A training method, application method, device and medium for blood pressure measurement network

Abstract

The present invention discloses a training method, application method, device, and medium for a blood pressure measurement network. The method includes: obtaining a pulse wave signal and a cuff pressure signal for blood pressure measurement and determining the mean pressure position; shearing the original signal according to the mean pressure position to obtain a blood pressure correction training sample set; constructing a blood pressure recognition training sample set using the signals corresponding to the systolic pressure sampling position and the diastolic pressure sampling position; downsampling the blood pressure training sample set using a preset position information expansion method to obtain a target blood pressure recognition training sample set; and determining a fully trained first recognition model, a second recognition model, a third recognition model, and a correction model. The present invention solves the technical problem in the prior art of using directly obtained pulse wave signals as input to a blood pressure prediction network, which results in a large workload and inaccurate measurement results.

Description

Technical Field

[0001] The present invention relates to the technical field of medical equipment, and in particular to a training method, application method, equipment and medium for a blood pressure measurement network. Background Art

[0002] The oscillometric method is the most commonly used non-invasive blood pressure measurement method today. It determines blood pressure by detecting the vibrations generated by blood flow hitting the blood vessel walls. The envelope of the oscillatory wave generated by the pulsation of the blood vessels is found, and the blood pressure value is derived based on the relationship between the envelope and the arterial blood pressure. This method focuses on determining the corresponding positions of systolic, diastolic, and mean pressures from the pulse wave curve. The amplitude coefficient method is commonly used to obtain these corresponding positions. However, the application of the coefficient method to the oscillometric method relies on statistical results and requires constant manual adjustment, which is labor-intensive. Other techniques often calculate blood pressure values ​​by supplementing them with other physiological signals, such as ECG or PPG signals. Alternatively, pulse wave features obtained by the oscillometric method are selectively extracted and used as input to a deep neural network, which is then trained to directly predict blood pressure values. This method relies on a strong correlation between the extracted features and the blood pressure value, which is labor-intensive and cannot effectively integrate the corresponding pressure signal of the pulse wave signal, resulting in inaccurate measurement results. Summary of the Invention

[0003] The purpose of the present invention is to overcome the above-mentioned technical deficiencies and provide a training method, application method, device and medium for a blood pressure measurement network to solve the technical problems in the prior art of using directly obtained pulse wave signals as input to a blood pressure prediction network, which results in heavy workload and inaccurate measurement results.

[0004] In order to achieve the above technical objectives, the present invention adopts the following technical solutions:

[0005] In a first aspect, the present invention provides a method for training a blood pressure measurement network, comprising:

[0006] Obtaining pulse wave signals and cuff pressure signals for blood pressure measurement;

[0007] Determining the mean pressure position based on the degree of influence of the oscillation of the blood pressure pulse wave on the frequency of change of the original blood pressure signal; and shearing the original blood pressure signal according to the mean pressure position to obtain a blood pressure correction training sample set;

[0008] collecting diastolic pressure and systolic pressure based on a preset Korotkoff sound method, determining a systolic pressure sampling position corresponding to the systolic pressure on the cuff pressure signal and a diastolic pressure sampling position corresponding to the diastolic pressure on the cuff pressure signal; and constructing a blood pressure recognition training sample set using the signals corresponding to the systolic pressure sampling position and the diastolic pressure sampling position;

[0009] downsampling the blood pressure training sample set using a preset position information expansion method to obtain a target blood pressure recognition training sample set;

[0010] The corresponding male data in the blood pressure training sample set is input into the constructed first recognition model for training to obtain a fully trained first recognition model; the corresponding female data in the blood pressure training sample set is input into the constructed second recognition model for training to obtain a fully trained second recognition model; the blood pressure training sample set is completely input into the constructed third recognition model for training to obtain a fully trained third recognition model; and the blood pressure correction training sample set is input into the constructed correction model for training to obtain a fully trained correction model.

[0011] In some embodiments, before acquiring the pulse wave signal and the cuff pressure signal, the method includes:

[0012] Use the preset upward-breathing blood pressure measurement method to obtain the original blood pressure measurement signal and mark the corresponding gender information;

[0013] performing band-pass filtering on the original signal to obtain the pulse wave signal;

[0014] The original signal is low-pass filtered to obtain the cuff pressure signal.

[0015] In some embodiments, determining the mean pressure position based on the degree of influence of the oscillation of the blood pressure pulse wave on the frequency of change of the original blood pressure signal includes:

[0016] Performing smoothing filtering on the original signal to obtain a smoothed signal;

[0017] Extracting the slope of the smoothed signal to determine a slope set;

[0018] The position corresponding to the point with the maximum slope in the slope set is determined as the average pressure position.

[0019] In some embodiments, the blood pressure training sample set includes a systolic pressure training sample set constructed using the signal corresponding to the systolic pressure sampling position and a diastolic pressure training sample set constructed using the signal corresponding to the diastolic pressure sampling position, and downsampling the blood pressure training sample set using a preset position information expansion method to obtain a target blood pressure identification training sample set includes:

[0020] Downsampling the systolic pressure training sample set to obtain a first initial systolic pressure training sample set; and downsampling the diastolic pressure training sample set to obtain a first initial diastolic pressure training sample set;

[0021] Taking the mean pressure position as a cutting boundary, constructing a systolic pressure position information expansion model corresponding to the first initial systolic pressure training sample set and a diastolic pressure position information expansion model corresponding to the first initial diastolic pressure training sample set;

[0022] Based on the systolic pressure position information expansion model and the diastolic pressure position information expansion model, and according to the ratio between the length of the original signal and the downsampling ratio, a corresponding target systolic pressure training sample set and a target diastolic pressure training sample set are respectively determined.

[0023] In some embodiments, the step of clipping the original signal according to the mean pressure position to obtain a blood pressure correction training sample set includes:

[0024] Taking the mean pressure position as the center, the original signal is intercepted toward the beginning and end with a preset step length to determine a blood pressure correction training sample set.

[0025] In a second aspect, the present invention further provides an application method of a blood pressure measurement network, comprising:

[0026] Obtaining a pre-processed pulse wave test signal of blood pressure measurement and a corrected test signal corresponding to a mean pressure position;

[0027] Inputting the pulse wave test signal into a fully trained first recognition model, a second recognition model, and a third recognition model, respectively, to determine a first blood pressure prediction result, a second blood pressure prediction result, and a third blood pressure prediction result, respectively; and inputting the pulse wave test signal and the correction test signal into a fully trained correction model to determine a correction result; wherein the fully trained first recognition model, the second recognition model, the third recognition model, and the correction model are determined according to any one of the above-mentioned training methods for the blood pressure measurement network;

[0028] Compare the similarity between the first blood pressure prediction result, the second blood pressure prediction result and the third blood pressure prediction result and the correction result, and determine that the one of the first blood pressure prediction result, the second blood pressure prediction result and the third blood pressure prediction result that is closest to the correction result is the blood pressure prediction result.

[0029] In some embodiments, the correction result includes a systolic pressure coefficient and a diastolic pressure coefficient determined according to the magnitude relationship between the diastolic pressure, the systolic pressure, and the mean pressure; wherein the diastolic pressure coefficient and the systolic pressure coefficient can be respectively expressed by the following formulas:

[0030]

[0031]

[0032] Among them, k Drepresents the diastolic pressure coefficient, k s represents the systolic pressure coefficient, t D Indicates the diastolic pressure sampling position, t S Indicates the systolic blood pressure sampling position, t M Indicates the mean pressure position.

[0033] In some embodiments, the first blood pressure prediction result includes a first systolic pressure prediction result and a first diastolic pressure prediction result, the second blood pressure prediction result includes a second systolic pressure prediction result and a second diastolic pressure prediction result, and the third blood pressure prediction result includes a third systolic pressure prediction result and a third diastolic pressure prediction result; after determining the blood pressure prediction results, the method further includes:

[0034] The systolic pressure prediction result and the diastolic pressure prediction result in the blood pressure prediction result are mapped onto the cuff pressure signal curve to determine the predicted target systolic pressure and target diastolic pressure.

[0035] In a third aspect, the present invention further provides an electronic device, comprising: a processor and a memory;

[0036] The memory stores a computer-readable program executable by the processor;

[0037] When the processor executes the computer-readable program, it implements the steps of the training method of the blood pressure measurement network described above, and / or the application method of the blood pressure measurement network described above.

[0038] In a fourth aspect, the present invention also provides a computer-readable storage medium, which stores one or more programs, and the one or more programs can be executed by one or more processors to implement the steps in the training method of the blood pressure measurement network as described above, and / or the application method of the blood pressure measurement network as described above.

[0039] Compared with the prior art, the training method, application method, device and medium of the blood pressure measurement network provided by the present invention first obtain the pulse wave signal of the blood pressure measurement; determine the mean pressure position according to the degree of influence of the oscillation of the blood pressure pulse wave on the frequency of change of the original blood pressure signal; and obtain the blood pressure correction training sample set by shearing the original signal according to the mean pressure position; so that the position of the mean pressure is more accurate; then collect the diastolic pressure and systolic pressure based on the preset Korotkoff sound method, determine the systolic pressure sampling position corresponding to the systolic pressure on the pulse wave signal and the diastolic pressure sampling position corresponding to the diastolic pressure; and construct the signals corresponding to the systolic pressure sampling position and the diastolic pressure sampling position as the blood pressure recognition training sample set; adopt the preset downsampling method to The blood pressure training sample set is downsampled to obtain a target blood pressure recognition training sample set; by converting two simple diastolic pressure information and systolic pressure information into signals containing more information, and expanding them according to the characteristics of the data, it is ensured that the model can apply more variable-length original signals; finally, the corresponding male data in the blood pressure training sample set is input into the constructed first recognition model, the corresponding female data in the blood pressure training sample set is input into the constructed second recognition model, the blood pressure training sample set is fully input into the constructed third recognition model, and the blood pressure correction training sample set is input into the constructed correction model to determine the fully trained first recognition model, second recognition model, third recognition model and correction model. Using gender information as one of the training conditions can obtain model parameters that are more suitable for different genders, thereby obtaining more accurate blood pressure detection results in the absence of gender information.

[0040] Furthermore, the output results of the first recognition model, the second recognition model and the third recognition model are compared with the output results of the correction model, and the result of the recognition model closest to the correction result is used as the prediction result. A better recognition model result can be selected to obtain a more accurate prediction value. At the same time, the systolic pressure coefficient and diastolic pressure coefficient are output by the correction model, and the influence of the strong correlation between mean pressure, diastolic pressure and systolic pressure on blood pressure prediction can be combined. The blood pressure prediction results are considered from the global data, so that the prediction results are more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 is a flow chart of an embodiment of a method for training a blood pressure measurement network provided by the present invention;

[0042] Figure 2 is a flowchart of an embodiment of step S102 in the blood pressure measurement network training method provided by the present invention;

[0043] Figure 3 This is a flow chart of an implementation of the application method of the blood pressure measurement network provided by the present invention;

[0044] Figure 4 It is a schematic diagram of the operating environment of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0045] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0046] Blood pressure measurements obtained by professionals using the Korotkoff sound method are often used as the gold standard. However, in actual clinical practice, other non-invasive measurement methods that are easier to use, safer, and more reliable are needed. The oscillometric method is the most commonly used non-invasive blood pressure measurement method today. It determines blood pressure by detecting the vibrations generated by blood flow hitting the blood vessel walls. The envelope of the oscillometric wave generated by the pulsation of the blood vessel walls is found, and the relationship between the envelope and arterial blood pressure is used to obtain the blood pressure value. This method focuses on determining the corresponding positions of systolic, diastolic, and mean pressures from the pulse wave curve. The amplitude coefficient method is commonly used, but this method's coefficients rely on medical data statistics and cannot account for all individual differences. The oscillometric method can also be divided into up-puff and out-puff measurement methods based on the phases, which also include step measurement and continuous measurement methods. With the development of deep learning, some studies have used deep learning to extract pulse waveform features obtained by the oscillometric method and combine them with Gaussian process regression or directly use deep learning networks to obtain blood pressure values.

[0047] The present invention provides a training method for a blood pressure measurement network. The method uses software to connect a cuff to the upper arm of a subject, and controls a linear pump to inflate the cuff at a specified speed. The cuff also collects measurement results at a frequency of 500 Hz. Figure 1 ,include:

[0048] S101, obtaining a pulse wave signal and a cuff pressure signal for blood pressure measurement;

[0049] S102, determining the mean pressure position based on the degree of influence of the oscillation of the blood pressure pulse wave on the frequency of change of the original blood pressure signal; and clipping the original signal based on the mean pressure position to obtain a blood pressure correction training sample set;

[0050] S103: collecting diastolic pressure and systolic pressure based on a preset Korotkoff sound method, determining a systolic pressure sampling position corresponding to the systolic pressure on the cuff pressure signal and a diastolic pressure sampling position corresponding to the diastolic pressure on the cuff pressure signal; and constructing a blood pressure recognition training sample set using the signals corresponding to the systolic pressure sampling position and the diastolic pressure sampling position;

[0051] S104, downsampling the blood pressure training sample set using a preset position information expansion method to obtain a target blood pressure recognition training sample set;

[0052] S105. Input the corresponding male data in the blood pressure training sample set into the constructed first recognition model for training to obtain a fully trained first recognition model; input the corresponding female data in the blood pressure training sample set into the constructed second recognition model for training to obtain a fully trained second recognition model; input the entire blood pressure training sample set into the constructed third recognition model for training to obtain a fully trained third recognition model; and input the blood pressure correction training sample set into the constructed correction model for training to obtain a fully trained correction model.

[0053] In this embodiment, the pulse wave signal of the blood pressure measurement is first obtained; the mean pressure position is determined according to the degree of influence of the oscillation of the blood pressure pulse wave on the frequency of change of the original blood pressure signal; and the original signal is sheared according to the mean pressure position to obtain a blood pressure correction training sample set; so that the position of the mean pressure is more accurate; then the diastolic pressure and systolic pressure are collected based on the preset Korotkoff sound method, and the systolic pressure sampling position corresponding to the systolic pressure on the pulse wave signal and the diastolic pressure sampling position corresponding to the diastolic pressure are determined; and the signals corresponding to the systolic pressure sampling position and the diastolic pressure sampling position are constructed as a blood pressure recognition training sample set; and the blood pressure training sample set is downsampled using the preset position information expansion method. Processing is performed to obtain a target blood pressure recognition training sample set; by converting two simple diastolic pressure information and systolic pressure information into signals containing more information, and expanding them according to the characteristics of the data, it is ensured that the model can apply more variable-length original signals; finally, the corresponding male data in the blood pressure training sample set is input into the constructed first recognition model, the corresponding female data in the blood pressure training sample set is input into the constructed second recognition model, the blood pressure training sample set is fully input into the constructed third recognition model, and the blood pressure correction training sample set is input into the constructed correction model to determine the fully trained first recognition model, second recognition model, third recognition model and correction model. Using gender information as one of the training conditions can obtain model parameters that are more suitable for different genders, thereby obtaining more accurate blood pressure detection results in the absence of gender information.

[0054] It should be noted that in a complete data, t D to t S is necessary information, and the rest can be pruned appropriately to construct new data. In a specific embodiment, the length of o(t) is 45s, starting from D t At 19s, S tBy the time the signal ends at 3 seconds, the two data segments have 18 and 4 possible splitting methods, respectively, for a total of 19*3=57 possible splits for the data segment o(t). This amplification method yields 57 new data segments. This amplification method allows the model to learn various scenarios involving data of varying lengths and encompasses a wider range of blood pressure measurement scenarios. The expanded datasets are fed into the model as input, with the output being the combined two-dimensional signals DP and SP. To enhance the model's predictive performance for different genders, all data are fed into the first recognition model, male data into the second, and female data into the third. After training, three recognition models are obtained.

[0055] In some embodiments, before acquiring the pulse wave signal and the cuff pressure signal, the method includes:

[0056] Use the preset upward-breathing blood pressure measurement method to obtain the original blood pressure measurement signal and mark the corresponding gender information;

[0057] performing band-pass filtering on the original signal to obtain the pulse wave signal;

[0058] The original signal is low-pass filtered to obtain the cuff pressure signal.

[0059] In this embodiment, it should be noted that the pulse wave is detected using the sliding window method to obtain the pulse wave peak, and the pulse wave peak sequence is arranged. When the peak maximum is found, the inflation cut-off pressure value is calculated using an empirical formula. During real-time monitoring, when it is found that the cuff pressure value has reached the cut-off pressure value, inflation is stopped and deflation is started, and intelligent inflation is completed. The subject's gender and age are recorded, and the subject's blood pressure is measured using the Korotkoff sound method to obtain the blood pressure standard value of the data. Each subject repeats the above signal acquisition process three times. At the same time, during the inflation process, the collected signal is respectively passed through a low-pass filter and a band-pass filter to obtain the real-time cuff pressure value p and pulse wave value o.

[0060] In some embodiments, see Figure 2 The method of determining the mean pressure position according to the influence of the oscillation of the blood pressure pulse wave on the frequency of change of the original blood pressure signal includes:

[0061] S201, performing smoothing filtering on the original signal to obtain a smoothed signal;

[0062] S202, extracting the slope of the smoothed signal to determine a slope set;

[0063] S203: Determine the position corresponding to the point with the maximum slope in the slope set as the average pressure position.

[0064] In this embodiment, during the inflation process, the original signal v(t) collected ideally should be a straight line rising at a specified speed. However, due to the influence of the superimposed pulse wave and the elasticity of the blood vessels, the actual original signal v(t) is an S-shaped arc with superimposed pulses. Therefore, smoothing filtering is performed to obtain a smooth S-shaped curve signal. Due to individual differences, the smoothed signal curve has a variety of shapes, and the upper and lower curvatures are not consistent.

[0065] In the traditional oscillometric method, the highest point of the pulse wave peak is defined as the corresponding mean pressure position. However, in reality, this position may be offset due to the delay of the inflation strategy and the filter. Since the shape of the original signal v(t) is mainly affected by the pulse wave, the larger the pulse wave oscillation, the faster the curve changes. Therefore, considering the influence of the pulse wave oscillation on the slope of the smoothed signal, by analyzing the slope change of the smoothed signal s(t), the maximum slope position is defined, that is, the mean pressure position, which is defined as t M .

[0066] In some embodiments, the blood pressure training sample set includes a systolic pressure training sample set constructed using the signal corresponding to the systolic pressure sampling position and a diastolic pressure training sample set constructed using the signal corresponding to the diastolic pressure sampling position, and downsampling the blood pressure training sample set using a preset position information expansion method to obtain a target blood pressure identification training sample set includes:

[0067] Downsampling the systolic pressure training sample set to obtain a first initial systolic pressure training sample set; and downsampling the diastolic pressure training sample set to obtain a first initial diastolic pressure training sample set;

[0068] Taking the mean pressure position as a cutting boundary, constructing a systolic pressure position information expansion model corresponding to the first initial systolic pressure training sample set and a diastolic pressure position information expansion model corresponding to the first initial diastolic pressure training sample set;

[0069] Based on the systolic pressure position information expansion model and the diastolic pressure position information expansion model, and according to the ratio between the length of the original signal and the downsampling ratio, a corresponding target systolic pressure training sample set and a target diastolic pressure training sample set are respectively determined.

[0070] In this embodiment, compared with other fields of deep learning, it is more difficult to obtain effective data for non-invasive blood pressure measurement and the amount of data is small, so some expansion operations are needed to the data. In order to better train the model, the diastolic pressure sampling position t is constructed separately. D and systolic blood pressure sampling location t SThe corresponding signals are used as label data. The two signals contain position information corresponding to o(t). Since the original data is long, combined with the relationship between systolic pressure, diastolic pressure and mean pressure, that is, t D <t M <t S , downsample the tag signal and cut it based on the average pressure position. The original construction formula is as follows:

[0071]

[0072]

[0073] Where r is the downsampling ratio, with an empirical value of 10. DP(i) is the diastolic position expansion model, and SP(i) is the systolic position expansion model. The signal lengths of DP(i) and SP(i) are both L / r, where L is the length of o(t). This conversion transforms the position information of systolic and diastolic pressures into a triangular wave with a width of 2α, increasing the proportion of position information in the entire signal and amplifying it. In a specific real-time example, if the inflation rate is set to 5 mmHg / s, the maximum time to complete the inflation process is 60 seconds. In this case, α is defined as 50, indicating 1 second of data.

[0074] In some embodiments, the step of clipping the original signal according to the mean pressure position to obtain a blood pressure correction training sample set includes:

[0075] Taking the mean pressure position as the center, the original signal is intercepted toward the beginning and end with a preset step length to determine a blood pressure correction training sample set.

[0076] In this embodiment, the original o(t) is converted to the average pressure position t M As the intermediate position, we go to the signal starting point and the signal ending point respectively, take the shortest distance as the radius, intercept the data, for example, if the time to the starting point is 30s and the time to the ending point is 25s, then we intercept 5s to 55s of the complete signal, and the processed signal length is 50s. The processed data set is used as the model input and input into the LSTM-based regression model, with k D and k S The output value is trained to obtain the correction model.

[0077] The embodiment of the present invention also provides an application method of a blood pressure measurement network, see Figure 3 ,include:

[0078] S301, obtaining a pre-processed pulse wave test signal of blood pressure measurement and a corrected test signal corresponding to a mean pressure position;

[0079] S302: Input the pulse wave test signal into a first, second, and third recognition models that have been fully trained, respectively, to determine a first blood pressure prediction result, a second blood pressure prediction result, and a third blood pressure prediction result, respectively; and input the pulse wave test signal and the correction test signal into a fully trained correction model to determine a correction result; wherein the first, second, third, and correction models that have been fully trained are determined according to any one of the training methods for the blood pressure measurement network described above;

[0080] S303. Compare the similarities between the first blood pressure prediction result, the second blood pressure prediction result, and the third blood pressure prediction result and the correction result, and determine that the one of the first blood pressure prediction result, the second blood pressure prediction result, and the third blood pressure prediction result that is closest to the correction result is the blood pressure prediction result.

[0081] In this embodiment, the output results of the first recognition model, the second recognition model and the third recognition model are compared with the output results of the correction model, and the result of the recognition model closest to the correction result is used as the prediction result. A better recognition model result can be selected to obtain a more accurate prediction value. At the same time, the systolic pressure coefficient and diastolic pressure coefficient are output by the correction model. The influence of the strong correlation between mean pressure, diastolic pressure and systolic pressure on blood pressure prediction can be combined, and the blood pressure prediction results can be considered from the global data, so that the prediction results are more accurate.

[0082] It should be noted that in step S301, obtaining the preprocessed pulse wave test signal of the blood pressure measurement and the corrected test signal corresponding to the mean pressure position includes bandpass filtering and low-pass filtering of the original signal, obtaining the diastolic pressure sampling position and the systolic pressure sampling position to construct a test sample set, and downsampling the test sample set to expand the position information.

[0083] In some embodiments, the correction result includes a systolic pressure coefficient and a diastolic pressure coefficient determined according to the magnitude relationship between the diastolic pressure, the systolic pressure, and the mean pressure; wherein the diastolic pressure coefficient and the systolic pressure coefficient can be respectively expressed by the following formulas:

[0084]

[0085]

[0086] Among them, k D represents the diastolic pressure coefficient, k s represents the systolic pressure coefficient, t D Indicates the diastolic pressure sampling position, t S Indicates the systolic blood pressure sampling position, t M Indicates the mean pressure position.

[0087] In this embodiment, the collected original signal v(t) is passed through a bandpass filter to obtain a pulse wave signal o(t), which is then passed through a low-pass filter to obtain the corresponding pressure value p(t). The sampling points corresponding to the systolic and diastolic pressures measured by the Korotkoff sound method on p(t) are defined as t S and t D , then the diastolic pressure coefficient is

[0088]

[0089] The systolic blood pressure coefficient is

[0090]

[0091] In some embodiments, the first blood pressure prediction result includes a first systolic pressure prediction result and a first diastolic pressure prediction result, the second blood pressure prediction result includes a second systolic pressure prediction result and a second diastolic pressure prediction result, and the third blood pressure prediction result includes a third systolic pressure prediction result and a third diastolic pressure prediction result; after determining the blood pressure prediction results, the method further includes:

[0092] The systolic pressure prediction result and the diastolic pressure prediction result in the blood pressure prediction result are mapped onto the cuff pressure signal curve to determine the predicted target systolic pressure and target diastolic pressure.

[0093] In this embodiment, the prediction results obtained from the processed test set are mapped to the cuff pressure signal curve to restore the true systolic and diastolic pressures.

[0094] like Figure 4 As shown, based on the above-mentioned blood pressure measurement network training and application method, the present invention also provides an electronic device, which can be a computing device such as a mobile terminal, desktop computer, notebook, PDA, server, etc. The electronic device includes a processor 410, a memory 420, and a display 430. Figure 4 Only some of the components of the electronic device are shown, but it should be understood that it is not required to implement all of the shown components, and more or fewer components may be implemented instead.

[0095] In some embodiments, the memory 420 may be an internal storage unit of the electronic device, such as a hard disk or memory of the electronic device. In other embodiments, the memory 420 may also be an external storage device of the electronic device, such as a plug-in hard disk equipped on the electronic device, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. Furthermore, the memory 420 may also include both an internal storage unit of the electronic device and an external storage device. The memory 420 is used to store application software and various types of data installed in the electronic device, such as program codes installed in the electronic device. The memory 420 can also be used to temporarily store data that has been output or is to be output. In one embodiment, the memory 420 stores a training and application program 440 of a blood pressure measurement network, which can be executed by the processor 410, thereby realizing the training and application method of the blood pressure measurement network in each embodiment of the present application.

[0096] In some embodiments, the processor 410 may be a central processing unit (CPU), a microprocessor, or other data processing chip, configured to execute program codes or process data stored in the memory 420, such as executing training and application methods of a blood pressure measurement network.

[0097] In some embodiments, display 430 can be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. Display 430 is used to display information about the training and application devices in the blood pressure measurement network and to display a visual user interface. Components 410-430 of the electronic device communicate with each other via a system bus.

[0098] Of course, those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing related hardware (such as a processor, controller, etc.) through a computer program. The program can be stored in a computer-readable storage medium, and when executed, the program can include the processes in the above-described method embodiments. The storage medium can be a memory, a magnetic disk, an optical disk, etc.

[0099] The specific embodiments of the present invention described above do not limit the scope of protection of the present invention. Any other corresponding changes and modifications made based on the technical concept of the present invention should be included in the scope of protection of the claims of the present invention.

Claims

1. A method for training a blood pressure measurement network, characterized in that: include: Obtaining pulse wave signals and cuff pressure signals for blood pressure measurement; Determining the mean pressure position according to the degree of influence of the oscillation of the blood pressure pulse wave on the frequency of change of the original blood pressure signal, and shearing the original blood pressure signal according to the mean pressure position to obtain a blood pressure correction training sample set; collecting diastolic pressure and systolic pressure based on a preset Korotkoff sound method, determining a systolic pressure sampling position corresponding to the systolic pressure on the cuff pressure signal, and a diastolic pressure sampling position corresponding to the diastolic pressure on the cuff pressure signal, and constructing a blood pressure training sample set using the signals corresponding to the systolic pressure sampling position and the diastolic pressure sampling position; downsampling the blood pressure training sample set using a preset position information expansion method to obtain a target blood pressure training sample set; Inputting the male data corresponding to the blood pressure training sample set into the constructed first recognition model for training to obtain a fully trained first recognition model, inputting the female data corresponding to the blood pressure training sample set into the constructed second recognition model for training to obtain a fully trained second recognition model, inputting the entire blood pressure training sample set into the constructed third recognition model for training to obtain a fully trained third recognition model, and inputting the blood pressure correction training sample set into the constructed correction model for training to obtain a fully trained correction model; Obtaining a pre-processed pulse wave test signal of blood pressure measurement and a corrected test signal corresponding to a mean pressure position; Inputting the pulse wave test signal into a first, second, and third recognition models that have been fully trained, respectively, to determine a first blood pressure prediction result, a second blood pressure prediction result, and a third blood pressure prediction result, respectively; and inputting the pulse wave test signal and the correction test signal into a fully trained correction model to determine a correction result; comparing the similarities between the first blood pressure prediction result, the second blood pressure prediction result, and the third blood pressure prediction result and the correction result, and determining that the one of the first blood pressure prediction result, the second blood pressure prediction result, and the third blood pressure prediction result that is closest to the correction result is the blood pressure prediction result; The correction result includes a systolic pressure coefficient and a diastolic pressure coefficient determined according to the magnitude relationship between the diastolic pressure, the systolic pressure, and the mean pressure; wherein the diastolic pressure coefficient and the systolic pressure coefficient can be respectively expressed by the following formulas: in, represents the diastolic pressure coefficient, represents the systolic blood pressure coefficient, Indicates the diastolic pressure sampling position, Indicates the systolic blood pressure sampling position, Indicates the mean pressure position.

2. The blood pressure measurement network training method according to claim 1, characterized in that: Before obtaining the pulse wave signal and the cuff pressure signal, the method includes: Use the preset upward-breathing blood pressure measurement method to obtain the original blood pressure measurement signal and mark the corresponding gender information; performing band-pass filtering on the original signal to obtain the pulse wave signal; The original signal is low-pass filtered to obtain the cuff pressure signal.

3. The blood pressure measurement network training method according to claim 2, characterized in that: Determining the mean pressure position according to the degree of influence of the oscillation of the blood pressure pulse wave on the frequency of change of the original blood pressure signal includes: Performing smoothing filtering on the original signal to obtain a smoothed signal; Extracting the slope of the smoothed signal to determine a slope set; The position corresponding to the point with the maximum slope in the slope set is determined as the average pressure position.

4. The blood pressure measurement network training method according to claim 2, characterized in that: The blood pressure training sample set includes a systolic pressure training sample set constructed using the signal corresponding to the systolic pressure sampling position and a diastolic pressure training sample set constructed using the signal corresponding to the diastolic pressure sampling position, and the blood pressure training sample set is downsampled using a preset position information expansion method to obtain a target blood pressure training sample set, including: Downsampling the systolic pressure training sample set to obtain a first initial systolic pressure training sample set; and downsampling the diastolic pressure training sample set to obtain a first initial diastolic pressure training sample set; Taking the mean pressure position as a cutting boundary, constructing a systolic pressure position information expansion model corresponding to the first initial systolic pressure training sample set and a diastolic pressure position information expansion model corresponding to the first initial diastolic pressure training sample set; Based on the systolic pressure position information expansion model and the diastolic pressure position information expansion model, and according to the ratio between the length of the original signal and the downsampling ratio, a corresponding target systolic pressure training sample set and a target diastolic pressure training sample set are respectively determined.

5. The blood pressure measurement network training method according to claim 1, characterized in that: The step of clipping the original signal according to the mean pressure position to obtain a blood pressure correction training sample set includes: Taking the mean pressure position as the center, the original signal is intercepted toward the beginning and end with a preset step length to determine a blood pressure correction training sample set.

6. The blood pressure measurement network training method according to claim 1, characterized in that: The first blood pressure prediction result includes a first systolic pressure prediction result and a first diastolic pressure prediction result, the second blood pressure prediction result includes a second systolic pressure prediction result and a second diastolic pressure prediction result, and the third blood pressure prediction result includes a third systolic pressure prediction result and a third diastolic pressure prediction result; After determining the blood pressure prediction result, the method further includes: The systolic pressure prediction result and the diastolic pressure prediction result in the blood pressure prediction result are mapped onto the cuff pressure signal curve to determine the predicted target systolic pressure and target diastolic pressure.

7. An electronic device comprising: A processor and a memory; the memory stores a computer-readable program that can be executed by the processor; it is characterized in that when the processor executes the computer-readable program, it implements the steps in the training method of the blood pressure measurement network according to any one of claims 1-6.

8. A computer-readable storage medium, characterized in that The computer-readable storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the steps in the blood pressure measurement network training method according to any one of claims 1 to 6.

Images