Blood pressure prediction method based on PPG signal, electronic device, and storage medium

Abstract

This application provides a blood pressure prediction method and electronic device based on PPG signals, as well as a computer-readable storage medium. The method includes: converting the PPG signal to be measured into model input parameters; inputting the model input parameters into a trained classification model; and outputting M diastolic pressure binary classification results and N systolic pressure binary classification results through M diastolic pressure classifiers and N systolic pressure binary classification results, respectively. The M diastolic pressure classifiers correspond to M consecutive diastolic pressure values, and the N systolic pressure classifiers correspond to N systolic pressure values. The method determines the predicted diastolic pressure value based on the M diastolic pressure binary classification results and the predicted systolic pressure value based on the N systolic pressure binary classification results. This solution addresses the problems associated with relying solely on regression or classification methods for blood pressure prediction.

Description

Technical Field

[0001] This application relates to the field of electrophysiological signal processing technology, and in particular to a blood pressure prediction method based on PPG signals, an electronic device, and a computer-readable storage medium. Background Technology

[0002] In recent years, deep learning has been increasingly widely used, and blood pressure estimation can also be achieved using deep learning techniques. Generally, blood pressure estimation tasks are implemented using regression methods and classification methods. Regression methods treat blood pressure as a continuous value, directly predicting the blood pressure value based on the input. Classification methods, on the other hand, divide blood pressure into different categories, predicting the probability of belonging to each category, and selecting the category with the highest probability as the predicted blood pressure value. Regression methods typically require more hyperparameter tuning and assume that blood pressure variations are consistent across different blood pressure levels, which contradicts common sense. Classification methods, however, ignore the inherent continuity of blood pressure, because in conventional classification tasks, categories are completely independent. Summary of the Invention

[0003] The purpose of this application is to provide a blood pressure prediction method and electronic device based on PPG signals, as well as a computer-readable storage medium, to solve the problems existing in the prior art when relying solely on regression or classification methods to predict blood pressure.

[0004] On the one hand, this application provides a blood pressure prediction method based on PPG signals, including:

[0005] Convert the PPG signal to be measured into model input parameters;

[0006] The input parameters of the model are input into the trained classification model, and the classification model outputs M diastolic blood pressure binary classification results through M diastolic blood pressure classifiers and N systolic blood pressure binary classification results through N systolic blood pressure classifiers; wherein, the M diastolic blood pressure classifiers correspond to M consecutive diastolic blood pressure values, and the N systolic blood pressure classifiers correspond to N systolic blood pressure values.

[0007] The predicted diastolic blood pressure value is determined based on the M binary classification results of diastolic blood pressure, and the predicted systolic blood pressure value is determined based on the N binary classification results of systolic blood pressure.

[0008] In one embodiment, converting the PPG signal to be measured into model input parameters includes:

[0009] Determine the mean and standard deviation of multiple sampling points in the PPG signal to be tested;

[0010] The average number is subtracted from the multiple sampling points of the PPG signal to be tested, and then divided by the standard deviation to obtain the model input parameters.

[0011] In one embodiment, converting the PPG signal to be measured into model input parameters includes:

[0012] The PPG signal to be tested is filtered to obtain the filtered PPG signal;

[0013] The filtered PPG signal is differentiated to obtain a first-order PPG signal and a second-order PPG signal; wherein the first-order PPG signal is the first derivative of the filtered PPG signal, and the second-order PPG signal is the second derivative of the filtered PPG signal.

[0014] The model input parameters are determined based on the PPG signal to be tested, the first-order PPG signal, and the second-order PPG signal.

[0015] In one embodiment, filtering the PPG signal to be tested to obtain a filtered PPG signal includes:

[0016] The PPG signal to be tested is bandpass filtered by a second-order Butterworth filter to obtain the first PPG signal.

[0017] The PPG signal to be tested is low-pass filtered by a second-order Butterworth filter to obtain a second PPG signal.

[0018] The first PPG signal and the second PPG signal are added together to obtain the filtered PPG signal.

[0019] In one embodiment, determining the model input parameters based on the PPG signal to be measured, the first-order PPG signal, and the second-order PPG signal includes:

[0020] Determine the first mean and the first standard deviation of multiple sampling points in the PPG signal to be tested, and subtract the first mean from each of the multiple sampling points of the PPG signal to be tested, and divide by the first standard deviation to obtain the first sub-input parameter;

[0021] Determine the second mean and second standard deviation of multiple sampling points in the first-order PPG signal, and subtract the second mean from each sampling point of the first-order PPG signal and divide by the second standard deviation to obtain the second sub-input parameter;

[0022] The third mean and third standard deviation of multiple sampling points in the second-order PPG signal are determined, and the third mean is subtracted from each sampling point of the second-order PPG signal and divided by the third standard deviation to obtain the third sub-input parameter.

[0023] The model input parameters are constructed based on the first sub-input parameter, the second sub-input parameter, and the third sub-input parameter.

[0024] In one embodiment, determining the predicted diastolic blood pressure value based on the M diastolic blood pressure binary classification results and determining the predicted systolic blood pressure value based on the N systolic blood pressure binary classification results includes:

[0025] Determine whether each of the M diastolic blood pressure binary classification results is greater than a specified probability threshold. Set the diastolic blood pressure binary classification results that are greater than the specified probability threshold to 1, and set the diastolic blood pressure binary classification results that are not greater than the specified probability threshold to 0.

[0026] The predicted diastolic blood pressure value is obtained by summing the M binary classification results of diastolic blood pressure and the lower limit of diastolic blood pressure.

[0027] Determine whether each of the N systolic blood pressure binary classification results is greater than the specified probability threshold, set the systolic blood pressure binary classification results that are greater than the specified probability threshold to 1, and set the systolic blood pressure binary classification results that are not greater than the specified probability threshold to 0;

[0028] The predicted systolic blood pressure value is obtained by summing the N binary classification results of systolic blood pressure and the lower limit of systolic blood pressure.

[0029] In one embodiment, the classification model is trained in the following manner:

[0030] S1: Input the sample input parameters from the training set into the neural network model to obtain M diastolic blood pressure binary classification results output by M diastolic blood pressure classifiers and N systolic blood pressure binary classification results output by N systolic blood pressure classifiers; wherein, the sample input parameters carry diastolic blood pressure label sequences and systolic blood pressure label sequences, the diastolic blood pressure label sequences include labels corresponding to M consecutive diastolic blood pressure values, and the systolic blood pressure label sequences include labels corresponding to N consecutive systolic blood pressure values;

[0031] S2: Calculate the prediction loss based on the M diastolic blood pressure binary classification results, the N systolic blood pressure binary classification results, the diastolic blood pressure label sequence, and the systolic blood pressure label sequence, and adjust the model parameters of the neural network model based on the prediction loss;

[0032] S3: Repeat the process from S1 to S2 until the neural network model converges to obtain the classification model.

[0033] In one embodiment, the step of obtaining the training set includes:

[0034] Multiple sample PPG signals are converted into sample input parameters respectively;

[0035] Based on the diastolic blood pressure value and ordered array of diastolic blood pressure corresponding to each sample input parameter, the diastolic blood pressure label sequence of each sample input parameter is calculated.

[0036] Based on the systolic blood pressure value and ordered array of systolic blood pressure corresponding to each sample input parameter, the systolic blood pressure label sequence of each sample input parameter is calculated.

[0037] The training set is constructed based on input parameters from multiple samples carrying diastolic and systolic blood pressure label sequences.

[0038] In one embodiment, calculating the prediction loss based on the M diastolic blood pressure binary classification results, the N systolic blood pressure binary classification results, the diastolic blood pressure label sequence, and the systolic blood pressure label sequence includes:

[0039] Based on the M diastolic blood pressure binary classification results and the diastolic blood pressure label sequence, calculate the first sub-prediction loss;

[0040] Based on the N systolic blood pressure binary classification results and the systolic blood pressure label sequence, calculate the second sub-prediction loss;

[0041] The prediction loss is obtained by weighted summing of the first sub-prediction loss and the second sub-prediction loss.

[0042] In one embodiment, calculating the first sub-prediction loss based on the M diastolic blood pressure binary classification results and the diastolic blood pressure label sequence includes:

[0043]

[0044] Where, loss DBP α is the first sub-prediction loss; F represents the number of sample input parameters in a batch; KD is the total number of diastolic blood pressure value prediction tasks, equivalent to the number of diastolic blood pressure classifiers M; k This represents the weighting coefficient for the prediction task of the k-th diastolic blood pressure value; This represents the binary classification result of diastolic blood pressure after setting it to 0 or 1 for the prediction task of the diastolic blood pressure value of the input parameter of the i-th sample at the k-th value; The label represents the k-th diastolic blood pressure value in the diastolic blood pressure label sequence of the i-th sample input parameter; This represents the diastolic pressure binary classification result output by the k-th diastolic pressure classifier for the input parameters of the i-th sample.

[0045] In one embodiment, calculating the second sub-prediction loss based on the N systolic blood pressure binary classification results and the systolic blood pressure label sequence includes:

[0046]

[0047] Where, lossSBP The second sub-prediction loss; F represents the number of sample input parameters in a batch; KS is the total number of systolic blood pressure value prediction tasks, equivalent to the number of systolic blood pressure classifiers N; β k This represents the weighting coefficient for the prediction task of the kth systolic blood pressure value; This represents the binary classification result of systolic blood pressure after setting it to 0 or 1 for the prediction task of the systolic blood pressure value of the input parameter of the i-th sample at the k-th value; The label represents the k-th systolic blood pressure value in the systolic blood pressure label sequence of the i-th sample input parameters; This represents the systolic blood pressure binary classification result output by the k-th systolic blood pressure classifier for the input parameters of the i-th sample.

[0048] Furthermore, this application provides an electronic device, the electronic device comprising:

[0049] processor;

[0050] Memory used to store processor-executable instructions;

[0051] The processor is configured to execute the aforementioned blood pressure prediction method based on PPG signals.

[0052] In addition, this application provides a computer-readable storage medium storing a computer program that can be executed by a processor to perform the above-described blood pressure prediction method based on PPG signals.

[0053] This application's solution utilizes a classification model trained through ordered regression. During the prediction process, it outputs M binary classification results for diastolic blood pressure corresponding to M consecutive diastolic blood pressure values, and outputs N binary classification results for systolic blood pressure corresponding to N consecutive systolic blood pressure values. This process overcomes the erroneous assumption of regression methods that the blood pressure variation patterns are consistent across different blood pressure levels, and also solves the problem of ordinary classification methods ignoring the continuity of blood pressure. Therefore, based on the M binary classification results for diastolic blood pressure and the N binary classification results for systolic blood pressure, the predicted values ​​for diastolic and systolic blood pressure can be determined more accurately. Attached Figure Description

[0054] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly described below.

[0055] Figure 1 A schematic diagram illustrating an application scenario of a blood pressure prediction method based on PPG signals provided in an embodiment of this application;

[0056] Figure 2 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application;

[0057] Figure 3 A schematic flowchart of a blood pressure prediction method based on PPG signals provided in an embodiment of this application;

[0058] Figure 4 Provided for an embodiment of this application Figure 3 A detailed flowchart of step 310 is shown below;

[0059] Figure 5 Provided for an embodiment of this application Figure 3 A detailed flowchart of step 330 is shown below;

[0060] Figure 6 A flowchart illustrating a training method for a classification model provided in an embodiment of this application;

[0061] Figure 7 A flowchart illustrating a method for constructing a training set according to an embodiment of this application;

[0062] Figure 8 A block diagram of a blood pressure prediction device based on PPG signals provided in an embodiment of this application. Detailed Implementation

[0063] The technical solutions in the embodiments of this application will now be described with reference to the accompanying drawings.

[0064] Similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0065] Figure 1 This is a schematic diagram illustrating an application scenario of the blood pressure prediction method based on PPG signals provided in this application embodiment. For example... Figure 1 As shown, the application scenario includes a client 20 and a server 30. The client 20 can be a smartwatch, bracelet, or other wristband device used to collect the wearer's PPG signal and transmit the collected PPG signal to the server 30. The server 30 can be a host, server, server cluster, or cloud computing center, which can predict the wearer's blood pressure based on the PPG signal.

[0066] like Figure 2 As shown, this embodiment provides an electronic device 1, including: at least one processor 11 and a memory 12. Figure 2Taking a processor 11 as an example, the processor 11 and the memory 12 are connected via a bus 10. The memory 12 stores instructions that can be executed by the processor 11. The instructions are executed by the processor 11 to enable the electronic device 1 to perform all or part of the process of the method in the embodiments described below. In one embodiment, the electronic device 1 may be the server 30 described above, used to execute a blood pressure prediction method based on PPG signals.

[0067] The memory 12 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable red-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0068] This application also provides a computer-readable storage medium storing a computer program that can be executed by a processor 11 to perform the blood pressure prediction method based on PPG signals provided in this application.

[0069] See Figure 3 This is a flowchart illustrating a blood pressure prediction method based on PPG signals provided in an embodiment of this application. Figure 3 As shown, the method may include the following steps 310-330.

[0070] Step 310: Convert the PPG signal to be measured into model input parameters.

[0071] Among them, the PPG (Photoplethysmography) signal to be measured is the PPG signal that needs to be processed to predict blood pressure.

[0072] The electronic device implementing this solution can acquire the PPG signal to be measured from the wristband device. The PPG signal to be measured can be denoted as {x1, x2, ..., x...} n}, where x iThis represents the i-th sampling point. In this scheme, PPG signals of a specified duration or a specified length (number of sampling points) can be processed. Therefore, after acquiring the PPG signal from the wristband device, it can be truncated by specifying a duration or a length to obtain the PPG signal to be tested. Here, the specified duration or length can be configured as needed. For example, if the sampling frequency of the PPG signal is 25 Hz and the specified duration is 30 seconds, then the PPG signal to be tested contains 750 sampling points. Alternatively, specifying a length of 750 allows specifying that the PPG signal to be tested contains 750 sampling points.

[0073] By converting and processing the PPG signal to be tested, the model input parameters can be obtained.

[0074] In one embodiment, the mean and standard deviation of multiple sampling points in the PPG signal under test can be determined. After calculating the mean and standard deviation, the mean can be subtracted from each of the multiple sampling points in the PPG signal under test to obtain multiple difference values. Furthermore, each of the multiple difference values ​​can be divided by the standard deviation, thereby achieving normalization processing of the multiple sampling points. At this point, the processed multiple sampling points can be used as the model input parameters.

[0075] Step 320: Input the model input parameters into the trained classification model. The classification model outputs M diastolic blood pressure binary classification results through M diastolic blood pressure classifiers and N systolic blood pressure binary classification results through N systolic blood pressure classifiers. Among them, the M diastolic blood pressure classifiers correspond to M consecutive diastolic blood pressure values, and the N systolic blood pressure classifiers correspond to N systolic blood pressure values.

[0076] Here, the classification model is pre-trained using an ordered regression method. The classification model includes a feature extraction network and M diastolic blood pressure classifiers and N systolic blood pressure classifiers.

[0077] After inputting the model input parameters into the trained classification model, feature parameters can be extracted from the input parameters using the feature extraction network within the classification model. These feature parameters are then processed by M diastolic blood pressure classifiers and N systolic blood pressure classifiers, resulting in M ​​diastolic blood pressure binary classification results and N systolic blood pressure binary classification results. The M consecutive diastolic blood pressure values ​​corresponding to the M diastolic blood pressure classifiers can be configured empirically. For example, if the range of human diastolic blood pressure is between 50 and 119, then 70 diastolic blood pressure classifiers are set, corresponding to diastolic blood pressure values ​​from 50 to 119. The N consecutive systolic blood pressure values ​​corresponding to the N systolic blood pressure classifiers can also be configured empirically. For example, if the range of human systolic blood pressure is between 75 and 189, then 115 systolic blood pressure classifiers are set, corresponding to systolic blood pressure values ​​from 75 to 189.

[0078] The diastolic pressure binary classification result output by any diastolic pressure classifier indicates the probability that the actual diastolic pressure value corresponding to the measured PPG signal includes the diastolic pressure value corresponding to that classifier (if the actual diastolic pressure value is greater than or equal to the diastolic pressure value, then the actual diastolic pressure value is considered to include that diastolic pressure value). For example, the diastolic pressure binary classification result output by the diastolic pressure classifier corresponding to the diastolic pressure value 55 is {0.65, 0.35}, where 0.65 is the probability that the actual diastolic pressure value includes the diastolic pressure value 55, and 0.35 is the probability that the actual diastolic pressure value does not include the diastolic pressure value 55.

[0079] The systolic blood pressure binary classification result output by any systolic blood pressure classifier indicates the probability that the actual systolic blood pressure value corresponding to the measured PPG signal includes the systolic blood pressure value corresponding to that classifier (if the actual systolic blood pressure value is greater than or equal to the systolic blood pressure value, then the actual systolic blood pressure value is considered to include that systolic blood pressure value). For example, the systolic blood pressure binary classification result output by the systolic blood pressure classifier corresponding to the systolic blood pressure value 85 is {0.8, 0.2}, where 0.8 is the probability that the actual systolic blood pressure value includes the systolic blood pressure value 85, and 0.2 is the probability that the actual systolic blood pressure value does not include the systolic blood pressure value 85.

[0080] Step 330: Determine the predicted diastolic blood pressure value based on M binary classification results of diastolic blood pressure, and determine the predicted systolic blood pressure value based on N binary classification results of systolic blood pressure.

[0081] After obtaining M binary classification results for diastolic blood pressure, these results can be converted to obtain the predicted diastolic blood pressure value. Similarly, after obtaining N binary classification results for systolic blood pressure, these results can be converted to obtain the predicted systolic blood pressure value.

[0082] Through the above measures, this scheme utilizes a classification model trained by ordered regression. During the prediction process, it outputs M binary classification results for diastolic blood pressure corresponding to M consecutive diastolic blood pressure values, and N binary classification results for systolic blood pressure corresponding to N consecutive systolic blood pressure values. This process overcomes the erroneous assumption of regression methods that blood pressure variation patterns are consistent across different blood pressure levels, and also solves the problem of ordinary classification methods ignoring the continuity of blood pressure. Therefore, based on the M binary classification results for diastolic blood pressure and the N binary classification results for systolic blood pressure, the predicted values ​​for diastolic and systolic blood pressure can be determined more accurately. Compared to existing regression and classification methods, this scheme can significantly reduce the MAE (mean absolute error) index in the prediction process.

[0083] In one embodiment, see Figure 4 This is provided as an embodiment of the present application. Figure 3 A detailed flowchart of step 310 is shown below. Figure 4 As shown, when executing step 310, steps 311 to 313 can be executed.

[0084] Step 311: Filter the PPG signal to be tested to obtain the filtered PPG signal.

[0085] The filtered PPG signal and the PPG signal under test have the same number of sampling points.

[0086] In one embodiment, a second-order Butterworth filter is used to bandpass filter the PPG signal under test, with the bandpass frequency ranging from 0.5 to 8 Hz. The first PPG signal can be obtained through bandpass filtering.

[0087] The PPG signal under test is low-pass filtered using a second-order Butterworth filter with a low-pass frequency of 0.15 Hz. This low-pass filtering yields the second PPG signal.

[0088] Adding the first PPG signal to the second PPG signal yields the filtered PPG signal. When adding the first and second PPG signals, the i-th sample point in the first PPG signal needs to be added to the i-th sample point in the second PPG signal to obtain the i-th sample point in the filtered PPG signal.

[0089] Step 312: Take the derivative of the filtered PPG signal to obtain the first-order PPG signal and the second-order PPG signal; wherein, the first-order PPG signal is the first derivative of the filtered PPG signal, and the second-order PPG signal is the second derivative of the filtered PPG signal.

[0090] After obtaining the filtered PPG signal, its first derivative can be calculated to obtain the first-order PPG signal. Furthermore, its second derivative can be calculated to obtain the second-order PPG signal.

[0091] For example, a first-order PPG signal can be represented by the following formula (1):

[0092] PPG_VP[i]=PPG_filterd[i]-PPG_filterd[i-1] (1)

[0093] Where PPG_VP[i] represents the i-th sampling point in the first-order PPG signal; PPG_filtered[i] represents the i-th sampling point in the filtered PPG signal; and PPG_filtered[i-1] represents the (i-1)-th sampling point in the filtered PPG signal. Here, the value of i ranges from 2 to n, and n is the total number of sampling points in the filtered PPG signal.

[0094] The first-order PPG signal containing n-1 sampling points can be calculated using formula (1). The first sampling point is repeated once, thereby aligning the length of the first-order PPG signal with the length of the filtered PPG signal.

[0095] For example, a second-order PPG signal can be represented by the following formula (2):

[0096] PPG_AP[i]=PPG_VP[i]-PPG_VP[i-1] (2)

[0097] Where PPG_AP[i] represents the i-th sampling point in the second-order PPG signal; PPG_VP[i] represents the i-th sampling point in the first-order PPG signal; and PPG_VP[i-1] represents the (i-1)-th sampling point in the first-order PPG signal. Here, the value of i ranges from 2 to n, and n is the total number of sampling points in the filtered PPG signal.

[0098] The second-order PPG signal containing n-1 sampling points can be calculated using formula (2). The first sampling point is repeated once, thereby aligning the length of the second-order PPG signal with the length of the first-order PPG signal.

[0099] Step 313: Determine the model input parameters based on the PPG signal to be measured, the first-order PPG signal, and the second-order PPG signal.

[0100] After obtaining the first-order PPG signal and the second-order PPG signal, the measured PPG signal, the first-order PPG signal and the second-order PPG signal can be fused to obtain the model input parameters.

[0101] The above measures can be used to convert the PPG signal to be measured into model input parameters.

[0102] In one embodiment, when performing step 313, a first mean and a first standard deviation of multiple sampling points in the PPG signal to be tested can be determined. The first mean is the average of the multiple sampling points in the PPG signal to be tested, and the first standard deviation is the standard deviation of the multiple sampling points in the PPG signal to be tested.

[0103] The first average is subtracted from the multiple sampling points of the PPG signal to be tested, and then divided by the first standard deviation, thereby achieving normalization of the multiple sampling points. At this time, the multiple processed sampling points can be used to form the first sub-input parameters.

[0104] For a first-order PPG signal, the second mean and second standard deviation of multiple sampling points in the first-order PPG signal can be determined. The second mean is the average of the multiple sampling points in the first-order PPG signal, and the second standard deviation is the standard deviation of the multiple sampling points in the first-order PPG signal.

[0105] The second average is subtracted from the multiple sampling points of the first-order PPG signal and divided by the second standard deviation, thereby achieving normalization of the multiple sampling points. At this point, the multiple processed sampling points can be used to form the second sub-input parameters.

[0106] For a second-order PPG signal, the third mean and third standard deviation of multiple sampling points in the second-order PPG signal can be determined. The third mean is the average of the multiple sampling points in the second-order PPG signal, and the third standard deviation is the standard deviation of the multiple sampling points in the second-order PPG signal.

[0107] The third mean is subtracted from each of the multiple sampling points of the second-order PPG signal, and then divided by the third standard deviation, thereby achieving normalization of the multiple sampling points. At this point, the processed multiple sampling points can be used to form the third sub-input parameter.

[0108] The first, second, and third sub-input parameters are sequences of equal length and can each be considered as a 1*n vector. Combining them yields a 3*n matrix, which serves as the model input parameters. Here, n represents the total number of sampling points in any sub-input parameter.

[0109] In one embodiment, see Figure 5 This is provided as an embodiment of the present application. Figure 3 A detailed flowchart of step 330 is shown below. Figure 5 As shown, when executing step 330, steps 331 to 334 can be executed.

[0110] Step 331: Determine whether each of the M diastolic blood pressure binary classification results is greater than a specified probability threshold. Set the diastolic blood pressure binary classification results that are greater than the specified probability threshold to 1, and set the diastolic blood pressure binary classification results that are not greater than the specified probability threshold to 0.

[0111] For M diastolic blood pressure binary classification results, we can determine whether the probability of "yes" in each result is greater than a specified probability threshold. The probability of "yes" is the probability that the actual diastolic blood pressure value includes the diastolic blood pressure value corresponding to the binary classification result. The specified probability threshold can be configured empirically. For example, the specified probability threshold is 0.5.

[0112] If the probability of "yes" in any binary classification result of diastolic blood pressure is greater than a specified probability threshold, the binary classification result of diastolic blood pressure can be set to 1, indicating that the actual diastolic blood pressure value includes the diastolic blood pressure value corresponding to the binary classification result of diastolic blood pressure.

[0113] If the probability of "yes" in any diastolic blood pressure binary classification result is not greater than a specified probability threshold, the diastolic blood pressure binary classification result can be set to 0, indicating that the actual diastolic blood pressure value does not include the diastolic blood pressure value corresponding to that binary classification result. After setting it to 0 or 1, the diastolic blood pressure binary classification result will only contain a single value.

[0114] For example, this process can be represented by the following formula (3):

[0115]

[0116] Where, p i This represents the probability of "yes" in the binary classification result of diastolic blood pressure output by the i-th diastolic blood pressure classifier; This indicates the binary classification result of diastolic blood pressure after being set to 0 or 1.

[0117] Step 332: Sum the M binary classification results of diastolic blood pressure and the lower limit of diastolic blood pressure to obtain the predicted value of diastolic blood pressure.

[0118] The lower limit of diastolic blood pressure is the minimum value within the range of human diastolic blood pressure. It can be configured based on experience. For example, the lower limit of diastolic blood pressure is 50.

[0119] For example, the summation process can be represented by the following formula (4):

[0120]

[0121] Where pred_DBP is the predicted diastolic blood pressure value; M is the total number of binary classification results for diastolic blood pressure, for example, M is 70; This indicates the binary classification result of diastolic blood pressure after being set to 0 or 1.

[0122] Step 333: Determine whether each of the N systolic blood pressure binary classification results is greater than a specified probability threshold. Set the systolic blood pressure binary classification results that are greater than the specified probability threshold to 1, and set the systolic blood pressure binary classification results that are not greater than the specified probability threshold to 0.

[0123] For N systolic blood pressure binary classification results, we can determine whether the probability of "yes" in each result is greater than a specified probability threshold. The probability of "yes" is the probability that the actual systolic blood pressure value includes the systolic blood pressure value corresponding to the binary classification result. The specified probability threshold can be configured empirically. For example, the specified probability threshold is 0.5.

[0124] If the probability of "yes" in any systolic blood pressure binary classification result is greater than a specified probability threshold, the systolic blood pressure binary classification result can be set to 1, indicating that the actual systolic blood pressure value includes the systolic blood pressure value corresponding to the systolic blood pressure binary classification result.

[0125] If the probability of "yes" in any binary classification result of systolic blood pressure is not greater than a specified probability threshold, the binary classification result of systolic blood pressure can be set to 0, indicating that the actual systolic blood pressure value does not include the systolic blood pressure value corresponding to that binary classification result. After setting it to 0 or 1, the binary classification result of systolic blood pressure will only contain a single value.

[0126] For example, this process can be represented by the following formula (5):

[0127]

[0128] Where, p i This represents the probability of "yes" in the binary classification result of systolic blood pressure output by the i-th systolic blood pressure classifier; This indicates the binary classification result of systolic blood pressure after being set to 0 or 1.

[0129] Step 334: Sum the N systolic blood pressure binary classification results and the lower limit of systolic blood pressure to obtain the predicted systolic blood pressure value.

[0130] The lower limit of systolic blood pressure is the minimum value within the range of human systolic blood pressure, which can be configured based on experience. For example, the lower limit of systolic blood pressure is 75.

[0131] For example, the summation process can be represented by the following formula (6):

[0132]

[0133] Where pred_SBP is the predicted systolic blood pressure value; N is the total number of binary classification results for systolic blood pressure, for example, N is 115; This indicates the binary classification result of systolic blood pressure after being set to 0 or 1.

[0134] Through the above measures, M diastolic blood pressure binary classification results can be merged into a diastolic blood pressure prediction value, and N systolic blood pressure binary classification results can be merged into a systolic blood pressure prediction value.

[0135] In one embodiment, a classification model can be trained before performing the blood pressure prediction method in steps 310 to 330. See also Figure 6 This is a flowchart illustrating a training method for a classification model provided in an embodiment of this application. Figure 6 As shown, the method may include the following steps S1 to S3.

[0136] Step S1: Input the sample input parameters from the training set into the neural network model to obtain M diastolic blood pressure binary classification results output by M diastolic blood pressure classifiers and N systolic blood pressure binary classification results output by N systolic blood pressure classifiers. The sample input parameters carry diastolic blood pressure label sequences and systolic blood pressure label sequences. The diastolic blood pressure label sequence includes labels corresponding to M consecutive diastolic blood pressure values, and the systolic blood pressure label sequence includes labels corresponding to N consecutive systolic blood pressure values.

[0137] The form of the sample input parameters in the training set depends on the form of the model input parameters in the application stage. If the model input parameters in the application stage are 1*n vectors obtained by converting the PPG signal to be tested, then the sample input parameters are 1*n vectors; if the model input parameters in the application stage are 3*n matrices obtained by converting the PPG signal to be tested, the first-order PPG signal, and the second-order PPG signal, then the sample input parameters are 3*n matrices.

[0138] The M labels in the diastolic blood pressure label sequence are either 1 or 0. Here, the number of M is based on the number of pre-configured diastolic blood pressure classifiers in the neural network model. If any label is 1, it means that the diastolic blood pressure value corresponding to the sample input parameter includes the diastolic blood pressure value corresponding to that label. If any label is 0, it means that the diastolic blood pressure value corresponding to the sample input parameter does not include the diastolic blood pressure value corresponding to that label. Since the M labels in the diastolic blood pressure label sequence correspond to consecutive diastolic blood pressure values, the diastolic blood pressure label sequence of any sample input parameter includes several consecutive 1s followed by several 0s.

[0139] The N labels in the systolic blood pressure label sequence are either 1 or 0. Here, the number of N is based on the number of systolic blood pressure classifiers pre-configured in the neural network model. If any label is 1, it means that the systolic blood pressure value corresponding to the sample input parameter includes the systolic blood pressure value corresponding to that label. If any label is 0, it means that the systolic blood pressure value corresponding to the sample input parameter does not include the systolic blood pressure value corresponding to that label. Since the N labels in the systolic blood pressure label sequence correspond to consecutive systolic blood pressure values, the systolic blood pressure label sequence of any sample input parameter includes several consecutive 1s followed by several 0s.

[0140] The neural network model includes a feature extraction network, M diastolic blood pressure classifiers, and N systolic blood pressure classifiers. Here, the feature extraction network can be an existing convolutional neural network, recurrent neural network, or other network model, or it can be a custom network model.

[0141] For example, the structure of a neural network model is shown in Table 1 below:

[0142]

[0143] Table 1

[0144] The hyperparameters of convolutional layers are represented in the order of kernel size, stride, and number of output channels; the parameters of pooling layers are represented in the order of kernel size and stride; and the parameters of GRU (Gated Recurrent Unit) are represented in the order of input feature size, output feature size, and number of layers.

[0145] The following sequences of components are connected in series: Conv2D-1, MaxPool2D-1, Conv2D-2, MaxPool2D-2, Conv2D-3, MaxPool2D-3, Conv2D-4, MaxPool2D-4, Reshape, Bidirectional GRU, Bidirectional GRU, Expand_dims, Transpose, and Conv2D-4. Conv2D-4 is then connected to FC-DBP-1 and FC-SBP-1, which are connected in parallel. FC-DBP-1 is followed by M parallel FC-DBP-2 units. FC-SBP-1 is followed by N parallel FC-SBP-2 units. Each of the M parallel FC-DBP-2 units is followed by a sigmoid function. Each of the N parallel FC-SBP-2 units is followed by a sigmoid function. Here, each FC-DBP-2 unit and its subsequent sigmoid function constitutes a diastolic blood pressure classifier. An FC-SBP-2 and its subsequent sigmoid function constitute a systolic blood pressure classifier.

[0146] In the input dimension, N represents the batch size (the number of input parameters in a batch during training), H represents the height of the input parameters, W represents the width of the input parameters, and C represents the number of channels of the input parameters. In the output dimension, N represents the batch size (the number of output parameters in a batch during training), H represents the height of the output parameters, W represents the width of the output parameters, and C represents the number of channels of the output parameters.

[0147] Step S2: Calculate the prediction loss based on the M diastolic blood pressure binary classification results, N systolic blood pressure binary classification results, diastolic blood pressure label sequence, and systolic blood pressure label sequence, and adjust the model parameters of the neural network model based on the prediction loss.

[0148] Step S3: Repeat the process from S1 to S2 until the neural network model converges and a classification model is obtained.

[0149] Based on the M diastolic blood pressure binary classification results and the diastolic blood pressure label sequence, the first sub-prediction loss can be calculated. For example, the calculation process of the first sub-prediction loss can be represented by the following formula (7):

[0150]

[0151] Where, loss DBP α is the first sub-prediction loss; F represents the number of sample input parameters in a batch; KD is the total number of diastolic blood pressure value prediction tasks, equivalent to the number of diastolic blood pressure classifiers M; k This represents the weighting coefficient for the prediction task of the k-th diastolic blood pressure value; This represents the binary classification result of diastolic blood pressure after setting it to 0 or 1 for the prediction task of the diastolic blood pressure value of the input parameter of the i-th sample at the k-th value; Let represent the label of the k-th diastolic blood pressure value in the diastolic blood pressure label sequence of the i-th sample input parameter; here, Indicates when and Take 1 when they are the same, when and Do not both be 0; This represents the probability that the input parameter of the i-th sample is "yes" in the binary classification result of diastolic blood pressure output by the k-th diastolic blood pressure classifier.

[0152] Weighting coefficient α for each diastolic blood pressure value prediction task k It can be configured as needed. In one embodiment, all weight coefficients can be the same.

[0153] Based on the N systolic blood pressure binary classification results and the systolic blood pressure label sequence, the second sub-prediction loss can be calculated. For example, the calculation process of the second sub-prediction loss can be represented by the following formula (8):

[0154]

[0155] Where, loss SBP The second sub-prediction loss; F represents the number of sample input parameters in a batch; KS is the total number of systolic blood pressure value prediction tasks, equivalent to the number of systolic blood pressure classifiers N; β k This represents the weighting coefficient for the prediction task of the kth systolic blood pressure value; This represents the binary classification result of systolic blood pressure after setting it to 0 or 1 for the prediction task of the systolic blood pressure value of the input parameter of the i-th sample at the k-th value; Here, represents the label of the k-th systolic blood pressure value in the systolic blood pressure label sequence of the i-th sample input parameter; Indicates when and Take 1 when they are the same, when and Do not both be 0; This represents the probability that the input parameter of the i-th sample is "yes" in the binary classification result of systolic blood pressure output by the k-th systolic blood pressure classifier.

[0156] Weighting coefficient β for each systolic blood pressure value prediction task k It can be configured as needed. In one embodiment, all weight coefficients can be the same.

[0157] After obtaining the first sub-prediction loss and the second sub-prediction loss, they can be weighted and summed to obtain the prediction loss. For example, the calculation process can be represented by the following formula (9):

[0158] loss = loss SBP +λloss DBP (9)

[0159] Where loss represents the prediction loss; DBP For the first child prediction loss; loss SBP λ is the loss for the second sub-prediction; λ is the loss balancing coefficient, which can be initially set to 1 and adjusted during multiple training rounds to ensure that the loss for the first sub-prediction and the loss for the second sub-prediction are on the same order of magnitude.

[0160] After obtaining the prediction loss, the model parameters of the neural network model can be adjusted based on this prediction loss. After adjustment, return to step S1 for a new round of training. After repeated iterations, when the number of iterations reaches a preset threshold, or when the prediction loss region stabilizes, it can be determined that the neural network model has converged. At this point, the converged neural network model can be used as a classification model.

[0161] Using the above measures, a classification model can be trained using the ordered regression method.

[0162] In one embodiment, a training set can be constructed before training the classification model. See also Figure 7 This is a flowchart illustrating a method for constructing a training set according to an embodiment of this application. Figure 7 As shown, the training set can be constructed through the following steps 710 to 740.

[0163] Step 710: Convert the multiple sample PPG signals into sample input parameters respectively.

[0164] For each sample PPG signal, it can be converted into sample input parameters.

[0165] Here, the conversion method is determined as needed, and the conversion method must be the same as that used in the application phase.

[0166] Step 720: Based on the diastolic blood pressure value and ordered array of diastolic blood pressure corresponding to each sample input parameter, calculate the diastolic blood pressure label sequence for each sample input parameter.

[0167] The diastolic blood pressure value corresponding to each sample input parameter represents the actual diastolic blood pressure indicated by the sample PPG signal corresponding to that sample input parameter. An ordered array of diastolic blood pressure values ​​records the various diastolic blood pressure values ​​within the range of human diastolic blood pressure. For example, if the range of human diastolic blood pressure is between 50 and 119, then the ordered array of diastolic blood pressure can be denoted as Rank_DBP = {rd1, rd2, ..., rd...} KD}={50,51,…,119}, where KD is 70.

[0168] For the i-th sample input parameter, its corresponding diastolic blood pressure value is w. i DBP The conversion process can then be represented by the following formula (10):

[0169]

[0170] in, This represents the k-th label in the diastolic pressure label sequence of the i-th sample input parameter; rd represents the diastolic blood pressure value of the i-th sample input parameter; k This represents the k-th element in the ordered array of diastolic pressure.

[0171] Step 730: Based on the systolic blood pressure value and ordered array of systolic blood pressure corresponding to each sample input parameter, calculate the systolic blood pressure label sequence for each sample input parameter.

[0172] The systolic blood pressure value corresponding to each sample input parameter represents the actual systolic blood pressure indicated by the sample PPG signal corresponding to that input parameter. The ordered systolic blood pressure array records the various systolic blood pressure values ​​within the range of human systolic blood pressure. For example, if the range of human systolic blood pressure is between 75 and 189, then the ordered systolic blood pressure array can be denoted as Rank_SBP = {rs1, rs2, ..., rs...} KS}={75,76,…,189}, where KS is 115.

[0173] For the i-th sample input parameter, its corresponding systolic blood pressure value is w. i SBP The conversion process can then be represented by the following formula (11):

[0174]

[0175] in, This represents the k-th label in the systolic blood pressure label sequence of the i-th sample input parameter; rs represents the systolic blood pressure value of the i-th sample input parameter; k This represents the k-th element in the ordered array.

[0176] Step 740: Construct a training set based on the input parameters of multiple samples carrying diastolic blood pressure label sequences and systolic blood pressure label sequences.

[0177] After converting multiple sample PPG signals carrying diastolic and systolic blood pressure values ​​into sample input parameters carrying diastolic and systolic blood pressure label sequences, a training set can be constructed based on the multiple sample input parameters for subsequent training of the classification model.

[0178] Figure 8 This is a block diagram of a blood pressure prediction device based on PPG signals according to an embodiment of the present invention, as shown below. Figure 8 As shown, the device may include:

[0179] The conversion module 810 is used to convert the PPG signal to be measured into model input parameters;

[0180] The classification module 820 is used to input the model input parameters into the trained classification model, and output M diastolic blood pressure binary classification results through M diastolic blood pressure classifiers and N systolic blood pressure binary classification results through N systolic blood pressure classifiers; wherein, the M diastolic blood pressure classifiers correspond to M consecutive diastolic blood pressure values, and the N systolic blood pressure classifiers correspond to N systolic blood pressure values.

[0181] The determination module 830 is used to determine the predicted diastolic blood pressure value based on the M diastolic blood pressure binary classification results and to determine the predicted systolic blood pressure value based on the N systolic blood pressure binary classification results.

[0182] The specific implementation process of the functions and roles of each module in the above-mentioned device can be found in the implementation process of the corresponding steps in the blood pressure prediction method based on PPG signals, and will not be repeated here.

[0183] The apparatuses and methods disclosed in the several embodiments provided in this application can also be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatuses, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0184] In addition, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0185] If a function is implemented as a software module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

Claims

1. A blood pressure prediction method based on PPG signals, characterized in that, include: Convert the PPG signal to be measured into model input parameters; The model input parameters are input into a trained classification model. The classification model outputs M diastolic blood pressure binary classification results through M diastolic blood pressure classifiers and N systolic blood pressure binary classification results through N systolic blood pressure classifiers. The classification model is a model trained by ordered regression. The M diastolic blood pressure classifiers correspond to M consecutive diastolic blood pressure values, and the N systolic blood pressure classifiers correspond to N systolic blood pressure values. The diastolic blood pressure binary classification result output by any diastolic blood pressure classifier indicates the probability that the actual diastolic blood pressure value corresponding to the PPG signal under test is greater than or equal to the diastolic blood pressure value corresponding to that classifier. The systolic blood pressure binary classification result output by any systolic blood pressure classifier indicates the probability that the actual systolic blood pressure value corresponding to the PPG signal under test is greater than or equal to the systolic blood pressure value corresponding to that classifier. The predicted diastolic blood pressure value is determined based on the M binary classification results of diastolic blood pressure, and the predicted systolic blood pressure value is determined based on the N binary classification results of systolic blood pressure. The step of determining the predicted diastolic blood pressure value based on the M diastolic blood pressure binary classification results and determining the predicted systolic blood pressure value based on the N systolic blood pressure binary classification results includes: Determine whether each of the M diastolic blood pressure binary classification results is greater than a specified probability threshold. Set the diastolic blood pressure binary classification results that are greater than the specified probability threshold to 1, and set the diastolic blood pressure binary classification results that are not greater than the specified probability threshold to 0. The predicted diastolic blood pressure value is obtained by summing the M binary classification results of diastolic blood pressure and the lower limit of diastolic blood pressure. Determine whether each of the N systolic blood pressure binary classification results is greater than the specified probability threshold, set the systolic blood pressure binary classification results that are greater than the specified probability threshold to 1, and set the systolic blood pressure binary classification results that are not greater than the specified probability threshold to 0; The predicted systolic blood pressure value is obtained by summing the N binary classification results of systolic blood pressure and the lower limit of systolic blood pressure.

2. The method according to claim 1, characterized in that, The process of converting the PPG signal to be measured into model input parameters includes: Determine the mean and standard deviation of multiple sampling points in the PPG signal to be tested; The model input parameters are obtained by subtracting the mean from multiple sampling points of the PPG signal to be tested and dividing by the standard deviation.

3. The method according to claim 1, characterized in that, The process of converting the PPG signal to be measured into model input parameters includes: The PPG signal to be tested is filtered to obtain the filtered PPG signal; The filtered PPG signal is differentiated to obtain a first-order PPG signal and a second-order PPG signal; wherein the first-order PPG signal is the first derivative of the filtered PPG signal, and the second-order PPG signal is the second derivative of the filtered PPG signal. The model input parameters are determined based on the PPG signal to be tested, the first-order PPG signal, and the second-order PPG signal.

4. The method according to claim 3, characterized in that, The step of filtering the PPG signal to be tested to obtain the filtered PPG signal includes: The PPG signal to be tested is bandpass filtered by a second-order Butterworth filter to obtain the first PPG signal. The PPG signal to be tested is low-pass filtered by a second-order Butterworth filter to obtain a second PPG signal. The first PPG signal and the second PPG signal are added together to obtain the filtered PPG signal.

5. The method according to claim 3, characterized in that, The determination of the model input parameters based on the PPG signal to be measured, the first-order PPG signal, and the second-order PPG signal includes: Determine the first mean and the first standard deviation of multiple sampling points in the PPG signal to be tested, and subtract the first mean from each of the multiple sampling points of the PPG signal to be tested, and divide by the first standard deviation to obtain the first sub-input parameter; Determine the second mean and second standard deviation of multiple sampling points in the first-order PPG signal, and subtract the second mean from each sampling point of the first-order PPG signal and divide by the second standard deviation to obtain the second sub-input parameter; The third mean and third standard deviation of multiple sampling points in the second-order PPG signal are determined, and the third mean is subtracted from each sampling point of the second-order PPG signal and divided by the third standard deviation to obtain the third sub-input parameter. The model input parameters are constructed based on the first sub-input parameter, the second sub-input parameter, and the third sub-input parameter.

6. The method according to claim 1, characterized in that, The classification model is trained in the following manner: S1: Input the sample input parameters from the training set into the neural network model to obtain M diastolic blood pressure binary classification results output by M diastolic blood pressure classifiers and N systolic blood pressure binary classification results output by N systolic blood pressure classifiers; wherein, the sample input parameters carry diastolic blood pressure label sequences and systolic blood pressure label sequences, the diastolic blood pressure label sequences include labels corresponding to M consecutive diastolic blood pressure values, and the systolic blood pressure label sequences include labels corresponding to N consecutive systolic blood pressure values; S2: Calculate the prediction loss based on the M diastolic blood pressure binary classification results, the N systolic blood pressure binary classification results, the diastolic blood pressure label sequence, and the systolic blood pressure label sequence, and adjust the model parameters of the neural network model based on the prediction loss; S3: Repeat the process from S1 to S2 until the neural network model converges to obtain the classification model.

7. The method according to claim 6, characterized in that, The steps for obtaining the training set include: Multiple sample PPG signals are converted into sample input parameters respectively; Based on the diastolic blood pressure value and ordered array of diastolic blood pressure corresponding to each sample input parameter, the diastolic blood pressure label sequence of each sample input parameter is calculated. Based on the systolic blood pressure value and ordered array of systolic blood pressure corresponding to each sample input parameter, the systolic blood pressure label sequence of each sample input parameter is calculated. The training set is constructed based on input parameters from multiple samples carrying diastolic and systolic blood pressure label sequences.

8. The method according to claim 6, characterized in that, The step of calculating the prediction loss based on the M diastolic blood pressure binary classification results, the N systolic blood pressure binary classification results, the diastolic blood pressure label sequence, and the systolic blood pressure label sequence includes: Based on the M diastolic blood pressure binary classification results and the diastolic blood pressure label sequence, calculate the first sub-prediction loss; Based on the N systolic blood pressure binary classification results and the systolic blood pressure label sequence, calculate the second sub-prediction loss; The prediction loss is obtained by weighted summing of the first sub-prediction loss and the second sub-prediction loss.

9. The method according to claim 8, characterized in that, The step of calculating the first sub-prediction loss based on the M diastolic blood pressure binary classification results and the diastolic blood pressure label sequence includes: Where, loss DBP The first sub-prediction loss; F represents the number of sample input parameters in a batch; KD is the total number of diastolic blood pressure value prediction tasks, which is equivalent to the number of diastolic blood pressure classifiers M; This represents the weighting coefficient for the prediction task of the k-th diastolic blood pressure value; This represents the binary classification result of diastolic blood pressure after setting it to 0 or 1 for the prediction task of the diastolic blood pressure value of the input parameter of the i-th sample at the k-th value; The label represents the k-th diastolic blood pressure value in the diastolic blood pressure label sequence of the i-th sample input parameter; This represents the diastolic pressure binary classification result output by the k-th diastolic pressure classifier for the input parameters of the i-th sample.

10. The method according to claim 8, characterized in that, The step of calculating the second sub-prediction loss based on the N systolic blood pressure binary classification results and the systolic blood pressure label sequence includes: Where, loss SBP The second sub-prediction loss; F represents the number of sample input parameters in a batch; KS is the total number of systolic blood pressure value prediction tasks, which is equivalent to the number of systolic blood pressure classifiers N; This represents the weighting coefficient for the prediction task of the kth systolic blood pressure value; This represents the binary classification result of systolic blood pressure after setting it to 0 or 1 for the prediction task of the systolic blood pressure value of the input parameter of the i-th sample at the k-th value; The label represents the k-th systolic blood pressure value in the systolic blood pressure label sequence of the i-th sample input parameters; This represents the systolic blood pressure binary classification result output by the k-th systolic blood pressure classifier for the input parameters of the i-th sample.

11. An electronic device, characterized in that, The electronic device includes: processor; Memory used to store processor-executable instructions; The processor is configured to execute the blood pressure prediction method based on PPG signals as described in any one of claims 1-10.

12. A computer-readable storage medium, characterized in that, The storage medium stores a computer program that can be executed by a processor to perform the blood pressure prediction method based on PPG signals as described in any one of claims 1-10.

Images