A blood pressure monitoring method based on dynamic blood pressure monitoring data feature extraction

Abstract

The application discloses a blood pressure monitoring method based on dynamic blood pressure monitoring data feature extraction, comprising the following steps: acquiring dynamic blood pressure monitoring data; extracting multi-dimensional features of the dynamic blood pressure monitoring data, wherein the multi-dimensional features include time domain features, frequency domain features, nonlinear dynamic features and correlation relationship features; constructing a classification model, training the classification model based on the multi-dimensional features; and classifying and monitoring the dynamic blood pressure monitoring data based on the trained classification model. The application comprehensively considers time domain, frequency domain and nonlinear features, adopts a multi-dimensional feature fusion mode, and can more comprehensively describe the features of the blood pressure signal. The time domain features can reflect the statistical characteristics of the blood pressure signal, the frequency domain features can reveal the frequency components of the signal, and the nonlinear features can capture the complex dynamic changes of the signal. By comprehensively utilizing these features, the dynamic change law of the blood pressure data can be more accurately described.

Description

Technical Field

[0001] This invention belongs to the field of biomedical engineering technology, and in particular relates to a blood pressure monitoring method based on feature extraction of dynamic blood pressure monitoring data. Background Technology

[0002] Ambulatory blood pressure monitoring (ABPM) involves the continuous monitoring and recording of an individual's blood pressure over a long period to obtain more accurate and comprehensive blood pressure information. Traditional manual blood pressure measurement involves doctors manually measuring a patient's systolic and diastolic blood pressure using a stethoscope and sphygmomanometer. While simple and easy to perform, this method only measures short-term blood pressure values ​​and cannot reflect changes in a patient's blood pressure throughout their daily life. ABPM, on the other hand, can record a patient's blood pressure continuously or intermittently, including values ​​during the day, night, at rest, and during activity. However, the data generated by ABPM is massive and complex, making the extraction and analysis of useful features a critical task.

[0003] Currently, the processing and analysis of ambulatory blood pressure data typically employs methods for feature extraction from the time and frequency domains. While these methods have been widely used in medical research and have yielded some results, they also have limitations. Due to the diversity of ambulatory blood pressure data, feature extraction solely from the time and frequency domains may not fully reflect the characteristics of the samples, potentially leading to poor performance on certain datasets or tasks. Therefore, a blood pressure monitoring method based on feature extraction from ambulatory blood pressure monitoring data is needed, capable of extracting more comprehensive feature information from blood pressure data to support the analysis, assessment, and prediction of blood pressure status. Summary of the Invention

[0004] The purpose of this invention is to provide a blood pressure monitoring method based on feature extraction from ambulatory blood pressure monitoring data. This method employs a multi-feature fusion extraction approach to fully extract the rich information from blood pressure data, thereby improving the accuracy of blood pressure status assessment and monitoring. This better reflects the comprehensiveness of ambulatory blood pressure data and addresses the problems existing in the prior art.

[0005] To achieve the above objectives, the present invention provides a blood pressure monitoring method based on feature extraction from ambulatory blood pressure monitoring data, comprising the following steps:

[0006] Obtain ambulatory blood pressure monitoring data;

[0007] Extract multi-dimensional features from the dynamic blood pressure monitoring data, including time-domain features, frequency-domain features, nonlinear dynamic features, and correlation features;

[0008] Construct a classification model and train the classification model based on the multi-dimensional features;

[0009] Based on the trained classification model, the dynamic blood pressure monitoring data is classified and monitored.

[0010] Optionally, the dynamic blood pressure monitoring data includes four dimensions: systolic blood pressure, diastolic blood pressure, mean arterial pressure, and heart rate.

[0011] Optionally, before extracting the multi-dimensional features of the ambulatory blood pressure monitoring data, the method further includes: performing outlier removal and signal smoothing on the ambulatory blood pressure monitoring data, and storing the processed ambulatory blood pressure monitoring data in matrix form.

[0012] Optionally, the time-domain features include: the average value of the 24-hour monitoring range, the standard deviation of the 24-hour monitoring range, the coefficient of variation of the 24-hour monitoring range, the average true variability of the 24-hour monitoring range, and the relative variability index of the 24-hour monitoring range.

[0013] Optionally, the frequency domain features include: extremely low frequency power in the 24-hour monitoring range, low frequency power in the 24-hour monitoring range, high frequency power in the 24-hour monitoring range, total power in the 24-hour monitoring range, normalized low frequency power in the 24-hour monitoring range, normalized high frequency power in the 24-hour monitoring range, and the ratio of low frequency to high frequency power in the 24-hour monitoring range.

[0014] Optionally, the process of obtaining the frequency domain features includes: processing the heart rate sequence based on the Welch method to obtain the power spectral density; obtaining the extremely low frequency power, low frequency power, and high frequency power within the 24-hour monitoring range based on the power spectral density; and obtaining the total power, normalized low frequency power, normalized high frequency power, and low frequency to high frequency power ratio within the 24-hour monitoring range based on the functional relationship between the extremely low frequency power, low frequency power, and high frequency power within the 24-hour monitoring range.

[0015] Optionally, the nonlinear dynamic characteristics include: approximate entropy of the 24-hour monitoring range, fuzzy entropy, Lempel-Ziv complexity, and zero complexity.

[0016] Optionally, the process of obtaining the approximate entropy includes: setting a similarity judgment threshold, obtaining the number of target data points in the time series whose distance from other data points is less than the similarity judgment threshold, constructing a functional relationship based on the obtained number and the length and dimension of the signal, and obtaining the approximate entropy by subtraction;

[0017] The process of obtaining the fuzzy entropy includes: setting a preset window size, dividing the time series into several subsequences, obtaining the distance between each subsequence and all other subsequences, constructing a distance matrix, and obtaining fuzzy membership degree based on the distance matrix; increasing the window size to obtain the corresponding fuzzy membership degree; and obtaining the fuzzy entropy based on the difference between the fuzzy membership degree before and after the increase.

[0018] Optionally, the correlation features include: the ratio of systolic blood pressure to diastolic blood pressure Hjroth parameter positivity, the ratio of systolic blood pressure to diastolic blood pressure Hjroth parameter mobility, the ratio of systolic blood pressure to diastolic blood pressure Hjroth parameter complexity, the ratio of mean arterial pressure to heart rate Hjroth parameter positivity, the ratio of mean arterial pressure to heart rate Hjroth parameter mobility, and the ratio of mean arterial pressure to heart rate Hjroth parameter complexity.

[0019] Optionally, the process of obtaining the correlation features includes: obtaining the positivity, mobility, and complexity of the corresponding Hjorth parameters based on the functional relationship between the standard deviation of the original signal, the standard deviation of the first difference of the original signal, and the standard deviation of the second difference of the original signal; based on the positivity of the Hjorth parameters, obtaining the positivity of the Hjorth parameters for systolic and diastolic blood pressure, mean arterial pressure, and heart rate, respectively, and then obtaining the ratio of systolic to diastolic Hjorth parameter positivity and the ratio of mean arterial pressure to heart rate Hjorth parameter positivity; based on the mobility of the Hjorth parameters... The mobility of the Hjroth parameter was obtained for systolic and diastolic blood pressure, mean arterial pressure, and heart rate, respectively. The ratios of systolic and diastolic Hjroth parameter mobility and mean arterial pressure to heart rate Hjroth parameter mobility were then obtained. Based on the complexity of the Hjroth parameter, the complexity of the Hjroth parameter for systolic and diastolic blood pressure, mean arterial pressure, and heart rate was obtained, respectively. The ratios of systolic and diastolic Hjroth parameter complexity and mean arterial pressure to heart rate Hjroth parameter complexity were then obtained.

[0020] The technical effects of this invention are as follows:

[0021] 1. This invention comprehensively considers time-domain, frequency-domain, and nonlinear characteristics, employing a multi-dimensional feature fusion approach to more comprehensively describe the characteristics of blood pressure signals. Time-domain features reflect the statistical properties of blood pressure signals, frequency-domain features reveal the frequency components of the signal, and nonlinear characteristics capture the complex dynamic changes of the signal. By comprehensively utilizing these features, the dynamic variation patterns of blood pressure data can be described more accurately.

[0022] 2. This invention employs the Hjroth parameter to reflect the correlation between blood pressure and heart rate. The Hjroth parameter, by calculating parameters such as activity, mobility, and complexity, provides a method for assessing the association between blood pressure and heart rate from different perspectives. By introducing the Hjroth parameter and its ratio, this invention provides a new perspective and indicator for the analysis of ambulatory blood pressure monitoring data, further enriching the methods for assessing the correlation between blood pressure and heart rate. Attached Figure Description

[0023] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0024] Figure 1 This is a schematic diagram of the blood pressure monitoring method based on feature extraction of dynamic blood pressure monitoring data in an embodiment of the present invention;

[0025] Figure 2 This is a schematic diagram of the moving average smoothing result of the sample in the embodiment of the present invention; wherein, (a) is a schematic diagram of heart rate filtering, (b) is a schematic diagram of systolic blood pressure filtering, (c) is a schematic diagram of diastolic blood pressure filtering, and (d) is a schematic diagram of mean arterial pressure filtering.

[0026] Figure 3 The power spectrum of the sample in this embodiment of the invention is extracted using the Welch method in the frequency domain. Among them, (a) is the heart rate spectrum distribution, (b) is the systolic blood pressure spectrum distribution, (c) is the diastolic blood pressure spectrum distribution, and (d) is the mean arterial pressure spectrum distribution. Detailed Implementation

[0027] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0028] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0029] Example 1

[0030] like Figure 1 As shown, this embodiment provides a blood pressure monitoring method based on feature extraction from dynamic blood pressure monitoring data, including the following steps:

[0031] Step 1: Acquire Ambulatory Blood Pressure Monitoring Data: Simultaneously acquire four-dimensional ambulatory blood pressure monitoring data, store it in matrix form, and perform outlier removal and signal smoothing. The smoothing results are as follows: Figure 2 As shown.

[0032] In a specific implementation, the database contains 262 individuals, including 169 in Group 1 and 93 in Group 2. Each individual has four indicators: systolic blood pressure, diastolic blood pressure, mean arterial pressure, and heart rate. These indicators are stored in a matrix format, represented as: x l (n), l=1,2,3,4; n=1,2,3,…,N, where i represents the data dimension and N represents the length of the monitoring data sequence. The preprocessing steps are to first remove special symbols and outliers from the monitoring signal, and then use a moving average digital filtering method to eliminate noise and irregularities for x. l Given the filter window size M, the digital filter is calculated as follows: (n)

[0033]

[0034] Where, x l (n) represents the original data of the l-th dimension sequence, y l (t) represents the filtered output signal in the l-th dimension.

[0035] Step 2, Multi-dimensional Feature Extraction:

[0036] Calculate multidimensional feature indicators for the preprocessed dynamic blood pressure monitoring data from step 1: time domain features, frequency domain features, nonlinear dynamic features, and correlation features.

[0037] Specifically, the average value, standard deviation, coefficient of variation, mean true variability, and relative variability index of the 24-hour monitoring range are extracted as time-domain features to describe the time-varying characteristics of the signal; the extremely low-frequency power, low-frequency power, high-frequency power, total power, normalized low-frequency power, normalized high-frequency power, and low-frequency to high-frequency power ratio of the 24-hour monitoring range are extracted as frequency-domain features to describe the power distribution characteristics of the signal; and the 24-hour monitoring range... The approximate entropy of the monitoring range, the fuzzy entropy of the 24-hour monitoring range, the Lempel-Ziv complexity of the 24-hour monitoring range, and the zero complexity are used as nonlinear dynamic features to describe the complexity and nonlinear characteristics of the signal. Based on the Hjorth parameter, the Hjroth positivity ratio of systolic blood pressure to diastolic blood pressure, the Hjroth mobility ratio of systolic blood pressure to diastolic blood pressure, the Hjroth complexity ratio of systolic blood pressure to diastolic blood pressure, the Hjroth positivity ratio of mean arterial pressure to heart rate, the Hjroth mobility ratio of mean arterial pressure to heart rate, and the Hjroth complexity ratio of mean arterial pressure to heart rate are extracted as correlation indicators to describe the correlation characteristics of blood pressure and heart rate signals.

[0038] The time-domain features include: the average value of the 24-hour monitoring range, the standard deviation of the 24-hour monitoring range, the coefficient of variation of the 24-hour monitoring range, the average true variability of the 24-hour monitoring range, and the relative variability index of the 24-hour monitoring range.

[0039] The average value of the 24-hour monitoring range is calculated as follows:

[0040]

[0041] The standard deviation of the 24-hour monitoring range is calculated as follows:

[0042]

[0043] The coefficient of variability for the 24-hour monitoring range is calculated as follows:

[0044]

[0045] The average true variability of the 24-hour monitoring range is calculated as follows:

[0046]

[0047] The relative variability index of the 24-hour monitoring range is calculated as follows:

[0048]

[0049] Where k represents the proportionality constant, with a value of 1, and m represents the fitting coefficient of the nonlinear regression curve, which is calculated as the fitting coefficient of the natural logarithm of the standard deviation divided by the natural logarithm of the mean.

[0050] Frequency domain extraction, such as Figure 3 As shown, the power spectral density of the heart rate sequence is extracted using the Welch method. Fourier transform is performed at a sampling frequency of 4Hz, and the data of length L is then processed. l (m) Divide the spectrum into segments, each of length M. Calculate the frequency spectrum of each segment separately, and then average them as follows:

[0051]

[0052] Where U is the normalization factor, w(n) is the window function, L is the length of each sub-segment, M is the number of sub-segments, f is the frequency, and T is the sampling time interval.

[0053] The energy spectra of blood pressure and heart rate were calculated using the power spectrum estimation method, and then divided into the following frequency ranges for the 24-hour monitoring range: very low frequency power (VLF = 0.003-0.04 Hz), low frequency power (LF = 0.04-0.15 Hz), high frequency power (HF = 0.15-0.4 Hz), and total power spectrum. Next, the standardized low frequency power, standardized high frequency power, and low frequency to high frequency power ratio for the 24-hour monitoring range were calculated in standardized units, resulting in seven frequency domain characteristics.

[0054] The calculation of extremely low frequency power within the 24-hour monitoring range is as follows:

[0055]

[0056] The low-frequency power calculation for the 24-hour monitoring range is as follows:

[0057]

[0058] The high-frequency power calculation for the 24-hour monitoring range is as follows:

[0059]

[0060] The total power within the 24-hour monitoring range is calculated as follows:

[0061] TP i =VLF i +LF i +HF i ,

[0062] The standardized low-frequency power calculation for the 24-hour monitoring range is as follows:

[0063]

[0064] The standardized high-frequency power calculation for the 24-hour monitoring range is as follows:

[0065]

[0066] The low-frequency to high-frequency power ratio for the 24-hour monitoring range is calculated as follows:

[0067]

[0068] Where i = 1, 2, 3, 4 represents the dimension.

[0069] Nonlinear dynamic characteristics include the approximate entropy of the 24-hour monitoring range, fuzzy entropy, Lempel-Ziv complexity, and C0 complexity.

[0070] The approximate entropy calculation method is as follows:

[0071] Given a phase space dimension m of 1 and a similarity tolerance r of 0.15.

[0072] Define C m (i,r) represents the number of data points in the time series whose distance to the i-th data point is less than r, specifically calculated as follows:

[0073]

[0074] Wherein, Θ(r-|y i-1 -y j-1 |) is an indicator function, when |y i-1 -y j-1 When | ≤ r, the value is 1; otherwise, the value is 0. m represents the dimension, r represents the similarity threshold, and N represents the signal length. Define φ. m The r function is as follows:

[0075]

[0076] The approximate entropy of the l-th dimension of the nonlinear dynamics term can then be obtained from the following equation:

[0077] ApEn l (m,r,N)=φ m (r)-φ m+1 (r); m≥0,

[0078] The method for calculating fuzzy entropy is as follows:

[0079] Given a window size m = 1, a threshold coefficient r = 0.25, and a fuzzy membership dimension n = 2.

[0080] (1) Divide the time series into n-m+1 subsequences Y with a window size of m. i (t)=(y i (t),y i+1 ,…,y i+m-1 (t));

[0081] (2) Calculate the distance between each subsequence and all other subsequences, construct the distance matrix, and list the results in table d. ij =max{|y i+k (t)-y j+k (t)|}, which is the maximum absolute value of the interpolation of corresponding elements of the two vectors, where k=0,1,…,m-1;

[0082] (3) Calculate the fuzzy membership degree based on the distance d. Calculate the mean of all membership degrees except itself.

[0083] (4) Increase the window size m to m+1 and repeat the above three steps. Then the fuzzy entropy of the l-th dimension data can be calculated by the following formula: FuEn l (t)=lnΦ m (t)-lnΦ m+1 (t).

[0084] The Lempel-Ziv complexity calculation method is as follows:

[0085] Represent sequence x as a 01 string, where elements greater than the average are mapped to 1, and elements less than the average are mapped to 0. Initialize the pattern counter variable c to 1, and initialize sequence Q to empty. Initialize sequence S to the first element of x. Initialize sequence Q to empty.

[0086] Enter the loop and iterate from the second element of x: a. Append the current element to the end of sequence Q; b. Concatenate sequences S and Q to get SQ; c. Remove the last character of SQ to get SQv; d. Check if Q is a substring of SQv. If Q is not a substring, it means Q is a new pattern, so assign SQ to S. Clear sequence Q. Increment the pattern count variable c by 1.

[0087] Repeat the above steps until all elements in S have been processed.

[0088] calculate Where N is the sequence length, the LZ complexity of the l-th dimension can be expressed by the following formula:

[0089] The C0 complexity is calculated as follows:

[0090] For a given time series Y of length N l(n) is subjected to a discrete Fourier transform to obtain Y l (k). The mean square value is defined as:

[0091]

[0092] Introducing a parameter r, we retain the spectrum exceeding the mean square value by r times, while setting the rest to 0, that is...

[0093]

[0094] Pair Perform inverse Fourier transform to Then the l-th dimension sequence The complexity can be obtained from the following formula:

[0095]

[0096] The correlation characteristics include: the positive ratio of systolic blood pressure to diastolic blood pressure Hjroth parameter, the mobility ratio of systolic blood pressure to diastolic blood pressure Hjroth parameter, the complexity ratio of systolic blood pressure to diastolic blood pressure Hjroth parameter, the positive ratio of mean arterial pressure to heart rate Hjroth parameter, the mobility ratio of mean arterial pressure to heart rate Hjroth parameter, and the complexity ratio of mean arterial pressure to heart rate Hjroth parameter.

[0097] The calculation method for the 24-hour monitoring range Hjorth parameter is as follows:

[0098] Ac l =σ,

[0099]

[0100]

[0101] Where σ represents the standard deviation of the original signal. σ' represents the standard deviation of the first-order difference of the original signal. σ″ represents the standard deviation of the second-order difference of the original signal. Ac l Mo l Co l These represent the positivity, mobility, and complexity of the Hjorth parameter in the l-th dimension, respectively.

[0102] The correlation based on the Hjorth parameter is calculated as follows:

[0103]

[0104] Where Ac1 represents the positivity of the systolic blood pressure Hjroth parameter, and Ac2 represents the positivity of the diastolic blood pressure Hjroth parameter.

[0105]

[0106] Where Mo1 represents the mobility of the systolic Hjroth parameter, and Mo2 represents the mobility of the diastolic Hjroth parameter.

[0107]

[0108] Co1 represents the complexity of the systolic Hjroth parameter, and Co2 represents the complexity of the diastolic Hjroth parameter.

[0109]

[0110] Ac3 represents the activity of the mean arterial pressure Hjroth parameter, and Ac4 represents the activity of the heart rate Hjroth parameter.

[0111]

[0112] Wherein, Mo3 represents the mobility of the mean arterial pressure Hjroth parameter, and Mo4 represents the mobility of the heart rate Hjroth parameter.

[0113]

[0114] Where Co3 represents the complexity of the mean arterial pressure Hjroth parameter, and Co4 represents the complexity of the heart rate Hjroth parameter.

[0115] Step 3: Classification and Recognition

[0116] A classification model is built based on a supervised classification learner. The features fused in step 2 are used to train the supervised learning classifier. Based on the differences between the training mode and the test mode regarding blood pressure monitoring indicators, the blood pressure monitoring data is classified.

[0117] The feature vectors are cross-validated using multiple general-purpose supervised learning classifiers. The dataset is divided into training and test sets. The classifier model is trained using the training set and evaluated and validated using the test set. Based on the cross-validation results, the performance and accuracy of the classifier are evaluated.

[0118] Based on the multi-dimensional feature vectors extracted in step 2, there are 5 time-domain features, 7 frequency-domain features, and 4 non-linear features. There are four sequences: systolic blood pressure, diastolic blood pressure, mean arterial pressure, and heart rate. Each sequence has 16 feature vectors (5 time-domain features + 7 frequency-domain features + 4 non-linear features), resulting in a total of 64 feature vectors across the four sequences. Considering the correlations in step 2, there are a total of 70 feature vectors (64 feature vectors + 6 correlation feature vectors). The validation method uses 10-fold cross-validation and averaging. The original dataset is randomly divided into ten equal subsets, one as the validation set and the remaining nine as the training set, resulting in ten models. Each model is trained and validated using different data. The advantage is that every sample is used for both validation and training, thus allowing for a more accurate evaluation of model performance.

[0119] Table 1 shows the recognition rate (CCR) of the recognition experiment in this embodiment. The experimental data illustrates the superiority of this embodiment. As can be seen from the table, the method of this embodiment can achieve a high recognition rate under multiple cross-validation tests, and has high robustness.

[0120] Table 1

[0121] 0 91.9％ 89.6％ 91.7％ 93.8％ 1 90.4％ 85.9％ 89.9％ 94.3％ 2 91.5％ 89.0％ 91.3％ 93.8％ 3 91.5％ 91.9％ 91.6％ 91.2％ 4 85.6％ 84.4％ 85.4％ 86.4％ 5 91.9％ 91.1％ 91.8％ 92.5％ 6 90.4％ 85.9％ 89.9％ 94.3％ 7 90.7％ 89.6％ 90.6％ 91.7％ 8 88.9％ 85.2％ 88.5％ 92.0％ 9 92.2％ 88.1％ 91.9％ 96.0％ average value 90.5％ 88.1％ 90.3％ 92.6％

[0122] The above description is merely a preferred embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A blood pressure monitoring method based on feature extraction from ambulatory blood pressure monitoring data, characterized in that, Includes the following steps: Obtain ambulatory blood pressure monitoring data; Extract multi-dimensional features from the dynamic blood pressure monitoring data, including time-domain features, frequency-domain features, nonlinear dynamic features, and correlation features; Construct a classification model and train the classification model based on the multi-dimensional features; Based on the trained classification model, the dynamic blood pressure monitoring data is classified and monitored. The correlation features include: the positive ratio of systolic blood pressure to diastolic blood pressure Hjroth parameter, the mobility ratio of systolic blood pressure to diastolic blood pressure Hjroth parameter, the complexity ratio of systolic blood pressure to diastolic blood pressure Hjroth parameter, the positive ratio of mean arterial pressure to heart rate Hjroth parameter, the mobility ratio of mean arterial pressure to heart rate Hjroth parameter, and the complexity ratio of mean arterial pressure to heart rate Hjroth parameter. The process of obtaining the correlation features includes: obtaining the positivity, mobility, and complexity of the corresponding Hjorth parameters based on the functional relationship between the standard deviation of the original signal, the standard deviation of the first-order difference of the original signal, and the standard deviation of the second-order difference of the original signal; based on the positivity of the Hjorth parameters, obtaining the positivity of the Hjorth parameters for systolic and diastolic blood pressure, mean arterial pressure, and heart rate, respectively, and then obtaining the ratio of systolic to diastolic Hjorth parameter positivity and the ratio of mean arterial pressure to heart rate Hjorth parameter positivity; based on the mobility of the Hjorth parameters... The Hjroth parameter mobility of systolic and diastolic blood pressure, mean arterial pressure, and heart rate were obtained, respectively. The ratios of systolic and diastolic Hjroth parameter mobility and mean arterial pressure to heart rate Hjroth parameter mobility were then obtained. Based on the complexity of the Hjroth parameter, the complexity of systolic and diastolic Hjroth parameters, mean arterial pressure, and heart rate Hjroth parameter were obtained, respectively. The ratios of systolic and diastolic Hjroth parameter complexity and mean arterial pressure to heart rate Hjroth parameter complexity were then obtained. The time-domain features include: the average value of the 24-hour monitoring range, the standard deviation of the 24-hour monitoring range, the coefficient of variation of the 24-hour monitoring range, the average true variability of the 24-hour monitoring range, and the relative variability index of the 24-hour monitoring range; The frequency domain characteristics include: extremely low frequency power in the 24-hour monitoring range, low frequency power in the 24-hour monitoring range, high frequency power in the 24-hour monitoring range, total power in the 24-hour monitoring range, normalized low frequency power in the 24-hour monitoring range, normalized high frequency power in the 24-hour monitoring range, and the ratio of low frequency to high frequency power in the 24-hour monitoring range. The nonlinear dynamic characteristics include: approximate entropy, fuzzy entropy, Lempel-Ziv complexity, and zero complexity for a 24-hour monitoring range.

2. The blood pressure monitoring method based on feature extraction from dynamic blood pressure monitoring data according to claim 1, characterized in that, The dynamic blood pressure monitoring data includes four dimensions: systolic blood pressure, diastolic blood pressure, mean arterial pressure, and heart rate.

3. The blood pressure monitoring method based on feature extraction of dynamic blood pressure monitoring data according to claim 2, characterized in that, Before extracting the multi-dimensional features of the ambulatory blood pressure monitoring data, the process further includes: removing outliers and smoothing the signal of the ambulatory blood pressure monitoring data, and storing the processed ambulatory blood pressure monitoring data in matrix form.

4. The blood pressure monitoring method based on feature extraction from dynamic blood pressure monitoring data according to claim 1, characterized in that, The process of obtaining the frequency domain features includes: processing the heart rate sequence based on the Welch method to obtain the power spectral density; obtaining the extremely low frequency power, low frequency power, and high frequency power within the 24-hour monitoring range based on the power spectral density; and obtaining the total power, normalized low frequency power, normalized high frequency power, and low frequency to high frequency power ratio within the 24-hour monitoring range based on the functional relationship between the extremely low frequency power, low frequency power, and high frequency power within the 24-hour monitoring range.

5. The blood pressure monitoring method based on feature extraction from ambulatory blood pressure monitoring data according to claim 1, characterized in that, The process of obtaining the approximate entropy includes: setting a preset similarity judgment threshold, obtaining the number of target data points in the time series whose distance from other data points is less than the similarity judgment threshold, constructing a functional relationship based on the obtained number and the length and dimension of the signal, and obtaining the approximate entropy by subtraction; The process of obtaining the fuzzy entropy includes: setting a preset window size, dividing the time series into several subsequences, obtaining the distance between each subsequence and all other subsequences, constructing a distance matrix, and obtaining fuzzy membership degree based on the distance matrix; increasing the window size to obtain the corresponding fuzzy membership degree; and obtaining the fuzzy entropy based on the difference between the fuzzy membership degree before and after the increase.

Images