Method and device for predicting average arterial pressure based on full-connection neural network

Abstract

The invention discloses a method and device for predicting average arterial pressure based on a full-connection neural network, and relates to the technical field of biomedicine based on artificial intelligence. The method comprises the steps that blood pressure pulse wave signals of each patient are collected, and S-G filtering processing is conducted on the blood pressure pulse wave signals; constructing a blood pressure pulse wave envelope line through a double-Gaussian fitting algorithm, and estimating the average arterial pressure of each patient according to the blood pressure pulse wave envelope line; respectively calculating relevancy between the multiple human body physiological parameters of each patient and the average arterial pressure; the human physiological parameter variables with the relevancy larger than a preset value serve as the input quantity of a full-connection neural network, the predicted value of the average arterial pressure serves as the output quantity of the full-connection neural network, a training sample set and a verification sample set are constructed, and the full-connection neural network is trained; and predicting the average arterial pressure of the patient by using the trained full-connection neural network. The accuracy of the non-invasive blood pressure measuring equipment can be improved.

Description

Technical Field

[0001] The present invention relates to the field of biomedical technologies based on artificial intelligence, and particularly to a method and device for predicting mean arterial pressure based on a fully connected neural network. Background Art

[0002] Since the blood pressure pulse wave curve inside the non-invasive automatic blood pressure measuring device for verification cannot be traced, the current metrological verification regulations do not include the verification item of accuracy. And the mean arterial pressure (MAP), as a key influencing factor for the peak amplitude of the blood pressure pulse wave curve, directly determines the reliability of the blood pressure measurement result. MAP is an important input boundary condition in hemodynamic simulation modeling and plays an important role in evaluating arterial load and ventricle-artery coupling. These characteristics make MAP an indispensable core parameter in hemodynamic research and provide important support for the accuracy and reliability of the model.

[0003] MAP is the average value of arterial blood pressure in one cardiac cycle. The existing empirical calculation methods for MAP are calculated from systolic blood pressure (SBP), diastolic blood pressure (DBP), and a fixed proportional coefficient. Subsequently, some existing technologies proposed the MAP calculation graph method, which determines MAP by finding the scale points of systolic blood pressure and diastolic blood pressure in the calculation graph, but reading according to the graph has subjective deviation. Some existing technologies also proposed the pulse graph integration method to calculate systemic circulation MAP, which requires using a dedicated instrument, the sphygmograph, to record the pulse graph, and is not conducive to popularization and application. After that, some existing technologies proposed that the MAP can be calculated using the pulse graph area method and the quasi-linear elastic chamber, using different formulas to calculate MAP in the aorta and peripheral arteries, or calculating MAP using the pulse wave propagation time, and obtaining the relationship between the pulse wave propagation time and MAP through linear regression methods.

[0004] Although the existing technologies have proposed different methods to calculate MAP, none of them can obtain the accurate value of MAP through numerical information such as systolic blood pressure and diastolic blood pressure. Summary of the Invention

[0005] In view of the above-mentioned defects or deficiencies in the existing technology, the present invention provides a method and device for predicting mean arterial pressure based on a fully connected neural network. By inputting the selected human physiological parameters that are closely related to the mean arterial pressure into the fully connected neural network (FCNN), the mean arterial pressure (MAP) is predicted, which not only helps to improve the accuracy of non-invasive blood pressure measurement devices, but also significantly improves the quality of medical services.

[0006] One aspect of the present invention provides a method for predicting mean arterial pressure based on a fully connected neural network, including: collecting blood pressure pulse wave signals of each patient in a patient sample group, and performing S-G filtering processing on the blood pressure pulse wave signals; constructing a blood pressure pulse wave envelope line through a double Gaussian fitting algorithm, and estimating the mean arterial pressure of each patient in the patient sample group according to the blood pressure pulse wave envelope line; calculating the correlation degree between multiple human physiological parameters of each patient in the patient sample group and the mean arterial pressure respectively; using the human physiological parameter variables with a correlation degree greater than a preset value as the input quantity of the fully connected neural network, and using the predicted value of the mean arterial pressure as the output quantity of the fully connected neural network, constructing a training sample set and a verification sample set, and training the fully connected neural network; using the trained fully connected neural network to predict the mean arterial pressure of a patient.

[0007] Another aspect of the present invention further provides a device for predicting mean arterial pressure based on a fully connected neural network, including: a first module configured to collect blood pressure pulse wave signals of each patient in a patient sample group; a second module configured to construct a blood pressure pulse wave envelope line through a double Gaussian fitting algorithm, and estimate the mean arterial pressure of each patient in the patient sample group according to the blood pressure pulse wave envelope line; a third module configured to calculate the correlation degree between multiple human physiological parameters of each patient in the patient sample group and the mean arterial pressure respectively; a fourth module configured to use the human physiological parameter variables with a correlation degree greater than a preset value as the input quantity of the fully connected neural network, and use the predicted value of the mean arterial pressure as the output quantity of the fully connected neural network, construct a training sample set and a verification sample set, and train the fully connected neural network; a fifth module configured to use the trained fully connected neural network to predict the mean arterial pressure of a patient.

[0008] A method and device for predicting mean arterial pressure based on a fully connected neural network provided by the present invention utilize a blood pressure acquisition device that meets the requirements of the international blood pressure measurement device metrology regulation specification and has a high-precision sampling module, collect blood pressure data at the brachial artery of numerous patients through the auscultation method, which is the "gold standard" of blood pressure clinical measurement, then use Gaussian fitting to obtain the true mean arterial pressure MAP value of the blood pressure pulse wave envelope line, analyze the correlation between variables, design a fully connected neural network method, and accurately predict the mean arterial pressure MAP value after inputting human physiological parameter data closely related to the mean arterial pressure, thereby providing an optimal solution for accurately obtaining MAP. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] By reading the detailed description of the non-limiting embodiments with reference to the following drawings, other features, objects, and advantages of the present application will become more apparent: Figure 1It is a schematic flowchart of a method for predicting mean arterial pressure based on a fully connected neural network provided by an embodiment of the present application; Figure 2 It is a schematic logical diagram of a blood pressure data acquisition system provided by an embodiment of the present application; Figure 3 It is a schematic diagram of Gaussian fitting of a blood pressure pulse wave envelope provided by an embodiment of the present application; Figure 4 It is a heat map of variable correlation analysis in mean arterial pressure MAP prediction provided by an embodiment of the present application; Figure 5 It is a schematic diagram of a fully connected neural network FCNN provided by an embodiment of the present application; Figure 6 It is a schematic diagram of comparison and residual analysis of mean arterial pressure MAP prediction between the double Gaussian fitting algorithm, linear regression and empirical formula of the present invention; Figure 7 It is a schematic structural diagram of a device for predicting mean arterial pressure based on a fully connected neural network provided by another embodiment of the present application; Figure 8 It is a schematic structural diagram of an electronic device provided by another embodiment of the present application. Detailed implementation manners

[0010] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0011] The terms used in the embodiments of the present invention are only for the purpose of describing specific embodiments, and are not intended to limit the present invention. The singular forms "a", "the" and "said" used in the embodiments of the present invention are also intended to include the plural forms unless the context clearly indicates otherwise.

[0012] It should be understood that although the terms first, second, third, etc. may be used in the embodiments of the present invention to describe the acquisition modules, these acquisition modules should not be limited to these terms. These terms are only used to distinguish the acquisition modules from each other.

[0013] Depending on the context, as used herein, the term "if" can be interpreted as "when" or "while" or "in response to determining" or "in response to detecting". Similarly, depending on the context, the phrase "if determined" or "if detecting (stated condition or event)" can be interpreted as "when determined" or "in response to determining" or "when detecting (stated condition or event)" or "in response to detecting (stated condition or event)".

[0014] It should be noted that the orientation terms such as "upper", "lower", "left", and "right" described in the embodiments of the present invention are described from the angles shown in the drawings, and should not be construed as limitations on the embodiments of the present invention. In addition, in the context, it should also be understood that when it is mentioned that one element is formed "on" or "under" another element, it can not only be directly formed "on" or "under" another element, but also be indirectly formed "on" or "under" another element through an intermediate element.

[0015] See Figure 1 , an embodiment of the present application provides a method for predicting mean arterial pressure based on a fully connected neural network, including the following steps: Step S101, collect the blood pressure pulse wave signals of each patient in the patient sample population, and perform S-G filtering processing on the blood pressure pulse wave signals.

[0016] In this embodiment, human physiological parameters including blood pressure pulse wave signals are collected. The collection process strictly follows the requirements of the ISO 81060-1 standard, and the "gold standard" double-blind auscultation method of clinical blood pressure measurement is used for blood pressure measurement.

[0017] See Figure 2 , collect 3 groups of data on the same side of the arm of 1 patient. Two observers use their respective stethoscopes to simultaneously identify the Korotkoff sounds during the cuff deflation process on the same side of the arm, indirectly measure the blood pressure value through a tabletop mercury sphygmomanometer, and take the average of the two as the systolic blood pressure (SBP) and diastolic blood pressure (DBP) results. When the difference between the SBP or DBP obtained by the two observers is greater than 4 mmHg (1 mmHg = 0.133 kPa, the same below), the measurement data is invalid and needs to be discarded. The collection device of this embodiment includes two parts: a blood pressure data collection device and a mercury sphygmomanometer. When the observer listens to the Korotkoff sounds, the digital acquisition card outputs the pressure change in the gas path detected by the pressure transmitter, and saves and processes the corresponding blood pressure data signals through the host computer. Among them, the collected human physiological parameters include blood pressure pulse wave signals (also known as blood pressure pulse curves), name, gender, age, arm circumference, heart rate, SBP (systolic blood pressure), and DBP (diastolic blood pressure), etc.

[0018] More specifically, the blood pressure data acquisition device in this embodiment uses STM32F103C8T6 as the microprocessor. The selected ADS1255 is a 24-bit, 2-channel high-performance AD sampling chip with a sampling frequency of 200 Hz and an analog-to-digital conversion accuracy of not less than 16 bits. The brachial artery blood pressure signal collected by the system is composed of the pulse wave of the AC component and the cuff static pressure of the DC component. After the superimposed signal is processed by the filtering and amplification circuit, it is externally connected to a 24-bit AD conversion chip and transmitted to the MCU for preprocessing. The data is finally uploaded to the host computer through USB. The blood pressure data acquisition device complies with current domestic and foreign metrological calibration specification requirements such as IEC 80601-2-30, OIML-R149, and JJG 692-2010. The blood pressure data acquisition device needs to meet metrological requirements such as the static pressure measurement range, the maximum allowable error of the static pressure indication value, and the repeatability of the blood pressure indication value.

[0019] Furthermore, there is high-frequency noise in the blood pressure pulse wave signal collected by the host computer, and it is necessary to filter out the cuff static pressure interference using, for example, a 5 Hz low-pass filter. To further eliminate the slight jitter noise during the measurement process, filtering processing is also required. In this embodiment, the Savitzky-Golay (S-G) filtering algorithm is preferably used. S-G filtering is a convolution smoothing method based on least squares fitting. It maintains the local characteristics of the signal by performing polynomial fitting on the data within a sliding window. The moving window slides point by point along the entire signal and fits, and the value of this polynomial at the center point of the window is used as the result after filtering. The filtering effect can be evaluated by the signal-to-noise ratio (SNR), mean square error (MSE), peak signal-to-noise ratio (PSNR), and similarity The calculation formulas of which belong to the prior art and will not be elaborated here. SNR is used to measure the ratio of the useful components to the noise in the signal. MSE is the difference between the filtered signal and the true signal. PSNR evaluates the signal quality, and similarity is used to measure the similarity between two signals. Higher SNR, PSNR, and as well as lower MSE mean better filtering effect. In this embodiment, the effects of smoothing filtering and S-G filtering are compared. The results show that the S-G filtering method effectively removes noise and maximally retains key characteristics such as the peak and valley values and inflection points of the original pulse wave. The S-G filtering has a significantly higher signal-to-noise ratio, can more effectively retain the useful information in the signal, and at the same time reduce the influence of noise. At the same time, the S-G filtering has a lower MSE, better PSNR, and higher indicating that the S-G filtering method has a better effect.

[0020] Further, the wave peaks of the blood pressure pulse wave signal after S-G filtering are detected by the adaptive threshold method, the time intervals between adjacent wave peaks are calculated, and the average value of a continuous plurality of time intervals is taken as the period of the blood pressure pulse wave signal , and then the heart rate of each patient is calculated by the following formula : .

[0021] Step S102, construct the blood pressure pulse wave envelope by the double Gaussian fitting algorithm, and estimate the mean arterial pressure of each patient in the patient sample population according to the blood pressure pulse wave envelope.

[0022] It is generally considered that the cuff static pressure corresponding to the peak moment of the pulse wave signal is the value of the mean arterial pressure MAP. However, since the peak points of the pulse wave are relatively sparse, it is difficult to determine the accurate value of MAP only by finding the peak points of the pulse wave. Therefore, in this embodiment, the MAP value is determined by constructing an envelope for the blood pressure pulse wave.

[0023] Specifically, the blood pressure pulse wave curve during the cuff deflation process is composed of multiple peaks and periodic pulse wave curves with different shapes, and the peak of the pulse wave curve conforms to the rule of "small - large - small". The peak points of the pulse wave ( , ), where , the data points are concentrated near the mean value, and the probability of occurrence is smaller the farther away from the mean value, which generally conforms to the Gaussian distribution. Considering that the shape of the blood pressure pulse wave can generally be divided into the ascending branch and the descending branch, therefore, in this embodiment, a double Gaussian model is introduced to fit the blood pressure pulse wave envelope: ; where, , , represent the amplitude, mean value and standard deviation of the ascending branch of the blood pressure pulse wave, , , represent the amplitude, mean value and standard deviation of the descending branch of the blood pressure pulse wave, represents the abscissa value of each point on the blood pressure pulse wave envelope, represents the ordinate fitting value of each point on the blood pressure pulse wave envelope; According to the following formula, calculate the minimum residual sum of squares of the observed value and the fitting value of each point on the blood pressure pulse wave envelope to obtain the optimal parameters at this time , , , , , : ; where, represents the The observed value of the blood pressure pulse wave at the -th point, The fitted value of the blood pressure pulse wave at the represents the total number of data points on the envelope line of the blood pressure pulse wave; Taking the optimal parameter as the peak value of the envelope line of the blood pressure pulse wave, and taking the cuff pressure value corresponding to the peak moment of the envelope line of the blood pressure pulse wave as the mean arterial pressure.

[0024] In this embodiment, the double Gaussian fitting method is adopted. By the least squares method, the best parameters are found to minimize the sum of squared residuals. The double Gaussian fitting not only improves the fitting accuracy of complex waveforms, but also can separate different components in the systolic and diastolic periods to a certain extent, so as to more accurately estimate the MAP value.

[0025] This embodiment also compares double Gaussian fitting, single Gaussian fitting, triple Gaussian fitting, and quadruple Gaussian fitting. The results are as follows: (1) Compared with single Gaussian fitting, the normalized mean square error (NMSE) of double Gaussian fitting is reduced by 0.042, and the linear fitting degree R2 is increased to 0.9558. It is superior to triple Gaussian and quadruple Gaussian fitting in terms of the Akaike information criterion AIC and Bayesian information criterion BIC, and achieves a better balance between fitting accuracy and model complexity. (2) Refer to Figure 3 ., the peak value of the envelope line of double Gaussian fitting and the corresponding time coordinates of multi-Gaussian fitting are approximately the same, ensuring the accuracy of MAP acquisition, and at the same time avoiding the overfitting risk of multi-Gaussian fitting. After the blood pressure pulse wave curve passes through the double Gaussian fitting envelope line, the correlation analysis is carried out between the collected human body information and the 6 parameters of the double Gaussian function. The results show that the amplitude parameter of the first Gaussian function has a correlation coefficient of 0.41 with SBP (systolic blood pressure), and the amplitude parameter of the second Gaussian function has a correlation coefficient of 0.48 with SBP (systolic blood pressure). In contrast, the correlation between DBP (diastolic blood pressure) and Gaussian function parameters is weak, because DBP (diastolic blood pressure) mainly affects the diastolic characteristics of the pulse wave and has little impact on the overall shape of the envelope line.

[0026] Step S103, calculate the correlation degree between each patient's multiple human physiological parameters and the mean arterial pressure in the patient sample population respectively.

[0027] This step analyzes and compares the relationships between human physiological parameters such as gender, age, arm circumference, heart rate, blood pressure, etc. and the mean arterial pressure MAP through the Pearson correlation coefficient . Specifically, calculate the Pearson correlation coefficient between each patient's multiple human physiological parameters and the mean arterial pressure in the patient sample population according to the following formula : ; Among them, represents the observed value of a certain human physiological parameter of the -th patient, represents the mean arterial pressure of the -th patient, represents the average value of a certain human physiological parameter of all patients, represents the average value of the mean arterial pressure of all patients. By comparing the Pearson correlation coefficients of various physiological parameters, the correlation between multiple human physiological parameters and mean arterial pressure can be obtained.

[0028] Step S104: Use the human physiological parameter variables with a correlation greater than the preset value as the input of the fully connected neural network, and use the predicted value of the mean arterial pressure as the output of the fully connected neural network to construct a training sample set and a validation sample set, and train the fully connected neural network.

[0029] Figure 4 is a heat map reflecting the correlation between information. Through this heat map, the correlation between the patient's physiological characteristics and MAP can be analyzed. From the heat correlation map and coefficients in the upper right triangle, the correlation coefficients between MAP and SBP, DBP are 0.89 and 0.87, showing a very strong correlation. The histogram on the diagonal shows the distribution characteristics of each variable, while the scatter plot in the lower left triangle shows the specific relationship between each variable. Based on the above analysis, in this embodiment, 4 physiological parameters, namely SBP, DBP, age, and arm circumference, are used as significant influencing factors of the mean arterial pressure MAP and used as input variables for the subsequent fully connected neural network.

[0030] See Figure 5 , and use a fully connected neural network FCNN to predict the mean arterial pressure MAP. The input layer of the fully connected neural network receives 4 features, namely SBP, DBP, arm circumference, and heart rate. The neural network includes an input layer, 5 hidden layers, and an output layer. The number of neurons in the hidden layers is 128, 64, 32, 16, and 8 respectively. Each layer is followed by a ReLU activation function to introduce non-linearity. During training, the mean squared error loss function is used as the loss function: ; Among them, represents the true value of the -th sample; represents the predicted value of the -th sample; n is the number of samples. The Adam optimizer is used for parameter update, the learning rate is set to 0.01, and L2 regularization is introduced to prevent overfitting. The sample number ratio of the training set to the validation set is 9:1, and finally a trained fully connected neural network is obtained.

[0031] Step S105, predicting the mean arterial pressure of the patient using the trained fully-connected neural network.

[0032] Specifically, predicting the mean arterial pressure of the patient using the trained fully-connected neural network. Figure 6 The implementation effects of predicting MAP by the fully-connected neural network FCNN, predicting MAP by the linear regression method, and calculating MAP by the empirical formula were compared. By analyzing the relationship between the MAP obtained by FCNN, linear regression, and the empirical formula and the true MAP, as well as the residual scatter plot, it can be seen that the prediction result of FCNN has the smallest residual with the true value and the highest prediction accuracy. The Bland-Altman consistency test shows that the maximum absolute difference between the predicted value of FCNN and the true value is 10.63, the average difference is 0.43, and only 10 cases exceed the 95% confidence interval, indicating good consistency. The maximum absolute differences of linear regression and the empirical formula are 12.63 and 11.67 respectively, and the average differences are 0.85 and -2.36 respectively. Among them, 17 cases of the empirical formula exceed the confidence interval, indicating poor consistency. The ROC curve graph shows that the AUC of FCNN and the linear fitting method is 0.96, which is better than 0.95 of the empirical formula, and the classification performance is better.

[0033] The method for predicting mean arterial pressure based on a fully-connected neural network provided in this embodiment collects blood pressure data at the brachial artery of numerous patients through the auscultation method, which is the "gold standard" for clinical blood pressure measurement. Then, the Gaussian fitting is used to fit the envelope of the blood pressure pulse wave to obtain the true mean arterial pressure MAP value. By analyzing the correlation between various variables, a fully-connected neural network is designed. After inputting the human physiological parameter data closely related to the mean arterial pressure, it can accurately predict the mean arterial pressure MAP value.

[0034] See Figure 7 , another embodiment of the present invention also provides a device 200 for predicting mean arterial pressure based on a fully-connected neural network, including a first module 201, a second module 202, a third module 203, a fourth module 204, and a fifth module 205. The device 200 can execute the method for predicting mean arterial pressure based on a fully-connected neural network in the method embodiment.

[0035] Specifically, the device 200 for predicting mean arterial pressure based on a fully-connected neural network includes: The first module 201 is configured to collect the blood pressure pulse wave signals of each patient in the patient sample population; The second module 202 is configured to construct the envelope of the blood pressure pulse wave through the double-Gaussian fitting algorithm and estimate the mean arterial pressure of each patient in the patient sample population according to the envelope of the blood pressure pulse wave; The third module 203 is configured to calculate the correlation between multiple human physiological parameters of each patient in the patient sample group and the mean arterial pressure respectively; The fourth module 204 is configured to use the human physiological parameter variables with a correlation greater than a preset value as the input of the fully connected neural network, and use the predicted value of the mean arterial pressure as the output of the fully connected neural network to construct a training sample set and a validation sample set, and train the fully connected neural network; The fifth module 205 is configured to predict the mean arterial pressure of the patient using the trained fully connected neural network.

[0036] It should be noted that the device 200 for predicting the mean arterial pressure based on the fully connected neural network provided in this embodiment corresponds to the technical solutions that can be used to execute the method embodiments. Its implementation principle and technical effects are similar to those of the method, and will not be elaborated here.

[0037] Figure 8 FIG. is a schematic structural diagram of an electronic device 300 provided in another embodiment of the present invention. The electronic device is used to implement the method for predicting the mean arterial pressure based on the fully connected neural network in the method embodiments. The electronic device 300 in the embodiments of the present invention may include, but is not limited to, a WAF device. Figure 8 The illustrated electronic device is only an example and should not impose any limitation on the functions and usage scope of the embodiments of the present invention.

[0038] As Figure 8 shown, the electronic device 300 may include a processing device (such as a central processing unit, a graphics processing unit, etc.) 301, which can execute various appropriate actions and processes according to the program stored in the read-only memory (ROM) 302 or the program loaded from the storage device 308 into the random access memory (RAM) 303 to implement the method of the embodiments described in the present invention. In the RAM 303, various programs and data required for the operation of the electronic device 300 are also stored. The processing device 301, the ROM 302, and the RAM 303 are connected to each other through a bus 304. The input / output (I / O) interface 305 is also connected to the bus 304.

[0039] Generally, the following devices may be connected to the I / O interface 305: an input device 306 including, for example, a touch screen, a touchpad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; an output device 307 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 308 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 309. The communication device 309 may allow the electronic device 300 to communicate with other devices wirelessly or wiredly to exchange data. Although Figure 8An electronic device 300 having various devices is shown, but it should be understood that it is not required to implement or have all the shown devices. Instead, more or fewer devices may be implemented or had.

[0040] The above description is only a preferred embodiment of the present invention. Those skilled in the art should understand that the scope of disclosure involved in the present invention is not limited to the technical solutions formed by the specific combination of the above technical features, and should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above disclosed concept. For example, the technical solutions formed by mutually replacing the above features with the technical features (but not limited to) having similar functions disclosed in the present invention.

Claims

1. A method for predicting mean arterial pressure based on a fully connected neural network, characterized in that: The steps include: Collecting the blood pressure pulse wave signal of each patient in the patient sample group, and performing SG filtering on the blood pressure pulse wave signal; The blood pressure pulse wave envelope is constructed by a double Gaussian fitting algorithm, and the mean arterial pressure of each patient in the patient sample group is estimated according to the blood pressure pulse wave envelope; Calculate the correlation between multiple physiological parameters and mean arterial pressure for each patient in the patient sample group; Taking the human physiological parameter variables with correlation greater than the preset value as the input of the fully connected neural network, taking the predicted value of the mean arterial pressure as the output of the fully connected neural network, constructing a training sample set and a verification sample set, and training the fully connected neural network; The trained fully connected neural network was used to predict the mean arterial pressure of the patients.

2. The method for predicting mean arterial pressure based on a fully connected neural network according to claim 1, characterized in that: The step of constructing a blood pressure pulse wave envelope using a double Gaussian fitting algorithm and estimating the mean arterial pressure of each patient in the patient sample group according to the blood pressure pulse wave envelope includes: The blood pressure pulse wave envelope is fitted by the following double Gaussian model: ； in, , , It represents the amplitude, mean and standard deviation of the ascending branch of the blood pressure pulse wave. , , represents the amplitude, mean and standard deviation of the descending branch of the blood pressure pulse wave, Represents the horizontal coordinate value of each point on the blood pressure pulse wave envelope. It represents the ordinate fitting value of each point on the blood pressure pulse wave envelope; According to the following formula, the minimum residual sum of squares of the observed value and the fitted value of each point on the blood pressure pulse wave envelope is calculated to obtain the optimal parameter at this time: , , , , , : ； in, Indicates The observed value of the blood pressure pulse wave at each point, Indicates The fitted value of the blood pressure pulse wave at each point, Indicates the total number of data points on the blood pressure pulse wave envelope; The optimal parameters As the peak value of the blood pressure pulse wave envelope, the cuff pressure value corresponding to the peak moment of the blood pressure pulse wave envelope is taken as the mean arterial pressure.

3. The method for predicting mean arterial pressure based on a fully connected neural network according to claim 1, characterized in that: The steps of respectively calculating the correlation between a plurality of human physiological parameters and mean arterial pressure of each patient in the patient sample group include: The Pearson correlation coefficient between multiple physiological parameters and mean arterial pressure of each patient in the patient sample group was calculated according to the following formula: : ； in, Indicates The observed value of a certain physiological parameter of a patient, Indicates The mean arterial pressure of the patients It represents the average value of a physiological parameter of all patients. It represents the average value of mean arterial pressure of all patients.

4. The method for predicting mean arterial pressure based on a fully connected neural network according to claim 3, characterized in that: Also includes: The peak of the filtered blood pressure pulse wave signal is detected by the adaptive threshold method, the time interval between adjacent peaks is calculated, and the average value of multiple consecutive time intervals is taken as the period of the blood pressure pulse wave signal. ; The heart rate of each patient was calculated by : .

5. The method for predicting mean arterial pressure based on a fully connected neural network according to claim 1, characterized in that: The fully connected neural network includes an input layer, 5 hidden layers and an output layer; The number of neurons in the five hidden layers is 128, 64, 32, 16 and 8 respectively, and a ReLU activation function is connected after each hidden layer; The loss function adopts the mean square error loss function.

6. A device for predicting mean arterial pressure based on a fully connected neural network, characterized in that: include: A first module is configured to collect a blood pressure pulse wave signal of each patient in a patient sample group; The second module is configured to construct a blood pressure pulse wave envelope by a double Gaussian fitting algorithm, and estimate the mean arterial pressure of each patient in the patient sample group according to the blood pressure pulse wave envelope; The third module is configured to respectively calculate the correlation between a plurality of human physiological parameters and the mean arterial pressure of each patient in the patient sample group; The fourth module is configured to use the human physiological parameter variable with a correlation greater than a preset value as the input of the fully connected neural network, use the predicted value of the mean arterial pressure as the output of the fully connected neural network, construct a training sample set and a verification sample set, and train the fully connected neural network; The fifth module is configured to predict the patient's mean arterial pressure using the trained fully connected neural network.

7. The device for predicting mean arterial pressure based on a fully connected neural network according to claim 6, characterized in that: The second module is further configured as follows: The blood pressure pulse wave envelope is fitted by the following double Gaussian model: ； in, , , It represents the amplitude, mean and standard deviation of the ascending branch of the blood pressure pulse wave. , , represents the amplitude, mean and standard deviation of the descending branch of the blood pressure pulse wave, Represents the horizontal coordinate value of each point on the blood pressure pulse wave envelope. It represents the ordinate fitting value of each point on the blood pressure pulse wave envelope; According to the following formula, the minimum residual sum of squares of the observed value and the fitted value of each point on the blood pressure pulse wave envelope is calculated to obtain the optimal parameter at this time: , , , , , : ； in, Indicates The observed value of the blood pressure pulse wave at each point, Indicates The fitted value of the blood pressure pulse wave at each point, Indicates the total number of data points on the blood pressure pulse wave envelope; The optimal parameters As the peak value of the blood pressure pulse wave envelope, the cuff pressure value corresponding to the peak moment of the blood pressure pulse wave envelope is taken as the mean arterial pressure.

8. The device for predicting mean arterial pressure based on a fully connected neural network according to claim 6, characterized in that: The third module is further configured as follows: The Pearson correlation coefficient between multiple physiological parameters and mean arterial pressure of each patient in the patient sample group was calculated according to the following formula: : ； in, Indicates The observed value of a certain physiological parameter of a patient, Indicates The mean arterial pressure of the patients It represents the average value of a physiological parameter of all patients. It represents the average value of mean arterial pressure of all patients.

9. The device for predicting mean arterial pressure based on a fully connected neural network according to claim 8, characterized in that: It also includes a heart rate calculation module configured as follows: The peak of the filtered blood pressure pulse wave signal is detected by the adaptive threshold method, the time interval between adjacent peaks is calculated, and the average value of multiple consecutive time intervals is taken as the period of the blood pressure pulse wave signal. ; The heart rate of each patient was calculated by : .

10. The device for predicting mean arterial pressure based on a fully connected neural network according to claim 6, characterized in that: The fully connected neural network includes an input layer, 5 hidden layers and an output layer; The number of neurons in the five hidden layers is 128, 64, 32, 16 and 8 respectively, and a ReLU activation function is connected after each hidden layer; The loss function adopts the mean square error loss function.

Images