Blood pressure prediction method and electronic device

Abstract

This application relates to the field of electronic device technology, providing a blood pressure prediction method and electronic device. The method includes: acquiring the user's current photoplethysmography (PPG) signal in response to a blood pressure prediction command; extracting features from the PPG signal to determine multiple blood pressure-related features corresponding to the PPG signal; fusing the multiple blood pressure-related features to generate target input features corresponding to the PPG signal; and inputting the target input features into a blood pressure prediction model to determine the user's diastolic and systolic blood pressure. Therefore, by integrating a PPG signal detection module into the electronic device to acquire PPG signals and predicting blood pressure based on multiple blood pressure-related signal features corresponding to the PPG signal, users can conveniently measure their blood pressure at any time, improving the convenience of blood pressure measurement, and ensuring the accuracy and reliability of blood pressure prediction.

Description

Technical Field

[0001] This application belongs to the field of electronic equipment technology, and in particular relates to a blood pressure prediction method, electronic equipment, chip system and computer-readable storage medium. Background Technology

[0002] Blood pressure refers to the pressure of blood pumped by the heart within the blood vessels. If blood pressure cannot be maintained at a normal level, problems such as insufficient nutrition to the body and organ dysfunction can occur. In particular, when blood pressure is too high, arteriosclerosis can develop, leading to various diseases. Blood pressure is a key factor in determining health risks, and regular blood pressure checks are necessary, especially in the context of increasing life expectancy and an aging population. Traditional methods of blood pressure measurement use mercury sphygmomanometers or electronic sphygmomanometers. However, the disadvantage of traditional methods is that individuals need to go to the hospital or purchase measuring instruments to obtain blood pressure data, which is not only inconvenient but also costly.

[0003] With the widespread adoption and rapid development of electronic devices such as smartphones, smartwatches, and fitness trackers, their functions are becoming increasingly powerful. Providing blood pressure measurement functionality in these portable electronic devices, allowing users to conveniently measure their blood pressure anytime, anywhere, is a problem that urgently needs to be solved. Summary of the Invention

[0004] This application provides a blood pressure prediction method, electronic device, chip system, computer-readable storage medium, and computer program product, which can solve the problem of the urgent need to provide blood pressure measurement function in commonly used portable electronic devices, so that users can conveniently measure blood pressure at any time, thereby improving the convenience of blood pressure measurement.

[0005] In a first aspect, embodiments of this application provide a blood pressure prediction method, comprising: in response to a blood pressure prediction command, acquiring the user's current photoplethysmography (PPG) signal; extracting features from the PPG signal to determine multiple blood pressure-related features corresponding to the PPG signal; fusing the multiple blood pressure-related features to generate target input features corresponding to the PPG signal; and inputting the target input features into a blood pressure prediction model to determine the user's diastolic and systolic blood pressure.

[0006] Thus, by acquiring the user's PPG signal through a PPG sensor in an electronic device and extracting various signal features from this signal to identify multiple blood pressure-related signal features, these features are then fused as input to a blood pressure prediction model to predict the user's blood pressure. Therefore, by integrating a PPG signal detection module into the electronic device for PPG signal acquisition and predicting blood pressure based on the multiple blood pressure-related signal features corresponding to the PPG signal, users can conveniently measure their blood pressure at any time, improving the ease of blood pressure measurement, while also ensuring the accuracy and reliability of blood pressure prediction.

[0007] In one possible implementation of the first aspect, the aforementioned blood pressure-related features include at least two of the following features: morphological features, spectral features, temporal features, and invasive artery blood pressure (ABP) features corresponding to the PPG signal.

[0008] Thus, by extracting various signal features from the user's PPG signal, the morphological features, spectral features, time-domain features, ABP features, and other signal features related to the user's blood pressure are obtained. The fused features of these features are then input into the blood pressure prediction model to predict diastolic and systolic blood pressure. Since these signal features are closely related to the user's blood pressure characteristics, blood pressure prediction using these features further improves the accuracy and reliability of blood pressure prediction.

[0009] Optionally, in another possible implementation of the first aspect, the blood pressure prediction model includes a backbone network, a first subnetwork, and a second subnetwork; correspondingly, inputting the target input features into the blood pressure prediction model to determine the user's diastolic and systolic blood pressure includes:

[0010] The target input features are input into the backbone network to generate shared output features corresponding to the PPG signal;

[0011] The shared output features are input into the first subnetwork to determine diastolic pressure;

[0012] The shared output features are input into the second subnetwork to determine the systolic pressure.

[0013] Therefore, since blood pressure measurement in the human body usually requires the prediction of diastolic and systolic blood pressure, and the relationships between diastolic and systolic blood pressure and different features may be different, a blood pressure prediction model can be constructed based on a multi-task model. After shared feature extraction through the backbone network, different sub-networks can be used to predict diastolic and systolic blood pressure separately. Based on the two different prediction tasks of diastolic and systolic blood pressure prediction, further adaptive feature extraction can be performed through different sub-networks to obtain more accurate diastolic and systolic blood pressure prediction results, thereby further improving the accuracy and reliability of blood pressure prediction.

[0014] Optionally, in another possible implementation of the first aspect, the backbone network includes N multi-scale modules, where N is a positive integer; correspondingly, the above-mentioned inputting the target input features into the backbone network to generate shared output features corresponding to the PPG signal includes:

[0015] The target input features are input into the backbone network, and the multi-scale modules are used to extract multi-scale features from the target input features in sequence, so as to perform multi-scale feature extraction on the target input features N times and generate shared output features.

[0016] In this way, by stacking multiple multi-scale modules in the backbone network of the blood pressure prediction model to increase the network depth and extract multi-scale features from the target input features, the model's ability to learn key information from the input features is improved, thereby further enhancing the accuracy and reliability of blood pressure prediction.

[0017] Optionally, in another possible implementation of the first aspect, the target input features are input into the backbone network, and multi-scale feature extraction is performed sequentially on the target input features using various multi-scale modules to perform N multi-scale feature extractions on the target input features, generating shared output features, including:

[0018] The i-th intermediate feature is input into the (i+1)-th multi-scale module, and the (i+1)-th multi-scale feature extraction is performed on the target input feature to generate the (i+1)-th intermediate feature. Here, the 0th intermediate feature is the target input feature, the Nth intermediate feature is the shared output feature, and i is an integer greater than or equal to 0 and less than N.

[0019] In this way, by performing residual connections with the original input features of each multi-scale module during each multi-scale feature extraction, the model's learning ability of the original input features is improved, preventing the loss of key information in the original input features during blood pressure prediction, thereby further improving the accuracy and reliability of blood pressure prediction.

[0020] Optionally, in another possible implementation of the first aspect, each of the above multi-scale modules includes multiple convolutional kernels of different sizes, a first attention module, and a residual connection module; correspondingly, the above-mentioned inputting the i-th intermediate feature into the (i+1)-th multi-scale module to perform the (i+1)-th multi-scale feature extraction on the target input feature to generate the (i+1)-th intermediate feature includes:

[0021] The i-th intermediate feature is input into each convolution kernel of the (i+1)-th multi-scale module for convolution processing to generate convolution features of different sizes corresponding to the i-th intermediate feature.

[0022] Each convolutional feature is input into the first attention module of the (i+1)th multi-scale module to generate the first attention weights corresponding to each convolutional feature, and the convolutional features are fused according to the first attention weights corresponding to each convolutional feature to generate the (i+1)th multi-scale fused feature.

[0023] The i-th intermediate feature and the (i+1)-th multi-scale fused feature are input into the residual connection module of the (i+1)-th multi-scale module to generate the (i+1)-th intermediate feature.

[0024] In this way, multi-scale feature extraction is performed on the target input features by using multiple convolutional kernels of different sizes, and the weights corresponding to features of different scales are learned through the attention mechanism. Furthermore, the original input features of each multi-scale module are fused with the extracted multi-scale features through residual connections. This allows the attention mechanism to assign higher weights to features that are highly correlated with blood pressure, and the residual connections to improve the network's learning ability for input features, thereby further improving the accuracy and reliability of blood pressure prediction.

[0025] Optionally, in another possible implementation of the first aspect, the size of each of the above convolutional kernels is determined based on the interval of the PPG signal.

[0026] Thus, by determining the size of each convolutional kernel in the multi-scale module based on the interval of the PPG signal, the multi-scale features extracted by the multi-scale module are better matched with the key features of the PPG signal, thereby further improving the accuracy and reliability of blood pressure prediction.

[0027] Optionally, in another possible implementation of the first aspect, the first sub-network includes a first hybrid attention module, a first convolutional layer, and a first fully connected layer, and the second sub-network includes a second hybrid attention module, a second convolutional layer, and a second fully connected layer.

[0028] Optionally, in another possible implementation of the first aspect, the above-mentioned inputting shared output features into the first sub-network to determine diastolic pressure includes:

[0029] The shared output features are input into the first hybrid attention module to determine the first hybrid attention weight matrix corresponding to the shared output features;

[0030] The shared output features are weighted according to the first hybrid attention weight matrix to generate the first sub-output features corresponding to the shared output features;

[0031] The first sub-output feature is input into the first convolutional layer and the first fully connected layer to determine the diastolic pressure.

[0032] Thus, after extracting the shared output features corresponding to the PPG signal, the shared output features are input into the first sub-network for predicting diastolic blood pressure to further extract features from the shared output features. The hybrid attention mechanism is used to determine the hybrid attention weight matrix of the shared output features in the diastolic blood pressure prediction task, so that features with higher correlation to diastolic blood pressure have higher weights. This results in the first output sub-features containing more key information related to diastolic blood pressure, thereby further improving the accuracy and reliability of diastolic blood pressure prediction.

[0033] Optionally, in another possible implementation of the first aspect, the above-mentioned inputting shared output features into a second sub-network to determine the systolic pressure includes:

[0034] The shared output features are input into the second hybrid attention module to determine the second hybrid attention weight matrix corresponding to the shared output features;

[0035] The shared output features are weighted according to the second hybrid attention weight matrix to generate the second sub-output features corresponding to the shared output features;

[0036] The second sub-output feature is input into the second convolutional layer and the second fully connected layer to determine the shrinkage pressure.

[0037] Thus, after extracting the shared output features corresponding to the PPG signal, the shared output features are input into the second sub-network for predicting systolic blood pressure to further extract features from the shared output features. The hybrid attention mechanism is used to determine the hybrid attention weight matrix of the shared output features in the systolic blood pressure prediction task, so that features with higher correlation to systolic blood pressure have higher weights. This makes the second output sub-features contain more key information related to systolic blood pressure, thereby further improving the accuracy and reliability of diastolic blood pressure prediction.

[0038] Secondly, embodiments of this application provide a blood pressure prediction device, comprising: a first acquisition module, configured to acquire a user's current PPG signal in response to a blood pressure prediction command; a first determination module, configured to extract features from the PPG signal to determine multiple blood pressure-related features corresponding to the PPG signal; a first generation module, configured to fuse the multiple blood pressure-related features to generate target input features corresponding to the PPG signal; and a second determination module, configured to input the target input features into a blood pressure prediction model to determine the user's diastolic and systolic blood pressure.

[0039] Thus, by acquiring the user's PPG signal through a PPG sensor in an electronic device and extracting various signal features from this signal to identify multiple blood pressure-related signal features, these features are then fused as input to a blood pressure prediction model to predict the user's blood pressure. Therefore, by integrating a PPG signal detection module into the electronic device for PPG signal acquisition and predicting blood pressure based on the multiple blood pressure-related signal features corresponding to the PPG signal, users can conveniently measure their blood pressure at any time, improving the ease of blood pressure measurement, while also ensuring the accuracy and reliability of blood pressure prediction.

[0040] In one possible implementation of the second aspect, the aforementioned blood pressure-related features include at least two of the following features: morphological features, spectral features, time-domain features, and ABP features corresponding to the PPG signal.

[0041] Thus, by extracting various signal features from the user's PPG signal, the morphological features, spectral features, time-domain features, ABP features, and other signal features related to the user's blood pressure are obtained. The fused features of these features are then input into the blood pressure prediction model to predict diastolic and systolic blood pressure. Since these signal features are closely related to the user's blood pressure characteristics, blood pressure prediction using these features further improves the accuracy and reliability of blood pressure prediction.

[0042] Optionally, in another possible implementation of the second aspect, the blood pressure prediction model includes a backbone network, a first subnetwork, and a second subnetwork; correspondingly, the second determining module includes:

[0043] The first generation unit is used to input the target input features into the backbone network to generate shared output features corresponding to the PPG signal;

[0044] The first determining unit is used to input the shared output features into the first sub-network to determine diastolic pressure;

[0045] The second determining unit is used to input the shared output features into the second sub-network to determine the systolic pressure.

[0046] Therefore, since blood pressure measurement in the human body usually requires the prediction of diastolic and systolic blood pressure, and the relationships between diastolic and systolic blood pressure and different features may be different, a blood pressure prediction model can be constructed based on a multi-task model. After shared feature extraction through the backbone network, different sub-networks can be used to predict diastolic and systolic blood pressure separately. Based on the two different prediction tasks of diastolic and systolic blood pressure prediction, further adaptive feature extraction can be performed through different sub-networks to obtain more accurate diastolic and systolic blood pressure prediction results, thereby further improving the accuracy and reliability of blood pressure prediction.

[0047] Optionally, in another possible implementation of the second aspect, the backbone network comprises N multi-scale modules, where N is a positive integer; correspondingly, the first generation unit is specifically used for:

[0048] The target input features are input into the backbone network, and the multi-scale modules are used to extract multi-scale features from the target input features in sequence, so as to perform multi-scale feature extraction on the target input features N times and generate shared output features.

[0049] In this way, by stacking multiple multi-scale modules in the backbone network of the blood pressure prediction model to increase the network depth and extract multi-scale features from the target input features, the model's ability to learn key information from the input features is improved, thereby further enhancing the accuracy and reliability of blood pressure prediction.

[0050] Optionally, in another possible implementation of the second aspect, the first generating unit described above is further configured to:

[0051] The i-th intermediate feature is input into the (i+1)-th multi-scale module, and the (i+1)-th multi-scale feature extraction is performed on the target input feature to generate the (i+1)-th intermediate feature. Here, the 0th intermediate feature is the target input feature, the Nth intermediate feature is the shared output feature, and i is an integer greater than or equal to 0 and less than N.

[0052] In this way, by performing residual connections with the original input features of each multi-scale module during each multi-scale feature extraction, the model's learning ability of the original input features is improved, preventing the loss of key information in the original input features during blood pressure prediction, thereby further improving the accuracy and reliability of blood pressure prediction.

[0053] Optionally, in another possible implementation of the second aspect, each of the above multi-scale modules includes multiple convolutional kernels of different sizes, a first attention module, and a residual connection module; correspondingly, the above first generation unit is further used for:

[0054] The i-th intermediate feature is input into each convolution kernel of the (i+1)-th multi-scale module for convolution processing to generate convolution features of different sizes corresponding to the i-th intermediate feature.

[0055] Each convolutional feature is input into the first attention module of the (i+1)th multi-scale module to generate the first attention weights corresponding to each convolutional feature, and the convolutional features are fused according to the first attention weights corresponding to each convolutional feature to generate the (i+1)th multi-scale fused feature.

[0056] The i-th intermediate feature and the (i+1)-th multi-scale fused feature are input into the residual connection module of the (i+1)-th multi-scale module to generate the (i+1)-th intermediate feature.

[0057] In this way, multi-scale feature extraction is performed on the target input features by using multiple convolutional kernels of different sizes, and the weights corresponding to features of different scales are learned through the attention mechanism. Furthermore, the original input features of each multi-scale module are fused with the extracted multi-scale features through residual connections. This allows the attention mechanism to assign higher weights to features that are highly correlated with blood pressure, and the residual connections to improve the network's learning ability for input features, thereby further improving the accuracy and reliability of blood pressure prediction.

[0058] Alternatively, in another possible implementation of the second aspect, the size of each of the above convolutional kernels is determined based on the interval of the PPG signal.

[0059] Thus, by determining the size of each convolutional kernel in the multi-scale module based on the interval of the PPG signal, the multi-scale features extracted by the multi-scale module are better matched with the key features of the PPG signal, thereby further improving the accuracy and reliability of blood pressure prediction.

[0060] Optionally, in another possible implementation of the second aspect, the first sub-network includes a first hybrid attention module, a first convolutional layer, and a first fully connected layer, and the second sub-network includes a second hybrid attention module, a second convolutional layer, and a second fully connected layer.

[0061] Optionally, in another possible implementation of the second aspect, the aforementioned first determining unit is specifically used for:

[0062] The shared output features are input into the first hybrid attention module to determine the first hybrid attention weight matrix corresponding to the shared output features;

[0063] The shared output features are weighted according to the first hybrid attention weight matrix to generate the first sub-output features corresponding to the shared output features;

[0064] The first sub-output feature is input into the first convolutional layer and the first fully connected layer to determine the diastolic pressure.

[0065] Thus, after extracting the shared output features corresponding to the PPG signal, the shared output features are input into the first sub-network for predicting diastolic blood pressure to further extract features from the shared output features. The hybrid attention mechanism is used to determine the hybrid attention weight matrix of the shared output features in the diastolic blood pressure prediction task, so that features with higher correlation to diastolic blood pressure have higher weights. This results in the first output sub-features containing more key information related to diastolic blood pressure, thereby further improving the accuracy and reliability of diastolic blood pressure prediction.

[0066] Optionally, in another possible implementation of the second aspect, the aforementioned second determining unit is specifically used for:

[0067] The shared output features are input into the second hybrid attention module to determine the second hybrid attention weight matrix corresponding to the shared output features;

[0068] The shared output features are weighted according to the second hybrid attention weight matrix to generate the second sub-output features corresponding to the shared output features;

[0069] The second sub-output feature is input into the second convolutional layer and the second fully connected layer to determine the shrinkage pressure.

[0070] Thus, after extracting the shared output features corresponding to the PPG signal, the shared output features are input into the second sub-network for predicting systolic blood pressure to further extract features from the shared output features. The hybrid attention mechanism is used to determine the hybrid attention weight matrix of the shared output features in the systolic blood pressure prediction task, so that features with higher correlation to systolic blood pressure have higher weights. This makes the second output sub-features contain more key information related to systolic blood pressure, thereby further improving the accuracy and reliability of diastolic blood pressure prediction.

[0071] Thirdly, embodiments of this application provide an electronic device, including: one or more processors, and a memory; the memory is coupled to one or more processors, the memory is used to store computer program code, the computer program code including computer instructions, and the one or more processors call the computer instructions to cause the electronic device to perform the blood pressure prediction method as described above.

[0072] Fourthly, embodiments of this application provide a chip system applied to an electronic device, the chip system including one or more processors, the one or more processors being used to invoke computer instructions to cause the electronic device to perform the blood pressure prediction method as described above.

[0073] Fifthly, embodiments of this application provide a computer-readable storage medium including instructions that, when executed on an electronic device, cause the electronic device to perform the blood pressure prediction method as described above.

[0074] Sixthly, embodiments of this application provide a computer program product that, when run on an electronic device, causes the electronic device to perform the blood pressure prediction method as described above.

[0075] The technical effects achieved by the second, third, fourth, fifth, and sixth aspects mentioned above are similar to those achieved by the corresponding technical means in the first aspect mentioned above, and will not be repeated here. Attached Figure Description

[0076] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0077] Figure 1 This is a schematic flowchart of a blood pressure prediction method provided in an embodiment of this application;

[0078] Figure 2 This is a schematic flowchart of a blood pressure prediction method provided in another embodiment of this application;

[0079] Figure 3 This is a schematic diagram of the structure of a blood pressure prediction model provided in one embodiment of this application;

[0080] Figure 4 This is a schematic diagram of the structure of a multi-scale module provided in one embodiment of this application;

[0081] Figure 5 This is a schematic diagram of the structure of a hybrid attention module provided in one embodiment of this application;

[0082] Figure 6 This is an overall algorithm flowchart of a blood pressure prediction method provided in an embodiment of this application;

[0083] Figure 7 This is a schematic diagram of the structure of a blood pressure prediction device provided in an embodiment of this application;

[0084] Figure 8 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0085] The blood pressure prediction method, device, electronic equipment, chip system, storage medium, and computer program provided in this application will be described in detail below with reference to the accompanying drawings.

[0086] Please refer to Figure 1 , Figure 1 This is a flowchart illustrating a blood pressure prediction method provided in an embodiment of this application. The method may include some or all of the following:

[0087] Step 101: In response to the blood pressure prediction command, obtain the user's current PPG signal.

[0088] It should be noted that the blood pressure prediction method of this application embodiment can be executed by the blood pressure prediction device of this application embodiment. The blood pressure prediction device of this application embodiment can be configured in any electronic device to execute the blood pressure prediction method of this application embodiment. For example, the blood pressure prediction device of this application embodiment can be configured in electronic devices such as mobile phones, computers, and wearable devices to measure the user's blood pressure. As an example, the blood pressure prediction device of this application embodiment can be configured in a wearable device with PPG signal detection function (such as a smartwatch, smart bracelet, etc.) to monitor the user's blood pressure in real time, or to measure the user's blood pressure according to the user's operation instructions.

[0089] Among them, the blood pressure prediction command can refer to any command that can trigger the electronic device to activate the blood pressure prediction function.

[0090] As an example, assuming the blood pressure prediction method of this application is applied to a smartwatch, if the smartwatch can monitor the user's blood pressure at a preset frequency (such as once per hour, once per day, etc.) according to default settings or user settings, then a blood pressure prediction command can be obtained each time the measurement time is reached; if the smartwatch can monitor the user's blood pressure at preset times (such as 8:00 AM and 10:00 PM daily, etc.) according to default settings or user settings, then a blood pressure prediction command can be obtained each time the preset time is reached; if the smartwatch has a blood pressure prediction control or has an application that provides blood pressure prediction function (such as a health application, a sports application, etc.) installed, then a blood pressure prediction command can also be obtained when the user triggers the blood pressure prediction control or starts the blood pressure prediction function in the application.

[0091] It should be noted that the above examples are merely illustrative and should not be considered as limitations on this application. In actual use, the type of blood pressure prediction instruction may be related to the function of the electronic device, the application scenario of the blood pressure prediction method in the embodiments of this application, etc., and the embodiments of this application do not limit this.

[0092] Among them, photoplethysmography (PPG) is portable and inexpensive, and can be applied to blood pressure monitoring. PPG uses a light-emitting diode (LED) to illuminate human skin and measures the changes in reflected light intensity caused by blood flow. The resulting periodic waveform contains information about the pulsating changes in blood volume during the heart's beat cycle. Blood volume is largest during heart contraction and smallest during diastole. Therefore, systolic and diastolic blood pressure can be obtained by relating pulse-flow information to blood pressure. Therefore, in this embodiment, a PPG sensor can be configured in the electronic device to collect the user's PPG signal.

[0093] In this embodiment of the application, when a blood pressure prediction command is obtained, the PPG sensor in the electronic device can be driven to collect the user's current PPG signal, so as to use the user's current PPG signal to predict the user's blood pressure.

[0094] Step 102: Extract features from the PPG signal to determine multiple blood pressure-related features corresponding to the PPG signal.

[0095] Among them, blood pressure-related features may include at least two of the following features: morphological features, spectral features, time-domain features, and ABP features corresponding to PPG signals.

[0096] In this embodiment of the application, after obtaining the user's current PPG signal, feature extraction can be performed on the PPG signal to obtain various signal features related to the user's blood pressure, namely blood pressure-related features, so as to expand the diversity and richness of the input information of the subsequent blood pressure prediction model, thereby further improving the accuracy and reliability of blood pressure prediction.

[0097] As one possible implementation, morphological features can incorporate both waveform and periodic characteristics of the PPG signal. As an example, morphological features can be obtained by calculating the recurrence graph of the PPG signal, which can be derived from the trajectory distance of the time series.

[0098] As one possible implementation, the spectral characteristics can include the energy distribution and dominant frequency characteristics of the PPG signal. As an example, the spectral characteristics can be obtained by calculating the spectrum of the PPG signal; for instance, the spectrum can be obtained by calculating the short-time Fourier transform of the PPG signal.

[0099] As one possible implementation, the time-domain features can be obtained by including the PPG signal, the first derivative and the second derivative of the PPG signal. The first derivative and the second derivative of the PPG signal contain feature points related to physiological signals and blood pressure (for example, the zero point of the first derivative corresponds to the peak position of the original PPG signal), thus being related to blood pressure prediction.

[0100] As one possible implementation, ABP features are highly correlated with the blood pressure value to be predicted. This can be achieved by constructing an autoencoder that converts PPG signals into ABP features. The autoencoder can be trained by minimizing the reconstruction error. The PPG signal can then be input into the trained autoencoder for encoding to generate latent features related to ABP, i.e., ABP features. As an example, the network structure of the aforementioned autoencoder can be a U-net network structure; this application does not limit this implementation.

[0101] It should be noted that the methods for calculating various blood pressure-related features listed above are merely exemplary and should not be considered as limitations on this application. In actual use, the specific methods for extracting various blood pressure-related features can be determined according to actual needs and specific application scenarios, and the embodiments of this application do not limit this.

[0102] Step 103: Fuse multiple blood pressure-related features to generate target input features corresponding to the PPG signal.

[0103] In this embodiment of the application, after extracting multiple blood pressure-related features corresponding to the PPG signal, the various blood pressure-related features can be fused to generate target input features corresponding to the PPG signal, which are then input into the blood pressure prediction model to predict the user's diastolic and systolic blood pressure.

[0104] One possible approach is to map various blood pressure-related features to a unified dimension through convolutional and fully connected layers, thereby fusing these features to generate the target input features for a blood pressure prediction model.

[0105] Step 104: Input the target input features into the blood pressure prediction model to determine the user's diastolic and systolic blood pressure.

[0106] In this embodiment of the application, since human blood pressure can include two values, diastolic pressure and systolic pressure, a blood pressure prediction model that can predict the user's diastolic and systolic pressure can be pre-trained. After the target input features are input into the blood pressure prediction model, the blood pressure prediction model can output the user's diastolic and systolic pressure.

[0107] As one possible approach, after determining the user's diastolic and systolic blood pressure, the diastolic and systolic blood pressure can be notified to the user through screen display or voice broadcast, making it very convenient for the user to measure blood pressure.

[0108] The blood pressure prediction method provided in this application acquires the user's PPG signal through a PPG sensor in an electronic device, extracts multiple signal features from the PPG signal to identify several blood pressure-related signal features, and then fuses these multiple blood pressure-related features as input to a blood pressure prediction model to predict the user's blood pressure. Therefore, by integrating a PPG signal detection module into the electronic device to acquire PPG signals and predict blood pressure based on the multiple blood pressure-related signal features corresponding to the PPG signal, users can conveniently measure their blood pressure at any time, improving the convenience of blood pressure measurement, and ensuring the accuracy and reliability of blood pressure prediction.

[0109] Please refer to Figure 2 , Figure 2 This is a flowchart illustrating a blood pressure prediction method according to another embodiment of this application. The method may include some or all of the following:

[0110] Step 201: In response to the blood pressure prediction command, obtain the user's current PPG signal.

[0111] Step 202: Feature extraction is performed on the PPG signal to determine multiple blood pressure-related features corresponding to the PPG signal.

[0112] Step 203: Fuse multiple blood pressure-related features to generate target input features corresponding to the PPG signal.

[0113] The specific implementation process and principle of steps 201-203 above can be found in the detailed description of the above embodiments, and will not be repeated here.

[0114] Step 204: Input the target input features into the backbone network of the blood pressure prediction model to generate shared output features corresponding to the PPG signal.

[0115] As one possible implementation, such as Figure 3The diagram shows a structural schematic of a blood pressure prediction model provided in an embodiment of this application. The network structure of the blood pressure prediction model in this embodiment can be a deep neural network structure based on multi-task learning. The blood pressure prediction model can include a backbone network, a first sub-network, and a second sub-network. The backbone network is used to extract shared features for blood pressure prediction, the first sub-network is used to predict diastolic blood pressure, and the second sub-network is used to predict systolic blood pressure. Since blood pressure measurement in humans typically requires prediction of both diastolic and systolic blood pressure, and the relationships between diastolic and systolic blood pressure and different features may differ, a blood pressure prediction model can be constructed based on a multi-task model. After extracting shared features through the backbone network and obtaining shared output features, different sub-networks can be used to predict diastolic and systolic blood pressure separately. Based on these two different prediction tasks (diastolic and systolic blood pressure prediction), further adaptive feature extraction is performed through different sub-networks to obtain more accurate diastolic and systolic blood pressure prediction results, further improving the accuracy and reliability of blood pressure prediction.

[0116] Furthermore, multiple multi-scale modules can be stacked in the backbone network of the blood pressure prediction model to increase the network depth and perform multi-scale feature extraction on the target input features, thereby improving the model's ability to learn key information of the input features and further enhancing the accuracy and reliability of blood pressure prediction. Specifically, in one possible implementation of this application embodiment, the backbone network may include N multi-scale modules, where N is a positive integer; correspondingly, step 204 may include:

[0117] The target input features are input into the backbone network, and the multi-scale modules are used to extract multi-scale features from the target input features in sequence, so as to perform multi-scale feature extraction on the target input features N times and generate shared output features.

[0118] As one possible implementation, such as Figure 3 As shown, the backbone network of the blood pressure prediction model can include N multi-scale modules. After the target input features are input into the blood pressure prediction model, these N multi-scale modules can sequentially extract multi-scale features from the target input features to generate shared output features.

[0119] Furthermore, by performing residual connections with the original input features of each multi-scale module during each multi-scale feature extraction, the model's learning ability on the original input features can be improved, preventing the loss of key information in the original input features during blood pressure prediction, thereby further improving the accuracy and reliability of blood pressure prediction. That is, in one possible implementation of this application embodiment, the above-mentioned input of the target input features into the backbone network, and the sequential multi-scale feature extraction of the target input features using each multi-scale module to generate shared output features through N multi-scale feature extractions, may include:

[0120] The i-th intermediate feature is input into the (i+1)-th multi-scale module, and the (i+1)-th multi-scale feature extraction is performed on the target input feature to generate the (i+1)-th intermediate feature. Here, the 0th intermediate feature is the target input feature, the Nth intermediate feature is the shared output feature, and i is an integer greater than or equal to 0 and less than N.

[0121] As one possible implementation, after inputting the target input features into the blood pressure prediction model, the first multi-scale module of the backbone network can be used to perform the first multi-scale feature extraction on the target input features, generating the first intermediate feature. Then, the first intermediate feature can be input into the second multi-scale module, and multi-scale feature extraction can be performed on the first intermediate feature to achieve the second multi-scale feature extraction on the target input features. This process is repeated, with the intermediate feature output from the previous multi-scale module used as the input of the next multi-scale module each time, to achieve N multi-scale feature extractions on the target input features and generate shared output features.

[0122] Furthermore, multiple convolutional kernels of different sizes can be used to extract multi-scale features from the target input features. An attention mechanism is used to learn the weights corresponding to features at different scales. Residual connections are used to fuse the original input features of each multi-scale module with the extracted multi-scale features. This allows the attention mechanism to assign higher weights to features highly correlated with blood pressure, and the residual connections enhance the network's learning ability for input features, thereby further improving the accuracy and reliability of blood pressure prediction. Specifically, in one possible implementation of this application, each multi-scale module includes multiple convolutional kernels of different sizes, a first attention module, and a residual connection module. Correspondingly, the i-th intermediate feature is input into the (i+1)-th multi-scale module to perform the (i+1)-th multi-scale feature extraction on the target input features to generate the (i+1)-th intermediate feature, including:

[0123] The i-th intermediate feature is input into each convolution kernel of the (i+1)-th multi-scale module for convolution processing to generate convolution features of different sizes corresponding to the i-th intermediate feature.

[0124] Each convolutional feature is input into the first attention module of the (i+1)th multi-scale module to generate the first attention weights corresponding to each convolutional feature, and the convolutional features are fused according to the first attention weights corresponding to each convolutional feature to generate the (i+1)th multi-scale fused feature.

[0125] The i-th intermediate feature and the (i+1)-th multi-scale fused feature are input into the residual connection module of the (i+1)-th multi-scale module to generate the (i+1)-th intermediate feature.

[0126] As one possible implementation, such as Figure 4 The diagram shown is a structural schematic of a multi-scale module provided in an embodiment of this application. It is assumed that... Figure 4 The multi-scale module in the blood pressure prediction model is the (i+1)th multi-scale module. Its corresponding input feature is the i-th intermediate feature, and its corresponding output feature is the (i+1)-th intermediate feature. For ease of description, the following descriptions will use input features and output features. After inputting the input features into the multi-scale module, the input features can be convolved by K (K is an integer greater than 1) convolutional kernels of different sizes (convolution kernel 1...convolution kernel K) in the multi-scale module to generate K convolutional features of different sizes (i.e., convolutional feature 1...convolutional feature K). Then, the convolutional features can be fused and input into the first attention module of the multi-scale module to perform pooling, fully connected, and softmax connections on each convolutional feature to generate the first attention weights corresponding to each convolutional feature. Then, the convolutional features are weighted and summed according to the first attention weights corresponding to each convolutional feature to generate the multi-scale fused feature corresponding to the multi-scale module. Finally, the multi-scale fused feature is residually connected to the input features of the multi-scale module to generate the output feature corresponding to the multi-scale module.

[0127] As an example, such as Figure 4 As shown, the first attention module in a multi-scale module may include pooling layers (such as global average pooling layers), fully connected layers (such as dense layers), and softmax layers, but is not limited to these.

[0128] As one possible implementation, the size of each convolutional kernel in the multi-scale module can be determined based on the interval of the PPG signal. By determining the size of each convolutional kernel in the multi-scale module based on the interval of the PPG signal, the multi-scale features extracted by the multi-scale module can be better matched with the key features of the PPG signal, thereby further improving the accuracy and reliability of blood pressure prediction.

[0129] Step 205: Input the shared output features into the first sub-network of the blood pressure prediction model to determine diastolic blood pressure.

[0130] As one possible implementation, such as Figure 3 As shown, the first sub-network of the blood pressure prediction model may include a first hybrid attention module, a first convolutional layer, and a first fully connected layer, and the second sub-network may include a second hybrid attention module, a second convolutional layer, and a second fully connected layer.

[0131] Furthermore, after extracting the shared output features corresponding to the PPG signal, the shared output features can be input into the first sub-network used to predict diastolic blood pressure for further feature extraction. A hybrid attention mechanism is then used to determine the hybrid attention weight matrix of the shared output features in the diastolic blood pressure prediction task, so that features more correlated with diastolic blood pressure have higher weights. This results in the first output sub-features containing more key information related to diastolic blood pressure, thereby further improving the accuracy and reliability of diastolic blood pressure prediction. In one possible implementation of this application embodiment, step 205 may include:

[0132] The shared output features are input into the first hybrid attention module to determine the first hybrid attention weight matrix corresponding to the shared output features;

[0133] The shared output features are weighted according to the first hybrid attention weight matrix to generate the first sub-output features corresponding to the shared output features;

[0134] The first sub-output feature is input into the first convolutional layer and the first fully connected layer to determine the diastolic pressure.

[0135] As one possible implementation, such as Figure 5 The diagram shown is a structural schematic of a hybrid attention module provided in an embodiment of this application. Assuming... Figure 5The hybrid attention module shown is the first hybrid attention module in the first sub-network. After inputting the shared output features into the first hybrid attention module of the first sub-network, global average pooling can be performed on the shared output features along the channel dimension to determine the first channel attention weight matrix corresponding to the shared output features. Then, pointwise (PW) convolution operation is performed on the shared output features to determine the first spatial attention weight matrix corresponding to the shared output layer features. Next, the first channel attention weight matrix corresponding to the shared output features is multiplied by the first spatial attention weight matrix to generate the first hybrid attention weight matrix corresponding to the shared output features. The first hybrid attention weight matrix has the same dimension as the shared output features, that is, the first hybrid attention weight matrix contains the hybrid attention weight corresponding to each element in the shared output features. Finally, each element in the shared output features is weighted according to the first hybrid attention weight matrix to generate the first sub-output features corresponding to the shared output features. Then, the first sub-output features are input into the first convolutional layer and the first fully connected layer in the first sub-network to predict and output the user's diastolic blood pressure.

[0136] Step 206: Input the shared output features into the second sub-network of the blood pressure prediction model to determine the systolic blood pressure.

[0137] As one possible implementation, after extracting the shared output features corresponding to the PPG signal, the shared output features can be input into a second sub-network for predicting systolic blood pressure to further extract features from the shared output features. A hybrid attention mechanism can then be used to determine the hybrid attention weight matrix of the shared output features in the systolic blood pressure prediction task, giving higher weights to features more correlated with systolic blood pressure. This results in the second output sub-features containing more key information related to systolic blood pressure, thereby further improving the accuracy and reliability of diastolic blood pressure prediction. Specifically, in one possible implementation of this application embodiment, step 206 may include:

[0138] The shared output features are input into the second hybrid attention module to determine the second hybrid attention weight matrix corresponding to the shared output features;

[0139] The shared output features are weighted according to the second hybrid attention weight matrix to generate the second sub-output features corresponding to the shared output features;

[0140] The second sub-output feature is input into the second convolutional layer and the second fully connected layer to determine the shrinkage pressure.

[0141] As one possible implementation, such as Figure 5 As shown, assuming Figure 5The hybrid attention module shown is the second hybrid attention module in the second sub-network. After inputting the shared output features into the second hybrid attention module of the second sub-network, global average pooling is performed on the shared output features along the channel dimension to determine the second channel attention weight matrix corresponding to the shared output features. Then, PW convolution is performed on the shared output features to determine the second spatial attention weight matrix corresponding to the shared output layer features. Next, the second channel attention weight matrix corresponding to the shared output features is multiplied by the second spatial attention weight matrix to generate the second hybrid attention weight matrix corresponding to the shared output features. The second hybrid attention weight matrix has the same dimension as the shared output features, that is, the second hybrid attention weight matrix contains the hybrid attention weight corresponding to each element in the shared output features. Finally, each element in the shared output features is weighted according to the second hybrid attention weight matrix to generate the second sub-output features corresponding to the shared output features. Then, the second sub-output features are input into the second convolutional layer and the second fully connected layer in the second sub-network to predict and output the user's systolic pressure.

[0142] As an example, such as Figure 6 The diagram shown is an overall algorithm flowchart of a blood pressure prediction method provided in this application embodiment. After obtaining the user's current PPG signal, multiple features are first extracted from the PPG signal to generate morphological features, spectral features, temporal features, and ABP features corresponding to the PPG signal. These four features are then fused and input into the backbone network of the blood pressure prediction model. The backbone network uses N multi-scale modules to perform N multi-scale feature extractions on the input features to generate shared output features. The shared output features are then input into a first sub-network and a second sub-network, respectively. The first sub-network predicts the user's diastolic blood pressure, and the second sub-network predicts the user's systolic blood pressure. The first sub-network may include a first hybrid attention module, a first convolutional layer, and a first fully connected layer; the second sub-network may include a second hybrid attention module, a second convolutional layer, and a second fully connected layer.

[0143] It should be noted that, Figure 6 The embodiment shown is a specific implementation scheme proposed in this application. The specific implementation process and principle can be referred to in the detailed description of the above embodiment, and will not be repeated here.

[0144] The blood pressure prediction method provided in this application typically requires the prediction of both diastolic and systolic blood pressure during human blood pressure measurement. Since the relationships between diastolic and systolic blood pressure and different features may vary, a blood pressure prediction model can be constructed based on a multi-task model. After shared feature extraction through the backbone network, different sub-networks can be used to predict diastolic and systolic blood pressure separately. Based on these two different prediction tasks (diastolic and systolic blood pressure prediction), further adaptive feature extraction is performed through different sub-networks to obtain more accurate diastolic and systolic blood pressure prediction results, thereby improving the accuracy and reliability of blood pressure prediction.

[0145] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0146] Corresponding to the blood pressure prediction method described in the above embodiments, Figure 7 A structural block diagram of a blood pressure prediction device provided in an embodiment of this application is shown. For ease of explanation, only the parts related to the embodiments of this application are shown.

[0147] Reference Figure 7 The device 70 includes:

[0148] The first acquisition module 71 is used to acquire the user's current PPG signal in response to the blood pressure prediction command;

[0149] The first determining module 72 is used to extract features from the PPG signal to determine multiple blood pressure-related features corresponding to the PPG signal;

[0150] The first generation module 73 is used to fuse multiple blood pressure-related features to generate target input features corresponding to the PPG signal;

[0151] The second determining module 74 is used to input the target input features into the blood pressure prediction model to determine the user's diastolic and systolic blood pressure.

[0152] In practical use, the blood pressure prediction device provided in this application embodiment can be configured in any electronic device, especially a wearable device, to perform the aforementioned blood pressure prediction method.

[0153] The blood pressure prediction device provided in this application acquires the user's PPG signal through a PPG sensor in an electronic device, extracts various signal features from the PPG signal to determine multiple signal features related to the user's blood pressure, and then fuses these multiple blood pressure-related features as input to a blood pressure prediction model to predict the user's blood pressure. Therefore, by integrating a PPG signal detection module into the electronic device to acquire PPG signals and predict blood pressure based on multiple blood pressure-related signal features corresponding to the PPG signal, users can conveniently measure their blood pressure at any time, improving the convenience of blood pressure measurement, and ensuring the accuracy and reliability of blood pressure prediction.

[0154] In one possible implementation of this application, the aforementioned blood pressure-related features include at least two of the following features: morphological features, spectral features, time-domain features, and ABP features corresponding to the PPG signal.

[0155] Thus, by extracting various signal features from the user's PPG signal, the morphological features, spectral features, time-domain features, ABP features, and other signal features related to the user's blood pressure are obtained. The fused features of these features are then input into the blood pressure prediction model to predict diastolic and systolic blood pressure. Since these signal features are closely related to the user's blood pressure characteristics, blood pressure prediction using these features further improves the accuracy and reliability of blood pressure prediction.

[0156] Furthermore, in another possible implementation of this application, the blood pressure prediction model includes a backbone network, a first subnetwork, and a second subnetwork; correspondingly, the second determining module 74 includes:

[0157] The first generation unit is used to input the target input features into the backbone network to generate shared output features corresponding to the PPG signal;

[0158] The first determining unit is used to input the shared output features into the first sub-network to determine diastolic pressure;

[0159] The second determining unit is used to input the shared output features into the second sub-network to determine the systolic pressure.

[0160] Furthermore, in another possible implementation of this application, the backbone network includes N multi-scale modules, where N is a positive integer; correspondingly, the first generation unit is specifically used for:

[0161] The target input features are input into the backbone network, and the multi-scale modules are used to extract multi-scale features from the target input features in sequence, so as to perform multi-scale feature extraction on the target input features N times and generate shared output features.

[0162] Furthermore, in yet another possible implementation of this application, the first generating unit is further configured to:

[0163] The i-th intermediate feature is input into the (i+1)-th multi-scale module, and the (i+1)-th multi-scale feature extraction is performed on the target input feature to generate the (i+1)-th intermediate feature. Here, the 0th intermediate feature is the target input feature, the Nth intermediate feature is the shared output feature, and i is an integer greater than or equal to 0 and less than N.

[0164] Furthermore, in another possible implementation of this application, each of the above-mentioned multi-scale modules includes multiple convolutional kernels of different sizes, a first attention module, and a residual connection module; correspondingly, the above-mentioned first generation unit is also used for:

[0165] The i-th intermediate feature is input into each convolution kernel of the (i+1)-th multi-scale module for convolution processing to generate convolution features of different sizes corresponding to the i-th intermediate feature.

[0166] Each convolutional feature is input into the first attention module of the (i+1)th multi-scale module to generate the first attention weights corresponding to each convolutional feature, and the convolutional features are fused according to the first attention weights corresponding to each convolutional feature to generate the (i+1)th multi-scale fused feature.

[0167] The i-th intermediate feature and the (i+1)-th multi-scale fused feature are input into the residual connection module of the (i+1)-th multi-scale module to generate the (i+1)-th intermediate feature.

[0168] Furthermore, in another possible implementation of this application, the size of each of the above convolutional kernels is determined based on the interval of the PPG signal.

[0169] Furthermore, in another possible implementation of this application, the first sub-network includes a first hybrid attention module, a first convolutional layer, and a first fully connected layer, and the second sub-network includes a second hybrid attention module, a second convolutional layer, and a second fully connected layer.

[0170] Furthermore, in yet another possible implementation of this application, the aforementioned first determining unit is specifically used for:

[0171] The shared output features are input into the first hybrid attention module to determine the first hybrid attention weight matrix corresponding to the shared output features;

[0172] The shared output features are weighted according to the first hybrid attention weight matrix to generate the first sub-output features corresponding to the shared output features;

[0173] The first sub-output feature is input into the first convolutional layer and the first fully connected layer to determine the diastolic pressure.

[0174] Furthermore, in yet another possible implementation of this application, the aforementioned second determining unit is specifically used for:

[0175] The shared output features are input into the second hybrid attention module to determine the second hybrid attention weight matrix corresponding to the shared output features;

[0176] The shared output features are weighted according to the second hybrid attention weight matrix to generate the second sub-output features corresponding to the shared output features;

[0177] The second sub-output feature is input into the second convolutional layer and the second fully connected layer to determine the shrinkage pressure.

[0178] It should be noted that the information interaction and execution process between the above-mentioned devices / units are based on the same concept as the method embodiments of this application. For details on their specific functions and technical effects, please refer to the method embodiments section, and they will not be repeated here.

[0179] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0180] To implement the above embodiments, this application also proposes an electronic device.

[0181] Figure 8 This is a schematic diagram of the structure of an electronic device according to an embodiment of this application.

[0182] See Figure 8The electronic device 100 may include a processor 110, an external memory interface 120, an internal memory 121, a universal serial bus (USB) interface 130, a charging management module 140, a power management module 141, a battery 142, an antenna 1, an antenna 2, a mobile communication module 150, a wireless communication module 160, an audio module 170, a speaker 170A, a receiver 170B, a microphone 170C, a headphone jack 170D, a sensor module 180, buttons 190, a motor 191, an indicator 192, a camera 193, a display screen 194, and a subscriber identification module (SIM) card interface 195, etc. The sensor module 180 may include a pressure sensor 180A, a gyroscope sensor 180B, a barometric pressure sensor 180C, a magnetic sensor 180D, an accelerometer sensor 180E, a distance sensor 180F, a proximity sensor 180G, a fingerprint sensor 180H, a temperature sensor 180J, a touch sensor 180K, an ambient light sensor 180L, a bone conduction sensor 180M, etc.

[0183] As one possible implementation, the electronic device in this application embodiment can be a wearable device, such as a smartwatch or smart bracelet, and the sensor module 180 may also include a PPG sensor for collecting the user's PPG signal.

[0184] It is understood that the structures illustrated in the embodiments of this application do not constitute a specific limitation on the electronic device 100. In other embodiments of this application, the electronic device 100 may include more or fewer components than illustrated, or combine some components, or split some components, or have different component arrangements. The illustrated components may be implemented in hardware, software, or a combination of software and hardware.

[0185] Processor 110 may include one or more processing units, such as: application processor (AP), modem processor, graphics processing unit (GPU), image signal processor (ISP), controller, memory, video codec, digital signal processor (DSP), baseband processor, and / or neural network processing unit (NPU), etc. Different processing units may be independent devices or integrated into one or more processors.

[0186] The controller can be the nerve center and command center of the electronic device 100. The controller can generate operation control signals according to the instruction opcode and timing signals to complete the control of fetching and executing instructions.

[0187] The processor 110 may also include a memory for storing instructions and data. In some embodiments, the memory in the processor 110 is a cache memory. This memory can store instructions or data that the processor 110 has just used or that are used repeatedly. If the processor 110 needs to use the instruction or data again, it can retrieve it directly from this memory. This avoids repeated accesses, reduces the waiting time of the processor 110, and thus improves the efficiency of the system.

[0188] In some embodiments, the processor 110 may include one or more interfaces, such as an inter-integrated circuit (I2C) interface, an inter-integrated circuit sound (I2S) interface, a pulse code modulation (PCM) interface, a universal asynchronous receiver / transmitter (UART) interface, a mobile industry processor interface (MIPI), a general-purpose input / output (GPIO) interface, a subscriber identity module (SIM) interface, and / or a universal serial bus (USB) interface, etc.

[0189] It is understood that the interface connection relationships between the modules illustrated in the embodiments of this application are merely illustrative and do not constitute a structural limitation on the electronic device 100. In other embodiments of this application, the electronic device 100 may also employ different interface connection methods or combinations of multiple interface connection methods as described in the above embodiments.

[0190] Internal memory 121 can be used to store computer-executable program code, which includes instructions. Processor 110 executes various functional applications and data processing of electronic device 100 by running the instructions stored in internal memory 121. Internal memory 121 may include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback, image playback, etc.), etc. The data storage area may store data created by electronic device 100 during use (such as audio data, phonebook, etc.). Furthermore, internal memory 121 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, universal flash storage (UFS), etc.

[0191] It should be noted that the implementation process and technical principles of the electronic device in this embodiment are explained in the foregoing description of the blood pressure prediction method in the embodiments of this application, and will not be repeated here.

[0192] This application also provides a chip system for use in an electronic device. The chip system includes one or more processors, which are used to invoke computer instructions to cause the electronic device to perform the steps in the above-described method embodiments.

[0193] This application also provides a computer-readable storage medium including instructions that, when executed on an electronic device, cause the electronic device to perform the steps described in the various method embodiments above.

[0194] This application also provides a computer program product that, when run on an electronic device, causes the electronic device to perform the steps described in the various method embodiments above.

[0195] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of this application can be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include at least: any entity or device capable of carrying computer program code to a device / electronic device, a recording medium, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media. Examples include USB flash drives, portable hard drives, magnetic disks, or optical disks. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electrical carrier signals or telecommunication signals.

[0196] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0197] In the above embodiments, specific details such as particular system structures and techniques have been set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application can also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary details.

[0198] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0199] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0200] As used in this application specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if detected [the described condition or event]" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once detected [the described condition or event]," or "in response to detection [the described condition or event]."

[0201] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0202] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0203] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0204] In the embodiments provided in this application, it should be understood that the disclosed devices / electronic devices and methods can be implemented in other ways. For example, the device / electronic device embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual couplings or direct couplings or communication connections may be through some interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0205] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0206] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A method for predicting blood pressure, characterized in that, include: In response to a blood pressure prediction command, the user's current photoplethysmography (PPG) signal is acquired. Feature extraction is performed on the PPG signal to determine multiple blood pressure-related features corresponding to the PPG signal, wherein the blood pressure-related features include at least two of the following features: morphological features, spectral features, time-domain features, and invasive continuous arterial blood pressure (ABP) features corresponding to the PPG signal; Multiple blood pressure-related features are fused to generate the target input features corresponding to the PPG signal; The target input features are input into the backbone network of the blood pressure prediction model to generate shared output features corresponding to the PPG signal. The backbone network includes N multi-scale modules, each containing multiple convolutional kernels of different sizes, a first attention module, and a residual connection module. The convolutional kernels of the nth multi-scale module generate multiple convolutional features corresponding to the input features of the nth multi-scale module. The first attention module of the nth multi-scale module generates attention weights for each convolutional feature and fuses the convolutional features according to their attention weights. The residual connection module of the nth multi-scale module fuses the fused features output by the first attention module of the nth multi-scale module with the input features of the nth multi-scale module. N is a positive integer, and n is an integer greater than or equal to 1 and less than or equal to N. The shared output features are input into the first sub-network of the blood pressure prediction model to determine the user's diastolic blood pressure. The first sub-network includes a first hybrid attention module, a first convolutional layer, and a first fully connected layer. The first hybrid attention module is used to determine the first hybrid attention weight matrix of the shared output features in the diastolic blood pressure prediction task. The first hybrid attention weight matrix is ​​determined based on the first channel attention weight matrix and the first spatial attention weight of the shared output features in the diastolic blood pressure prediction task. The shared output features are input into the second sub-network of the blood pressure prediction model to determine the user's systolic blood pressure. The second sub-network includes a second hybrid attention module, a second convolutional layer, and a second fully connected layer. The second hybrid attention module is used to determine the second hybrid attention weight matrix of the shared output features in the systolic blood pressure prediction task. The second hybrid attention weight matrix is ​​determined based on the second channel attention weight matrix and the second spatial attention weight of the shared output features in the systolic blood pressure prediction task.

2. The method as described in claim 1, characterized in that, The step of inputting the target input features into the backbone network of the blood pressure prediction model to generate the shared output features corresponding to the PPG signal includes: The target input features are input into the backbone network, and the multi-scale modules are used to extract multi-scale features from the target input features in sequence, so as to perform multi-scale feature extraction on the target input features N times and generate the shared output features.

3. The method as described in claim 2, characterized in that, The step of inputting the target input features into the backbone network and sequentially extracting multi-scale features from the target input features using each of the multi-scale modules, to perform N multi-scale feature extractions on the target input features and generate the shared output features, includes: The i-th intermediate feature is input into the (i+1)-th multi-scale module, and the (i+1)-th multi-scale feature extraction is performed on the target input feature to generate the (i+1)-th intermediate feature, wherein the 0-th intermediate feature is the target input feature, the N-th intermediate feature is the shared output feature, and i is an integer greater than or equal to 0 and less than N.

4. The method as described in claim 3, characterized in that, The step of inputting the i-th intermediate feature into the (i+1)-th multi-scale module and performing the (i+1)-th multi-scale feature extraction on the target input feature to generate the (i+1)-th intermediate feature includes: The i-th intermediate feature is input into each of the convolution kernels of the (i+1)-th multi-scale module for convolution processing to generate multiple convolution features of different sizes corresponding to the i-th intermediate feature; Each of the convolutional features is input into the first attention module of the (i+1)th multi-scale module to generate the first attention weights corresponding to each of the convolutional features, and the convolutional features are fused according to the first attention weights corresponding to each of the convolutional features to generate the (i+1)th multi-scale fused feature. The i-th intermediate feature and the (i+1)-th multi-scale fusion feature are input into the residual connection module of the (i+1)-th multi-scale module to generate the (i+1)-th intermediate feature.

5. The method as described in claim 1, characterized in that, The size of each convolutional kernel is determined based on the interval of the PPG signal.

6. The method according to any one of claims 1-5, characterized in that, The step of inputting the shared output features into the first sub-network of the blood pressure prediction model to determine the user's diastolic blood pressure includes: The shared output features are input into the first hybrid attention module to determine the first hybrid attention weight matrix corresponding to the shared output features; The shared output features are weighted according to the first hybrid attention weight matrix to generate the first sub-output features corresponding to the shared output features; The first sub-output feature is input into the first convolutional layer and the first fully connected layer to determine the diastolic pressure.

7. The method according to any one of claims 1-5, characterized in that, The step of inputting the shared output features into the second sub-network of the blood pressure prediction model to determine the user's systolic blood pressure includes: The shared output features are input into the second hybrid attention module to determine the second hybrid attention weight matrix corresponding to the shared output features; The shared output features are weighted according to the second hybrid attention weight matrix to generate the second sub-output features corresponding to the shared output features; The second sub-output feature is input into the second convolutional layer and the second fully connected layer to determine the shrinkage pressure.

8. An electronic device, characterized in that, The electronic device includes: one or more processors, and a memory; the memory is coupled to the one or more processors, the memory being used to store computer program code, the computer program code including computer instructions, and the one or more processors invoking the computer instructions to cause the electronic device to perform the method as described in any one of claims 1-7.

9. A chip system, characterized in that, The chip system is applied to an electronic device, the chip system including one or more processors, the one or more processors being used to invoke computer instructions to cause the electronic device to perform the method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes instructions that, when executed on an electronic device, cause the electronic device to perform the method as described in any one of claims 1-7.

Images