitution into the independent variable), and the actual blood pressure values ​​corresponding to each of the multiple sample pulse wave signals can be used as labels (for substitution into the dependent variable). The initial candidate model corresponding to the candidate feature combination is iteratively trained, and during the iterative training process, the trained candidate model is evaluated using the root mean square error and the determination coefficient. When the root mean square error is less than the first preset value and the determination coefficient is greater than the second preset value, the iteration is stopped, and the candidate model at the time of stopping the iteration is output. The first preset value is 5, 10 or others. The second preset value is 0.8, 0.9 or others.

[0080] When multiple blood pressure prediction models are screened out from the candidate models corresponding to different candidate feature combinations, the evaluation scores of the respective candidate models can be calculated, and then multiple blood pressure prediction models can be screened out based on the evaluation scores of the candidate models corresponding to the different candidate feature combinations.

[0081] Each of the multiple blood pressure prediction models can predict multiple sample blood pressure values ​​corresponding to multiple sample pulse wave signals. It can be understood that the sample feature value of each target feature can be extracted based on each sample pulse wave signal to obtain the sample feature value corresponding to each sample pulse wave signal, and the sample feature value of the target feature applicable to each blood pressure prediction model among the sample feature values ​​corresponding to each sample pulse wave signal is input into the blood pressure prediction model, so that the sample blood pressure value corresponding to the sample pulse wave signal of the blood pressure prediction model can be predicted.

[0082] When performing linear regression model training to obtain the weights corresponding to multiple blood pressure prediction models, a multivariate linear regression model can be constructed with the blood pressure predicted by the multiple blood pressure prediction models as the independent variable and the blood pressure predicted by the monitoring watch as the dependent variable. The multiple sample blood pressure values ​​predicted by the multiple blood pressure prediction models and corresponding to the multiple sample pulse wave signals are used as input data (for substituting into the independent variable), and the actual blood pressure values ​​corresponding to the multiple sample pulse wave signals are used as labels (for substituting into the dependent variable). The multivariate linear regression model is trained. After the training is completed, the coefficients corresponding to the blood pressures predicted by the multiple blood pressure prediction models in the obtained multivariate linear regression model are used as the weights corresponding to the multiple blood pressure prediction models.

[0083] In this embodiment, candidate models corresponding to different candidate feature combinations are obtained through linear regression model training, and then multiple blood pressure prediction models are screened out. Then, linear regression model training is performed based on multiple sample blood pressure values ​​predicted by the multiple blood pressure prediction models and corresponding to multiple sample pulse wave signals, as well as actual blood pressure values ​​corresponding to the multiple sample pulse wave signals. In this way, the weights corresponding to the multiple blood pressure prediction models can be obtained more accurately, creating conditions for the subsequent accurate prediction of blood pressure values.

[0084] In an exemplary embodiment, step 106 may include: filtering the pulse wave signal to obtain a filtered pulse wave signal; performing outlier data processing on the filtered pulse wave signal, and then performing local abnormal data processing based on a sliding window to obtain a preprocessed pulse wave signal; extracting the characteristic value of each target feature from the preprocessed pulse wave signal.

[0085] Among them, outlier data processing can include outlier data removal processing and interpolation processing performed in sequence. Since outlier data may be generated during the pulse wave signal collection process due to wrist shaking, irregular heart rate, etc., outlier data processing can improve the accuracy of the collected pulse wave signal. Outlier data removal processing can be implemented using the Z-Score method (Z score method), the interquartile range method or other algorithms.

[0086] When processing local abnormal data based on a sliding window, a fixed-length window can be slid on the pulse wave signal to detect and process the local abnormal data within the window. The fixed length can be 1 heartbeat cycle, 2 heartbeat cycles or other. The local abnormal data can be outlier data within the window.

[0087] In this embodiment, the pulse wave signal is filtered, outlier data is processed, and local abnormal data is processed based on a sliding window, which can improve the signal quality of the pulse wave signal. Based on the preprocessed pulse wave signal, the characteristic value of each target feature can be accurately extracted, providing a basis for subsequent accurate prediction of blood pressure values.

[0088] In an exemplary embodiment, the step of filtering the pulse wave signal to obtain a filtered pulse wave signal may include: bandpass filtering the pulse wave signal to obtain an intermediate pulse wave signal; lowpass filtering the pulse wave signal to obtain a reference pressure change signal and a static pressure; and obtaining a filtered pulse wave signal based on the intermediate pulse wave signal, the reference pressure change signal and the static pressure.

[0089] The reference pressure change signal can reflect the average pressure change of the airbag in the monitoring watch on the blood vessel wall. The wrist shaking, breathing, and muscle exertion of the subject during the acquisition process may cause the average pressure of the airbag on the blood vessel wall to change. The reference pressure change signal can be used to remove the baseline drift in the intermediate pulse wave signal.

[0090] The static pressure signal refers to the pressure value generated by the balloon on the blood vessel wall after the initial inflation is completed. Initial inflation means that the balloon is inflated to a fixed pressure so that it can stick to the skin. The static pressure signal is almost unchanged during the acquisition process and can be regarded as a DC component. The static pressure signal can be used to evaluate the availability of the intermediate pulse wave signal; specifically, the static pressure signal can be used to evaluate the status of the balloon during the measurement process (for example, whether there is a leak). If the static pressure signal remains unchanged, it can be considered that the status of the balloon during the measurement is normal and the intermediate pulse wave signal is available, so that the baseline drift of the intermediate pulse wave signal can be removed based on the reference pressure change signal to obtain a filtered pulse wave signal.

[0091] Bandpass filtering is used to retain signals in a specific frequency band and filter out signals outside the specific frequency band. The specific frequency band is the effective frequency range of the pulse wave signal, for example, it can be 0.5Hz (Hertz) to 4Hz. Low-pass filtering is used to retain signals that are not higher than a preset frequency and filter out signals that are higher than a preset frequency. The preset frequency can be less than the above-mentioned specific frequency band. By performing a low-pass filter at a first preset frequency on the pulse wave signal, a reference pressure change signal can be obtained, and by performing a low-pass filter at a second preset frequency on the pulse wave signal, a static pressure signal can be obtained. The first preset frequency can be greater than the second preset frequency, for example, the first preset frequency can be 0.3Hz, and the second preset frequency can be 0.05Hz.

[0092] In this embodiment, the pulse wave signal is bandpass filtered to extract the effective part of the pulse wave signal to obtain an intermediate pulse wave signal, and the pulse wave signal is low-pass filtered to extract the reference pressure change signal and the static pressure signal. The reference pressure change signal and the static pressure signal can represent the low-frequency information, and the intermediate pulse wave signal is further processed based on the reference pressure change signal and the static pressure signal to improve the signal quality of the filtered pulse wave signal.

[0093] In a specific embodiment, the blood pressure prediction method may include a training step and a prediction step, see Figure 2 The training steps flow chart shown in Figure 3 As shown in the flowchart of prediction steps, the above-mentioned blood pressure prediction method may specifically include the following steps.

[0094] In the training step, the computer device can obtain multiple sample pulse wave signals collected by the monitoring watch, as well as the actual blood pressure value corresponding to each sample pulse wave signal.

[0095] The computer device can preprocess each sample pulse wave signal, specifically, can perform filtering processing, outlier data processing, and local abnormal data processing based on a sliding window on each sample pulse wave signal in turn to obtain a preprocessed sample pulse wave signal.

[0096] The computer equipment can perform feature engineering processing, specifically, based on each preprocessed sample pulse wave signal, extract the characteristic values ​​of multiple preset features, and obtain the characteristic values ​​of each sample pulse signal corresponding to the multiple preset features; and perform different combinations of multiple preset features to obtain different candidate feature combinations.

[0097] The computer device can perform linear regression learning. Specifically, for each candidate feature combination in different candidate feature combinations, linear regression model training can be performed based on the feature values ​​of multiple sample pulse wave signals corresponding to preset features in the candidate feature combination, and the actual blood pressure values ​​corresponding to the multiple sample pulse wave signals, to obtain a candidate model corresponding to the candidate feature combination.

[0098] The computer device can screen out multiple blood pressure prediction models from the candidate models corresponding to different candidate feature combinations, and perform linear regression model training based on the multiple sample blood pressure values ​​predicted by the multiple blood pressure prediction models and corresponding to the multiple sample pulse wave signals, as well as the actual blood pressure values ​​corresponding to the multiple sample pulse wave signals, to obtain the weights corresponding to the multiple blood pressure prediction models.

[0099] In the prediction step, the computer device can obtain the pulse wave signal collected by the monitoring watch, determine the target features applicable to each of the multiple trained blood pressure prediction models, and preprocess the pulse wave signal. Specifically, the pulse wave signal can be filtered, outlier data processed, and local abnormal data processed based on a sliding window in turn to obtain a preprocessed pulse wave signal. Based on the preprocessed pulse wave signal, the characteristic value of each target feature is extracted, and the blood pressure values ​​predicted by the multiple blood pressure prediction models are determined according to the characteristic values ​​of the applicable target features through multiple blood pressure prediction models; based on the blood pressure values ​​predicted by the multiple blood pressure prediction models and their corresponding weights, the blood pressure value predicted by the monitoring watch is determined, and the blood pressure value may include systolic pressure SBP and diastolic pressure DBP.

[0100] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0101] Based on the same inventive concept, the embodiment of the present application also provides a blood pressure prediction device for implementing the blood pressure prediction method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more blood pressure prediction device embodiments provided below can refer to the limitations of the blood pressure prediction method above, and will not be repeated here.

[0102] In an exemplary embodiment, Figure 4 As shown, a blood pressure prediction device 400 is provided, comprising: an acquisition module 410, a feature processing module 420 and a blood pressure prediction module 430, wherein:

[0103] The acquisition module 410 is used to acquire the pulse wave signal collected by the monitoring watch.

[0104] The feature processing module 420 is used to determine the target features applicable to each of the multiple trained blood pressure prediction models; and extract the feature value of each target feature based on the pulse wave signal.

[0105] The blood pressure prediction module 430 is used to determine the blood pressure values ​​predicted by each of the multiple blood pressure prediction models according to the characteristic values ​​of the respective applicable target features through the multiple blood pressure prediction models; and determine the blood pressure value predicted by the monitoring watch based on the blood pressure values ​​predicted by each of the multiple blood pressure prediction models and their respective corresponding weights.

[0106] In an exemplary embodiment, the target feature is a feature among multiple preset features, and the acquisition module 410 is also used to acquire multiple sample pulse wave signals collected by the monitoring watch, and the actual blood pressure value corresponding to each sample pulse wave signal; the actual blood pressure value is obtained by measuring with a sphygmomanometer; the feature processing module 420 is also used to extract the feature values ​​of each of the multiple preset features based on each sample pulse wave signal, and obtain the feature values ​​of each sample pulse signal corresponding to the multiple preset features; the multiple preset features are combined differently to obtain different candidate feature combinations; each candidate feature combination includes at least two preset features; the blood pressure prediction device 400 also includes a training module, which is used to perform model training according to the feature values ​​of each candidate feature combination in the different candidate feature combinations corresponding to the multiple sample pulse wave signals, and the actual blood pressure values ​​corresponding to the multiple sample pulse wave signals, to obtain multiple blood pressure prediction models and the weights corresponding to each blood pressure prediction model, the multiple blood pressure prediction models respectively correspond to at least a part of the different candidate feature combinations, and the preset features in the candidate feature combination corresponding to each blood pressure prediction model are used as applicable target features.

[0107] In an exemplary embodiment, the multiple preset features include morphological features, time domain features and dynamic features; the morphological features include at least one of peak features, trough features, extreme value features and envelope features; the time domain features include heart rate; the dynamic features include at least one of change trend features and variability features.

[0108] In an exemplary embodiment, the training module is also used to perform linear regression model training for each candidate feature combination in the different candidate feature combinations based on the feature values ​​of the multiple sample pulse wave signals corresponding to the preset features in the candidate feature combination, and the actual blood pressure values ​​corresponding to the multiple sample pulse wave signals, to obtain the candidate model corresponding to the candidate feature combination; screen out multiple blood pressure prediction models from the candidate models corresponding to the different candidate feature combinations, and use the preset features in the candidate feature combination corresponding to each blood pressure prediction model as applicable target features; perform linear regression model training based on the multiple sample blood pressure values ​​predicted by the multiple blood pressure prediction models and corresponding to the multiple sample pulse wave signals, and the actual blood pressure values ​​corresponding to the multiple sample pulse wave signals, to obtain the weights corresponding to the multiple blood pressure prediction models.

[0109] In an exemplary embodiment, the feature processing module 420 is also used to filter the pulse wave signal to obtain a filtered pulse wave signal; after performing outlier data processing on the filtered pulse wave signal, local abnormal data processing is performed based on a sliding window to obtain a preprocessed pulse wave signal; based on the preprocessed pulse wave signal, the feature value of each target feature is extracted.

[0110] In an exemplary embodiment, the feature processing module 420 is also used to perform bandpass filtering on the pulse wave signal to obtain an intermediate pulse wave signal; perform low-pass filtering on the pulse wave signal to obtain a reference pressure change signal and a static pressure signal; and obtain a filtered pulse wave signal based on the intermediate pulse wave signal, the reference pressure change signal and the static pressure signal.

[0111] Each module in the above-mentioned blood pressure prediction device can be implemented in whole or in part by software, hardware and a combination thereof. Each of the above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to each of the above modules.

[0112] In an exemplary embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as shown in FIG. Figure 5 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. Among them, the processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data that needs to be stored when executing the above-mentioned blood pressure prediction model. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a blood pressure prediction method is implemented.

[0113] Those skilled in the art will understand that Figure 5 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0114] In one embodiment, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above method embodiments when executing the computer program.

[0115] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0116] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0117] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0118] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (AI) processor, etc., but are not limited to this.

[0119] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0120] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A blood pressure prediction method, characterized in that: The method comprises: Obtain the pulse wave signal collected by the monitoring watch; Determine the applicable target features for each of the multiple trained blood pressure prediction models; Based on the pulse wave signal, extracting a feature value of each target feature; Determining the blood pressure values ​​predicted by each of the multiple blood pressure prediction models according to the characteristic values ​​of the respective applicable target characteristics by using the multiple blood pressure prediction models; Based on the blood pressure values ​​predicted by the multiple blood pressure prediction models and their corresponding weights, the blood pressure value predicted by the monitoring watch is determined.

2. The method according to claim 1, characterized in that The target feature is a feature among a plurality of preset features, and the training steps of training the plurality of blood pressure prediction models include: Acquire multiple sample pulse wave signals collected by the monitoring watch, and the actual blood pressure value corresponding to each sample pulse wave signal; the actual blood pressure value is measured by a sphygmomanometer; Based on each sample pulse wave signal, extracting characteristic values ​​of each of the plurality of preset features, and obtaining characteristic values ​​of each sample pulse signal corresponding to the plurality of preset features; Performing different combinations of the plurality of preset features to obtain different candidate feature combinations; each candidate feature combination includes at least two preset features; Model training is performed based on the feature values ​​of each candidate feature combination in the different candidate feature combinations corresponding to each of the multiple sample pulse wave signals, and the actual blood pressure values ​​corresponding to each of the multiple sample pulse wave signals, to obtain multiple blood pressure prediction models and weights corresponding to each blood pressure prediction model, the multiple blood pressure prediction models respectively corresponding to at least a part of the different candidate feature combinations, and the preset features in the candidate feature combination corresponding to each blood pressure prediction model are used as applicable target features.

3. The method according to claim 2, characterized in that The multiple preset features include morphological features, time domain features and dynamic features; the morphological features include at least one of peak features, trough features, extreme value features and envelope features; the time domain features include heart rate; the dynamic features include at least one of change trend features and variability features.

4. The method according to claim 2, characterized in that: The method of performing model training according to the feature values ​​of each candidate feature combination in the different candidate feature combinations corresponding to each of the multiple sample pulse wave signals and the actual blood pressure values ​​corresponding to each of the multiple sample pulse wave signals to obtain multiple blood pressure prediction models and a weight corresponding to each blood pressure prediction model includes: For each candidate feature combination in the different candidate feature combinations, a linear regression model training is performed based on the feature values ​​of the plurality of sample pulse wave signals corresponding to the preset features in the candidate feature combination and the actual blood pressure values ​​corresponding to the plurality of sample pulse wave signals to obtain a candidate model corresponding to the candidate feature combination; Screening out a plurality of blood pressure prediction models from the candidate models corresponding to the different candidate feature combinations, wherein a preset feature in the candidate feature combination corresponding to each blood pressure prediction model is used as an applicable target feature; A linear regression model is trained based on the multiple sample blood pressure values ​​predicted by each of the multiple blood pressure prediction models and corresponding to the multiple sample pulse wave signals, as well as the actual blood pressure values ​​corresponding to each of the multiple sample pulse wave signals, to obtain the weights corresponding to each of the multiple blood pressure prediction models.

5. The method according to any one of claims 1 to 4, characterized in that: The step of extracting a characteristic value of each target characteristic based on the pulse wave signal comprises: Filtering the pulse wave signal to obtain a filtered pulse wave signal; After performing outlier data processing on the filtered pulse wave signal, local abnormal data processing is performed based on a sliding window to obtain a preprocessed pulse wave signal; Based on the preprocessed pulse wave signal, the feature value of each target feature is extracted.

6. The method according to claim 5, characterized in that The filtering process of the pulse wave signal to obtain a filtered pulse wave signal includes: performing bandpass filtering on the pulse wave signal to obtain an intermediate pulse wave signal; Performing low-pass filtering on the pulse wave signal to obtain a reference pressure change signal and a static pressure signal; A filtered pulse wave signal is obtained based on the intermediate pulse wave signal, the reference pressure change signal and the static pressure signal.

7. A blood pressure prediction device, characterized in that: The device comprises: An acquisition module is used to acquire the pulse wave signal collected by the monitoring watch; A feature processing module, used to determine target features applicable to each of the multiple trained blood pressure prediction models; and extract a feature value of each target feature based on the pulse wave signal; The blood pressure prediction module is used to determine the blood pressure values ​​predicted by each of the multiple blood pressure prediction models according to the characteristic values ​​of the respective applicable target features through the multiple blood pressure prediction models; and determine the blood pressure values ​​predicted by the monitoring watch based on the blood pressure values ​​predicted by each of the multiple blood pressure prediction models and their respective corresponding weights.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images