Method and apparatus for blood pressure prediction, electronic device, and storage medium

Abstract

Provided are a method and apparatus (600) for blood pressure prediction, an electronic device (800), and a computer-readable storage medium (1000). The method comprises: acquiring pulse information of a first user (S110); inputting the pulse information into a first prediction model corresponding to the first user to obtain predicted blood pressure information corresponding to the pulse information (S120); and determining a blood pressure prediction value of the first user on the basis of the predicted blood pressure information (S130), wherein the first prediction model is obtained by training a blood pressure prediction model on the basis of sample data of the first user, and the blood pressure prediction model is obtained by training an initial model on the basis of sample data of a plurality of users.

Description

Blood pressure prediction methods, devices, electronic equipment and storage media Technical Field

[0001] Embodiments of this disclosure relate to a blood pressure prediction method, apparatus, electronic device, and computer-readable storage medium. Background Technology

[0002] Currently, all non-invasive blood pressure sensors on the market that meet medical standards are cuff-type. However, cuff-type blood pressure sensors cannot perform continuous blood pressure monitoring and measurement, and they cause significant discomfort due to blood flow obstruction in the arm during use, making it impossible for hypertensive patients to know their blood pressure anytime, anywhere. Cuffless blood pressure measurement devices, as a blood pressure measurement device that combines continuous detection and wearing comfort, offer great convenience in practical use and have significant medical and commercial value. Summary of the Invention

[0003] At least one embodiment of this disclosure provides a blood pressure prediction method, comprising: acquiring pulse information of a first user; inputting the pulse information into a first prediction model corresponding to the first user to obtain predicted blood pressure information corresponding to the pulse information; and determining a predicted blood pressure value for the first user based on the predicted blood pressure information; wherein the first prediction model is obtained by training a blood pressure prediction model based on sample data of the first user, and the blood pressure prediction model is obtained by training an initial model based on sample data of multiple users.

[0004] For example, in the blood pressure prediction method provided in at least one example of the above embodiments of this disclosure, during the training process of the blood pressure prediction model, multiple weight parameters in the initial model are updated based on the sample data of the multiple users, while keeping the value of at least one target weight node in the initial model unchanged, so as to obtain the blood pressure prediction model; during the training process of the first prediction model, the at least one target weight node in the blood pressure prediction model is updated based on the sample data of the first user, so as to obtain the first prediction model.

[0005] For example, in the blood pressure prediction method provided in at least one example of the above embodiments of this disclosure, when the type of the initial model is a multiple linear regression model, the at least one target weight node is at least one constant zero weight node in the multiple linear regression model.

[0006] For example, in a blood pressure prediction method provided in at least one example of the above embodiments of this disclosure, obtaining the pulse information of a first user includes: acquiring a pressure signal from the wrist of the first user using a signal acquisition module; converting the pressure signal into a frequency domain signal sequence using a fast Fourier transform; and extracting amplitude information with a frequency within a predetermined range from the frequency domain signal sequence as the pulse information of the first user.

[0007] For example, in at least one example of the blood pressure prediction method provided in the above embodiments of this disclosure, the pressure signal is converted into a frequency domain signal sequence using Fast Fourier Transform, including: performing noise reduction processing on the pressure signal to obtain a noise-reduced pressure signal; performing downsampling processing on the noise-reduced pressure signal to obtain a downsampled pressure signal; and using Fast Fourier Transform to convert the downsampled pressure signal into the frequency domain signal sequence.

[0008] For example, in at least one example of the blood pressure prediction method provided in the above embodiments of this disclosure, both the pulse information and the predicted blood pressure information are frequency domain signal sequences; wherein, determining the predicted blood pressure value of a first user based on the predicted blood pressure information includes: converting the predicted blood pressure information into a time domain signal sequence using the inverse of the fast Fourier transform; and extracting the systolic blood pressure and diastolic blood pressure from the time domain signal sequence using a peak-finding algorithm as the predicted blood pressure value of the first user.

[0009] For example, in a blood pressure prediction method provided in at least one example of the above embodiments of this disclosure, systolic and diastolic blood pressure are extracted from the time-domain signal sequence using a peak-finding algorithm, including: extracting the maximum value other than a predetermined amplitude from the frequency-domain signal sequence of the predicted blood pressure information as the fundamental frequency; filtering at least one maximum and at least one minimum value from the time-domain signal sequence of the predicted blood pressure information, wherein the distance between any of the maximum and minimum values ​​is greater than N times the wavelength corresponding to the fundamental frequency, wherein N is greater than 0.8 and less than 0.95; and obtaining the systolic and diastolic blood pressure based on the at least one maximum and the at least one minimum value.

[0010] For example, in at least one example of the blood pressure prediction method provided in the embodiments of this disclosure, the initial model, the blood pressure prediction model and the first prediction model are all of the type of multiple linear regression model; or the initial model, the blood pressure prediction model and the first prediction model are all of the type of one-dimensional convolutional neural network model.

[0011] For example, in at least one example of the blood pressure prediction method provided in the above embodiments of this disclosure, the method further includes: if the error between the predicted blood pressure value of the first user and the actual blood pressure value of the first user is greater than a threshold, then the first prediction model is optimized and trained using the predicted blood pressure value and the actual blood pressure value.

[0012] For example, in at least one example of the blood pressure prediction method provided in the embodiments of this disclosure, the signal acquisition module includes at least one pressure sensor; acquiring the pressure signal of the first user's wrist using the signal acquisition module includes: acquiring the pressure signal of the first user's wrist using the at least one pressure sensor; extracting the sensor with the largest amplitude among the at least one pressure sensor as the target sensor, wherein the pressure signal of the first user's wrist includes at least the pressure information acquired by the target sensor.

[0013] At least one embodiment of this disclosure provides a model training method, comprising: acquiring sample data from multiple users, wherein the sample data includes pulse information and blood pressure information corresponding to the pulse information; training an initial model based on the sample data from the multiple users to obtain a blood pressure prediction model; acquiring sample data from a first user; and training the blood pressure prediction model based on the sample data from the first user to obtain a first prediction model corresponding to the first user.

[0014] At least one embodiment of this disclosure provides a blood pressure prediction method, comprising: collecting first pulse information of a first user using a first blood pressure device, and determining a first blood pressure prediction value of the first user based on the first pulse information using a first prediction model corresponding to the first user; receiving a blood pressure measurement value input by the first user using the first blood pressure device, and updating the model parameters of the first prediction model based on the blood pressure measurement value and the first blood pressure prediction value, wherein the blood pressure measurement value is obtained by the first user using a second blood pressure device; collecting second pulse information of the first user using the first blood pressure device, and determining a second blood pressure prediction value of the first user based on the second pulse information using the updated first prediction model.

[0015] For example, in a blood pressure prediction method provided in at least one example of the above embodiments of this disclosure, the first blood pressure device receives a blood pressure measurement value input by the first user and updates the model parameters of the first prediction model based on the blood pressure measurement value and the first blood pressure prediction value, including: in response to the first user triggering a correction mode of the first blood pressure device, receiving a blood pressure measurement value input by the first user, and updating the model parameters of the first prediction model based on the blood pressure measurement value and the first blood pressure prediction value.

[0016] For example, in a blood pressure prediction method provided in at least one example of the above embodiments of this disclosure, the first user triggers a calibration mode of the first blood pressure device, including: in response to a first operation by the first user, the first blood pressure device outputs a first prompt message, wherein the first prompt message is used to indicate that the calibration mode has been triggered.

[0017] For example, in a blood pressure prediction method provided in at least one example of the above embodiments of this disclosure, the first user triggers a correction mode of the first blood pressure device, including: in response to the first blood pressure prediction value being outside a predetermined blood pressure range, the first blood pressure device outputs a second prompt message, wherein the predetermined blood pressure range is determined based on the first user's historical blood pressure prediction values, and the second prompt message is used to prompt the first user to activate the correction mode; in response to a second operation by the first user, the first blood pressure device triggers the correction mode.

[0018] For example, in a blood pressure prediction method provided in at least one example of the above embodiments of this disclosure, in response to the first user triggering a calibration mode of the first blood pressure device, receiving a blood pressure measurement value input by the first user includes: in response to the first user triggering the calibration mode, the first blood pressure device outputs a third prompt message, wherein the third prompt message is used to prompt the first user to input the blood pressure measurement value; and receiving the blood pressure measurement value input by the first user.

[0019] At least one embodiment of this disclosure provides a blood pressure prediction device, including a first acquisition unit, a prediction unit, and a determination unit. The first acquisition unit is configured to acquire pulse information of a first user; the prediction unit is configured to input the pulse information into a first prediction model corresponding to the first user to obtain predicted blood pressure information corresponding to the pulse information; the determination unit is configured to determine the predicted blood pressure value of the first user based on the predicted blood pressure information; wherein, the first prediction model is obtained by training a blood pressure prediction model based on sample data of the first user, and the blood pressure prediction model is obtained by training an initial model based on sample data of multiple users.

[0020] At least one embodiment of this disclosure provides a model training apparatus, including a second acquisition unit, a first training unit, a third acquisition unit, and a second training unit. The second acquisition unit is configured to acquire sample data from multiple users, wherein the sample data includes pulse information and predicted blood pressure information corresponding to the pulse information. The first training unit is configured to train an initial model based on the sample data from the multiple users to obtain a blood pressure prediction model. The third acquisition unit is configured to acquire sample data from a first user. The second training unit is configured to train the blood pressure prediction model based on the sample data from the first user to obtain a first prediction model corresponding to the first user.

[0021] At least one embodiment of this disclosure provides an electronic device, including a processor; a memory storing one or more computer program modules; wherein the one or more computer program modules are configured to be executed by the processor to implement the blood pressure prediction method or model training method provided in any embodiment of this disclosure.

[0022] At least one embodiment of this disclosure provides a computer-readable storage medium storing non-transitory computer-readable instructions, which, when executed by a computer, can implement the blood pressure prediction method or model training method provided in any embodiment of this disclosure. Attached Figure Description

[0023] To more clearly illustrate the technical solutions of the embodiments of this disclosure, the accompanying drawings of the embodiments will be briefly described below. Obviously, the drawings described below only relate to some embodiments of this disclosure and are not intended to limit this disclosure.

[0024] Figure 1 shows a flowchart of a blood pressure prediction method provided in at least one embodiment of the present disclosure;

[0025] Figure 2 shows a schematic diagram of a cuffless blood pressure prediction device provided in at least one embodiment of the present disclosure;

[0026] Figure 3A shows a schematic diagram of a pressure signal provided by at least one embodiment of the present disclosure;

[0027] Figure 3B shows a schematic diagram of the pressure signal after noise reduction processing provided in at least one embodiment of the present disclosure;

[0028] Figure 3C shows a schematic diagram of a frequency domain signal sequence provided in at least one embodiment of the present disclosure;

[0029] Figure 3D shows a schematic diagram of the frequency domain principal components extracted from the frequency domain signal sequence of Figure 3C according to at least one embodiment of the present disclosure;

[0030] Figure 3E shows a schematic diagram of a frequency domain signal sequence of a predicted blood pressure waveform provided in at least one embodiment of the present disclosure;

[0031] Figure 3F shows a schematic diagram of the time-domain information of a blood pressure waveform provided in at least one embodiment of the present disclosure.

[0032] Figure 4A illustrates a schematic diagram of the model optimization process provided in at least one embodiment of this disclosure;

[0033] Figure 4B illustrates a schematic diagram of the model prediction process provided in at least one embodiment of this disclosure;

[0034] Figure 5A shows a schematic diagram of a model training method provided in at least one embodiment of the present disclosure;

[0035] Figure 5B shows a schematic diagram of another blood pressure prediction method provided by at least one embodiment of the present disclosure;

[0036] Figure 6 shows a schematic block diagram of a blood pressure prediction device provided in at least one embodiment of the present disclosure;

[0037] Figure 7 shows a schematic block diagram of a model training apparatus provided in at least one embodiment of the present disclosure;

[0038] Figure 8A shows a schematic block diagram of an electronic device provided in at least one embodiment of the present disclosure;

[0039] Figure 8B shows a schematic block diagram of another electronic device provided in at least one embodiment of the present disclosure; and

[0040] Figure 9 shows a schematic diagram of a computer-readable storage medium provided in at least one embodiment of the present disclosure. Detailed Implementation

[0041] To make the objectives, technical solutions, and advantages of the embodiments of this disclosure clearer, the technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this disclosure. All other embodiments obtained by those skilled in the art based on the described embodiments of this disclosure without creative effort are within the scope of protection of this disclosure.

[0042] Unless otherwise defined, the technical or scientific terms used in this disclosure shall have the ordinary meaning understood by one of ordinary skill in the art to which this disclosure pertains. The terms “first,” “second,” and similar terms used in this disclosure do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Similarly, the terms “an,” “a,” or “the,” and similar terms do not indicate a quantity limitation, but rather indicate the presence of at least one. The terms “including,” “comprising,” or “containing,” and similar terms mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. The terms “connected,” “linked,” or similar terms are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. The terms “upper,” “lower,” “left,” and “right,” etc., are used only to indicate relative positional relationships, and these relative positional relationships may change accordingly when the absolute position of the described objects changes.

[0043] Although cuffless blood pressure monitors can continuously monitor and measure blood pressure and are comfortable to wear, there are still some issues that need improvement. For example, they are not personalized enough and cannot balance accuracy and personalization.

[0044] At least one embodiment of this disclosure provides a blood pressure prediction method, a blood pressure prediction device, an electronic device, and a computer-readable storage medium. The blood pressure prediction method includes: acquiring pulse information of a first user; inputting the pulse information into a first prediction model corresponding to the first user to obtain predicted blood pressure information corresponding to the pulse information; and determining a predicted blood pressure value for the first user based on the predicted blood pressure information; wherein the first prediction model is obtained by training a blood pressure prediction model based on sample data of the first user, and the blood pressure prediction model is obtained by training an initial model based on sample data of multiple users.

[0045] This blood pressure prediction method first trains a blood pressure prediction model using multi-user sample data. Then, it uses the sample data of the first user to perform transfer learning and training on the blood pressure prediction model to obtain a first prediction model corresponding to the first user. This first prediction model not only ensures accuracy and stability but also offers greater personalization, making it more suitable for the first user. Therefore, based on this first prediction model, personalized predictions can be made for the first user, ensuring that the prediction results better match the user's actual situation, balancing accuracy and personalization.

[0046] Figure 1 shows a flowchart of a blood pressure prediction method provided by at least one embodiment of the present disclosure.

[0047] As shown in Figure 1, the method may include steps S110 to S130.

[0048] Step S110: Obtain the pulse information of the first user.

[0049] Step S120: Input the pulse information into the first prediction model corresponding to the first user to obtain the predicted blood pressure information corresponding to the pulse information. The first prediction model is obtained by training the blood pressure prediction model based on the sample data of the first user. The blood pressure prediction model is obtained by training the initial model based on the sample data of multiple users.

[0050] Step S130: Determine the predicted blood pressure value for the first user based on the predicted blood pressure information.

[0051] For example, the blood pressure prediction method of this disclosure can be used in a cuffless blood pressure prediction device (or a cuffless blood pressure measurement device).

[0052] Figure 2 shows a schematic diagram of a cuffless blood pressure prediction device provided in at least one embodiment of the present disclosure.

[0053] As shown in Figure 2, the cuffless blood pressure prediction device 210 includes a signal acquisition module 211 and a microcontroller 212. The signal acquisition module 211 may include a pressure sensor array 2111, and the microcontroller 212 includes a signal processing module 2121 and a prediction algorithm module 2122. The cuffless blood pressure prediction device 210 can communicate with a computer 220 via wired or wireless means. The computer 220 includes an algorithm optimization module 221. The functions of each of the above modules will be described in the following embodiments. The cuffless blood pressure prediction device shown in Figure 2 is only one example of this disclosure. In other embodiments, any component of the cuffless blood pressure prediction device may be replaced with other forms, or may include more or fewer modules. For example, the microcontroller 212 may be replaced with other data processing devices, or the algorithm optimization module 221 may be included in the cuffless blood pressure prediction device.

[0054] For example, in step S110, the pulse information may be the pulse information at the user's wrist, which can be detected using the pressure sensor array 2111.

[0055] For example, obtaining the pulse information of the first user includes: acquiring the pressure signal of the first user's wrist using a signal acquisition module; converting the pressure signal into a frequency domain signal sequence using a fast Fourier transform; and extracting the amplitude information with a frequency within a predetermined range from the frequency domain signal sequence as the pulse information of the first user.

[0056] For example, the signal acquisition module includes at least one pressure sensor. Acquiring the pressure signal of a first user's wrist using the signal acquisition module includes: acquiring the pressure signal of the first user's wrist using at least one pressure sensor; and extracting the sensor with the largest amplitude among the at least one pressure sensor as the target sensor, wherein the pressure signal of the first user's wrist includes at least the pressure information acquired by the target sensor.

[0057] For example, the hardware device used in the signal acquisition module 211 is a pressure sensor array 2111. This array can be a 3×3 or 2×3 arrangement and can operate robustly in various environments. By placing the pressure sensor array at the radial artery in the wrist, it can sense the vibration of the pulse accompanying the heartbeat. The sensor with the largest amplitude can be extracted as the one closest to the radial artery for subsequent processing, thus allowing the extracted pressure signal to more accurately reflect the human pulse.

[0058] Figure 3A shows a schematic diagram of a pressure signal provided by at least one embodiment of the present disclosure.

[0059] For example, as shown in Figure 3A, a pressure sensor array can continuously acquire pressure signals. In one example, the pressure sensor array can sample the signal over a 20-second period, maintaining a sampling rate of 1000Hz, resulting in a pressure signal containing 20,000 sampling points. The pressure sensor array then transmits the acquired signal to a microcontroller for further signal processing. The values ​​of the parameters mentioned above are merely examples; in practical applications, they can be set to other values. For instance, the sampling duration can be set to less than or more than 20 seconds, such as 30 seconds; the sampling rate can also be set to less than or more than 1000Hz.

[0060] For example, converting a pressure signal into a frequency domain signal sequence using Fast Fourier Transform can include: denoising the pressure signal to obtain a denoised pressure signal; downsampling the denoised pressure signal to obtain a downsampled pressure signal; and converting the downsampled pressure signal into a frequency domain signal sequence using Fast Fourier Transform.

[0061] For example, the hardware device corresponding to the signal processing module 2121 is a microcontroller in the cuffless blood pressure prediction device. The function of the signal processing module 2121 can be to perform noise reduction processing on the signal from the signal acquisition module and perform downsampling processing, which can reduce the number of sampling points to be processed, and then perform Fast Fourier Transform (FFT) processing to obtain the frequency domain signal, and extract the low frequency components of the frequency domain signal to obtain the frequency domain principal components as the pulse information of the first user.

[0062] For example, the noise reduction process may include: (1) using a Butterworth low-pass filter to filter out high-frequency noise components, which are often generated by electronic components; and (2) using a Butterworth high-pass filter to filter out low-frequency noise components, which are often generated by human movement or breathing rhythms. A Butterworth filter is a type of electronic filter, also known as a maximum flatness filter. After noise reduction, the frequency response curve in the passband is maximized flatness without ripple, while it gradually decreases to zero in the stopband. Figure 3B shows a schematic diagram of the pressure signal after noise reduction processing provided by at least one embodiment of this disclosure.

[0063] For example, if the sampling rate is 1000Hz, the downsampling process can be to downsample the signal obtained from the original 1000Hz sampling rate to 50Hz. After downsampling, the signal with a duration of 20s contains 1000 sampling points.

[0064] For example, the Fast Fourier Transform (FFT) is an efficient and fast computation method that uses a computer to calculate the Discrete Fourier Transform (DFT). The FFT can convert a time-domain signal of a certain length (such as a signal directly acquired by a sensor) into a frequency-domain signal sequence, which contains the amplitude and phase information of the original signal at each frequency domain. Performing the FFT on a 20-second time-domain signal containing 1000 sampling points yields a frequency-domain signal sequence of length 1000. Figure 3C shows a schematic diagram of a frequency-domain signal sequence provided by at least one embodiment of this disclosure. As shown in Figure 3C, this frequency-domain signal sequence, for example, contains amplitude and phase information within the frequency range of 0–25 Hz, where Frequency represents frequency and Amplitude represents amplitude. Then, the amplitude of the first few points is extracted from this frequency-domain signal sequence. The extracted amplitude information corresponds to a lower frequency and can be used as the principal components of the pulse signal in the frequency domain. Figure 3D illustrates a schematic diagram of extracting the frequency domain principal component from the frequency domain signal sequence of Figure 3C according to at least one embodiment of this disclosure. As shown in Figure 3D, for example, the amplitude of the first 200 points extracted from the frequency domain signal sequence contains amplitude information within the frequency range of 0 to 10 Hz. This information is used as the frequency domain principal component, which can be used as the pulse information of the first user and input into the prediction algorithm module. By extracting amplitude information within a predetermined frequency range from the frequency domain signal sequence, the information density can be increased and the computational efficiency can be improved.

[0065] For example, the initial model, blood pressure prediction model, and first prediction model are all of the type of multiple linear regression model; or the initial model, blood pressure prediction model, and first prediction model are all of the type of one-dimensional convolutional neural network model.

[0066] For example, before implementing a blood pressure prediction method, a first prediction model corresponding to the first user can be pre-trained based on sample data. The training process includes, for example, acquiring sample data from multiple users, wherein the sample data includes pulse information and blood pressure information corresponding to the pulse information; training an initial model based on the sample data from multiple users to obtain a blood pressure prediction model; acquiring sample data from the first user; and training the blood pressure prediction model based on the sample data from the first user to obtain a first prediction model corresponding to the first user.

[0067] For example, each user's sample data may include the user's pulse information and the corresponding blood pressure information. The pulse information in each user's sample data may be obtained by processing the user's wrist pressure signal through noise reduction, downsampling, fast Fourier transform, and information extraction (extracting the principal components in the frequency domain).

[0068] For example, blood pressure information in the sample data can be frequency domain information of the blood pressure waveform (hereinafter also referred to as the frequency domain signal sequence of the blood pressure waveform or blood pressure frequency domain information) or time domain information of the blood pressure waveform (hereinafter also referred to as the time domain signal sequence of the blood pressure waveform or blood pressure time domain information). For example, blood pressure information in the sample data can be detected using an accurate blood pressure measurement device (such as an invasive blood pressure monitoring device or a cuff blood pressure measurement device).

[0069] For example, during the training of a blood pressure prediction model, multiple weight parameters in the initial model are updated based on sample data from multiple users, while keeping the value of at least one target weight node in the initial model unchanged, to obtain the blood pressure prediction model. During the training of the first prediction model, this at least one target weight node in the blood pressure prediction model is updated based on sample data from the first user, to obtain the first prediction model.

[0070] For example, multiple users can include the same or different types of people. When multiple users include different types of people, they can include people of different age groups, people of different genders, etc. When multiple users include the same type of people, the type of these multiple users can be the same as the type of the first user.

[0071] For example, the initial model can be a multiple linear regression model, and the trained blood pressure prediction model and the first prediction model are also multiple linear regression models. The following embodiments use a multiple linear regression model as an example, but this disclosure is not limited to this, and the initial model can also be other types of models.

[0072] For example, if the initial model is a multiple linear regression model, the at least one target weight node can be at least one constant zero weight node in the multiple linear regression model, that is, the values ​​of each weight in the node remain zero during the training of the blood pressure prediction model and do not participate in the training of the blood pressure prediction model.

[0073] For example, in the training phase of a blood pressure prediction model, a multiple linear regression model is built and trained using data from multiple individuals to form a blood pressure prediction model (or initial blood pressure prediction model). This model can predict blood pressure waveforms in a population and can therefore be called a general prediction model. M personalized nodes with constant zero weights can be reserved in the multiple linear regression model. These M personalized nodes with constant zero weights do not participate in the training of the blood pressure prediction model but are reserved for use during model optimization. Here, M is an integer greater than or equal to 1; for example, M can be equal to 2, meaning 2 personalized nodes with constant zero weights are reserved. The training process of the blood pressure prediction model can be executed by the prediction algorithm module 2122 shown in Figure 2.

[0074] For example, the linear regression weight matrix W of the initial model can be:

[0075] Among them, w 11 ~w n+2,n It has multiple weights.

[0076] For example, during the training phase of a blood pressure prediction model, two personalized nodes with constant zero weights are reserved in W. W can be modified as follows:

[0077] Thus, the mapping formula from pulse information x (such as the frequency domain principal component obtained by processing pressure signals through noise reduction, downsampling, FFT, etc.) to blood pressure information y (such as the frequency domain information of blood pressure waveform) is as follows:

[0078] Here, x and y can both be row vectors of length n, where n is an integer greater than 1. x and y are shown below:

[0079] Where, x1~x n Let y1 be the amplitude of the pulse signal x, and y2 be the amplitude of the pulse signal x. n This refers to multiple blood pressure values ​​in the blood pressure information y.

[0080] For example, at the start of blood pressure prediction model training, each weight in the weight matrix W shown in formula (2) has an initial value. The model is trained in multiple rounds. In each round, based on formula (3), the pulse information x from the sample data and the weight matrix W are used to predict the corresponding blood pressure information y. At this time, the blood pressure information y can be the frequency domain information of the blood pressure waveform. If the blood pressure information in the sample data is the frequency domain information of the blood pressure waveform, the model error can be calculated based on the predicted blood pressure waveform frequency domain information y and the frequency domain information of the blood pressure waveform in the sample data. If the blood pressure information in the sample data is the time domain information of the blood pressure waveform, the predicted blood pressure waveform frequency domain information y can be converted into the time domain information y of the blood pressure waveform by performing an inverse fast Fourier transform, and then the model error can be calculated based on the predicted blood pressure waveform time domain information y and the time domain information of the blood pressure waveform in the sample data. In some embodiments, the blood pressure information in the sample data can also be blood pressure values ​​measured using an accurate blood pressure measuring device. In this case, the predicted systolic and diastolic blood pressure can be further extracted from the predicted blood pressure waveform time-domain information y. Then, the model error is calculated based on the predicted blood pressure waveform time-domain information y and the predicted systolic / diastolic blood pressure, compared with the blood pressure information in the sample data (such as the actual blood pressure waveform time-domain information and the actual systolic / diastolic blood pressure). After calculating the model error, the gradient of the other weights (excluding reserved nodes) is calculated using this model error. The error is compensated according to the gradient, and the other weights (excluding reserved nodes) are updated. After multiple rounds of training, the blood pressure waveform predicted by the model is made close to the actual blood pressure waveform in the sample data. After training, a blood pressure prediction model is obtained.

[0081] For example, during the training phase of the first prediction model, the blood pressure prediction model is trained in a personalized manner using sample data from the first user to form the first prediction model. This first prediction model is suitable for making more accurate blood pressure predictions for a single user (such as the first user), and therefore can also be called a dedicated prediction model or a personalized prediction model.

[0082] For example, the training process of the first prediction model can be executed by the algorithm optimization module 221 shown in Figure 2. The algorithm optimization module refers to personalized algorithm optimization for the user. Before the user wears the cuffless blood pressure prediction device in this embodiment for the first time, the algorithm optimization can be performed on a computer, and the optimized personalized prediction algorithm model can be written into the microcontroller 212 of the cuffless blood pressure prediction device.

[0083] For example, before a user uses a cuffless blood pressure prediction device for the first time, the user wears the device to collect a certain amount of pressure signal data. Simultaneously, the actual blood pressure waveform and systolic / diastolic pressure values ​​measured by the accurate blood pressure measurement device are obtained, and the blood pressure prediction model is trained individually. Based on the small amount of data from individual users, the blood pressure prediction model trained with a large amount of data is combined for transfer learning and further training.

[0084] For example, the transfer learning and training process is as follows: Activate the two reserved personalized nodes with constant zero weights in the multiple linear regression model, making their weights variable; fix the other nodes of the multiple linear regression model, which do not participate in training. Perform multiple predictions using the blood pressure prediction model, and calculate the mean squared error (RMSE) by comparing the predicted blood pressure waveform with the actual blood pressure waveform measured by an accurate blood pressure measuring device each time. The RMSE is calculated using the following formula:

[0085] Where n is a positive number greater than 1, It can be a predicted blood pressure value, y i It can be the actual blood pressure value.

[0086] For example, the mean square error is propagated backward along the multiple linear regression model, the gradient of the weights of the personalized nodes is calculated, the error is compensated according to the gradient, and the weights of the personalized nodes are updated. The backward propagation and gradient compensation operations are performed multiple times to make the blood pressure waveform predicted by the prediction algorithm model close to the actual blood pressure waveform, and finally a personalized prediction model (i.e., the first prediction model) is obtained.

[0087] For example, in the above embodiment, the weight of the reserved personalized node is set to 0. However, this disclosure is not limited to this. In other embodiments, the weight of the reserved node can also be set to other values ​​besides 0, such as 1.

[0088] According to the blood pressure prediction method of this disclosure, a blood pressure prediction model is first trained using multi-user sample data, and then the blood pressure prediction model is transferred and trained using the sample data of the first user to obtain a first prediction model corresponding to the first user. The first prediction model obtained in this way can not only guarantee accuracy and stability, but also be more personalized and more suitable for the first user.

[0089] According to the blood pressure prediction method of this disclosure, the first prediction model adopts a multiple linear regression model, which has a fast calculation speed and is relatively easy to integrate into wearable devices.

[0090] For example, in some embodiments, the first prediction model may also employ a one-dimensional convolutional neural network model, which has a higher computational cost than a multiple linear regression model.

[0091] For example, the trained first prediction model can be stored in the data processing device (such as a microcontroller) of the cuffless blood pressure prediction device. In application, the signal acquisition device of the cuffless blood pressure prediction device can collect the pressure signal from the first user's wrist and process it to obtain pulse information. This pulse information is then input into the first prediction model to obtain its output, which includes predicted blood pressure information. This predicted blood pressure information can be, for example, the frequency domain information of the predicted blood pressure waveform. Figure 3E shows a schematic diagram of the frequency domain signal sequence of the predicted blood pressure waveform provided in at least one embodiment of this disclosure. As shown in Figure 3E, the gray solid line represents the true value of the blood pressure waveform frequency domain information, and the black dashed line represents the predicted value. After obtaining the predicted blood pressure waveform frequency domain information, the predicted blood pressure value for the first user can be obtained based on this information. Thus, based on the first prediction model, personalized predictions can be made for the first user, making the prediction results more consistent with the actual situation of the first user, balancing accuracy and personalization.

[0092] For example, both pulse information and predicted blood pressure information are frequency domain signal sequences. In step S130, the blood pressure value of the first user is obtained based on the predicted blood pressure information, including: converting the predicted blood pressure information into a time domain signal sequence using the inverse of the fast Fourier transform; and extracting the systolic and diastolic blood pressure from the time domain signal sequence using a peak-finding algorithm as the predicted blood pressure value for the first user. Step S130 can be executed, for example, by the prediction algorithm module 2122 shown in Figure 2.

[0093] For example, the inverse fast Fourier transform (IFFT) can be calculated using the following formula:

[0094] For example, Figure 3F shows a schematic diagram of the time-domain information of a blood pressure waveform provided in at least one embodiment of this disclosure. As shown in Figure 3F, the frequency domain signal sequence shown in Figure 3E is converted into a time-domain signal sequence using the inverse of the fast Fourier transform. Here, the vertical axis represents blood pressure, the horizontal axis represents time, the gray solid line represents the true value of the blood pressure waveform time-domain information, and the black dashed line represents the predicted value of the blood pressure waveform time-domain information.

[0095] For example, extracting systolic and diastolic blood pressure from a time-domain signal sequence using a peak-finding algorithm includes: extracting the maximum value other than a predetermined amplitude from the frequency-domain signal sequence of predicted blood pressure information as the fundamental frequency; filtering at least one maximum and at least one minimum value from the time-domain signal sequence of predicted blood pressure information, wherein the distance between any maximum and any minimum value is greater than N times the wavelength corresponding to the fundamental frequency, where N is greater than 0.8 and less than 0.95; and obtaining systolic and diastolic blood pressure based on at least one maximum and at least one minimum value.

[0096] For example, the function of the prediction algorithm module includes taking the frequency domain principal components obtained by the signal processing module as input to the first prediction model, calculating the predicted value of the blood pressure waveform through the first prediction model, and extracting the systolic and diastolic blood pressure from the blood pressure waveform.

[0097] For example, the algorithm used in the first prediction model could be multiple linear regression. The input frequency domain principal components are one-dimensional multivariate vectors, and the output is also a one-dimensional multivariate vector, representing the predicted values ​​of the frequency domain components of the blood pressure waveform. The predicted values ​​of the blood pressure waveform's frequency domain components output by the multiple linear regression model can be further transformed using an inverse fast Fourier transform to convert the frequency domain signal back into a time domain signal, i.e., the time domain predicted values ​​of the blood pressure waveform. Simultaneously, a peak-finding algorithm extracts the systolic and diastolic blood pressure from the blood pressure waveform.

[0098] For example, the peak-finding algorithm can extract systolic and diastolic blood pressure from a blood pressure waveform. The algorithm can be as follows: extract the maximum value other than 0Hz amplitude from the predicted values ​​of the frequency domain components of the blood pressure waveform, i.e., the fundamental frequency of the signal; then, filter the maxima and minima in the time domain information of the blood pressure waveform, with the filtering condition that the distance between the maxima / minimum values ​​should be greater than 0.9 times the wavelength corresponding to the fundamental frequency, obtaining multiple maxima / minimum values. Averaging these multiple maxima and minima separately yields the systolic and diastolic blood pressure. Using the peak-finding algorithm, systolic and diastolic blood pressure can be extracted from the predicted blood pressure waveform simply and quickly, with fast calculation speed and high accuracy.

[0099] For example, the blood pressure prediction method of this disclosure may further include: if the error between the predicted blood pressure value of the first user and the actual blood pressure value of the first user is greater than a threshold, then the first prediction model is optimized and trained using the predicted blood pressure value and the actual blood pressure value, and the threshold value can be determined according to medical standards.

[0100] For example, after predicting the diastolic and systolic blood pressure of the first user, the error between the predicted systolic / diastolic blood pressure and the actual systolic / diastolic blood pressure can be calculated. If the error is greater than the medical standard, the process is returned to transfer learning and training to further optimize the personalized prediction algorithm model (i.e., the first prediction model).

[0101] For example, in some embodiments, in addition to pressure sensors, the signal acquisition module can also use sensors such as ECG (Electrocardiogram), PPG (Photoplethysmography), and patch strain sensors. The detection signals of these sensors can be used as inputs to the model for model training and prediction, which can make the prediction model more adaptable and stable.

[0102] Figure 3 illustrates a schematic diagram of the model optimization process provided in at least one embodiment of this disclosure.

[0103] As shown in Figure 3, the signal acquisition module can obtain the non-invasive sensor measurement signal of the first user, such as the pressure signal detected by a pressure sensor array. Then, the signal processing module performs noise reduction, filtering, and FFT processing on the detected signal to obtain the non-invasive signal spectrum. The prediction algorithm module uses a multiple linear regression model (such as a blood pressure prediction model) stored in the module to obtain the predicted blood pressure frequency domain information, performs inverse FFT operation to obtain the predicted blood pressure waveform value, and then uses a peak finding algorithm to extract the predicted SBP (systolic blood pressure) / DBP (diastolic blood pressure). The module also acquires the invasive sensor measurement signal of the first user and extracts the true SBP / DBP value from it. The algorithm optimization module can use the predicted blood pressure waveform value and the predicted SBP / DBP value, as well as the invasive sensor measurement signal and the true SBP / DBP value, to calculate the model error, and use the model error to optimize the blood pressure prediction model to obtain the first prediction model.

[0104] Figure 4 shows a schematic diagram of the model prediction process provided in at least one embodiment of the present disclosure.

[0105] As shown in Figure 4, after training the first prediction model, it can be used to predict the blood pressure of the first user. The signal acquisition module can obtain the non-invasive sensor measurement signal of the first user, such as the pressure signal detected by a pressure sensor array. Then, the signal processing module performs noise reduction, filtering, and FFT on the detected signal to obtain the non-invasive signal spectrum. The predicted blood pressure frequency domain information is obtained using the first prediction model stored in the prediction algorithm module, and inverse FFT operation is performed to obtain the predicted blood pressure waveform value. Then, the peak finding algorithm is used to extract the predicted value of SBP (systolic blood pressure) / DBP (diastolic blood pressure), which is used as the blood pressure prediction value for the first user.

[0106] For example, this disclosure provides at least one example of a blood pressure prediction algorithm based on a cuffless pressure sensor array signal. This algorithm can predict and diagnose blood pressure based on the pressure signal from the skin surface of the human wrist, thus enabling hypertension measurement. The hardware included in this disclosure is a cuffless blood pressure prediction device, comprising a pressure sensor array and a microcontroller, among other hardware modules, and may utilize a computer. Functionally, it comprises four main modules: a signal acquisition module, a signal processing module, a prediction algorithm module, and an algorithm optimization module. The signal acquisition module acquires signals from the wrist pressure sensor array. The signal processing module extracts, denoises, filters, and performs frequency domain conversion on the pressure signals. The prediction algorithm module uses data from multiple users during the development phase to build and train a general algorithm model, forming an initial prediction algorithm model. Then, in the algorithm optimization module, machine learning model transfer training is performed on an individual to form a personalized blood pressure prediction algorithm model.

[0107] For example, at least one embodiment of this disclosure provides a model training method, and Figure 5 shows a schematic diagram of the model training method provided by at least one embodiment of this disclosure.

[0108] As shown in Figure 5A, the model training method includes steps S511 to S514.

[0109] Step S511: Obtain sample data from multiple users, including pulse information and the corresponding predicted blood pressure information.

[0110] Step S512: Train the initial model based on sample data from multiple users to obtain a blood pressure prediction model.

[0111] Step S513: Obtain sample data of the first user.

[0112] Step S514: Train the blood pressure prediction model based on the sample data of the first user to obtain the first prediction model corresponding to the first user.

[0113] For the training method of this model, please refer to the relevant description in the above embodiments, which will not be repeated here.

[0114] This disclosure provides at least one embodiment of another blood pressure prediction method. Figure 5B shows a schematic diagram of another blood pressure prediction method provided by at least one embodiment of this disclosure.

[0115] As shown in Figure 5B, the blood pressure prediction method includes steps S521 to S523.

[0116] Step S521: Collect the first pulse information of the first user using the first blood pressure device, and determine the first blood pressure prediction value of the first user based on the first pulse information using the first prediction model corresponding to the first user.

[0117] Step S522: The first blood pressure device receives the blood pressure measurement value input by the first user, and updates the model parameters of the first prediction model based on the blood pressure measurement value and the first blood pressure prediction value. The blood pressure measurement value is obtained by the first user using the second blood pressure device.

[0118] Step S523: Collect the second pulse information of the first user using the first blood pressure device, and use the updated first prediction model to determine the second blood pressure prediction value of the first user based on the second pulse information.

[0119] For example, the first blood pressure device can be the cuffless blood pressure prediction device shown in Figure 2, which can execute steps S110 to S130 in any of the above embodiments. The first blood pressure device stores a trained first prediction model, which is obtained by training a blood pressure prediction model based on sample data from a first user. The blood pressure prediction model is obtained by training an initial model based on sample data from multiple users. For example, the second blood pressure device can be another blood pressure measurement device different from the first blood pressure device, such as an invasive blood pressure monitoring device or a cuff-type blood pressure measurement device.

[0120] For example, a first user can use a second blood pressure device to obtain an accurate blood pressure measurement, and within a predetermined time range (e.g., 10 minutes or 30 minutes), the first blood pressure device can predict the user's first blood pressure value. The first user can input the blood pressure measurement value detected by the second blood pressure device into the first blood pressure device, for example, through a keyboard, touchscreen display, or voice input. The first blood pressure device can then determine the error between the blood pressure measurement value and the first blood pressure prediction value, and use this error to update the model parameters of the first prediction model, resulting in an updated first prediction model. In this way, during the use of the first blood pressure device, the prediction model of the first blood pressure device can be updated according to the user's needs, making the prediction results of the first blood pressure device more accurate and better reflecting the user's current blood pressure status.

[0121] For example, the first blood pressure device may include a display device, which may be a display screen or a touch screen. In some embodiments, the first blood pressure device may also include a voice input / output device.

[0122] For example, in step S522, the first blood pressure device receives the blood pressure measurement value input by the first user and updates the model parameters of the first prediction model based on the blood pressure measurement value and the first blood pressure prediction value, including: in response to the first user triggering the calibration mode of the first blood pressure device, receiving the blood pressure measurement value input by the first user and updating the model parameters of the first prediction model based on the blood pressure measurement value and the first blood pressure prediction value.

[0123] For example, in some embodiments, a first user triggering a calibration mode of a first blood pressure device includes: in response to a first operation by the first user, the first blood pressure device outputting a first prompt message, wherein the first prompt message is used to indicate that the calibration mode has been triggered.

[0124] For example, the first blood pressure device may have a button for triggering a calibration mode. The first operation can be pressing the button to input a request to trigger the calibration mode. After receiving the first operation, the first blood pressure device can display a first mode prompt message on the display screen in the form of a pop-up window to inform the user that the calibration mode has been triggered. Alternatively, the first blood pressure device may have a voice recognition module. When the voice recognition module is activated, it can receive the user's voice input. The first operation can be, for example, a request from the user to activate the calibration mode via voice input. After receiving the first operation, the first blood pressure device can output a first mode prompt message via voice to inform the user that the calibration mode has been triggered.

[0125] For example, when the blood pressure value measured by the first user using the first blood pressure device differs significantly from its historical blood pressure value, the calibration function of the first blood pressure device can be triggered. Or, when the blood pressure value measured by the first user using the first blood pressure device differs significantly from the blood pressure value measured by other blood pressure measuring devices, the calibration function of the first blood pressure device can be triggered.

[0126] For example, in other embodiments, the first user triggers the calibration mode of the first blood pressure device, including: in response to the first blood pressure prediction value being outside a predetermined blood pressure range, the first blood pressure device outputs a second prompt message, wherein the predetermined blood pressure range is determined based on the first user's historical blood pressure prediction values, and the second prompt message is used to prompt the first user to activate the calibration mode; in response to the first user's second operation, the first blood pressure device triggers the calibration mode.

[0127] For example, the predetermined blood pressure range includes a predetermined systolic blood pressure range and a predetermined diastolic blood pressure range. The first blood pressure device can determine the predetermined systolic blood pressure range and predetermined diastolic blood pressure range for the first user based on multiple historical systolic blood pressures and multiple historical diastolic blood pressures, respectively. For example, if the minimum value of multiple historical systolic blood pressures is A1 and the maximum value is A2, the predetermined systolic blood pressure range can be [A1-k, A2+k], where k is a value greater than or equal to 0; if the minimum value of multiple historical diastolic blood pressures is B1 and the maximum value is B2, the predetermined systolic blood pressure range can be [B1-k, B2+k]. When the first blood pressure prediction value exceeds the predetermined blood pressure range, there may be measurement inaccuracies. In this case, the user can be prompted to activate the calibration mode to calibrate the first blood pressure device.

[0128] For example, in other embodiments, in response to a first user triggering a calibration mode of the first blood pressure device, receiving a blood pressure measurement value input by the first user includes: in response to the first user triggering the calibration mode, the first blood pressure device outputs a third prompt message, wherein the third prompt message is used to prompt the first user to input a blood pressure measurement value; and receiving the blood pressure measurement value input by the first user.

[0129] For example, when the calibration mode of the first blood pressure device is triggered, the display device can show input prompts in the form of pop-ups or similar means to prompt the user to input the blood pressure measurement value obtained using the second blood pressure device. Alternatively, when the calibration mode of the first blood pressure device is triggered, the voice output device can output voice prompts to prompt the user to input the blood pressure measurement value.

[0130] Figure 6 shows a schematic block diagram of a blood pressure prediction device 600 provided in at least one embodiment of the present disclosure.

[0131] For example, as shown in FIG6, the blood pressure prediction device 600 includes a first acquisition unit 610, a prediction unit 620, and a determination unit 630.

[0132] The first acquisition unit 610 is configured to acquire the pulse information of a first user. The first acquisition unit 610 may, for example, execute step S110 as described in FIG1.

[0133] The prediction unit 620 is configured to input the pulse information into a first prediction model corresponding to the first user to obtain predicted blood pressure information corresponding to the pulse information. The prediction unit 620 may, for example, execute step S120 as described in FIG1.

[0134] The determining unit 630 is configured to determine the predicted blood pressure value for the first user based on the predicted blood pressure information. The determining unit 630 may, for example, execute step S130 as described in FIG1.

[0135] For example, the function of the first acquisition unit 610 can correspond to the signal acquisition module and signal processing module in Figure 2, and the functions of the prediction unit 620 and the determination unit 630 can correspond to the prediction algorithm module in Figure 2.

[0136] Figure 7 shows a schematic block diagram of a model training apparatus 700 provided in at least one embodiment of the present disclosure.

[0137] As shown in Figure 7, the model training device 700 includes a second acquisition unit 710, a first training unit 720, a third acquisition unit 730, and a second training unit 740.

[0138] The second acquisition unit 710 is configured to acquire sample data from multiple users, wherein the sample data includes pulse information and predicted blood pressure information corresponding to the pulse information. The second acquisition unit 710 may, for example, execute step S511 as described in FIG5.

[0139] The first training unit 720 is configured to train an initial model based on sample data from the multiple users to obtain a blood pressure prediction model. The first training unit 720 may, for example, execute step S512 as described in Figure 5.

[0140] The third acquisition unit 730 is configured to acquire sample data of the first user. The third acquisition unit 730 may, for example, execute step S513 as described in FIG5.

[0141] The second training unit 740 is configured to train the blood pressure prediction model based on the sample data of the first user, thereby obtaining a first prediction model corresponding to the first user. The third acquisition unit 730 may, for example, execute step S514 as described in FIG5.

[0142] For example, the function of the first training unit 720 can correspond to the prediction algorithm module in Figure 2, and the function of the second training unit 740 can correspond to the algorithm optimization module in Figure 2.

[0143] For example, each of the above units / modules can be hardware, software, firmware, or any feasible combination thereof. For example, each of the above units / modules can be a dedicated or general-purpose circuit, chip, or device, or a combination of a processor and memory. The embodiments of this disclosure do not limit the specific implementation of the above units.

[0144] For example, each of the above units / modules may include code and programs stored in memory; the processor may execute the code and programs to implement some or all of the functions of the units / modules as described above. For example, each unit / module may be a dedicated hardware device used to implement some or all of the functions of the units / modules as described above. For example, each unit / module may be a circuit board or a combination of multiple circuit boards used to implement the functions as described above. In embodiments of this disclosure, the circuit board or the combination of multiple circuit boards may include: (1) one or more processors; (2) one or more non-transitory memories connected to the processor; and (3) processor-executable firmware stored in memory.

[0145] It should be noted that in the embodiments of this disclosure, each unit of the blood pressure prediction device 600 corresponds to each step of the aforementioned blood pressure prediction method. For details regarding the specific functions of the blood pressure prediction device 600, please refer to the relevant descriptions of the blood pressure prediction method; these will not be repeated here. The components and structure of the blood pressure prediction device 600 shown in Figure 7 are merely exemplary and not limiting. The blood pressure prediction device 600 may also include other components and structures as needed. The blood pressure prediction device 600 may include more or fewer circuits or units, and the connection relationships between the various circuits or units are not limited and can be determined according to actual needs. The specific configuration of each circuit or unit is not limited; it can be constructed from analog devices, digital chips, or other suitable methods according to circuit principles. The same applies to the model training device 700.

[0146] These components are interconnected via a bus system and / or other forms of connection mechanisms (not shown). For example, at least one embodiment of this disclosure also provides an electronic device including a processor and a memory storing one or more computer program modules. The one or more computer program modules are configured to be executed by the processor to implement the blood pressure prediction method or model training method described above.

[0147] Figure 8A is a schematic block diagram of an electronic device provided in some embodiments of this disclosure. As shown in Figure 8A, the electronic device 800 includes a processor 810 and a memory 820. Furthermore, the electronic device 800 may also include a signal acquisition device. The memory 820 stores non-transitory computer-readable instructions (e.g., one or more computer program modules). The processor 810 is used to execute the non-transitory computer-readable instructions, which, when executed by the processor 810, perform one or more steps in the blood pressure prediction method or model training method described above. The memory 820 and the processor 810 can be interconnected via a bus system and / or other forms of connection mechanisms (not shown). For specific implementations and related explanations of the various steps of the blood pressure prediction method or model training method, please refer to the embodiments of the blood pressure prediction method or model training method described above; repetitions will not be repeated here.

[0148] It should be noted that the components of the electronic device 800 shown in Figure 8A are merely exemplary and not restrictive. Depending on the actual application requirements, the electronic device 800 may also have other components.

[0149] For example, the processor 810 and the memory 820 can communicate with each other directly or indirectly.

[0150] For example, processor 810 and memory 820 can communicate via a network. The network can include wireless networks, wired networks, and / or any combination of wireless and wired networks. Processor 810 and memory 820 can also communicate with each other via a system bus, and this disclosure is not limiting in this regard.

[0151] For example, the processor 810 and memory 820 can be located on the server side (or in the cloud).

[0152] For example, processor 810 can control other components in electronic device 800 to perform desired functions. For example, processor 810 can be a central processing unit (CPU), a graphics processing unit (GPU), or other form of processing unit with data processing capabilities and / or program execution capabilities. For example, the central processing unit (CPU) can be an x86 or ARM architecture. Processor 810 can be a general-purpose processor or a special-purpose processor, and can control other components in electronic device 800 to perform desired functions.

[0153] For example, memory 820 may include any combination of one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and / or cache memory. Non-volatile memory may include, for example, read-only memory (ROM), hard disk, erasable programmable read-only memory (EPROM), portable compact disc read-only memory (CD-ROM), USB memory, flash memory, etc. One or more computer program modules may be stored on the computer-readable storage medium, and processor 810 may run one or more computer program modules to implement various functions of electronic device 800. Various application programs and various data, as well as various data used and / or generated by the application programs, may also be stored in the computer-readable storage medium.

[0154] It should be noted that, in the embodiments of this disclosure, the specific functions and technical effects of the electronic device 800 can be referred to the description of the blood pressure prediction method above, and will not be repeated here.

[0155] Figure 8B is a schematic block diagram of another electronic device provided in some embodiments of this disclosure. This electronic device 900 is, for example, suitable for implementing the blood pressure prediction method or model training method provided in the embodiments of this disclosure. The electronic device 900 can be a terminal device, etc. It should be noted that the electronic device 900 shown in Figure 8B is merely an example and does not impose any limitations on the functionality and scope of use of the embodiments of this disclosure.

[0156] As shown in Figure 8B, the electronic device 900 may include a processing unit (e.g., a central processing unit, a graphics processing unit, etc.) 910, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 920 or a program loaded from a storage device 980 into a random access memory (RAM) 930. The RAM 930 also stores various programs and data required for the operation of the electronic device 900. The processing unit 910, the ROM 920, and the RAM 930 are interconnected via a bus 940. An input / output (I / O) interface 950 is also connected to the bus 940.

[0157] Typically, the following devices can be connected to the I / O interface 950: input devices 960 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 970 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 980 including, for example, magnetic tapes, hard disks, etc.; and communication devices 990. The communication device 990 allows the electronic device 900 to communicate wirelessly or wiredly with other electronic devices to exchange data. Although Figure 8B shows an electronic device 900 with various devices, it should be understood that it is not required to implement or possess all the devices shown, and the electronic device 900 may alternatively implement or possess more or fewer devices.

[0158] For example, according to embodiments of this disclosure, the aforementioned blood pressure prediction method or model training method can be implemented as a computer software program. For instance, embodiments of this disclosure include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program including program code for performing the aforementioned blood pressure prediction method or model training method. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 990, or installed from a storage device 980, or installed from a ROM 920. When the computer program is executed by the processing device 910, the functions defined in the blood pressure prediction method or model training method provided in embodiments of this disclosure can be implemented.

[0159] At least one embodiment of this disclosure also provides a computer-readable storage medium storing non-transitory computer-readable instructions that, when executed by a computer, can implement the blood pressure prediction method or model training method described above.

[0160] Figure 9 is a schematic diagram of a storage medium provided in some embodiments of this disclosure. As shown in Figure 9, the storage medium 1000 stores non-transitory computer-readable instructions 1010. For example, when the non-transitory computer-readable instructions 1010 are executed by a computer, one or more steps in the blood pressure prediction method described above are performed.

[0161] For example, the storage medium 1000 can be applied in the aforementioned electronic device 800. For example, the storage medium 1000 can be the memory 820 in the electronic device 800 shown in FIG8A. For example, for relevant descriptions of the storage medium 1000, please refer to the corresponding description of the memory 820 in the electronic device 800 shown in FIG8A, which will not be repeated here.

[0162] The above description is merely a preferred embodiment of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features disclosed in this disclosure that have similar functions.

[0163] Furthermore, while the operations are described in a specific order, this should not be construed as requiring these operations to be performed in the specific order shown or in a sequential order. In certain environments, multitasking and parallel processing may be advantageous. Similarly, while several specific implementation details are included in the above discussion, these should not be construed as limiting the scope of this disclosure. Certain features described in the context of individual embodiments may also be implemented in combination in a single embodiment. Conversely, various features described in the context of a single embodiment may also be implemented individually or in any suitable sub-combination in multiple embodiments.

[0164] Although the subject matter has been described using language specific to structural features and / or methodological logic, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or actions described above. Rather, the specific features and actions described above are merely illustrative examples of implementing the claims.

[0165] The following points should be noted regarding this disclosure:

[0166] (1) The accompanying drawings of the embodiments of this disclosure only involve the structures involved in the embodiments of this disclosure. Other structures can be referred to the general design.

[0167] (2) Where there is no conflict, the embodiments of this disclosure and the features in the embodiments can be combined with each other to obtain new embodiments.

[0168] The above description is merely a specific embodiment of this disclosure, but the scope of protection of this disclosure is not limited thereto. The scope of protection of this disclosure should be determined by the scope of protection of the claims.

Claims

1. A method for predicting blood pressure, comprising: Obtain the pulse information of the first user; The pulse information is input into the first prediction model corresponding to the first user to obtain the predicted blood pressure information corresponding to the pulse information; Based on the predicted blood pressure information, the predicted blood pressure value for the first user is determined; The first prediction model is obtained by training a blood pressure prediction model based on the sample data of the first user, and the blood pressure prediction model is obtained by training an initial model based on the sample data of multiple users.

2. The method according to claim 1, wherein, During the training process of the blood pressure prediction model, multiple weight parameters in the initial model are updated based on the sample data of the multiple users, while keeping the value of at least one target weight node in the initial model unchanged, so as to obtain the blood pressure prediction model. During the training process of the first prediction model, at least one target weight node in the blood pressure prediction model is updated based on the sample data of the first user to obtain the first prediction model.

3. The method according to claim 2, wherein, When the initial model is a multiple linear regression model, the at least one target weight node is at least one constant zero weight node in the multiple linear regression model.

4. The method according to any one of claims 1-3, wherein, Obtain the pulse information of the first user, including: The pressure signal from the first user's wrist is acquired using a signal acquisition module; The pressure signal is converted into a frequency domain signal sequence using a fast Fourier transform. The amplitude information with a frequency within a predetermined range is extracted from the frequency domain signal sequence and used as the pulse information of the first user.

5. The method according to claim 4, wherein, The pressure signal is converted into a frequency domain signal sequence using a Fast Fourier Transform, including: The pressure signal is subjected to noise reduction processing to obtain a noise-reduced pressure signal; The noise-reduced pressure signal is downsampled to obtain a downsampled pressure signal. The downsampled pressure signal is converted into the frequency domain signal sequence using a fast Fourier transform.

6. The method according to any one of claims 1-4, wherein, Both the pulse information and the predicted blood pressure information are frequency domain signal sequences; Among these steps, determining the predicted blood pressure value for the first user based on predicted blood pressure information includes: The predicted blood pressure information is converted into a time-domain signal sequence using the inverse of the fast Fourier transform. Systolic and diastolic blood pressures are extracted from the time-domain signal sequence using a peak-finding algorithm and used as the predicted blood pressure values ​​for the first user.

7. The method according to claim 6, wherein, The systolic and diastolic blood pressures are extracted from the time-domain signal sequence using a peak-finding algorithm, including: Extract the maximum value other than the predetermined amplitude from the frequency domain signal sequence of the predicted blood pressure information, and use it as the fundamental frequency; At least one maximum and at least one minimum value are obtained by filtering from the time-domain signal sequence of the predicted blood pressure information, wherein the distance between any maximum and any minimum value is greater than N times the wavelength corresponding to the fundamental frequency, wherein N is greater than 0.8 and less than 0.95; The systolic blood pressure and the diastolic blood pressure are obtained based on the at least one maximum value and the at least one minimum value.

8. The method according to any one of claims 1-7, wherein, The initial model, the blood pressure prediction model, and the first prediction model are all of the type of multiple linear regression models; or The initial model, the blood pressure prediction model, and the first prediction model are all one-dimensional convolutional neural network models.

9. The method according to any one of claims 1-8, further comprising: If the error between the predicted blood pressure value and the actual blood pressure value of the first user is greater than a threshold, then the first prediction model is optimized and trained using the predicted blood pressure value and the actual blood pressure value.

10. The method according to claim 4, wherein, The signal acquisition module includes at least one pressure sensor; The signal acquisition module is used to acquire pressure signals from the first user's wrist, including: The pressure signal of the first user's wrist is acquired using the at least one pressure sensor; The sensor with the largest amplitude among the at least one pressure sensor is selected as the target sensor, wherein the pressure signal of the first user's wrist includes at least the pressure information collected by the target sensor.

11. A model training method, comprising: Acquire sample data from multiple users, wherein the sample data includes pulse information and blood pressure information corresponding to the pulse information; The initial model is trained based on the sample data of the multiple users to obtain a blood pressure prediction model; Obtain sample data from the first user; The blood pressure prediction model is trained based on the sample data of the first user to obtain the first prediction model corresponding to the first user.

12. A method for predicting blood pressure, comprising: The first pulse information of the first user is collected using a first blood pressure device, and the first prediction model corresponding to the first user is used to determine the first blood pressure prediction value of the first user based on the first pulse information. The first blood pressure device receives the blood pressure measurement value input by the first user, and updates the model parameters of the first prediction model based on the blood pressure measurement value and the first blood pressure prediction value, wherein the blood pressure measurement value is obtained by the first user using the second blood pressure device; The second pulse information of the first user is collected using a first blood pressure device, and the second blood pressure prediction value of the first user is determined based on the second pulse information using an updated first prediction model.

13. The method according to claim 12, wherein, The first blood pressure device receives the blood pressure measurement value input by the first user, and updates the model parameters of the first prediction model based on the blood pressure measurement value and the first predicted blood pressure value, including: In response to the first user triggering the calibration mode of the first blood pressure device, the system receives the blood pressure measurement value input by the first user and updates the model parameters of the first prediction model based on the blood pressure measurement value and the first blood pressure prediction value.

14. The method according to claim 13, wherein, The first user triggers the calibration mode of the first blood pressure device, including: In response to the first user's first operation, the first blood pressure device outputs a first prompt message, wherein the first prompt message is used to indicate that the correction mode has been triggered.

15. The method according to claim 13 or 14, wherein, The first user triggers the calibration mode of the first blood pressure device, including: In response to the first blood pressure prediction value being outside the predetermined blood pressure range, the first blood pressure device outputs a second prompt message, wherein the predetermined blood pressure range is determined based on the first user's historical blood pressure prediction values, and the second prompt message is used to prompt the first user to activate the correction mode; In response to the second operation by the first user, the first blood pressure device triggers the calibration mode.

16. The method according to any one of claims 13-15, wherein, In response to the first user triggering the calibration mode of the first blood pressure device, the device receives the blood pressure measurement value input by the first user, including: In response to the first user triggering the calibration mode, the first blood pressure device outputs a third prompt message, wherein the third prompt message is used to prompt the first user to input the blood pressure measurement value; Receive the blood pressure measurement value input by the first user.

17. A blood pressure prediction device, comprising: The first acquisition unit is configured to acquire the pulse information of the first user; The prediction unit is configured to input the pulse information into a first prediction model corresponding to the first user to obtain predicted blood pressure information corresponding to the pulse information; The determining unit is configured to determine the predicted blood pressure value of the first user based on the predicted blood pressure information; The first prediction model is obtained by training a blood pressure prediction model based on the sample data of the first user, and the blood pressure prediction model is obtained by training an initial model based on the sample data of multiple users.

18. A model training device, comprising: The second acquisition unit is configured to acquire sample data from multiple users, wherein the sample data includes pulse information and predicted blood pressure information corresponding to the pulse information; The first training unit is configured to train an initial model based on sample data from the multiple users to obtain a blood pressure prediction model; The third acquisition unit is configured to acquire sample data of the first user; The second training unit is configured to train the blood pressure prediction model based on the sample data of the first user, so as to obtain the first prediction model corresponding to the first user.

19. An electronic device comprising: processor; Memory, which stores one or more computer program modules; The one or more computer program modules are configured to be executed by the processor to implement the method according to any one of claims 1-16.

20. A computer-readable storage medium storing non-transitory computer-readable instructions that, when executed by a computer, can implement the method of any one of claims 1-16.

Images