Blood pressure measuring method, electronic equipment, computer storage medium and computer program product

Abstract

The embodiment of the invention discloses a blood pressure measurement method, electronic equipment, a computer storage medium and a computer program product, and the method comprises the steps: determining a blood pressure measurement result of a user according to obtained first PPG data through a pre-trained blood pressure prediction model; the first PPG data is obtained based on a photoelectric volume pulse wave PPG signal collected for a user, model parameters of the pre-trained blood pressure prediction model comprise a first parameter and a second parameter, the first parameter is obtained by training of first equipment, and the second parameter is obtained by carrying out transfer learning on the model continuously by second equipment under the condition that the first parameter is frozen. The first PPG data can be collected through the device with the PPG signal collection function, and the cost of blood pressure measurement is reduced; meanwhile, the second parameter in the pre-trained blood pressure prediction model is obtained through training of the second device, the time and resource consumption needed for retraining the blood pressure prediction model is reduced, and meanwhile the blood pressure measurement precision can be improved.

Description

Technical Field

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

[0002] Hypertension refers to long-term and persistent high blood pressure, and is a common chronic disease. Hypertension may lead to cardiovascular disease and affect health. For patients with hypertension, it is necessary to monitor blood pressure regularly and adjust the corresponding treatment plan according to blood pressure to avoid the occurrence of cardiovascular disease.

[0003] In the related technology, blood pressure can be measured by professional equipment such as cuffs and stethoscopes, but it needs to be operated and interpreted by medical staff, blood pressure measurement is complicated, and the efficiency of blood pressure measurement is low.

[0004] In order to reduce the complexity of blood pressure measurement and improve the efficiency of blood pressure measurement, the relevant technology provides a wearable device with blood pressure measurement function, such as a bracelet, watch, etc. The wearable device can collect photoelectric volume pulse wave (PPG) signals and electrocardiogram (ECG) signals, and determine the pulse conduction time based on the PPG signals and ECG signals, and then use the pulse conduction time to estimate the blood pressure.

[0005] However, wearable devices that can collect PPG and ECG signals are expensive, which increases the cost of blood pressure measurement. At the same time, since wearable devices may be affected by user-specific factors (such as the elastic modulus of blood vessels, the thickness of blood vessel walls, the state of movement, etc.) during signal collection, the collected signals are inaccurate, which further leads to low accuracy of blood pressure measurement. Summary of the invention

[0006] The blood pressure measurement method, electronic device, computer storage medium and computer program product provided in the present application solve the problems of high cost and low accuracy of blood pressure measurement.

[0007] In order to achieve the above objectives, this application adopts the following technical solutions: In a first aspect, the present application provides a blood pressure measurement method, which is applied to a first device and may include: determining a user's blood pressure measurement result based on an acquired first PPG data through a pre-trained blood pressure prediction model; the first PPG data is obtained based on a photoplethysmogram (PPG) signal collected from the user, and the model parameters of the pre-trained blood pressure prediction model include a first parameter and a second parameter, the first parameter is obtained by training the first device, and the second parameter is obtained by continuing to perform transfer learning on the model by the second device when the first parameter is frozen.

[0008] It should be understood that, on the one hand, compared with wearable devices capable of collecting PPG signals and ECG signals, the present application can collect the first PPG data through a device with a PPG signal collection function, thereby reducing the cost of blood pressure measurement; on the other hand, by using the second device to perform transfer learning on the model when the first parameter is frozen to obtain the second parameter, there is no need to repeatedly train all parameters, thereby reducing the time and resource consumption required to retrain the model. At the same time, by using the first parameter and the second parameter obtained by transfer learning as the model parameters of the pre-trained blood pressure prediction model, the performance of the model is improved, thereby improving the accuracy of blood pressure measurement.

[0009] In combination with the first aspect, in a possible implementation, the second parameter can be obtained in the following manner: obtaining second PPG data and gold label data corresponding to the second PPG data, and then sending the second PPG data and the gold label data to the second device, and further calibrating the second initial parameter in the pre-trained parameter according to the second PPG data and the gold label data through the second device to obtain the second parameter; wherein the second PPG data is obtained based on the PPG signal collected from the user, the gold label data is obtained based on the actual blood pressure collected from the user, the pre-trained parameters are parameters of a trained blood pressure prediction model, the trained blood pressure model is obtained based on the PPG training data and the standard blood pressure training data, the pre-trained parameters include a first initial parameter and a second initial parameter, and the first parameter is the first initial parameter.

[0010] It should be understood that since some layers in the trained blood pressure prediction model have learned the general feature representation through the training data set, there is no need to retrain the trained blood pressure prediction model. In this way, by calibrating the second initial parameter, the efficiency of fine-tuning can be improved while improving the accuracy of blood pressure measurement, and computing resources can be saved.

[0011] In combination with the first aspect, in a possible implementation, the second PPG data includes first data and second data, the first data is obtained based on a PPG signal collected from a user in a first state, and the second data is obtained based on a PPG signal collected from a user in a second state; the gold label data includes first standard data and second standard data, the first standard data is the standard blood pressure of the user in the first state, and the second standard data is the standard blood pressure of the user in the second state.

[0012] It should be understood that by collecting the user's PPG calibration data and gold label data under different conditions, the data dimension can be enriched, making the results output by the fine-tuned blood pressure prediction model more accurate and improving the accuracy of blood pressure measurement.

[0013] In combination with the first aspect, in a possible implementation, before taking the first data and the second data as the second PPG data and taking the first standard data and the second standard data as the gold label data, it also includes: judging whether the absolute value of the difference between the first standard data and the second standard data is greater than or equal to the preset difference; if the absolute value of the difference between the first standard data and the second standard data is greater than or equal to the preset difference, taking the first data and the second data as the second PPG data and taking the first standard data and the second standard data as the gold label data; if the absolute value of the difference between the first standard data and the second standard data is less than the preset difference, reacquiring new second data and new second standard data until the difference between the new second standard data and the first standard data is greater than or equal to the preset difference, taking the first data and the new second data as the second PPG data and taking the first standard data and the new second standard data as the gold label data.

[0014] It should be understood that in order to ensure that there is a difference between the first data and the second data and to enrich the data dimension, it is necessary to make the difference between the first standard data and the second standard data greater than or equal to the preset difference before determining the second PPG data and the gold label data.

[0015] In combination with the first aspect, in a possible implementation, a blood pressure measurement result of a user is determined based on first PPG data through a pre-trained blood pressure prediction model, including: determining a first test result based on the first PPG data through a pre-trained blood pressure prediction model; determining the blood pressure measurement result of the user based on the initial test result, a confidence decay algorithm and the first test result; wherein the initial test result is the initial test result output by the pre-trained blood pressure prediction model for the user.

[0016] It should be understood that the effectiveness of the calibration of the pre-trained blood pressure prediction model will gradually fail over time, resulting in reduced accuracy of blood pressure measurement. In order to ensure the accuracy of blood pressure measurement, this embodiment can use a confidence reduction algorithm to adjust the confidence of each blood pressure measurement result to ensure the accuracy of blood pressure measurement.

[0017] In combination with the first aspect, in a possible implementation, determining the blood pressure measurement result of the user according to the initial test result, the confidence decay algorithm, and the first test result includes: multiplying the first test result and the confidence decay algorithm to obtain a first product; multiplying the initial test result and the complement of the confidence decay algorithm to obtain a second product; and taking the sum of the first product and the second product as the blood pressure measurement result of the user. The complement of the confidence decay algorithm refers to 1-the set (or value) obtained by the confidence decay algorithm.

[0018] It should be understood that the confidence of each blood pressure measurement result is adjusted by using a confidence reduction algorithm to ensure the accuracy of blood pressure measurement.

[0019] In combination with the first aspect, in a possible implementation, the pre-trained blood pressure prediction model includes a feature extraction layer and a linear layer, the first parameter is a parameter of the feature extraction layer, and the second parameter is a parameter of the linear layer.

[0020] In a second aspect, the present application provides a blood pressure measurement method, which is applied to a second device and may include: determining a user's test result based on an acquired first PPG data through a pre-trained blood pressure prediction model; wherein the first PPG data is obtained by collecting a photoplethysmogram (PPG) signal of the user from the first device; the model parameters of the pre-trained blood pressure prediction model include a first parameter and a second parameter, the first parameter being obtained by training the first device, and the second parameter being obtained by continuing to perform transfer learning on the model by the second device when the first parameter is frozen.

[0021] It should be understood that, on the one hand, compared with wearable devices capable of collecting PPG signals and ECG signals, the present application can collect the first PPG data through a device with a PPG signal collection function, thereby reducing the cost of blood pressure measurement; on the other hand, based on transfer learning, the second parameter in the pre-trained blood pressure prediction model is obtained by training with a second device, without the need to repeatedly train all parameters, thereby reducing the time and resource consumption required to retrain the blood pressure prediction model, and at the same time can improve the accuracy of blood pressure measurement.

[0022] It should be noted that in the embodiment of the present application, the user's blood pressure can be measured by the first device, and the user's blood pressure can also be measured by the second device, which is not specifically limited here.

[0023] In combination with the second aspect, in a possible implementation, the second parameter is obtained by: receiving second PPG data sent by the first device and gold label data corresponding to the second PPG data, and then calibrating the second initial parameter in the pre-trained parameter according to the second PPG data and the gold label data to obtain the second parameter; wherein the second PPG data is obtained based on the PPG signal collected from the user by the first device, the gold label data is obtained based on the actual blood pressure collected from the user, the pre-trained parameters are parameters of a trained blood pressure prediction model, the trained blood pressure model is obtained based on the PPG training data and the standard blood pressure training data, the pre-trained parameters include a first initial parameter and a second initial parameter, and the first parameter is the first initial parameter.

[0024] It should be understood that since some layers in the trained blood pressure prediction model have learned the general feature representation through the training data set, there is no need to retrain the trained blood pressure prediction model. In this way, by calibrating the second initial parameter, the efficiency of fine-tuning can be improved while improving the accuracy of blood pressure measurement, and computing resources can be saved.

[0025] In a third aspect, the present application provides an electronic device, including: a processor and a memory; Among them, one or more computer programs are stored in the memory, and the one or more computer programs include instructions; when the instructions are executed by the processor, the electronic device performs the blood pressure measurement method as described in the first aspect, or performs the blood pressure measurement method as described in the second aspect.

[0026] In a fourth aspect, the present application provides a computer storage medium comprising computer instructions. When the computer instructions are executed on an electronic device, the electronic device executes the blood pressure measurement method as described in the first aspect, or executes the blood pressure measurement method as described in the second aspect.

[0027] In a fifth aspect, the present application provides a computer program product. When the computer program product runs on a computer, the computer executes the blood pressure measurement method as described in the first aspect, or executes the blood pressure measurement method as described in the second aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 A schematic diagram of a blood pressure measurement method provided in an embodiment of the present application; Figure 2a A schematic diagram of a blood pressure measurement model training process provided in an embodiment of the present application; Figure 2b A schematic diagram of a process for obtaining PPG standard data and gold label data provided in an embodiment of the present application; Figure 3a A schematic diagram of a linear attenuation curve provided in an embodiment of the present application; Figure 3b A schematic diagram of a nonlinear attenuation curve provided in an embodiment of the present application; Figure 3c A schematic diagram of another nonlinear attenuation curve provided in an embodiment of the present application; Figure 4 A schematic diagram of a test result comparison provided in an embodiment of the present application; Figure 5 A scenario diagram of a blood pressure measurement method provided in an embodiment of the present application. DETAILED DESCRIPTION

[0029] In order to make the description of the following embodiments clear and concise, a brief introduction to the related technology is first given: Blood pressure refers to the lateral pressure exerted on the blood vessel wall per unit area when blood flows in the blood vessels. When the heart contracts, blood is pumped into the aorta, forming a pulse wave. The pulse wave propagates outward along the arterial system. During the propagation process, the pressure exerted by the pulse wave on the blood vessels is called blood pressure. Blood pressure can include systolic pressure and diastolic pressure.

[0030] Systolic blood pressure (SBP) refers to the maximum pressure exerted by blood on the blood vessel wall when the heart contracts. Systolic blood pressure can be measured as the first number in a blood pressure reading, for example, the "120" in 120 / 80 mmHg is the systolic blood pressure.

[0031] Diastolic Blood Pressure (DBP) refers to the minimum pressure exerted by blood on the blood vessel wall when the heart relaxes. Diastolic pressure can be measured as the second number in a blood pressure reading, for example, the "80" in 120 / 80 mmHg is the diastolic pressure.

[0032] Normal blood pressure means that without the use of antihypertensive drugs, the systolic blood pressure is less than 120 mmHg and the diastolic blood pressure is less than 80 mmHg. Among them, mmHg (millimeters of mercury) is the unit of blood pressure.

[0033] Photoplethysmography (PPG) is a physiological signal that is used to characterize the change in blood volume in the blood vessels inside the measured tissue. When the heart contracts, blood is pumped into the arteries, causing the arteries to dilate, and the light intensity received by the photoelectric sensor decreases, generating a PPG signal peak; when the heart relaxes, the arteries contract, the light intensity received by the photoelectric sensor increases, and the PPG signal decreases. PPG signals can be used for, but are not limited to, heart rate monitoring, blood oxygen saturation detection (abbreviated as blood oxygen detection), blood pressure monitoring, cardiovascular health assessment (such as arterial compliance, arterial aging, etc.), and so on.

[0034] Electrocardiogram (ECG) is a graph that records the electrical activity of the heart. The ECG machine can capture the weak electrical signals generated by the heart during contraction and relaxation, and convert these signals into waveforms. These waveforms mainly include P waves, QRS complexes, and T waves, which represent the electrical activity of the atria and ventricles, respectively. The P wave represents the potential change of atrial muscle depolarization, the QRS complex represents the potential change of ventricular muscle depolarization, and the T wave represents the potential change during ventricular rapid repolarization.

[0035] The Moens-Korteweg equation is used to describe the speed at which pressure waves propagate in a liquid, such as water in a pipe.

[0036] The VGG (Visual Geometry Group) network is a deep convolutional neural network architecture.

[0037] ResNet (Residual Network) is a convolutional neural network (CNN) architecture in the field of deep learning. The core idea of ​​ResNet is to introduce residual blocks, which directly add input to output through skip connections, thereby solving the gradient vanishing and gradient exploding problems in deep neural networks.

[0038] EfficientNet is a convolutional neural network that optimizes depth, width, and resolution through a compound scaling strategy to achieve a balance between performance and computational efficiency.

[0039] Resting means that the user is in a still or calm state, such as sitting, lying, or standing, and there is no movement that causes a significant change in heart rate.

[0040] Transfer learning is a machine learning method that uses a model that has been trained on one task (source task) to solve another related but different task (target task). By transferring existing knowledge and experience, the time and resource consumption of retraining the model on the new task can be reduced, and the performance of the model may be improved.

[0041] The related technology provides a wearable device with the function of collecting PPG signals and ECG signals. The wearable device is used to collect PPG signals and ECG signals, and then blood pressure is estimated based on the PPG signals and ECG signals, thereby realizing blood pressure measurement.

[0042] However, the price of the first device that has the function of collecting both PPG and ECG signals is relatively expensive, which increases the cost of blood pressure measurement. In addition, the collected PPG and ECG signals may be affected by human body-specific factors (such as the elastic modulus of blood vessels, the thickness of blood vessel walls, exercise conditions, etc.), resulting in poor quality of the collected signals.

[0043] As an example, a physiological model of blood pressure is established based on the Moens-Korteweg equation, such as the following formula (1): ;Formula (1) in, is the elastic modulus of the blood vessel, is the thickness of the blood vessel wall, , is a constant, is the blood density, is the radius of the blood vessel.

[0044] As can be seen from formula (1), specific factors affecting blood pressure include but are not limited to the elastic modulus of the user's blood vessels, the thickness of the blood vessel wall, and the radius of the blood vessels. The method of measuring blood pressure using PPG signals and ECG signals in related technologies may result in low accuracy of blood pressure measurement.

[0045] Based on the above problems, combined with Figure 1 As shown, a blood pressure measurement method provided in an embodiment of the present application aims to reduce the cost of blood pressure measurement and improve the accuracy of blood pressure measurement through a training process, a fine-tuning process, and a testing process. For ease of explanation, the following is a combination of the training process, the fine-tuning process, and the testing process. Figure 1 Provide explanation.

[0046] Combination Figure 1 As shown in 10, the training process of blood pressure measurement provided in the embodiment of the present application may be: Step 1: Obtain PPG data and standard blood pressure data corresponding to the PPG data.

[0047] PPG data refers to a data set obtained by sampling, recording and processing the PPG signal. The PPG signal in the PPG data is in the format of a time series, which characterizes the heart beat and the change of blood volume. In a possible implementation, the PPG data can be obtained based on a first device with a PPG signal acquisition function.

[0048] The first device refers to an electronic device that has the ability to collect PPG signals and can be worn by a user, such as a watch, a bracelet, a portable detector, a patch-type detection device, etc., which is not specifically limited here. In the following embodiments, the first device can be exemplified by taking a watch as an example.

[0049] Standard blood pressure data refers to data including multiple standard blood pressures. Standard blood pressure (also called actual blood pressure) refers to the user's actual blood pressure value, and standard blood pressure includes systolic pressure and diastolic pressure. In a possible implementation, the user's standard blood pressure can be obtained by a sphygmomanometer, a blood pressure meter, or other blood pressure measuring equipment.

[0050] As an example, assuming the systolic / diastolic blood pressure is 140 / 90 mmHg, then the standard blood pressure is 140 / 90 mmHg; assuming the systolic / diastolic blood pressure is 110 / 70 mmHg, then the standard blood pressure is 110 / 70 mmHg.

[0051] Step 2: Determine the training data set based on the PPG data and standard blood pressure data.

[0052] The training data set refers to a data set used to train the blood pressure prediction model, and the training data set includes PPG training data and standard blood pressure training data corresponding to the PPG training data. As an example, assuming that the PPG training data includes PPG1 and PPG2, the standard blood pressure data includes standard blood pressure 1 corresponding to PPG1 and standard blood pressure 2 corresponding to PPG2. The training data set can be obtained by performing data alignment, data slicing, and other processing on the PPG data and the standard blood pressure data.

[0053] Data alignment means adjusting data from different sources, different formats or different time points to a unified format or time reference for comparison, analysis or fusion. In one possible implementation, PPG data and standard blood pressure data can be aligned one-to-one in time, that is, each PPG signal corresponds to a standard blood pressure, to ensure the accuracy and reliability of the subsequent training process.

[0054] As an example, assuming that the PPG data includes PPG1, PPG2, PPG3, PPG4, and PPG5, and the standard blood pressure data includes standard blood pressure 1, standard blood pressure 2, standard blood pressure 3, standard blood pressure 4, and standard blood pressure 5, by aligning the PPG data with the standard blood pressure data, PPG1 corresponds to standard blood pressure 1, PPG2 corresponds to standard blood pressure 2, PPG3 corresponds to standard blood pressure 3, PPG4 corresponds to standard blood pressure 4, and PPG5 corresponds to standard blood pressure 5.

[0055] Data slicing means extracting a portion of data from an initial data set for a specific analysis or processing task. In one possible implementation, the data set after the PPG data and the standard blood pressure data are aligned can be processed by data slicing to divide the training data set, the validation data set, and the test data set in proportion for training and evaluation.

[0056] As an example, assuming that the PPG data includes PPG1, PPG2, PPG3, PPG4, and PPG5, and the standard blood pressure data includes standard blood pressure 1, standard blood pressure 2, standard blood pressure 3, standard blood pressure 4, and standard blood pressure 5, then data slicing processing of the PPG data and the standard blood pressure data can obtain a training data set: the processed PPG data includes PPG2, PPG3, and PPG4, and the processed standard blood pressure data includes standard blood pressure 2, standard blood pressure 3, and standard blood pressure 4.

[0057] Step 3: Use the training data set to train the blood pressure prediction model.

[0058] The blood pressure prediction model (also called the initial blood pressure prediction model) refers to the initial network model used to predict blood pressure, such as Figure 2a As shown, the blood pressure prediction model may include a feature extraction layer and a linear layer.

[0059] The feature extraction layer refers to a layer used to extract features from PPG training data. The network structure of the feature extraction layer can be, but is not limited to, VGG, ResNet, EfficientNet, etc., and is not specifically limited here.

[0060] The linear layer is also called the fully connected layer (FCL) or dense layer. The linear layer is used to perform linear transformation on the input data. The linear layer is located after the feature extraction layer, and is used to receive these high-level abstract features and integrate them into the final output. In this embodiment, the output of the feature extraction layer is used as the input data of the linear layer, and the output of the feature extraction layer is integrated to obtain the training result. Among them, the training result includes systolic pressure and diastolic pressure.

[0061] In one possible implementation, each PPG signal in the PPG training data in the training data set can be input into the blood pressure prediction model to obtain a training result corresponding to each PPG signal, and then the blood pressure prediction model is optimized according to the multiple training results and the standard blood pressure training data to obtain a trained blood pressure prediction model.

[0062] It should be understood that supervised training can be used for the training of the blood pressure prediction model in the embodiments of the present application. Supervised training (also known as supervised learning) refers to the training process of adjusting the parameters of the blood pressure prediction model using a set of samples of known categories (i.e., training data sets) to achieve the required performance. Supervised learning enables more accurate prediction and classification to minimize prediction errors. Since each training result in supervised learning provides a corresponding standard blood pressure, the blood pressure prediction model can directly learn from errors and adjust its performance, which is conducive to the rapid convergence of the blood pressure prediction model and improves performance.

[0063] It should be noted that the training process of the blood pressure prediction model in the embodiment of the present application can be implemented based on the first device. Training the blood pressure prediction model based on the first device can make full use of existing resources and improve resource utilization; it can also reduce training delays, which is conducive to quickly adjusting and optimizing the blood pressure prediction model; at the same time, it can also avoid time and bandwidth consumption during data transmission, thereby improving training efficiency.

[0064] Step 4: Obtain and store the pre-trained parameters of the trained blood pressure prediction model.

[0065] The trained blood pressure prediction model refers to the network obtained by training the training data set. The pre-trained parameters refer to the parameters corresponding to the trained blood pressure prediction model, and the pre-trained parameters may include feature extraction layer parameters (also referred to as first initial parameters) and linear layer parameters (also referred to as second initial parameters). The feature extraction layer parameters refer to the parameters corresponding to the feature extraction layer, and the linear layer parameters refer to the parameters corresponding to the linear layer.

[0066] It should be noted that, during the PPG signal acquisition process, the acquired signal quality may be poor due to the influence of specific factors, which in turn leads to low accuracy of blood pressure measurement. In order to improve the accuracy of blood pressure measurement, after training the blood pressure prediction model, the trained blood pressure prediction model can also be fine-tuned.

[0067] Combination Figure 1 As shown in 20, the fine-tuning process (also referred to as a calibration process) of the blood pressure measurement provided in the embodiment of the present application can be: Step 1: Obtain PPG calibration data and gold standard data corresponding to the PPG calibration data.

[0068] PPG calibration data (also referred to as second PPG data) refers to data used to fine-tune the trained blood pressure prediction model. The PPG calibration data may include first calibration data and second calibration data. The first calibration data refers to the PPG data collected when the user is at rest, and the second calibration data refers to the PPG data collected when the user is exercising.

[0069] Gold standard data (i.e., standard blood pressure data) refers to data including multiple standard blood pressures, and gold standard data is used to characterize the actual blood pressure of the user. In a possible implementation, in the process of obtaining PPG calibration data, a sphygmomanometer, a blood pressure meter, or other equipment can be used to synchronously collect the user's actual systolic and diastolic blood pressures, and then the actual systolic and diastolic blood pressures are used as standard blood pressures, and the obtained multiple standard blood pressures are used as gold standard data.

[0070] As an example, combining Figure 2b As shown, the process of obtaining PPG calibration data and gold standard data can be: A1: Prompt the user to perform calibration.

[0071] It should be understood that in the embodiment of the present application, when the user uses the blood pressure measurement function for the first time or again after a preset time, in order to ensure the accuracy of the blood pressure measurement and improve the precision of the blood pressure measurement, the blood pressure measurement application in the first device can prompt the user to calibrate so as to fine-tune the trained blood pressure prediction model. Among them, the prompting method includes but is not limited to button display, full-screen display, vibration display, voice prompt, etc., which are not specifically limited here.

[0072] A2: The user is at rest.

[0073] It should be understood that the blood pressure conditions characterized by the user in different states may also be different. For example, the blood pressure of the user when exercising may be higher than that of the user when sleeping. In order to improve the accuracy of blood pressure measurement, calibration data of the user in resting (also referred to as the first state) and exercising (also referred to as the second state) can be obtained. Among them, the user can rest for a period of time to collect data so that the blood pressure is stable within a certain range and avoid poor quality of collected data caused by large fluctuations. For example, the user can rest for three minutes before collecting data.

[0074] A3: Collect first calibration data and first standard data corresponding to the first calibration data.

[0075] The first calibration data (also referred to as the first data) refers to the PPG data collected by the first device when the user is in a resting state, and the first standard data refers to the gold standard data corresponding to the first calibration data.

[0076] In the embodiment of the present application, the first calibration data can be collected by the first device, and the first standard data can be collected by a sphygmomanometer, a blood pressure meter or other devices. The first calibration data and the first standard data can be collected at the same time; the first calibration data can be collected first, and then the first standard data; the first standard data can be collected first, and then the first calibration data, which is not specifically limited here.

[0077] A4: User movement.

[0078] It should be understood that in order to improve the accuracy of blood pressure measurement, after collecting the first calibration data and the first standard data, the user can be prompted to exercise so that there is a difference between the blood pressure during exercise and the blood pressure at rest. For example, the user can exercise for one minute and then collect data.

[0079] A5: Collect second calibration data and second standard data corresponding to the second calibration data.

[0080] The second calibration data (also referred to as the second data) refers to the PPG data collected by the first device when the user is in motion (or after motion), and the second standard data refers to the gold standard data corresponding to the second calibration data.

[0081] In an embodiment of the present application, the second calibration data can be collected through the first device, and the second standard data can be collected through a sphygmomanometer, a blood pressure meter or other equipment. The second calibration data and the second standard data can be collected at the same time; the second calibration data can also be collected first, and then the second standard data; the second standard data can also be collected first, and then the second calibration data, which is not specifically limited here.

[0082] A6: Determine whether the absolute value of the difference between the systolic pressure in the first standard data and the systolic pressure in the second standard data is greater than or equal to a preset difference.

[0083] If the absolute value of the difference between the systolic pressure in the first standard data and the systolic pressure in the second standard data is greater than or equal to the preset difference, step A7 is executed; otherwise, step A4 is executed.

[0084] The preset difference refers to the absolute value of the difference between the systolic pressure of the first standard data and the systolic pressure of the second standard data that meet the output condition. For example, the preset difference is 5 mmHg.

[0085] It should be understood that in order to ensure that there is a difference between the data collected at rest and the data collected during exercise, the absolute value of the difference between the systolic pressure in the first standard data and the systolic pressure in the second standard data needs to be greater than or equal to the preset difference before output; otherwise, the user is required to exercise again and re-collect the second calibration data (also referred to as new second calibration data) and the second standard data (also referred to as new second standard data).

[0086] A7: The first calibration data and the second calibration data are used as PPG calibration data, and the first standard data and the second standard data are used as gold standard data.

[0087] It should be understood that in the embodiment of the present application, by collecting PPG calibration data and gold label data under different states of the user, the data dimension can be enriched, so that the results output by the fine-tuned blood pressure prediction model are more accurate, thereby improving the accuracy of blood pressure measurement.

[0088] Step 2: Upload the PPG calibration data and gold label data to the second device.

[0089] The second device refers to a device that has a communication connection with the first device and has more computing resources, such as a mobile phone, a computer, a tablet, etc., which is not specifically limited here. The second device and the first device can be connected via Bluetooth, a local area network, a WLAN, etc., which is not specifically limited here.

[0090] It should be understood that, since the computing resources of the first device are limited, in order to save resources, the fine-tuning process of the blood pressure prediction model can be performed by the second device. Therefore, in the embodiment of the present application, the collected PPG calibration data and gold standard data can be sent to the second device, and the second device can use the PPG calibration data and gold standard data to fine-tune the trained blood pressure prediction model.

[0091] Step 3: Use the PPG calibration data and gold standard data to fine-tune the linear layer in the trained blood pressure prediction model through the second device.

[0092] It should be understood that in the embodiment of the present application, the feature extraction layer can be frozen first, that is, the feature extraction layer is not fine-tuned and the feature extraction parameters remain unchanged.

[0093] It should be understood that since the feature extraction layer in the trained blood pressure prediction model has learned the general feature representation through the training data set, the general feature representation already contains rich information and prior knowledge for the new task (i.e., the fine-tuning process), so there is no need to fine-tune the trained blood pressure prediction model as a whole.

[0094] At the same time, since the linear layer is responsible for combining and classifying features according to the characteristics of the specific task, fine-tuning the linear layer can make it better adapt to the data distribution and label space of the new task without changing the overall structure or most parameters of the trained blood pressure prediction model. Among them, fine-tuning the linear layer requires less computing resources and time than retraining the entire model. On the basis of improving the accuracy of blood pressure measurement, it can improve the efficiency of fine-tuning and save computing resources.

[0095] In addition, retraining the entire model may lead to overfitting, so the risk of overfitting can be reduced by keeping the feature extraction parameters of the feature extraction layer in the trained blood pressure prediction model unchanged and leveraging its generalization ability.

[0096] Step 4: Use the feature extraction parameters of the feature extraction layer (also referred to as the first parameters or the first initial parameters) and the linear layer parameters of the fine-tuned linear layer (also referred to as the second parameters) as model parameters of the trained blood pressure prediction model.

[0097] The model parameters may be stored in the second device, or may be transmitted from the second device to the first device and stored by the first device, which is not specifically limited here.

[0098] It should be understood that, on the one hand, by collecting PPG calibration data and gold label data under different user states, the data dimension can be enriched, making the output of the fine-tuned blood pressure prediction model more accurate and improving the accuracy of blood pressure measurement. On the other hand, based on transfer learning, the linear layer in the blood pressure prediction model is fine-tuned by the second device without repeated training, reducing the time and resource consumption required to retrain the blood pressure prediction model, while improving the accuracy of blood pressure measurement.

[0099] After obtaining the fine-tuned blood pressure prediction model (referred to as the pre-trained blood pressure prediction model), combined with Figure 1 As shown in 30, the test process of blood pressure measurement provided in the embodiment of the present application can be: Step A: Acquire test data through the first device.

[0100] The test data (also referred to as the first PPG data) refers to the PPG data that the user needs to collect when measuring blood pressure.

[0101] Step B: Input the test data into the pre-trained blood pressure prediction model to obtain the test results.

[0102] The pre-trained blood pressure prediction model in this embodiment refers to a trained blood pressure prediction model after fine-tuning. The test result refers to the systolic and diastolic blood pressures predicted by the pre-trained blood pressure prediction model.

[0103] In a possible implementation, the blood pressure of the user can be predicted based on the test data using the pre-trained blood pressure prediction model stored on the first device to obtain a test result (also referred to as a blood pressure measurement result). The linear layer parameters of the fine-tuned linear layer can be sent to the first device via the second device for subsequent blood pressure measurement.

[0104] It should be noted that, since the second device only fine-tunes the linear layer in the blood pressure prediction model, the parameters of feature extraction on the first device and the second device are the same. In order to improve transmission efficiency, the second device can only transmit the linear layer parameters to the first device. In other possible implementations, the second device can also transmit the model parameters to the first device, which is not specifically limited here.

[0105] In another possible implementation, the first device may send the test data to the second device, and the pre-trained blood pressure prediction model on the second device measures the user's blood pressure according to the test data to obtain a test result. It should be understood that since the second device has higher computing power, the blood pressure measurement can be performed using the pre-trained blood pressure prediction model on the second device to increase the speed of blood pressure measurement.

[0106] Based on the above test process, further research found that the effectiveness of the calibration of the pre-trained blood pressure prediction model will gradually fail over time, resulting in reduced accuracy of blood pressure measurement. In order to ensure the accuracy of blood pressure measurement, the embodiment of the present application also provides a confidence reduction algorithm for adjusting the confidence of each test result, such as the following formula (2): ;Formula (2) in, To finally display the test results to the user (also known as blood pressure measurement results), is the current test result (also called the first test result), is the initial test result (also called the initial test result), It is the confidence decay algorithm.

[0107] In the embodiment of the present application, the attenuation mode of the confidence attenuation algorithm may include linear attenuation (such as Figure 3a as shown) and nonlinear attenuation (as Figure 3b and Figure 3c As shown), nonlinear attenuation includes but is not limited to exponential attenuation, power-law attenuation, etc., which are not specifically limited here.

[0108] Linear attenuation means that there is a linear relationship between the attenuation of the actual value (the current test result) and the initial value (the initial test result of the blood pressure prediction model) or a reference value, that is, the attenuation increases proportionally with the increase of time or distance.

[0109] Nonlinear attenuation means that the attenuation of the actual value is not linearly related to the initial value or a reference value, that is, the attenuation rate is not constant, but changes with the increase of time or distance.

[0110] Exponential decay is a special nonlinear decay mode in which the decay rate of the actual value is proportional to its current value, that is, the decay rate decreases as the actual value decreases.

[0111] Power-law decay is another special nonlinear decay mode, in which the decay rate of a physical quantity or signal is proportional to the power of the current value of the quantity.

[0112] Based on the blood pressure measurement method provided in the above embodiment, combined with Figure 4 As shown, Figure 4 a in the figure shows the first test results (L1 and L2) obtained by the blood pressure measurement method of the related art, L1 represents the fitting lines of multiple systolic pressures in the first test result, and L2 represents the fitting lines of multiple diastolic pressures in the first test result. Figure 4 b in the figure shows the second test result (L3 and L4) obtained by the blood pressure measurement method of the embodiment of the present application, L3 represents the fitting lines of multiple systolic pressures in the second test result, and L4 represents the fitting lines of multiple systolic pressures in the second test result. K1 represents multiple reference lines of actual systolic pressures, and K2 represents multiple reference lines of actual diastolic pressures. The actual systolic pressure and the actual diastolic pressure are obtained based on the blood pressure measurement of the user.

[0113] Among them, the correlation coefficients between the fitting line and the reference line of the first test result obtained by the blood pressure measurement method of the related art are r1=0.395 and r2=0.321 respectively; the correlation coefficients between the fitting line and the reference line of the second test result obtained by the blood pressure measurement method provided in the embodiment of the present application are r3=0.755 and r4=0.807 respectively. r1 represents the correlation coefficient between L1 and K1, r2 represents the correlation coefficient between L2 and K2, r3 represents the correlation coefficient between L3 and K1, and r4 represents the correlation coefficient between L4 and K2.

[0114] The correlation coefficient (r) is used to indicate the degree of linear correlation between two variables. The range of the correlation coefficient is [-1,1], where 1 indicates a perfect positive correlation, -1 indicates a perfect negative correlation, and 0 indicates no correlation. The correlation coefficient is an indicator of the strength and direction of the linear relationship between two variables. When r is close to 1 or -1, it indicates that there is a strong linear relationship between the two variables; when r is close to 0, it indicates that there is almost no linear relationship between the two variables.

[0115] Depend on Figure 4 It can be seen that the test results of the blood pressure measurement method provided in the embodiment of the present application are more highly correlated with the actual test results, that is, compared with the related art, the blood pressure measurement method provided in the embodiment of the present application improves the accuracy of blood pressure measurement.

[0116] Based on the blood pressure measurement method provided in the above embodiment, combined with Figure 5 As shown, the embodiment of the present application also provides a schematic diagram of a blood pressure measurement scenario, where the first device is a watch as an example, and the second device is a mobile phone as an example. The blood pressure measurement process may be: Step a: The user clicks on the blood pressure test through the mobile phone.

[0117] The mobile phone establishes a communication connection with the watch, the mobile phone can respond to the user's click operation, and send a test instruction to the watch via Bluetooth or other means, and the watch responds to the test instruction sent by the mobile phone.

[0118] Step b: The watch prompts the user to calibrate.

[0119] If the user measures blood pressure for the first time or measures blood pressure at a preset interval, calibration may be performed to improve the accuracy of blood pressure measurement. It should be understood that in one possible implementation, if the user measures again within the preset time, calibration may not be performed; the user may also set calibration for each measurement, and the number of calibrations or calibration conditions are not limited.

[0120] Step c: The watch collects the user's PPG calibration data and sends the PPG data to the mobile phone.

[0121] It should be noted that the watch sends the collected data to the mobile phone, which is calibrated by the mobile phone. After the calibration is completed, the parameters of the blood pressure prediction model can be saved or sent to the watch for saving, which is not specifically limited here. The calibration process can refer to the fine-tuning process of blood pressure measurement in the above embodiment, and will not be repeated here.

[0122] If the calibration fails, the watch can prompt the user to recalibrate; if the calibration is successful, the watch can prompt the user to measure blood pressure.

[0123] Step d: After the phone is calibrated, the watch prompts the user to measure blood pressure.

[0124] It should be noted that if a watch is used to measure blood pressure, the watch can output the test results based on the collected PPG data; if a mobile phone is used to measure blood pressure, the PPG data collected by the mobile phone can be sent to the mobile phone, and the mobile phone can output the test results based on the PPG data.

[0125] Step e: The watch or mobile phone displays the user's test results.

[0126] It should be understood that the embodiment of the present application can improve the accuracy of blood pressure measurement by measuring blood pressure after calibration.

[0127] The terms "first", "second", "third", etc. in the specification, claims and drawings of this application are used to distinguish different objects rather than to limit a specific order.

[0128] In the embodiments of the present application, words such as "exemplary" or "for example" are used to indicate examples, illustrations or descriptions. Any embodiment or design described as "exemplary" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary" or "for example" is intended to present related concepts in a specific way.

[0129] The technical solution of this embodiment, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) or a processor to execute all or part of the steps of the method described in each embodiment. The aforementioned storage medium includes: flash memory, mobile hard disk, read-only memory, random access memory, magnetic disk or optical disk, etc., various media that can store program codes.

[0130] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A blood pressure measurement method, characterized in that: The method is applied to a first device, comprising: Acquire first PPG data; the first PPG data is obtained based on a photoplethysmogram (PPG) signal collected from a user; The blood pressure measurement result of the user is determined according to the first PPG data through a pre-trained blood pressure prediction model; the model parameters of the pre-trained blood pressure prediction model include a first parameter and a second parameter, the first parameter is obtained by training the first device, and the second parameter is obtained by continuing to perform transfer learning on the model by the second device when the first parameter is frozen.

2. The blood pressure measurement method according to claim 1, characterized in that: The second parameter is obtained by: Acquire second PPG data and gold label data corresponding to the second PPG data; the second PPG data is obtained based on the PPG signal collected from the user, and the gold label data is obtained based on the actual blood pressure collected from the user; Sending the second PPG data and the golden label data to the second device; Through the second device, according to the second PPG data and the gold standard data, the second initial parameter in the pre-trained parameters is calibrated to obtain the second parameter; the pre-trained parameters are parameters of a trained blood pressure prediction model, and the trained blood pressure model is obtained based on PPG training data and standard blood pressure training data. The pre-trained parameters include a first initial parameter and a second initial parameter, and the first parameter is the first initial parameter.

3. The blood pressure measurement method according to claim 2, characterized in that: The second PPG data includes first data and second data, the first data is obtained based on a PPG signal collected from the user in a first state, and the second data is obtained based on a PPG signal collected from the user in a second state; The gold standard data includes first standard data and second standard data, the first standard data is the standard blood pressure of the user in the first state, and the second standard data is the standard blood pressure of the user in the second state.

4. The blood pressure measurement method according to claim 3, characterized in that: Before using the first data and the second data as the second PPG data, and using the first standard data and the second standard data as the gold label data, the method further includes: Determining whether an absolute value of a difference between the first standard data and the second standard data is greater than or equal to a preset difference; If the absolute value of the difference between the first standard data and the second standard data is greater than or equal to the preset difference, the first data and the second data are used as the second PPG data, and the first standard data and the second standard data are used as gold label data; If the absolute value of the difference between the first standard data and the second standard data is less than the preset difference, new second data and new second standard data are reacquired until the difference between the new second standard data and the first standard data is greater than or equal to the preset difference, and the first data and the new second data are used as the second PPG data, and the first standard data and the new second standard data are used as the gold label data.

5. The blood pressure measurement method according to claim 1, characterized in that: Determining the blood pressure measurement result of the user according to the first PPG data using the pre-trained blood pressure prediction model includes: Determining a first test result according to the first PPG data by using the pre-trained blood pressure prediction model; The blood pressure measurement result of the user is determined according to the initial test result, the confidence decay algorithm and the first test result; the initial test result is the initial test result output by the pre-trained blood pressure prediction model for the user.

6. The blood pressure measurement method according to claim 5, characterized in that: Determining the blood pressure measurement result of the user according to the initial test result, the confidence decay algorithm and the first test result includes: Multiplying the first test result and the confidence decay algorithm to obtain a first product; Multiplying the initial test result and the complement of the confidence decay algorithm to obtain a second product; The sum of the first product and the second product is used as the blood pressure measurement result of the user.

7. The blood pressure measurement method according to any one of claims 1 to 6, characterized in that: The pre-trained blood pressure prediction model includes a feature extraction layer and a linear layer, the first parameter is a parameter of the feature extraction layer, and the second parameter is a parameter of the linear layer.

8. A blood pressure measurement method, characterized in that: Applied to the second device, comprising: Acquire first PPG data; the first PPG data is obtained by collecting a photoplethysmogram (PPG) signal of a user from a first device; The blood pressure measurement result of the user is determined according to the first PPG data through a pre-trained blood pressure prediction model; the model parameters of the pre-trained blood pressure prediction model include a first parameter and a second parameter, the first parameter is obtained by training the first device, and the second parameter is obtained by continuing to perform transfer learning on the model by the second device when the first parameter is frozen.

9. The blood pressure measurement method according to claim 8, characterized in that: The second parameter is obtained by: receiving second PPG data and gold label data corresponding to the second PPG data sent by the first device, wherein the second PPG data is obtained based on the PPG signal collected from the user by the first device, and the gold label data is obtained based on the actual blood pressure collected from the user; According to the second PPG data and the gold standard data, the second initial parameter in the pre-training parameter is calibrated to obtain the second parameter; the pre-training parameter is the parameter of the trained blood pressure prediction model, and the trained blood pressure model is obtained based on the PPG training data and the standard blood pressure training data. The pre-training parameter includes a first initial parameter and a second initial parameter, and the first parameter is the first initial parameter.

10. An electronic device, characterized in that: include: Processor and memory; Wherein, one or more computer programs are stored in the memory, and the one or more computer programs include instructions; when the instructions are executed by the processor, the electronic device performs the blood pressure measurement method as described in any one of claims 1-7, or performs the blood pressure measurement method as described in any one of claims 8-9.

11. A computer storage medium, characterized in that: The method comprises computer instructions. When the computer instructions are executed on an electronic device, the electronic device executes the blood pressure measurement method according to any one of claims 1 to 7, or executes the blood pressure measurement method according to any one of claims 8 to 9.

12. A computer program product, characterized in that When the computer program product runs on a computer, the computer executes the blood pressure measurement method according to any one of claims 1 to 7, or executes the blood pressure measurement method according to any one of claims 8 to 9.

Images