Blood pressure measurement system and method based on oscillation wave dynamic characteristics and regression modeling

Abstract

The present invention provides a blood pressure measurement system and method based on the dynamic characteristics of oscillation waves and regression modeling, which relates to the field of wearable medical health monitoring technology. With signal processing methods as the main body, the time domain, frequency domain, time-frequency and statistical methods are used to extract the key features of the oscillation waves, and then a lightweight deep learning model is used to fine-tune the traditional features. Finally, a mapping relationship between the characteristic band changes and the blood pressure values is established through machine learning. It can not only predict the complete oscillation wave waveform through the characteristic changes of part of the oscillation wave to achieve low-pressure comfortable measurement, but also realize personalized blood pressure measurement by identifying the mutation points of the oscillation wave characteristics. The present invention not only ensures the interpretability and stability of the traditional method, but also uses lightweight deep learning fine-tuning to improve the feature recognition accuracy and individual adaptability, and can realize edge computing, thereby reducing measurement time, improving user comfort, measurement accuracy and system practicality.

Description

Technical Field

[0001] The present invention relates to the field of wearable medical health monitoring technology, and in particular to a blood pressure measurement system and method based on oscillation wave dynamic characteristics and regression modeling. Background Art

[0002] In the field of non-invasive blood pressure measurement, the traditional oscillometric method uses a single-balloon cuff and empirical formulas to calculate blood pressure values, but it has many drawbacks. The complex human blood circulation can easily induce hemodynamic fluctuations during measurement, affecting the oscillation wave shape and resulting in poor measurement accuracy. It also has poor adaptability to individual physiological differences and dynamic blood pressure fluctuations, and measurement accuracy drops significantly when faced with abnormal pulse waves or motion interference. Furthermore, the hardware cost is high. While multi-balloon or multi-sensor solutions can improve accuracy, they increase system complexity and manufacturing costs, hindering the large-scale promotion and application of the technology. These issues have become the main obstacles to the development of non-invasive blood pressure measurement technology.

[0003] Chinese patent CN116509354A utilizes machine learning algorithms to construct SBP and DBP prediction models. While this improves the specificity of blood pressure calculations, it relies on a large amount of labeled training data, resulting in complex feature extraction and dimensionality reduction operations and high computational cost, making it difficult to run in real time on resource-constrained wearable devices. Furthermore, the model's generalization is limited, resulting in poor prediction accuracy for patients with hypertension or hypotension. Chinese patent CN116570259A utilizes the discrete spectrum of the Schrödinger operator to decompose a single-cycle pulse wave, extracting the reconstructed signal's discrete spectral features and combining them with a machine learning model to predict blood pressure. While this model demonstrates some robustness under low signal-to-noise ratio conditions, it is highly dependent on pulse wave morphology. Atypical pulse wave signals can affect feature extraction accuracy, leading to increased blood pressure prediction errors. Furthermore, the model requires high computational resources, making it unsuitable for low-power wearable devices. Chinese patent CN119453960A expands the PPG signal dimensionality and combines it with a lightweight neural network model to improve blood pressure prediction accuracy. However, this model relies on simultaneously acquired aortic pressure data for training, resulting in insufficient robustness to motion artifacts. In practice, frequent calibration is required due to individual differences and dynamic blood pressure fluctuations, limiting its universality and practicality. The shortcomings of these patented technologies collectively reflect that existing blood pressure measurement technology needs to be improved in terms of continuity, comfort, real-time and personalized adaptability.

[0004] In summary, existing blood pressure measurement technologies have limitations in terms of continuity, comfort, real-time and personalized adaptability. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to address the deficiencies of the above-mentioned prior art and provide a blood pressure measurement system and method based on oscillation wave dynamic characteristics and regression modeling to achieve accurate estimation of blood pressure.

[0006] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0007] In one aspect, the present invention provides a blood pressure measurement system based on oscillation wave dynamic characteristics and regression modeling, comprising an oscillation wave database, a pressure control module, a data quality assessment module, a data preprocessing module, a feature extraction module, a deep learning fine-tuning module, and a blood pressure estimation module;

[0008] The database stores oscillation waves and corresponding blood pressure results of at least 3,000 subjects, wherein each subject needs to have multiple blood pressure measurement data in a resting state;

[0009] The pressure control module controls the inflation and deflation of the cuff through a micro-electromagnetic valve, ensuring that the pressure of the cuff can be adjusted in real time as needed; the pressure control module includes an MCU, a cuff pressure sensor, a cuff pressure AC isolation filter circuit, and a cuff pressure DC isolation filter circuit; the cuff pressure data on the subject's arm is continuously collected in real time through the cuff pressure sensor, and this is used as the basis for blood pressure estimation; the cuff pressure AC isolation filter circuit converts the AC signal of the cuff pressure into a DC signal, which is provided to the MCU for subsequent negative feedback control and regulation; the cuff pressure DC isolation filter circuit is responsible for isolating the DC signal of the cuff pressure, obtaining the AC signal of the cuff pressure, and collecting the real-time value of the cuff pressure oscillation wave;

[0010] The data quality assessment module is used to screen out signals with poor quality assessment and remove them;

[0011] The data preprocessing module is used to filter the acquired oscillation wave to remove common noise;

[0012] The feature extraction module uses traditional signal processing methods to extract time domain, frequency domain, time-frequency and statistical features from the pre-processed oscillation wave signal, and divides the signal into training set and test set;

[0013] The deep learning fine-tuning module uses a lightweight deep learning fine-tuning model to fine-tune the extracted features to achieve personalized blood pressure prediction;

[0014] The blood pressure estimation module uses a machine learning model to match signals in a database based on real-time input fine-tuning features to predict the complete oscillation waveform, while identifying the two single-cycle oscillation wave positions with the largest changes in the characteristic bands of the systolic, diastolic, and mean arterial pressure oscillation waves, and mapping the changes in the characteristic bands of the oscillation waves to blood pressure values.

[0015] Furthermore, the deep learning fine-tuning module first normalizes and preprocesses the time domain, frequency domain, time-frequency and statistical features extracted by the feature extraction module, and then inputs them into the lightweight deep learning fine-tuning model; the lightweight deep learning fine-tuning model is a fully connected network, including an input layer, one or two hidden layers, a Dropout layer and a fully connected output layer; the hidden layer is provided with an appropriate number of neurons and uses a ReLU activation function to capture the nonlinear relationship between features; the Dropout layer is added between the hidden layers to prevent overfitting; the fully connected output layer uses a linear activation function to output the fine-tuned feature representation;

[0016] The network of the lightweight deep learning fine-tuning model is trained using the training set data. Through forward propagation, loss calculation, backpropagation and parameter update, the network automatically learns how to optimize the extracted features. The fine-tuned model is evaluated using the test set to detect the generalization performance of the model and perform hyperparameter tuning as needed.

[0017] Furthermore, the blood pressure estimation module first loads the fine-tuned feature vectors of the lightweight deep learning fine-tuning model and the corresponding blood pressure data, divides the data into training and test sets to ensure that there are sufficient samples for each blood pressure level; uses the training set to train the support vector machine, random forest, and extreme gradient boosting models respectively, and tunes their respective hyperparameters; uses cross-validation to evaluate the performance of each model on the training set and select the optimal model parameter combination;

[0018] Apply the trained model to the test set and evaluate the model using mean square error, root mean square error, or mean absolute error metrics; compare the prediction accuracy of each model on the test set, select the best performing model, or use model ensemble methods to further improve prediction stability;

[0019] The selected model is used for actual prediction, and the fine-tuned features are input to output a complete oscillation waveform prediction. At the same time, the identified oscillation wave characteristic wave mutation positions of systolic pressure, diastolic pressure and mean arterial pressure and the corresponding cuff pressure are output to achieve non-invasive measurement of blood pressure; the prediction results are corrected according to personalized calibration data as needed.

[0020] On the other hand, the present invention also provides a blood pressure measurement method based on oscillation wave dynamic characteristics and regression modeling, which is implemented by the above-mentioned blood pressure measurement system based on oscillation wave dynamic characteristics and regression modeling, and includes the following steps:

[0021] Step 1: Construct an oscillation wave signal database, including the complete oscillation wave signal and its corresponding oscillation wave characteristic bands of systolic pressure, diastolic pressure, and mean arterial pressure; control the inflation and deflation of the cuff through a micro-electromagnetic valve to ensure that the cuff pressure is adjusted in real time as needed;

[0022] Step 2: Perform quality assessment on the oscillation wave signals in the database, including waveform smoothness and stationarity assessment based on the time domain and spectrum consistency assessment based on the frequency domain;

[0023] Step 3: Preprocess the oscillation wave signal that has passed the quality assessment, including low-pass filtering, high-pass filtering, and empirical wavelet transform, to optimize the noise and baseline drift in the oscillation wave signal;

[0024] Step 4: Perform characteristic band identification and use traditional signal processing methods to extract time domain, frequency domain, time-frequency and statistical features from the preprocessed oscillation wave signal, and divide it into training set and test set;

[0025] Step 5: Fine-tune the extracted traditional features using a lightweight deep learning fine-tuning model;

[0026] Step 6: Train the machine learning model to automatically learn the mapping relationship between diastolic pressure, mean arterial pressure and oscillation wave signals, output a complete oscillation wave waveform, and simultaneously learn the characteristic band changes of the oscillation waves of systolic pressure, diastolic pressure and mean arterial pressure, and map the characteristic band changes to blood pressure values to achieve non-invasive blood pressure measurement.

[0027] Furthermore, the specific method of step 1 is:

[0028] Step 1.1: Place the oscillatory wave signal device correctly on the subject's upper arm and secure it with a cuff;

[0029] Step 1.2: The cuff is gradually pressurized, and the oscillation wave data of the subject at different pressures is recorded and stored;

[0030] Step 1.3: Measure the subject's systolic and diastolic blood pressure using the built-in blood pressure measurement device;

[0031] Step 1.4: The cuff pressure sensor is responsible for continuously collecting cuff pressure data on the user's arm in real time;

[0032] Step 1.5: Obtain the DC and AC signals of the cuff pressure;

[0033] Step 1.5.1: The cuff pressure AC isolation and filtering circuit converts the cuff pressure AC signal into a DC signal and provides it to the MCU for real-time negative feedback control and regulation. The negative feedback control and regulation signal generated by the MCU is transmitted to the micro-solenoid valve. The micro-solenoid valve controls the inflation and deflation of the cuff, ensuring that the cuff pressure is adjusted in real time as needed.

[0034] Step 1.5.2: The cuff pressure DC isolation and filtering circuit is responsible for isolating the cuff pressure DC signal and obtaining the cuff pressure AC signal, which is used to collect the real-time value of the cuff pressure oscillation wave;

[0035] Step 1.6: Output the processed oscillation wave signal and cuff pressure signal;

[0036] Step 1.7: Fit the waveform features of the diastolic pressure and mean arterial pressure in the oscillation waveform by identifying them. If the fit is good, inflation is stopped. If the fit is poor, pressure is continued until a normal blood pressure measurement is completed, and the obtained complete oscillation waveform is uploaded to the database.

[0037] Furthermore, the specific method of step 2 is:

[0038] Step 2.1: Based on the time domain waveform smoothness assessment, detect spikes or noise in the waveform and ensure that the waveform transitions smoothly on the time axis;

[0039] Step 2.2: Perform spectrum consistency assessment based on the frequency domain to ensure that the acquired partial waveform is consistent with the overall waveform and that the frequency distribution of the fitted waveform is similar to that of the actual waveform.

[0040] Furthermore, the specific method of step 3 is:

[0041] Step 3.1: Use a low-pass filter to remove high-frequency noise from the original oscillation wave signal, thereby retaining the low-frequency components of the oscillation wave signal;

[0042] Step 3.2: Use a high-pass filter to remove the effect of baseline drift on the oscillation wave, thereby retaining the high-frequency components in the oscillation wave signal;

[0043] Step 3.3: Construct a bandpass filter based on the empirical wavelet transform to process each periodic oscillation wave to eliminate artifacts caused by various factors, as shown in the following formula:

[0044] (1);

[0045] in, a and b are scale and translation parameters, ψ is the mother wavelet function, f(t) is the original signal, and is a function of t.

[0046] Furthermore, the specific method of step 4 is:

[0047] Step 4.1: Prepare the data by loading the oscillatory wave signal and its corresponding characteristic bands of systolic pressure, diastolic pressure, and mean arterial pressure, and dividing the data into training and test sets.

[0048] Step 4.2: Use traditional signal processing methods to extract features from the pre-processed oscillation wave signal. The specific methods include:

[0049] Step 4.2.1: Time domain feature extraction;

[0050] Calculate the maximum value, minimum value, peak value, number of peaks / troughs, and second peak amplitude of the signal;

[0051] Calculate the RMS value and average power of the signal; the RMS value of the signal is:

[0052] (5);

[0053] Where T is the time length of the signal, and x(t) is the time domain representation of the signal, that is, the oscillating wave signal that changes with time;

[0054] The average power of the signal P avg for:

[0055] (6);

[0056] Extract waveform slope, pulse width and waveform asymmetry indicators;

[0057] Waveform slope: Define a single-cycle signal as x(n), where n is a discrete time point, n0 is the starting point of the single-cycle signal, and n peak is the peak of a single cycle waveform, n D For the dicrotic notch, n SecondPeak At the second peak, n valley The trough of the single-cycle waveform, i.e., the end point of the single-cycle waveform, is calculated as the slope of the line connecting two adjacent points of the above characteristic points;

[0058] Pulse width: Pulse width refers to the duration of the signal above a certain threshold. The threshold is set to V th , from the moment the signal rising edge exceeds the threshold t start When the falling edge falls below the threshold end The time interval between them is the pulse width PW, which is as follows:

[0059] PW=t end -t start (7);

[0060] Waveform asymmetry index: The asymmetry of the waveform is measured by calculating the energy ratio of the signal on the positive semi-axis and the negative semi-axis; the energy E of the signal on the positive semi-axis is + for:

[0061] (8);

[0062] The energy of the signal on the negative half axis E - for:

[0063] (9);

[0064] The waveform asymmetry index AI is:

[0065] (10);

[0066] Step 4.2.2: Frequency domain feature extraction;

[0067] Use Fourier transform to convert time domain signals into frequency domain signals;

[0068] Extract the main frequency, power spectrum density and spectral entropy indicators of the signal;

[0069] The specific calculation method is as follows:

[0070] Fourier transform: Convert the time domain signal x(t) into the frequency domain signal X(f), where f is the frequency, as shown in the following formula:

[0071] (11);

[0072] The dominant frequency of a signal: In the frequency domain signal X(f), the frequency corresponding to the frequency component with the largest amplitude is the dominant frequency, as shown in the following formula:

[0073] (12);

[0074] The power spectral density is:

[0075] (13);

[0076] Among them, X T (f) is the signal x(t) in the time interval The Fourier transform of , PSD(f) represents the power spectral density;

[0077] The spectral entropy is:

[0078] (14);

[0079] in, , Sum the square of the power amplitude of all frequency points in the signal, N is the number of frequency points of the frequency domain signal, X(f i ) is the frequency domain signal at frequency f i The value at , H represents the spectral entropy.

[0080] Step 4.2.3: Time-frequency feature extraction;

[0081] Wavelet transform is used to extract the characteristics of time domain signals at different scales; empirical mode decomposition is used to decompose time domain signals and extract the time-frequency characteristics of each intrinsic mode component;

[0082] Empirical mode decomposition is to decompose the signal x(t) into a series of intrinsic mode components IMF m (t), assuming that the decomposition results in M intrinsic modal components, then

[0083] (15);

[0084] For each intrinsic mode component IMF m (t), extract its time-frequency features;

[0085] Step 4.2.4: Statistical feature extraction;

[0086] Calculate the mean, standard deviation, skewness, and kurtosis statistics of the signal as follows:

[0087] mean μ :

[0088] (16);

[0089] Among them, x i is the discrete sample value of the signal, N is the number of samples;

[0090] Standard deviation σ :

[0091] (17);

[0092] Skewness:

[0093] (18);

[0094] Kurtosis:

[0095] (19).

[0096] Furthermore, the specific method of step 5 is:

[0097] Step 5.1: Perform normalization preprocessing on the time domain, frequency domain, time-frequency and statistical feature vectors obtained by traditional signal processing methods in step 4;

[0098] Step 5.2: Input the normalized traditional feature vector into the lightweight deep learning fine-tuning model as follows:

[0099] Step 5.2.1: Build a small fully connected network and take the normalized feature vector as input;

[0100] Step 5.2.2: Hidden layer: Construct one or two fully connected hidden layers, with an appropriate number of neurons in each layer. The activation function uses ReLU to capture the nonlinear relationship between features.

[0101] Step 5.2.3: Set the Dropout layer between the hidden layers to prevent overfitting;

[0102] Step 5.2.4: Finally, construct a fully connected output layer using a linear activation function to fine-tune features and generate the final regression output;

[0103] Step 5.3: Fine-tune the model on the training set data, and perform regression optimization using the features extracted by traditional methods and the changes in the characteristic bands of systolic blood pressure, diastolic blood pressure, and mean arterial pressure;

[0104] Step 5.4: Use the test set to evaluate the fine-tuned model, check the generalization performance of the model and tune the hyperparameters as needed.

[0105] Furthermore, the specific method of step 6 is:

[0106] Step 6.1: Data preparation;

[0107] Load the changes in the eigenvectors and corresponding eigenbands after fine-tuning the lightweight deep learning fine-tuning model in step 5, and divide them into training and test sets to ensure that there are sufficient samples for each blood pressure level;

[0108] Step 6.2: Model selection and feature optimization;

[0109] Choose from a set of machine learning algorithms, including support vector machines, random forests, and extreme gradient boosting;

[0110] Step 6.3: Model training;

[0111] Step 6.3.1: Use the training set to train the support vector machine, random forest, and extreme gradient boosting algorithm models, and tune their respective hyperparameters;

[0112] Step 6.3.2: Use cross-validation to evaluate the model performance on the training set and select the best model parameter combination;

[0113] Step 6.4: Model evaluation and selection;

[0114] Step 6.4.1: Apply the trained model to the test set and evaluate the machine learning model using mean squared error, root mean squared error, or mean absolute error.

[0115] Step 6.4.2: Compare the prediction accuracy of each machine learning model on the test set and select the best performing machine learning model or use model ensemble methods to further improve prediction stability;

[0116] Step 6.5: Model integration and output;

[0117] Step 6.5.1: Use the selected machine learning model for actual prediction, input the fine-tuned traditional features, and output the complete oscillation waveform and the identified characteristic band mutation points of systolic blood pressure, diastolic blood pressure, and mean arterial pressure;

[0118] Step 6.5.2: Simultaneously output the cuff pressure corresponding to the mutation point of the oscillation wave characteristic band to achieve non-invasive measurement of blood pressure;

[0119] Step 6.5.3: As needed, modify the measurement results based on the personalized calibration data to further improve measurement accuracy.

[0120] The beneficial effects of adopting the above-mentioned technical solution are as follows: the blood pressure measurement system and method based on the dynamic characteristics of oscillation waves and regression modeling provided by the present invention, by combining traditional signal processing methods and machine learning techniques, can effectively extract oscillation wave features and capture the temporal dependence of signals, predict complete oscillation wave waveforms, and simultaneously establish a mapping relationship between them and blood pressure values by accurately identifying the morphological mutation points of the oscillation wave characteristic bands. This hybrid model not only retains the interpretability and computational efficiency of traditional methods, but also improves the feature expression capability through deep learning fine-tuning, thereby significantly improving the accuracy of blood pressure measurement. This technical solution is particularly suitable for home monitoring and wearable devices, reducing the discomfort and measurement time of subjects when repeatedly measuring blood pressure, and providing a reliable tool for the long-term management of hypertensive patients. BRIEF DESCRIPTION OF THE DRAWINGS

[0121] Figure 1 A structural block diagram of a blood pressure measurement system based on oscillation wave dynamic characteristics and regression modeling provided by an embodiment of the present invention;

[0122] Figure 2 This is a block diagram of an implementation of a pressure control module provided in an embodiment of the present invention;

[0123] Figure 3 This is a block diagram of an implementation of a data quality assessment module provided in an embodiment of the present invention;

[0124] Figure 4 This is a block diagram of an implementation of a data preprocessing module provided in an embodiment of the present invention;

[0125] Figure 5 A structural block diagram of a deep learning fine-tuning module provided in an embodiment of the present invention;

[0126] Figure 6 A structural block diagram of a lightweight fully connected fine-tuning network provided by an embodiment of the present invention;

[0127] Figure 7 This is a structural block diagram of the blood pressure estimation module provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0128] The following embodiments of the present invention are described in further detail with reference to the accompanying drawings and examples. The following examples are used to illustrate the present invention but are not intended to limit the scope of the present invention.

[0129] In this embodiment, a blood pressure measurement system based on oscillation wave dynamic characteristics and regression modeling is provided. Figure 1 As shown, it includes an oscillation wave database, a pressure control module, a data quality assessment module, a data preprocessing module, a feature extraction module, a deep learning fine-tuning module, and a blood pressure estimation module.

[0130] The database stores the oscillation waves and corresponding blood pressure results of at least 3,000 subjects, with individual subjects required to have multiple resting blood pressure measurements. To construct the oscillation wave signal database, the oscillation wave signal device is first correctly placed on the subject's upper arm and secured with a cuff to ensure accuracy and comfort. The cuff is gradually pressurized, and the oscillation wave data at different pressures is recorded and stored. Simultaneously, the subject's systolic and diastolic blood pressures are measured using a built-in blood pressure measurement device.

[0131] The pressure control module controls the inflation and deflation of the cuff through a micro electromagnetic valve, ensuring that the pressure of the cuff can be adjusted in real time as needed. Figure 2 As shown, a cuff pressure sensor continuously collects cuff pressure data from the subject's arm in real time, which serves as the basis for blood pressure estimation. The cuff pressure AC isolation filter circuit converts the cuff pressure's AC signal into a DC signal, which is provided to the MCU for subsequent negative feedback control and regulation. The cuff pressure DC isolation filter circuit isolates the DC signal, deriving the AC signal and collecting the real-time value of the cuff pressure oscillation wave. Based on the acquired real-time oscillation wave, the feature extraction module extracts features, and the blood pressure estimation module fits the complete oscillation wave. The fitted curve is evaluated. If the fit is good, inflation and pressurization are stopped, and the fitted curve serves as the basis for the final measurement and blood pressure estimation. If the fit is poor, pressurization continues until a standard blood pressure measurement is completed. The acquired complete oscillation waveform and blood pressure measurement value are uploaded to the database.

[0132] The data quality assessment module is used to filter out signals with poor quality assessment and remove them. Figure 3As shown, the first step is to assess waveform smoothness in the time domain. First-order differentiation is performed on a portion of the acquired waveform data to calculate the rate of change. If the rate of change fluctuates within a relatively small range, the waveform is relatively smooth. Excessive fluctuations indicate the presence of unsmooth peaks. Second-order differentiation is performed on the waveform data to further detect sharp fluctuations. The rate of change fluctuates significantly, indicating the presence of undesirable peaks. A sliding average method is used to remove noise, and the remaining peaks are tested after denoising. If the waveform fluctuations are reduced and smooth, the waveform is considered acceptable. Next, a frequency domain-based spectral consistency assessment is performed. A Fourier transform is performed on the acquired waveform, converting it from the time domain to the frequency domain to obtain its frequency distribution. The spectrum is analyzed to identify the waveform's dominant frequency and its harmonic distribution, while observing for significant higher-order harmonics. This is used to determine the waveform's periodicity and regularity. The acquired spectrum is compared with the spectrum of the complete waveform in the database to check for consistency in the dominant frequency, harmonic amplitude, and distribution, thereby determining the quality of the waveform fit. Finally, the power spectrum density is calculated to analyze the energy distribution at different frequencies of the waveform, and to intuitively display whether there are abnormal frequency components or noise in the waveform.

[0133] The data preprocessing module is used to filter the acquired oscillation wave and remove common noise. Figure 4 As shown, it mainly includes low-pass and high-pass filtering. First, a low-pass filter is used to reduce the noise of the original oscillation wave signal. The main purpose of low-pass filtering is to remove high-frequency noise in the signal, such as power supply interference and high-frequency random noise, thereby retaining the low-frequency components in the oscillation wave signal. Then high-pass filtering is performed. The main purpose of high-pass filtering is to remove low-frequency components in the signal, such as low-frequency noise such as baseline drift and respiratory artifacts, thereby retaining the high-frequency components in the oscillation wave signal. Next, a bandpass filter is constructed based on the empirical wavelet transform to process the oscillation wave of each cycle to eliminate motion artifacts caused by various external factors. The calculation formula is as shown in formula (1):

[0134] (1);

[0135] in, a and b are scale and translation parameters, ψ is the mother wavelet function, f(t) is the original signal, and is a function of t.

[0136] The feature extraction module uses traditional signal processing methods to extract time domain, frequency domain, time-frequency and statistical features from the preprocessed oscillation wave signal, and divides the data into a training set and a test set.

[0137] The deep learning fine-tuning module uses a lightweight deep learning fine-tuning model to fine-tune the above traditional features to achieve personalized blood pressure prediction. Figure 5 The specific method is as follows:

[0138] Step 1: Feature normalization: Normalize the obtained traditional feature vectors to ensure that each feature is in the same numerical range to facilitate subsequent network learning.

[0139] Step 2: Build a lightweight fully connected fine-tuning network (MLP), such as Figure 6 shown.

[0140] Step 2.1: Input layer: Take the normalized feature vector as input.

[0141] Step 2.2: Hidden Layer: Construct one or two fully connected hidden layers, set an appropriate number of neurons in each layer (for example, 64 or 128), and use the ReLU activation function to capture the nonlinear relationship between features.

[0142] Step 2.3: Add a Dropout layer (e.g. Dropout rate 0.5) between hidden layers to prevent overfitting.

[0143] Step 2.4: Output layer: Construct a fully connected output layer, use a linear activation function, and output the fine-tuned feature representation.

[0144] Step 3: Training and evaluation.

[0145] Step 3.1: Use the training set data to train the lightweight fine-tuning network. Through steps such as forward propagation, loss calculation, backpropagation, and parameter update, the network automatically learns how to optimize the traditionally extracted features.

[0146] Step 3.2: Use the test set to evaluate the fine-tuned model, check the generalization performance of the model and tune the hyperparameters as needed.

[0147] The blood pressure estimation module uses a machine learning model to match the signals in the database based on the fine-tuning features of the real-time input, predict the complete oscillation wave waveform, and simultaneously identify the two single-cycle oscillation wave positions (mutation points) with the largest changes in the oscillation wave characteristic bands of systolic pressure, diastolic pressure and mean arterial pressure, and map the changes in the oscillation wave characteristic bands to blood pressure values.

[0148] The machine learning model is trained to automatically learn the mapping relationship between systolic pressure, diastolic pressure and oscillation wave signals, fit and output the complete oscillation wave waveform, and at the same time learn the mapping relationship between the fine-tuned traditional features and the oscillation wave characteristic band changes of systolic pressure, diastolic pressure and mean arterial pressure, fit the two single-cycle oscillation wave positions (mutation points) with the maximum change points of the characteristic band, find the corresponding cuff pressure, and realize non-invasive blood pressure measurement. The specific implementation process is as follows: Figure 7 shown.

[0149] First, load the fine-tuned feature vectors of the lightweight deep learning fine-tuned model and the corresponding blood pressure data (diastolic pressure, mean arterial pressure, and, if necessary, systolic pressure). Divide the data into training and test sets to ensure sufficient samples for each blood pressure level. Select a set of traditional machine learning algorithms, including support vector machines, random forests, and extreme gradient boosting.

[0150] Use the training set to train the support vector machine, random forest, and extreme gradient boosting models, tuning their respective hyperparameters (e.g., kernel function parameters in SVM, number of trees in random forest, learning rate and tree depth in extreme gradient boosting). Use cross-validation to evaluate the performance of the machine learning models on the training set and select the optimal model parameter combination.

[0151] Model evaluation and selection: Apply the trained machine learning model to the test set and use mean square error (MSE), root mean square error (RMSE), or mean absolute error (MAE) as indicators to evaluate the machine learning model. Compare the prediction accuracy of each machine learning model on the test set, select the best performing machine learning model, or use model ensemble methods to further improve prediction stability. The specific calculation formulas are shown in Equations (2), (3), and (4):

[0152] (2);

[0153] (3);

[0154] (4);

[0155] Among them, n is the number of samples in the test set, y i is the true blood pressure value of the i-th sample in the test set, is the predicted blood pressure value of the i-th sample in the test set.

[0156] Model integration and output: The selected model is used for actual prediction, and the fine-tuned traditional features are input to output a complete oscillation waveform prediction. At the same time, the identified systolic, diastolic and mean arterial pressure oscillation wave characteristic wave mutation positions and the corresponding cuff pressure are output to achieve non-invasive measurement of blood pressure; the prediction results are corrected as needed in combination with personalized calibration data to further improve the prediction accuracy.

[0157] The blood pressure measurement method based on oscillation wave dynamic characteristics and regression modeling of this embodiment is implemented by the blood pressure measurement system based on oscillation wave dynamic characteristics and regression modeling described above, and includes the following steps:

[0158] Step 1: Construct an oscillation wave signal database, including the complete oscillation wave signal and its corresponding oscillation wave characteristic bands of systolic pressure, diastolic pressure and mean arterial pressure; control the inflation and deflation of the cuff through a micro-electromagnetic valve to ensure that the cuff pressure is adjusted in real time as needed. The specific method is:

[0159] Step 1.1: Place the oscillatory wave signal device correctly on the subject's upper arm and secure it with a cuff to ensure accuracy and comfort.

[0160] Step 1.2: The cuff is gradually pressurized, and the oscillation wave data of the subject at different pressures is recorded and stored;

[0161] Step 1.3: Measure the subject's systolic and diastolic blood pressure using the built-in blood pressure measurement device;

[0162] Step 1.4: The cuff pressure sensor is responsible for continuously collecting cuff pressure data on the user's arm in real time;

[0163] Step 1.5: Obtain the DC and AC signals of the cuff pressure;

[0164] Step 1.5.1: The cuff pressure AC isolation and filtering circuit converts the cuff pressure AC signal into a DC signal and provides it to the MCU for real-time negative feedback control and regulation. The negative feedback control and regulation signal generated by the MCU is transmitted to the micro-solenoid valve. The micro-solenoid valve controls the inflation and deflation of the cuff, ensuring that the cuff pressure is adjusted in real time as needed.

[0165] Step 1.5.2: The cuff pressure DC isolation and filtering circuit is responsible for isolating the cuff pressure DC signal and obtaining the cuff pressure AC signal, which is used to collect the real-time value of the cuff pressure oscillation wave;

[0166] Step 1.6: Output the processed oscillation wave signal and cuff pressure signal;

[0167] Step 1.7: Fit the waveform features of the diastolic pressure and mean arterial pressure in the oscillation waveform by identifying them. If the fit is good, inflation is stopped. If the fit is poor, pressure is continued until a normal blood pressure measurement is completed, and the obtained complete oscillation waveform is uploaded to the database.

[0168] Step 2: Perform quality assessment on the oscillation wave signals in the database, including waveform smoothness and stationarity assessment based on the time domain and spectrum consistency assessment based on the frequency domain.

[0169] Step 2.1: Based on the time domain waveform smoothness assessment, detect spikes or noise in the waveform and ensure that the waveform transitions smoothly on the time axis;

[0170] Step 2.2: Perform spectrum consistency assessment based on the frequency domain to ensure that the acquired partial waveform is consistent with the overall waveform and that the frequency distribution of the fitted waveform is similar to that of the actual waveform.

[0171] Step 3: Preprocess the oscillation wave signal that has passed the quality assessment, including low-pass filtering, high-pass filtering, and empirical wavelet transform, to optimize the noise and baseline drift in the oscillation wave signal. The specific method is:

[0172] Step 3.1: Use a low-pass filter to remove high-frequency noise from the original oscillation wave signal, thereby retaining the low-frequency components of the oscillation wave signal;

[0173] Step 3.2: Use a high-pass filter to remove the effect of baseline drift on the oscillation wave, thereby retaining the high-frequency components in the oscillation wave signal;

[0174] Step 3.3: Construct a bandpass filter based on the empirical wavelet transform to process each periodic oscillation wave to eliminate artifacts caused by various factors, as shown in the following formula:

[0175] (1);

[0176] in, a and b are scale and translation parameters, ψ is the mother wavelet function, f(t) is the original signal, and is a function of t.

[0177] Step 4: Perform characteristic band identification and use traditional signal processing methods to extract time domain, frequency domain, time-frequency and statistical features from the pre-processed oscillation wave signal, and divide it into training set and test set. The specific method is:

[0178] Step 4.1: Prepare the data by loading the oscillation wave signal and its corresponding characteristic bands of systolic pressure, diastolic pressure, and mean arterial pressure, and dividing the data into training and test sets.

[0179] Step 4.2: Use traditional signal processing methods to extract features from the pre-processed oscillation wave signal. The specific methods include:

[0180] Step 4.2.1: Time domain feature extraction;

[0181] Calculate the maximum value, minimum value, peak value, number of peaks / troughs, and second peak amplitude of the signal; calculate the RMS value and average power of the signal; and extract waveform slope, pulse width, and waveform asymmetry indicators.

[0182] The specific calculation formula is as follows:

[0183] The root mean square value RMS of the signal:

[0184] (5);

[0185] Where T is the time length of the signal, and x(t) is the time domain representation of the signal, that is, the oscillating wave signal that changes with time.

[0186] The average power of the signal P avg :

[0187] (6);

[0188] Waveform slope: Define a single-cycle signal as x(n), where n is a discrete time point, n0 is the starting point of the single-cycle signal, and n peak is the peak of a single cycle waveform, n D For the dicrotic notch, n SecondPeak At the second peak, n valley The trough of the single-cycle waveform, that is, the end point of the single-cycle waveform, is calculated as the slope of the line between two adjacent points of the above characteristic points.

[0189] Pulse width: Pulse width refers to the duration of the signal above a certain threshold. The threshold is set to V th , from the moment the signal rising edge exceeds the threshold t start When the falling edge falls below the threshold end The time interval between them is the pulse width PW, which is as follows:

[0190] PW=t end -t start (7);

[0191] The waveform asymmetry index can measure the asymmetry of the waveform by calculating the energy ratio of the signal on the positive half axis and the negative half axis. + for:

[0192] (8);

[0193] The energy of the signal on the negative half axis E - for:

[0194] (9);

[0195] The waveform asymmetry index AI is:

[0196] (10);

[0197] Step 4.2.2: Frequency domain feature extraction;

[0198] Use Fourier transform to convert the time domain signal into frequency domain signal; extract the main frequency, power spectrum density and spectral entropy indicators of the signal.

[0199] The specific calculation method is as follows:

[0200] Fourier transform: Convert the time domain signal x(t) into the frequency domain signal X(f), where f is the frequency, as shown in the following formula:

[0201] (11);

[0202] The dominant frequency of a signal: In the frequency domain signal X(f), the frequency corresponding to the frequency component with the largest amplitude is the dominant frequency, as shown in the following formula:

[0203] (12);

[0204] Power spectral density:

[0205] (13);

[0206] Among them, X T (f) is the signal x(t) in the time interval The Fourier transform of , PSD(f) represents the power spectral density;

[0207] The spectral entropy is:

[0208] (14);

[0209] in, , Sum the square of the power amplitude of all frequency points in the signal, N is the number of frequency points of the frequency domain signal, X(f i ) is the frequency domain signal at frequency f i The value at , H represents the spectral entropy.

[0210] Step 4.2.3: Time-frequency feature extraction;

[0211] Wavelet transform is used to extract the characteristics of time domain signals at different scales; empirical mode decomposition method is used to decompose the time domain signals and extract the time-frequency characteristics of each intrinsic mode component.

[0212] Empirical mode decomposition is to decompose the signal x(t) into a series of intrinsic mode components IMF m (t), assuming that the decomposition results in M intrinsic modal components, then

[0213] (15);

[0214] For each intrinsic mode component IMF m (t), its time-frequency features can be extracted.

[0215] Step 4.2.4: Statistical feature extraction;

[0216] Compute the mean, standard deviation, skewness, and kurtosis statistics of a signal.

[0217] mean μ :

[0218] (16);

[0219] Among them, x i is the discrete sample value of the signal, N is the number of samples;

[0220] Standard deviation σ :

[0221] (17);

[0222] Skewness:

[0223] (18);

[0224] Kurtosis:

[0225] (19).

[0226] Step 5: Use the lightweight deep learning fine-tuning model to fine-tune the extracted traditional features. The specific method is:

[0227] Step 5.1: Perform normalization preprocessing on the time domain, frequency domain, time-frequency and statistical feature vectors obtained by traditional signal processing methods in step 4;

[0228] Step 5.2: Input the normalized traditional feature vector into the lightweight deep learning fine-tuning model as follows:

[0229] Step 5.2.1: Build a small fully connected network and take the normalized feature vector as input;

[0230] Step 5.2.2: Hidden layer: Construct one or two fully connected hidden layers, with an appropriate number of neurons in each layer. The activation function uses ReLU to capture the nonlinear relationship between features.

[0231] Step 5.2.3: Set the Dropout layer between the hidden layers to prevent overfitting;

[0232] Step 5.2.4: Finally, construct a fully connected output layer using a linear activation function to fine-tune features and generate the final regression output;

[0233] Step 5.3: Fine-tune the model on the training set data, and perform regression optimization using the features extracted by traditional methods and the changes in the characteristic bands of systolic blood pressure, diastolic blood pressure, and mean arterial pressure;

[0234] Step 5.4: Use the test set to evaluate the fine-tuned model, check the generalization performance of the model and tune the hyperparameters as needed.

[0235] Step 6: Train the machine learning model to automatically learn the mapping relationship between diastolic pressure, mean arterial pressure and oscillation wave signals, output the complete oscillation wave waveform, and simultaneously learn the characteristic band changes of oscillation waves of systolic pressure, diastolic pressure and mean arterial pressure, and map the characteristic band changes to blood pressure values to achieve non-invasive blood pressure measurement. The specific method is:

[0236] Step 6.1: Data preparation;

[0237] Load the changes in the eigenvectors and corresponding eigenbands after fine-tuning the lightweight deep learning fine-tuning model in step 5, and divide them into training and test sets to ensure that there are sufficient samples for each blood pressure level;

[0238] Step 6.2: Model selection and feature optimization;

[0239] Choose from a set of machine learning algorithms, including support vector machines, random forests, and extreme gradient boosting;

[0240] Step 6.3: Model training;

[0241] Step 6.3.1: Train the support vector machine, random forest, and extreme gradient boosting models using the training set, tuning their respective hyperparameters (e.g., kernel parameters for the support vector machine, number of trees for the random forest, learning rate and tree depth for the extreme gradient boosting).

[0242] Step 6.3.2: Use cross-validation to evaluate the model performance on the training set and select the best model parameter combination;

[0243] Step 6.4: Model evaluation and selection;

[0244] Step 6.4.1: Apply the trained model to the test set and evaluate the machine learning model using mean squared error (MSE), root mean squared error (RMSE), or mean absolute error (MAE).

[0245] Step 6.4.2: Compare the prediction accuracy of each machine learning model on the test set and select the best performing machine learning model or use model ensemble methods to further improve prediction stability;

[0246] Step 6.5: Model integration and output;

[0247] Step 6.5.1: Use the selected machine learning model for actual prediction, input the fine-tuned traditional features, and output the complete oscillation waveform and the identified characteristic band mutation points of systolic blood pressure, diastolic blood pressure, and mean arterial pressure;

[0248] Step 6.5.2: Simultaneously output the cuff pressure corresponding to the mutation point of the oscillation wave characteristic band to achieve non-invasive measurement of blood pressure;

[0249] Step 6.5.3: As needed, modify the measurement results based on the personalized calibration data to further improve measurement accuracy.

[0250] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or make equivalent replacements for some or all of the technical features therein. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope defined by the claims of the present invention.

Claims

1. A blood pressure measurement system based on oscillation wave dynamic characteristics and regression modeling, characterized by: It includes an oscillation wave database, a pressure control module, a data quality assessment module, a data preprocessing module, a feature extraction module, a deep learning fine-tuning module, and a blood pressure estimation module; The database stores oscillation waves and corresponding blood pressure results of at least 3,000 subjects, wherein each subject needs to have multiple blood pressure measurement data in a resting state; The pressure control module controls the inflation and deflation of the cuff through a micro-electromagnetic valve, ensuring that the pressure of the cuff can be adjusted in real time as needed; the pressure control module includes an MCU, a cuff pressure sensor, a cuff pressure AC isolation filter circuit, and a cuff pressure DC isolation filter circuit; the cuff pressure data on the subject's arm is continuously collected in real time through the cuff pressure sensor, and this is used as the basis for blood pressure estimation; the cuff pressure AC isolation filter circuit converts the AC signal of the cuff pressure into a DC signal, which is provided to the MCU for subsequent negative feedback control and regulation; the cuff pressure DC isolation filter circuit is responsible for isolating the DC signal of the cuff pressure, obtaining the AC signal of the cuff pressure, and collecting the real-time value of the cuff pressure oscillation wave; The data quality assessment module is used to screen out signals with poor quality assessment and remove them; The data preprocessing module is used to filter the acquired oscillation wave to remove common noise; The feature extraction module uses traditional signal processing methods to extract time domain, frequency domain, time-frequency and statistical features from the pre-processed oscillation wave signal, and divides the signal into training set and test set; The deep learning fine-tuning module uses a lightweight deep learning fine-tuning model to fine-tune the extracted features to achieve personalized blood pressure prediction; The blood pressure estimation module uses a machine learning model to match signals in a database based on real-time input fine-tuning features to predict the complete oscillation waveform, while identifying the two single-cycle oscillation wave positions with the largest changes in the characteristic bands of the systolic, diastolic, and mean arterial pressure oscillation waves, and mapping the changes in the characteristic bands of the oscillation waves to blood pressure values.

2. The blood pressure measurement system based on oscillation wave dynamic characteristics and regression modeling according to claim 1, characterized in that: The deep learning fine-tuning module first normalizes and preprocesses the time domain, frequency domain, time-frequency, and statistical features extracted by the feature extraction module, and then inputs them into the lightweight deep learning fine-tuning model; the lightweight deep learning fine-tuning model is a fully connected network, including an input layer, one or two hidden layers, a dropout layer, and a fully connected output layer; the hidden layer is provided with an appropriate number of neurons and uses a ReLU activation function to capture the nonlinear relationship between features; the dropout layer is added between the hidden layers to prevent overfitting; the fully connected output layer uses a linear activation function to output the fine-tuned feature representation; The network of the lightweight deep learning fine-tuning model is trained using the training set data. Through forward propagation, loss calculation, backpropagation and parameter update, the network automatically learns how to optimize the extracted features. The fine-tuned model is evaluated using the test set to detect the generalization performance of the model and perform hyperparameter tuning as needed.

3. The blood pressure measurement system based on oscillation wave dynamic characteristics and regression modeling according to claim 2, characterized in that: The blood pressure estimation module first loads the fine-tuned feature vectors of the lightweight deep learning fine-tuning model and the corresponding blood pressure data, divides the data into training and test sets to ensure that there are sufficient samples for each blood pressure level; uses the training set to train the support vector machine, random forest, and extreme gradient boosting models, respectively, and tune their respective hyperparameters; uses cross-validation to evaluate the performance of each model on the training set and select the optimal model parameter combination; Apply the trained model to the test set and evaluate the model using mean square error, root mean square error, or mean absolute error metrics; compare the prediction accuracy of each model on the test set, select the best performing model, or use model ensemble methods to further improve prediction stability; The selected model is used for actual prediction, and the fine-tuned features are input to output a complete oscillation waveform prediction. At the same time, the identified oscillation wave characteristic wave mutation positions of systolic pressure, diastolic pressure and mean arterial pressure and the corresponding cuff pressure are output to achieve non-invasive measurement of blood pressure; the prediction results are corrected according to personalized calibration data as needed.

4. A blood pressure measurement method based on oscillation wave dynamic characteristics and regression modeling, characterized by: The blood pressure measurement system based on oscillation wave dynamic characteristics and regression modeling according to claim 1 is implemented, comprising the following steps: Step 1: Construct an oscillation wave signal database, including the complete oscillation wave signal and its corresponding oscillation wave characteristic bands of systolic pressure, diastolic pressure, and mean arterial pressure; control the inflation and deflation of the cuff through a micro-electromagnetic valve to ensure that the cuff pressure is adjusted in real time as needed; Step 2: Perform quality assessment on the oscillation wave signals in the database, including waveform smoothness and stationarity assessment based on the time domain and spectrum consistency assessment based on the frequency domain; Step 3: Preprocess the oscillation wave signal that has passed the quality assessment, including low-pass filtering, high-pass filtering, and empirical wavelet transform, to optimize the noise and baseline drift in the oscillation wave signal; Step 4: Perform characteristic band identification and use traditional signal processing methods to extract time domain, frequency domain, time-frequency and statistical features from the preprocessed oscillation wave signal, and divide it into training set and test set; Step 5: Fine-tune the extracted traditional features using a lightweight deep learning fine-tuning model; Step 6: Train the machine learning model to automatically learn the mapping relationship between diastolic pressure, mean arterial pressure and oscillation wave signals, output a complete oscillation wave waveform, and simultaneously learn the characteristic band changes of the oscillation waves of systolic pressure, diastolic pressure and mean arterial pressure, and map the characteristic band changes to blood pressure values to achieve non-invasive blood pressure measurement.

5. The blood pressure measurement method based on oscillation wave dynamic characteristics and regression modeling according to claim 4, characterized in that: The specific method of step 1 is: Step 1.1: Place the oscillatory wave signal device correctly on the subject's upper arm and secure it with a cuff; Step 1.2: The cuff is gradually pressurized, and the oscillation wave data of the subject at different pressures is recorded and stored; Step 1.3: Measure the subject's systolic and diastolic blood pressure using the built-in blood pressure measurement device; Step 1.4: The cuff pressure sensor is responsible for continuously collecting cuff pressure data on the user's arm in real time; Step 1.5: Obtain the DC and AC signals of the cuff pressure; Step 1.5.1: The cuff pressure AC isolation and filtering circuit converts the cuff pressure AC signal into a DC signal and provides it to the MCU for real-time negative feedback control and regulation. The negative feedback control and regulation signal generated by the MCU is transmitted to the micro-solenoid valve. The micro-solenoid valve controls the inflation and deflation of the cuff, ensuring that the cuff pressure is adjusted in real time as needed. Step 1.5.2: The cuff pressure DC isolation and filtering circuit is responsible for isolating the cuff pressure DC signal and obtaining the cuff pressure AC signal, which is used to collect the real-time value of the cuff pressure oscillation wave; Step 1.6: Output the processed oscillation wave signal and cuff pressure signal; Step 1.7: Fit the waveform features of the diastolic pressure and mean arterial pressure in the oscillation waveform by identifying them. If the fit is good, inflation is stopped. If the fit is poor, pressure is continued until a normal blood pressure measurement is completed, and the obtained complete oscillation waveform is uploaded to the database.

6. The blood pressure measurement method based on oscillation wave dynamic characteristics and regression modeling according to claim 4, characterized in that: The specific method of step 2 is: Step 2.1: Based on the time domain waveform smoothness assessment, detect spikes or noise in the waveform and ensure that the waveform transitions smoothly on the time axis; Step 2.2: Perform spectrum consistency assessment based on the frequency domain to ensure that the acquired partial waveform is consistent with the overall waveform and that the frequency distribution of the fitted waveform is similar to that of the actual waveform.

7. The blood pressure measurement method based on oscillation wave dynamic characteristics and regression modeling according to claim 4, characterized in that: The specific method of step 3 is: Step 3.1: Use a low-pass filter to remove high-frequency noise from the original oscillation wave signal, thereby retaining the low-frequency components of the oscillation wave signal; Step 3.2: Use a high-pass filter to remove the effect of baseline drift on the oscillation wave, thereby retaining the high-frequency components in the oscillation wave signal; Step 3.3: Construct a bandpass filter based on the empirical wavelet transform to process each periodic oscillation wave to eliminate artifacts caused by various factors, as shown in the following formula: （1）； in, a and b are scale and translation parameters, ψ is the mother wavelet function, f(t) is the original signal, and is a function of t.

8. The blood pressure measurement method based on oscillation wave dynamic characteristics and regression modeling according to claim 4, characterized in that: The specific method of step 4 is: Step 4.1: Prepare the data by loading the oscillatory wave signal and its corresponding characteristic bands of systolic pressure, diastolic pressure, and mean arterial pressure, and dividing the data into training and test sets. Step 4.2: Use traditional signal processing methods to extract features from the pre-processed oscillation wave signal. The specific methods include: Step 4.2.1: Time domain feature extraction; Calculate the maximum value, minimum value, peak value, number of peaks / troughs, and second peak amplitude of the signal; Calculate the RMS value and average power of the signal; the RMS value of the signal is: （5）； Where T is the time length of the signal, and x(t) is the time domain representation of the signal, that is, the oscillating wave signal that changes with time; The average power of the signal P avg for: （6）； Extract waveform slope, pulse width and waveform asymmetry indicators; Waveform slope: Define a single-cycle signal as x(n), where n is a discrete time point, n0 is the starting point of the single-cycle signal, and n peak is the peak of a single cycle waveform, n D For the dicrotic notch, n SecondPeak At the second peak, n valley The trough of the single-cycle waveform, i.e., the end point of the single-cycle waveform, is calculated as the slope of the line connecting two adjacent points of the above characteristic points; Pulse width: Pulse width refers to the duration of the signal above a certain threshold. The threshold is set to V th , from the moment the signal rising edge exceeds the threshold t start When the falling edge falls below the threshold end The time interval between them is the pulse width PW, which is as follows: PW=t end -t start (7); Waveform asymmetry index: The asymmetry of the waveform is measured by calculating the energy ratio of the signal on the positive semi-axis and the negative semi-axis; the energy E of the signal on the positive semi-axis is + for: （8）； The energy of the signal on the negative half axis E - for: （9）； The waveform asymmetry index AI is: （10）； Step 4.2.2: Frequency domain feature extraction; Use Fourier transform to convert time domain signals into frequency domain signals; Extract the main frequency, power spectrum density and spectral entropy indicators of the signal; The specific calculation method is as follows: Fourier transform: Convert the time domain signal x(t) into the frequency domain signal X(f), where f is the frequency, as shown in the following formula: （11）； The dominant frequency of a signal: In the frequency domain signal X(f), the frequency corresponding to the frequency component with the largest amplitude is the dominant frequency, as shown in the following formula: （12）； The power spectral density is: （13）； Among them, X T (f) is the signal x(t) in the time interval The Fourier transform of , PSD(f) represents the power spectral density; The spectral entropy is: （14）； in, , Sum the square of the power amplitude of all frequency points in the signal, N is the number of frequency points of the frequency domain signal, X(f i ) is the frequency domain signal at frequency f i The value at , H represents the spectral entropy; Step 4.2.3: Time-frequency feature extraction; Wavelet transform is used to extract the characteristics of time domain signals at different scales; empirical mode decomposition is used to decompose time domain signals and extract the time-frequency characteristics of each intrinsic mode component; Empirical mode decomposition is to decompose the signal x(t) into a series of intrinsic mode components IMF m (t), assuming that the decomposition results in M intrinsic modal components, then （15）； For each intrinsic mode component IMF m (t), extract its time-frequency features; Step 4.2.4: Statistical feature extraction; Calculate the mean, standard deviation, skewness, and kurtosis statistics of the signal as follows: mean μ : （16）； Among them, x i is the discrete sample value of the signal, N is the number of samples; Standard deviation σ : （17）； Skewness: （18）； Kurtosis: （19）。 9. The blood pressure measurement method based on oscillation wave dynamic characteristics and regression modeling according to claim 8, characterized in that: The specific method of step 5 is: Step 5.1: Perform normalization preprocessing on the time domain, frequency domain, time-frequency and statistical feature vectors obtained by traditional signal processing methods in step 4; Step 5.2: Input the normalized traditional feature vector into the lightweight deep learning fine-tuning model as follows: Step 5.2.1: Build a small fully connected network and take the normalized feature vector as input; Step 5.2.2: Hidden layer: Construct one or two fully connected hidden layers, with an appropriate number of neurons in each layer. The activation function uses ReLU to capture the nonlinear relationship between features. Step 5.2.3: Set the Dropout layer between the hidden layers to prevent overfitting; Step 5.2.4: Finally, construct a fully connected output layer using a linear activation function to fine-tune features and generate the final regression output; Step 5.3: Fine-tune the model on the training set data, and perform regression optimization using the features extracted by traditional methods and the changes in the characteristic bands of systolic blood pressure, diastolic blood pressure, and mean arterial pressure; Step 5.4: Use the test set to evaluate the fine-tuned model, check the generalization performance of the model and tune the hyperparameters as needed.

10. The blood pressure measurement method based on oscillation wave dynamic characteristics and regression modeling according to claim 9, characterized in that: The specific method of step 6 is: Step 6.1: Data preparation; Load the changes in the eigenvectors and corresponding eigenbands after fine-tuning the lightweight deep learning fine-tuning model in step 5, and divide them into training and test sets to ensure that there are sufficient samples for each blood pressure level; Step 6.2: Model selection and feature optimization; Choose from a set of machine learning algorithms, including support vector machines, random forests, and extreme gradient boosting; Step 6.3: Model training; Step 6.3.1: Use the training set to train the support vector machine, random forest, and extreme gradient boosting algorithm models, and tune their respective hyperparameters; Step 6.3.2: Use cross-validation to evaluate the model performance on the training set and select the best model parameter combination; Step 6.4: Model evaluation and selection; Step 6.4.1: Apply the trained model to the test set and evaluate the machine learning model using mean squared error, root mean squared error, or mean absolute error. Step 6.4.2: Compare the prediction accuracy of each machine learning model on the test set and select the best performing machine learning model or use model ensemble methods to further improve prediction stability; Step 6.5: Model integration and output; Step 6.5.1: Use the selected machine learning model for actual prediction, input the fine-tuned traditional features, and output the complete oscillation waveform and the identified characteristic band mutation points of systolic blood pressure, diastolic blood pressure, and mean arterial pressure; Step 6.5.2: Simultaneously output the cuff pressure corresponding to the mutation point of the oscillation wave characteristic band to achieve non-invasive measurement of blood pressure; Step 6.5.3: Correct the measurement results as needed by combining the personalized calibration data.

Images