Blood pressure measurement system and method based on oscillatory wave feature extraction and supervision fusion

Abstract

The invention provides a blood pressure measurement system and method based on oscillatory wave feature extraction and supervision fusion, and relates to the technical field of wearable medical health monitoring. Firstly, an oscillatory wave database is constructed, the oscillatory wave database comprises oscillatory wave signals and blood pressure measurement values corresponding to the oscillatory wave signals, then original oscillatory wave signals of a to-be-recognized sample are obtained, and available oscillatory wave signals and unavailable oscillatory wave signals are judged according to sensor attachment and wire connection detection indexes; performing quality evaluation on the available oscillatory wave signals; preprocessing the oscillatory wave signal with good quality, respectively extracting time domain, frequency domain and time-frequency characteristics of the preprocessed oscillatory wave signal, and obtaining multi-index characteristics of the oscillatory wave signal; and finally, carrying out standardization processing and supervised feature fusion on the time domain, the frequency domain and the time-frequency features of the acquired oscillatory wave signal, and carrying out feature weighted fusion, thereby realizing non-invasive blood pressure measurement.

Description

Technical Field

[0001] The present invention relates to the technical field of wearable medical health monitoring, and particularly to a blood pressure measurement system and method based on oscillatory wave feature extraction and supervised fusion. Background Art

[0002] With the improvement of health awareness, in the field of wearable medical health monitoring, the demand for non-invasive blood pressure monitoring has increased. Currently, traditional non-invasive blood pressure measurement methods have many limitations and cannot meet the needs of precision medicine. On the one hand, physiological differences between individuals, such as different vascular elasticity and heart rate variability, make the measurement methods based on general models produce large errors and are difficult to accurately reflect the individual's blood pressure status; on the other hand, the existing technology does not process and analyze oscillatory wave signals finely enough, and does not fully consider these individual differences, making it difficult to comprehensively extract key features. These two factors jointly affect the accuracy and reliability of non-invasive blood pressure measurement technology and limit its application and development in personalized medicine.

[0003] Chinese Patent CN101548883A determines blood pressure by the amplitude change of oscillatory waves during inflation and deflation. Although the accuracy is improved by correcting blood pressure through a sound pulse or light pulse monitoring device, its essence is still an extended application of the amplitude coefficient method, with insufficient exploration of oscillatory wave signal characteristics and large measurement errors under complex physiological conditions. Chinese Patent CN119138869A obtains a preliminary blood pressure result based on the oscillometric method using a proportional coefficient, and then constructs a model using the extracted oscillatory wave characteristic parameters to calculate the final blood pressure. However, its characteristic dimension is single, and when facing complex physiological signals and environmental interference, the adaptability of the model and the reliability of the measurement are insufficient.

[0004] In summary, the existing technologies generally have problems such as single characteristic dimension, insufficient individual adaptability, and insufficient signal quality assessment, and it is difficult to achieve high-precision blood pressure measurement under complex physiological conditions and dynamic environments. The present invention breaks through the dependence on a single characteristic through multi-modal feature fusion, supervised learning, and personalized matching technologies, effectively solves the measurement deviation caused by individual waveform specificity, and improves the accuracy and robustness of blood pressure monitoring. Summary of the Invention

[0005] Aiming at the deficiencies of the existing technology, the present invention provides a blood pressure measurement system and method based on oscillatory wave feature extraction and supervised fusion, which realizes non-invasive blood pressure measurement based on oscillatory wave signals.

[0006] On the one hand, a blood pressure measurement system based on oscillatory wave feature extraction and supervised fusion includes: an oscillatory wave signal acquisition and available signal discrimination module, a preprocessing module, a multi-index feature extraction module, a multi-index feature normalization processing and supervised fusion module, and a non-invasive blood pressure measurement module.

[0007] The oscillation wave signal acquisition and available signal discrimination module is used to continuously acquire the original oscillation wave signal in real time, judge the oscillation wave signal according to the cuff fit and arm movement, discriminate available oscillation wave signals and unavailable oscillation wave signals, and evaluate the quality of available oscillation wave signals by fusing time domain and frequency domain features. Available oscillation waves with fusion features greater than a set threshold are called oscillation wave signals with good quality, and the remaining available oscillation waves are called oscillation wave signals with poor quality. The preprocessing module preprocesses the oscillation wave signal with good quality, removes the noise in the oscillation wave signal, and improves the quality of the oscillation wave signal; The multi-index feature extraction module extracts the time domain features of the oscillating wave signal based on the Hamilton algorithm, extracts the frequency domain features of the oscillating wave signal based on the continuous wavelet transform, and extracts the time-frequency features of the oscillating wave signal based on the Hilbert-Huang transform (HHT); The multi-index feature standardization processing and supervised fusion module performs standardization processing on the oscillation wave signal feature data, applies independent component analysis ICA to reduce the dimension of the standardized oscillation wave signal feature data, and uses supervised variational autoencoder to perform feature learning and fusion on the features selected by Fisher criterion; The non-invasive blood pressure measurement module is based on the fusion features extracted by the supervised variational autoencoder, performs feature importance evaluation based on reconstruction error contribution, performs weighted fusion of multi-index features based on XGBoost weight allocation, applies a pre-trained model for non-invasive blood pressure prediction to perform non-invasive blood pressure measurement, and finally completes blood pressure measurement.

[0008] On the other hand, a blood pressure measurement method based on oscillation wave feature extraction and supervised fusion is implemented based on the aforementioned blood pressure measurement system, comprising the following steps: Step 1: construct an oscillation wave database, wherein the oscillation wave database includes oscillation wave signals and their corresponding blood pressure measurement values; Step 1.1: Collect the subject's oscillation wave signal and its corresponding blood pressure measurement value as a supervision indicator; Step 1.2: Preprocess the oscillation wave signal in the oscillation wave database; Step 1.3: Extract the time domain, frequency domain and time-frequency features of the oscillation wave signal in the oscillation wave database, and select the supervision features related to blood pressure measurement; Step 1.4: Build and train a supervised variational autoencoder. Step 1.5: Calculate feature weights based on XGBoost and the oscillatory wave signal to be analyzed.

[0009] Step 2: Obtain the original oscillation wave signal of the sample to be identified, and distinguish the usable oscillation wave signal from the unusable oscillation wave signal based on the sensor fitting and wire connection detection indicators, and perform quality assessment on the usable oscillation wave signal; Step 2.1: Detect errors in the collected oscillatory wave signals, and distinguish available oscillatory wave signals from unavailable oscillatory wave signals; The unavailable oscillatory wave signals are those with distorted or completely disappeared waveforms, losing their effective information; Step 2.2: For the available oscillatory wave signals, comprehensively consider the indexes in both the time domain and the frequency domain to evaluate the quality of the signals as good or bad; Step 2.2.1: Conduct time-domain analysis on the oscillatory wave signals to determine the signal quality evaluation indexes in the time domain; In the time-domain analysis, the indexes of amplitude change degree, baseline drift degree, and pulse rate consistency degree are used to evaluate the quality of the oscillatory wave signals; Specifically: Perform variational mode decomposition and period segmentation on the oscillatory wave signals respectively to obtain the reconstructed signals and single-period signals representing the baseline drift of the signals; The amplitude change degree index is obtained through the amplitude difference of the signal waveforms in multiple cardiac cycles; the amplitude change degree index is obtained through the amplitude difference of the signal waveforms in multiple cardiac cycles; the amplitude change degree index of the oscillatory wave signal in the i-th cycle a ( i ) is shown in the following formula: ; where and are respectively the maximum and minimum values of the waveform amplitudes of the i-th cycle of the oscillatory wave signal after filtering; i ; The baseline drift degree index is obtained by windowing the reconstructed signal; the baseline drift degree index is obtained by windowing the reconstructed signal; windowing the reconstructed signal to obtain the baseline drift degree index of the oscillatory wave signal in the i-th cycle b ( i ) is shown in the following formula: ; where bWave i is the reconstructed signal of the i-th cycle of the oscillatory wave signal, b is the baseline standard value of the oscillatory wave signal in the stationary state; The pulse rate consistency degree index is calculated by calculating the consistency of the signal time lengths in multiple cycles; the pulse rate consistency degree index is calculated by calculating the consistency of the signal time lengths in multiple cycles; setting the cycle length of the oscillatory wave signal as l , obtaining the pulse rate consistency degree index of the oscillatory wave signal in the i-th cycle is shown in the following formula: ; where For the standard deviation of the length of the i th signal cycle calculated, and for the mean value of the length of the i th signal cycle calculated; Step 2.2.2: Conduct frequency-domain analysis on the oscillatory wave signal to determine the signal quality evaluation index in the frequency domain; In the said frequency-domain analysis, the frequency component index is used to evaluate the quality of the oscillatory wave signal; Specifically: The effective frequency band of the oscillatory wave signal is selected through fast Fourier transform to calculate the power spectral density to obtain the frequency component index; The frequency component index f ( i ) in the i-th cycle of the oscillatory wave signal is: ; where P is the power of the oscillatory wave signal, f represents the frequency, and the effective frequency band of the oscillatory wave signal is selected as 0.5 Hz to 10 Hz; Step 2.2.3: Calculate the entropy value based on the signal quality evaluation indexes in the time domain and the frequency domain to obtain the weight coefficient, and perform time-frequency domain multi-index weighted fusion on the oscillatory wave signal to obtain the fusion index; The said fusion index SQI ( i ) is as shown in the following formula: ; where , , , are respectively the entropy weight values of the evaluation indexes , , ; Step 2.2.4: Conduct threshold discrimination on the fusion index by setting a threshold, and classify the available oscillatory wave signals into signals with good quality and signals with poor quality; If the fusion index is greater than the set threshold, the available oscillatory wave signal is a signal with good quality, otherwise it is a signal with poor quality; For the signal with good quality, execute Step 3, and after filtering, denoising and baseline removal processing, further analyze it in the time domain, frequency domain and time-frequency domain; For the signal with poor quality, the oscillatory wave needs to be re-acquired.

[0010] Step 3: Preprocess the oscillatory wave signal with good quality, including noise removal, filtering and downsampling operations to remove the noise in the oscillatory wave signal; Step 3.1: Use a digital filter to remove the noise from the oscillatory wave signal; Specifically, it includes low-pass filtering and median filtering; Step 3.2: Use moving window averaging and a high-pass filter to filter, smooth the signal and remove baseline drift and low-frequency noise; Step 3.3: Perform downsampling using the extraction method, and selectively retain key feature samples through interpolation and multi-stage extraction; Step 3.4: Construct a supervised feature selection unit, align the preprocessed oscillatory wave signal with the blood pressure measurement values in the oscillatory wave database, and mark them as supervised samples.

[0011] Step 4: Extract time-domain, frequency-domain, and time-frequency features from the preprocessed oscillatory wave signal respectively to obtain multi-index features of the oscillatory wave signal; Step 4.1: Extract the time-domain features of the oscillatory wave signal based on the Hamilton algorithm; Step 4.2: Extract the frequency-domain features of the oscillatory wave signal based on the continuous wavelet transform (CWT); By performing the continuous wavelet transform (CWT) on the signal, the signal is decomposed into wavelet basis functions of different scales. The mathematical formula of CWT is as follows: ; where t is the time variable, x(t) is the input oscillatory wave signal; ψ ( t ) is the mother wavelet function; a is the scale parameter; b is the translation parameter; CWTx ( a , b ) is the wavelet coefficient; Step 4.3: Extract the time-frequency features of the oscillatory wave signal based on the Hilbert-Huang transform (HHT); The Hilbert-Huang transform (HHT) includes empirical mode decomposition (EMD) and Hilbert transform. The original blood pressure oscillatory wave signal is decomposed into several intrinsic mode functions (IMFs) using empirical mode decomposition (EMD); each IMF represents different frequency components in the signal; Perform the Hilbert transform on each IMF to calculate its instantaneous frequency and instantaneous amplitude. The formula of the Hilbert transform is: ; where: represents the Cauchy principal value, is the original oscillatory wave signal, is the integration variable; Step 4.4: Supervised feature selection based on the oscillatory wave database; For the preprocessed oscillatory wave signal, calculate the mutual information and F-value between the time-frequency features calculated using the blood pressure measurement values in the oscillatory wave database and SBP and DBP, and select the top 20 features strongly related to blood pressure: , ; where represents the mutual information, Yes And is the joint probability at that time, And are the marginal probabilities of X and Y respectively, 、 are the means of feature j in the high and low blood pressure groups, 、 are the variances.

[0012] Step 5: Standardize the time domain, frequency domain, and time-frequency features of the oscillatory wave signal obtained in Step 4 and perform supervised feature fusion; Step 5.1: Standardize the feature data of the oscillatory wave signal; Perform the following standardization processing on the time domain, frequency domain, and time-frequency features of the oscillatory wave signal: ; where x is the original feature value of the oscillatory wave signal, μ is the mean of the feature, σ is the standard deviation of the feature, and x1 is the standardized feature value; Step 5.2: Calculate the covariance matrix for the standardized oscillatory wave signal feature data; perform eigenvalue decomposition on the covariance matrix to obtain a series of eigenvalues and corresponding eigenvectors; use the Fisher score to rank the features and retain the top 80% of the features; Step 5.2.1: Calculate the covariance matrix: Calculate the covariance matrix for the standardized oscillatory wave signal feature data; Step 5.2.2: Eigenvalue decomposition: Perform eigenvalue decomposition on the covariance matrix to obtain a series of eigenvalues and corresponding eigenvectors; Step 5.2.3: Feature selection: Use the Fisher score to rank the features and retain the top 80% of the features; For each feature j , the Fisher score F j is defined as: ; where, K is the total number of classes; n k is the number of samples in class k ; μ k,j is the mean of the k th feature in class j ; μ j is the j th overall mean of the feature; σ k,j 2 is the k th in class jThe variance of a feature.

[0013] Step 5.2.4: Data Projection: Project the original oscillatory wave data onto the selected principal components to obtain the data after dimensionality reduction of the oscillatory wave signal features.

[0014] Step 5.3: Construct, train a supervised variational autoencoder, and obtain fused features; Step 5.3.1: Construct a supervised variational autoencoder and use the blood pressure measurement values in the oscillatory wave database as the supervision signal to train the supervised variational autoencoder; the variational autoencoder consists of an encoder and a decoder; the encoder is responsible for compressing the input features into a low-dimensional space, and the decoder is responsible for reconstructing the original features from the low-dimensional space; Step 5.3.2: Training of the supervised variational autoencoder: Use the dimensionality-reduced feature data as the input of the supervised variational autoencoder, and use the mean square error as the loss function to train the supervised variational autoencoder to minimize the difference between the input features and the reconstructed features; Step 5.3.3: Feature fusion representation: After training is completed, the output of the decoder, that is, the feature representation in the low-dimensional space, is used as the fused feature; The joint loss function of the supervised variational autoencoder is: ; where is the mean square error of feature reconstruction, is the mean square error of blood pressure prediction for the blood pressure values in the database, and β is a hyperparameter used to balance the relative importance of the mean square error of feature reconstruction and the mean square error of blood pressure prediction for the blood pressure values in the database in the joint loss function L.

[0015] Step 5.4: Feature weight calculation based on XGBoost; Step 5.4.1: Use the oscillatory wave data after preprocessing and feature extraction as the input and the corresponding blood pressure measurement values as the output to construct a training set; Step 5.4.2: Use the training set to train the XGBoost model to obtain a pre-trained model for non-invasive blood pressure prediction; During the training process, use the mean square error MSE as the loss function, and the formula is as follows: ; where n is the number of samples, is the true blood pressure value, is the blood pressure value predicted by the XGBoost model.

[0016] Step 5.4.3: After training is completed, obtain the weights of each feature to obtain a weighted feature vector; The XGBoost model outputs the importance scores of each feature. After normalizing these scores, the weights of each feature are obtained. 。

[0017] Step 6: Perform feature weighted fusion to achieve non-invasive blood pressure measurement; Step 6.1: Evaluate feature importance based on Fisher score; Step 6.1.1: Feature removal: Sequentially remove each fused feature from the input data, and use the remaining features to perform reconstruction through the trained supervised variational autoencoder; Step 6.1.2: Reconstruction error calculation: For the data after each feature removal, calculate the reconstruction error and compare it with the reconstruction error of the model when no fused feature is removed; Step 6.1.3: Importance evaluation: Reflect the importance of the feature according to the increase in the reconstruction error after feature removal; Step 6.2: Perform feature weighted fusion according to the feature importance evaluation results to obtain a weighted feature vector; Step 6.2.1: Feature weight assignment; The weight of a feature is proportional to its importance. Divide the importance score of each feature by the sum of the importance scores of all features to obtain the weight of this feature; Step 6.2.2: Perform feature weighted fusion based on weight assignment; Apply the calculated weight to each feature vector to obtain a weighted feature vector; Step 6.3: Perform non-invasive blood pressure measurement based on multi-index feature weighted fusion; Use the weighted feature vector obtained in Step 6.2 above and the feature weights of XGBoost obtained in Step 5.4 as the feature template of the current object to be detected; Adopt the weighted cosine similarity algorithm and Mahalanobis distance algorithm to perform SBP, DBP, and MAP recognition and final output in the pre-stored oscillatory wave feature database; Step 6.3.1: Calculate the weighted cosine similarity and weighted Mahalanobis distance between the sample to be recognized and the SBP waveform, DBP waveform, and MAP waveform samples; Specifically: Use the weighted feature vector and combine the XGBoost weights to calculate the weighted cosine similarity and weighted Mahalanobis distance between the sample to be recognized and each known SBP waveform, DBP waveform, and MAP waveform sample; Step 6.3.2: Based on the weighted cosine similarity and weighted Mahalanobis distance between the sample to be recognized and each known SBP waveform, DBP waveform, and MAP waveform sample, assign corresponding weight values and calculate the weighted similarity; Among them, for the sample A to be recognized and the known waveform sample B, the weighted cosine similarity is as described by the following formula: ; where is the weighted cosine similarity between sample A and sample B; is the feature weight output by the XGBoost model trained based on existing data; m is the number of feature dimensions; represents the weighted dot product; is the norm of the weighted vector A, is the norm of the weighted vector B.

[0018] The weighted Mahalanobis distance is as described by the following formula: ; where is the weighted Mahalanobis distance between sample A and sample B; W is the weight matrix, which is a diagonal matrix in the form of: ; Step 6.3.3: Identify SBP, DBP, and MAP based on the weighted similarity; By comparing the weighted similarities between the sample to be recognized and all known SBP waveform, DBP waveform, and MAP waveform samples, select the SBP, DBP, and MAP corresponding to the sample to be recognized with the lowest weighted similarity as the final result.

[0019] The beneficial effects of adopting the above technical solution are as follows: The present invention provides a blood pressure measurement system and method based on oscillatory wave feature extraction and supervised fusion. By constructing a large oscillatory wave database, storing a large number of oscillatory wave signals and blood pressure measurement values of subjects as reference standards, and through multi-index feature extraction, the signals are comprehensively analyzed from the time domain, frequency domain, and time-frequency domain. Using the supervised variational autoencoder and XGBoost algorithms, not only the features and patterns related to blood pressure are learned, but also accurate feature weights are calculated. The non-invasive blood pressure measurement is carried out by comprehensively considering the cosine similarity and Mahalanobis distance, and at the same time, the signal quality is strictly evaluated and screened. This series of operations effectively overcomes the measurement deviation caused by individual waveform specificity, greatly improves the accuracy and robustness of blood pressure monitoring, and injects new vitality into the development of non-invasive blood pressure measurement technology. Brief Description of the Drawings

[0020] Figure 1 is the structural block diagram of a blood pressure measurement system based on oscillatory wave feature extraction and supervised fusion provided by an embodiment of the present invention; Figure 2 is the flowchart of a blood pressure measurement method based on oscillatory wave feature extraction and supervised fusion provided by an embodiment of the present invention; Figure 3 is the flowchart of the "usable" oscillatory wave signal quality assessment provided by an embodiment of the present invention; Figure 4 Flow chart for time-domain, frequency-domain, and time-frequency feature extraction of the oscillatory wave signal provided by the embodiment of the present invention; Figure 5 Complete waveform diagram of the oscillatory wave signal provided by the embodiment of the present invention; Figure 6 Schematic diagram of the time-domain features of the oscillatory wave signal provided by the embodiment of the present invention; Figure 7 Flow chart for standardization processing and supervised feature fusion of multi-class feature data of the oscillatory wave signal provided by the embodiment of the present invention; Figure 8 Flow chart for feature weighted fusion of the oscillatory wave signal provided by the embodiment of the present invention. Detailed implementation manners

[0021] The following combines the accompanying drawings and embodiments to further describe in detail the specific implementation manners of the present invention. The following embodiments are used to illustrate the present invention, but are not used to limit the scope of the present invention.

[0022] On the one hand, a blood pressure measurement system based on oscillatory wave feature extraction and supervised fusion, as Figure 1 shown, includes: an oscillatory wave signal acquisition and available signal discrimination module, a preprocessing module, a multi-index feature extraction module, a multi-index feature standardization processing and supervised fusion module, and a non-invasive blood pressure measurement module; The oscillatory wave signal acquisition and available signal discrimination module is used to continuously collect the original oscillatory wave signal in real time, judge the oscillatory wave signal according to the cuff fitting and arm movement conditions, discriminate the available oscillatory wave signal and the unavailable oscillatory wave signal, evaluate the quality of the available oscillatory wave signal by fusing the time-domain and frequency-domain features, and call the available oscillatory wave with the fusion feature greater than the set threshold as a good-quality oscillatory wave signal, and the remaining available oscillatory waves as poor-quality oscillatory wave signals.

[0023] In this embodiment, the oscillatory wave signal acquisition and available signal discrimination module discriminates the acquired original oscillatory wave signal as an available or unavailable signal according to the cuff fitting and the movement of the arm. The unavailable oscillatory wave signal is caused by interference generated due to reasons such as the cuff being too tightly or loosely fitted, the user's arm moving or exerting force during the measurement process, resulting in the distortion or complete disappearance of the oscillatory wave waveform. Since the unavailable oscillatory wave signal has lost its effective information and cannot be simply restored, it needs to be acquired again. The available oscillatory wave signal indicates that its signal quality has not been completely overwhelmed by noise or interference, and the signal quality can be further distinguished as good or bad. For the available oscillatory wave signal, the indicators in both the time domain and the frequency domain are comprehensively considered to evaluate the quality of the signal. In time domain analysis, the amplitude change degree index, the baseline drift degree index, and the pulse rate consistency degree index are used to evaluate the signal quality. In frequency domain analysis, the frequency component index is used to evaluate the signal quality. The weight coefficients obtained by calculating the entropy values based on each index are used to weight and fuse the time-frequency multi-index features of the oscillatory wave signal to obtain a fusion index. Based on the fusion index, threshold discrimination is performed to classify the available oscillatory wave signal into signals with good and poor quality. For the signals with good quality, further analysis is carried out. For the signals with poor quality, they need to be acquired again.

[0024] The preprocessing module preprocesses the oscillatory wave signal with good quality to remove the noise in the oscillatory wave signal and improve the quality of the oscillatory wave signal. The multi-index feature extraction module extracts the time domain features of the oscillatory wave signal based on the Hamilton algorithm, extracts the frequency domain features of the oscillatory wave signal based on the continuous wavelet transform, and extracts the time-frequency features of the oscillatory wave signal based on the Hilbert-Huang transform (HHT). The multi-index feature standardization processing and supervised fusion module performs standardization processing on the feature data of the oscillatory wave signal. After the features of the oscillatory wave signal are extracted, the extracted feature data is first standardized to eliminate the dimension and scale differences. Subsequently, the Fisher criterion (Fisher Score) is applied to perform feature selection on the standardized feature data. By calculating the discrimination ability between each feature and the target variables (such as systolic blood pressure and diastolic blood pressure), the most discriminative feature subset is selected, thereby reducing the feature dimension and improving the accuracy and efficiency of the blood pressure estimation model. The variational autoencoder is used to further perform feature learning on the selected features. The non-invasive blood pressure measurement module, based on the features extracted from the variational autoencoder, evaluates the feature importance based on the reconstruction error contribution and performs multi-index feature weighted fusion based on the feature weight assignment. To more comprehensively measure the similarity between the oscillatory wave signal features, the cosine similarity and the Mahalanobis distance are combined by setting weight parameters. The cosine similarity is used to measure the direction consistency of the feature vectors, while the Mahalanobis distance is used to measure the difference of the feature vectors considering the data distribution. By optimizing the weight parameters, multi-angle evaluation of the feature similarity can be achieved, thereby improving the accuracy and robustness of the blood pressure estimation model.

[0025] In this embodiment, an oscillometric wave database should be established. First, place the oscillometric wave signal device correctly on the upper arm of the subject and fix it with a cuff to ensure accuracy and comfort. The cuff will gradually apply pressure, record the oscillometric wave data when the subject is under pressure, and store it. Then, analyze the data to obtain the blood pressure measurement value. It is necessary to ensure that there are 3000 pieces of oscillometric wave data with blood pressure measurement values stored in the database as a reference standard for subsequent machine learning.

[0026] In this embodiment, the preprocessing module performs filtering and denoising preprocessing on the oscillometric wave signals with good quality, eliminates the noise in the oscillometric wave signals, and improves the signal quality. First, use a digital filter, moving window average, and high-pass filter for filtering, smooth the signal, and remove baseline drift, low-frequency noise, and high-frequency noise, and filter out other interference components such as other burst noises, so as to reduce the noise interference in the signal and make the subsequent processing more accurate and reliable. Align the preprocessed oscillometric wave signals with the blood pressure data in the database and mark them as supervised samples for subsequent supervised learning.

[0027] The multi-index feature extraction module extracts the time-domain features of the oscillometric wave signals based on the Hamilton algorithm, extracts the frequency-domain features of the oscillometric wave signals based on the continuous wavelet transform, and extracts the time-frequency features of the oscillometric wave signals based on the Hilbert-Huang transform (HHT). The multi-index feature standardization processing and supervised fusion module performs standardization processing on the feature data of the oscillometric wave signals, applies independent component analysis (ICA) to reduce the dimension of the standardized oscillometric wave signal feature data, and uses a supervised variational autoencoder to perform feature learning and fusion on the features selected by the Fisher criterion. In this embodiment, the multi-index feature extraction module extracts time-domain, frequency-domain, and time-frequency features from the preprocessed oscillatory wave signal respectively to obtain multi-class index features of the oscillatory wave signal. The time-domain features of the oscillatory wave signal are extracted based on the Hamilton feature point extraction algorithm. The Hamilton algorithm is used to detect the starting point of the waveform period, and then the oscillatory wave waveform is segmented beat by beat to extract the time-domain waveform profile features. The frequency-domain features of the oscillatory wave signal are extracted based on continuous wavelet transform. The frequency bands are divided using the frequency distribution characteristics of the signal, and the spectral energy ratio of different sub-band frequency bands is used as the frequency-domain feature of the oscillatory wave signal. The time-frequency diagram features of the oscillatory wave signal are extracted based on Hilbert-Huang transform. The sliding window method is used for Hilbert-Huang transform (HHT) of the oscillatory wave signal. First, the intrinsic mode functions (IMFs) of the signal are obtained through empirical mode decomposition (EMD), and then the Hilbert transform is applied to obtain the time-frequency distribution. Next, the main energy distribution region in the time-frequency diagram is detected, and the closed curve of the edge contour of this region is extracted. Based on the closed curve, 8 local maximum values of the local peak points of the energy on the time-frequency distribution diagram are identified using the maximum value detection algorithm, and the selection basis is the sorting of their energy values. In this embodiment, the multi-index feature standardization processing and supervised fusion module performs standardization processing and supervised feature fusion on the multi-index feature data of the oscillatory wave signal. The standardization of the multi-index features of the oscillatory wave signal ensures that all features contribute approximately equally to the objective function by converting all features to the same scale, avoiding some features dominating in parameter updates. Independent components analysis (ICA) is used for the feature data of the standardized oscillatory wave signal. It aims to reduce the feature dimension while retaining the independent information in the data. ICA first preprocesses the standardized feature data to ensure that the data meets the analysis requirements. Then, the ICA algorithm is used to extract independent components from the mixed signal. This process does not involve the calculation of the covariance matrix, but directly applies the algorithm to separate the independent components in the original signal. By maximizing non-Gaussianity, ICA can effectively identify multiple source signals. After obtaining the independent components, the N most important independent components for the analysis can be selected according to actual needs. These components represent the main signal sources in the original data and can reflect the key features of the data. Finally, the original data is projected onto the selected independent components to obtain the feature data after dimensionality reduction. The Fisher criterion is applied to perform feature selection on the feature data of the standardized oscillatory wave signal, and the supervised variational autoencoder is used to further perform feature learning and fusion on the features after ICA dimensionality reduction. The variational autoencoder structure consists of an encoder and a decoder. The encoder is responsible for compressing the input features into a low-dimensional latent space and generating the distribution parameters (usually the mean and variance) of the points in this space. The decoder then attempts to reconstruct the original input features from this latent space.In the decoder part, the generated output is the feature representation sampled from the latent space, and these feature representations can be used as the basis for further analysis or fusion. Different from traditional autoencoders, variational autoencoders introduce probability in the latent space, making them more flexible in generating models and feature representations.

[0028] The non-invasive blood pressure measurement module is based on the fused features extracted by the supervised variational autoencoder, evaluates the feature importance based on the contribution of the reconstruction error, performs multi-index feature weighted fusion based on the XGBoost weight assignment, applies the pre-trained model for non-invasive blood pressure prediction to measure the non-invasive blood pressure, and finally completes the blood pressure measurement.

[0029] In this embodiment, the non-invasive blood pressure measurement module performs feature weighted fusion; based on the fused features extracted from the supervised variational autoencoder, it evaluates the feature importance based on the contribution of the reconstruction error; analyzes the change of the autoencoder reconstruction error after each fused feature is removed. The greater the contribution of the feature to the reconstruction error, the more important the feature is for the representation of the data, which is convenient for evaluating the importance of the fused features. For the data after each feature removal, calculate the reconstruction error and compare it with the reconstruction error of the original model (when no feature is removed). The increase in the reconstruction error after feature removal reflects the importance of the feature, and the features with a significant increase in the reconstruction error are considered more important for the model and data representation. In the process of feature fusion and similarity matching, an XGBoost model trained based on blood pressure and waveform data in the database is used to determine the weights of each feature to optimize the weighted similarity calculation, and the weighted similarity is used for non-invasive blood pressure measurement. The cosine similarity and Mahalanobis distance metric algorithms are used for blood pressure measurement and final output. By comparing the weighted similarities between the sample to be identified and all known SBP waveform and DBP waveform samples, the lower the weighted similarity, the closer the waveform is to SBP / DBP, and thus the higher the credibility of the blood pressure measurement. Effectively utilize the information in the weighted feature vector and select the sample to be identified with the lowest weighted similarity as the final SBP and DBP results.

[0030] On the other hand, a blood pressure measurement method based on oscillatory wave feature extraction and supervised fusion is implemented based on the aforementioned blood pressure measurement system, as Figure 2 shown, and includes the following steps: Step 1: Construct an oscillatory wave database, and the oscillatory wave database includes oscillatory wave signals and their corresponding blood pressure measurement values; In this embodiment, the oscillatory wave signals of at least 3000 subjects and their corresponding blood pressure measurement values are stored and processed as the reference standard; Step 1.1: Collect the oscillatory wave signals of at least 3000 subjects and their corresponding blood pressure measurement values as the supervision index, which is the reference standard for subsequent blood pressure prediction; Step 1.2: Preprocess the oscillatory wave signals in the oscillatory wave database; Step 1.3: Extract the time-domain, frequency-domain, and time-frequency features of the oscillatory wave signals in the oscillatory wave database, and select the supervised features related to blood pressure measurement; Step 1.4: Construct and train a supervised variational autoencoder; Step 1.5: Calculate the feature weights based on XGBoost and the oscillatory wave signals to be analyzed.

[0031] Step 2: Obtain the original oscillatory wave signals of the samples to be recognized, and based on the sensor fitting and wire connection detection indicators, distinguish the available oscillatory wave signals from the unavailable oscillatory wave signals, and evaluate the quality of the available oscillatory wave signals; Step 2.1: Detect errors in the collected oscillatory wave signals to distinguish the available oscillatory wave signals from the unavailable oscillatory wave signals; The unavailable oscillatory wave signals are oscillatory wave signals with distorted waveforms or completely disappeared, losing effective information; In this embodiment, the "unavailable" oscillatory wave signals refer to the interference caused by other reasons such as too tight or insecure cuff fitting, movement or force of the user's arm during measurement, resulting in distorted waveforms or complete disappearance of the oscillatory wave signals; the "unavailable" signals cannot be simply restored because they have lost effective information, so they need to be collected again; the "available" signals indicate that their signal quality has not been completely overwhelmed by noise or interference, and the quality of good and bad can be further distinguished. Good signals can be directly used for interpretation, while poor signals need to be collected again; Step 2.2: For the available oscillatory wave signals, comprehensively consider the indicators in both the time domain and the frequency domain to evaluate the quality of the signals as good or bad, such as Figure 3 shown; Step 2.2.1: Conduct time-domain analysis on the oscillatory wave signals to determine the signal quality evaluation indicators in the time domain; In the time-domain analysis, the amplitude change degree index, baseline drift degree index, and pulse rate consistency degree index are used to evaluate the quality of the oscillatory wave signals; Specifically: Perform variational mode decomposition and period segmentation on the oscillatory wave signals respectively to obtain the reconstructed signal and single-period signal representing the baseline drift of the signal; The amplitude change degree index is obtained through the amplitude difference of the signal waveforms in multiple cardiac cycles; the amplitude change degree index of the oscillatory wave signal is obtained through the amplitude difference of the signal waveforms in multiple cardiac cycles; the amplitude change degree index of the i-th cycle of the oscillatory wave signal a ( i ) is shown by the following formula: ; where and They are respectively the maximum and minimum values of the waveform amplitude of the i th cycle of the oscillatory wave signal after filtering processing; The baseline drift degree index is obtained by windowing the reconstructed signal; the baseline drift degree index is obtained by windowing the reconstructed signal; the reconstructed signal is windowed to obtain the baseline drift degree index within the i-th cycle of the oscillatory wave signal b ( i ), as shown in the following formula: ; where bWave i is the reconstructed signal of the i-th cycle of the oscillatory wave signal, b is the baseline standard value of the oscillatory wave signal in the stationary state; The pulse rate consistency index is calculated by calculating the consistency of the signal time lengths within multiple cycles; the pulse rate consistency index is calculated by calculating the consistency of the signal time lengths within multiple cycles; the period length of the oscillatory wave signal is set to be , and the pulse rate consistency index within the i-th cycle of the oscillatory wave signal is obtained , as shown in the following formula: ; where is the standard deviation of the calculated i th signal cycle length, ) is the mean value of the calculated i-th signal cycle length; Step 2.2.2: Perform frequency-domain analysis on the oscillatory wave signal to determine the signal quality evaluation index in the frequency domain; In the frequency-domain analysis, the frequency component index is used to evaluate the quality of the oscillatory wave signal; Specifically: The effective frequency band of the oscillatory wave signal is selected through fast Fourier transform, and the power spectral density is calculated to obtain the frequency component index; the frequency component index within the i-th cycle of the oscillatory wave signal f ( i ) is: ; where P is the power of the oscillatory wave signal, f represents the frequency, and the effective frequency band of the oscillatory wave signal is selected as 0.5 Hz to 10 Hz; Step 2.2.3: Calculate the entropy value based on the signal quality evaluation indexes in the time domain and frequency domain to obtain the weight coefficient, and perform time-frequency domain multi-index weighted fusion on the oscillatory wave signal to obtain the fusion index; The fusion index SQI ( i ) is as shown in the following formula: ; where 、 , , are the entropy weight values of the evaluation indicators , , respectively; Step 2.2.4: Perform threshold discrimination on the fusion index by setting a threshold, and classify the available oscillatory wave signals into signals with good quality and poor quality; If the fusion index is greater than the set threshold, the available oscillatory wave signal is a signal with good quality, otherwise it is a signal with poor quality; for the signals with good quality, execute Step 3. After filtering, denoising and baseline removal, further analyze in the time domain, frequency domain and time-frequency domain; for the signals with poor quality, the oscillatory waves need to be re-acquired.

[0032] Step 3: Preprocess the oscillatory wave signals with good quality, including noise removal, filtering and downsampling operations to remove the noise in the oscillatory wave signals; there are various noise sources in the oscillatory wave signals, including environmental noise, light scattering, motion artifacts, baseline drift and sensor noise. These noise components will reduce the accuracy and reliability of the signals. Therefore, in the preprocessing process, it is necessary to perform noise removal and filtering operations to reduce the influence of noise and extract the clean and accurate oscillatory wave signal features; Step 3.1: Use a digital filter to remove the noise in the oscillatory wave signals; specifically include low-pass filtering and median filtering; to filter out interference components such as high-frequency noise and other burst noises, thereby reducing the noise interference in the signals and making the subsequent processing more accurate and reliable; Step 3.2: Use moving window averaging and a high-pass filter to filter, smooth the signal and remove baseline drift and low-frequency noise; improve the stability and reliability of the signal and further improve the stability of the oscillatory wave signal; Step 3.3: Use the decimation method for downsampling, and selectively retain key feature samples through interpolation and multi-stage decimation; reduce the data volume and computational complexity, and at the same time maximize the retention of important information and reduce the computational load; Step 3.4: Construct a supervised feature selection unit, align the preprocessed oscillatory wave signals with the blood pressure measurement values in the oscillatory wave database, and mark them as supervised samples for subsequent supervised learning.

[0033] Step 4: Extract time-domain, frequency-domain and time-frequency features from the preprocessed oscillatory wave signals respectively to obtain multi-index features of the oscillatory wave signals, such as Figure 4 shown; Step 4.1: Extract the time-domain features of the oscillatory wave signals based on the Hamilton algorithm; The waveform of the oscillatory wave signal mainly contains two waveforms, namely the main wave and the dicrotic wave. The characteristics of the waveform, such as amplitude, slope, and time, contain a large amount of physiological information of the human body, can effectively reflect the rhythm pattern and functional state of the cardiovascular system, and contain important physiological information and clinical significance. The complete oscillatory wave waveform diagram is as shown in Figure 5 shown. Use the Hamilton algorithm to extract the waveform characteristics in the time domain from the single-cycle waveform of the oscillatory wave signal: First, perform a first-order difference on the sparse signal (that is, calculate the derivative of the signal) to reflect the change rate of the signal at each sampling point; square the derivative signal, which can further amplify the large waveform changes (such as QRS waves) in the signal and suppress smaller changes and noises; apply a sliding window with a fixed length to the squared signal for integration to obtain the energy distribution of the signal over a period of time. The size of the moving window is usually set according to the specific signal sampling frequency and the duration of the target waveform. This step can smooth the signal and highlight the key features in the oscillatory wave signal; by analyzing the local maximum, minimum, and change slope of the signal, determine the positions of the characteristic points; output the detected positions of the characteristic points and the corresponding time-domain characteristic information (such as the time points, amplitudes, periods, partial times, slopes of the wave peaks and wave valleys) for subsequent use; Step 4.2: Extract the frequency-domain characteristics of the oscillatory wave signal based on the continuous wavelet transform CWT; Divide the frequency band according to the frequency distribution characteristics of the signal, and use the spectral energy ratio of different sub-band frequency bands as the frequency-domain characteristics of the oscillatory wave signal; the frequency-domain characteristics of the oscillatory wave signal, by analyzing the distribution and components of the signal in the frequency domain, reveal the complexity and nuances of the dynamic regulation of the cardiovascular system. These characteristics reflect the responses of the heart and blood vessel systems to changes in physiological and pathological conditions. The specific method for extracting the frequency-domain characteristics of the oscillatory wave signal based on the continuous wavelet transform (CWT) is as follows: By performing the continuous wavelet transform CWT on the signal, the signal is decomposed into wavelet basis functions of different scales. The mathematical formula of CWT is as follows: ; where, t is the time variable, x(t) is the input oscillatory wave signal; ψ ( t ) is the mother wavelet function; a is the scale parameter, which controls the compression or expansion of the wavelet function and corresponds to the frequency; b is the translation parameter, which represents the position of the wavelet on the time axis; CWTx ( a , b ) is the wavelet coefficient, which reflects the frequency information of the signal at different time points and scales; The wavelet coefficient matrix calculated by CWT can reflect the change of the frequency components of the signal over time.

[0034] By analyzing wavelet coefficients at different scales, multi-scale frequency domain features of the signal can be extracted, including: Main frequency: Corresponding to the main frequency component of the signal oscillation, it can be determined by wavelet coefficients at different scales; Instantaneous frequency: The frequency variation of the signal at different time points. CWT can reflect the change of frequency over time; Energy distribution: The sum of squares of wavelet coefficients of the signal at different scales can reflect the energy distribution of the signal, which is used to capture the intensity of different frequency components; Local spectral characteristics: By analyzing wavelet coefficients at specific time points, the frequency domain features of specific events (such as QRS waves, P waves, T waves in electrocardiogram signals) can be identified; Step 4.3: Extract time-frequency features of the oscillatory wave signal based on the Hilbert-Huang transform HHT; The time-frequency diagram features of the oscillatory wave signal provide a method to simultaneously examine the changes of the signal in time and frequency, revealing the complexity and multi-dimensionality of the dynamic changes of the oscillatory wave. This analysis can capture the changes of frequency components of the oscillatory wave at different time points, thus providing more abundant and detailed information about the state of the cardiovascular system. The specific method for extracting the time-frequency diagram features of the oscillatory wave signal based on the Hilbert-Huang transform (HHT) is as follows: The Hilbert-Huang transform HHT includes empirical mode decomposition EMD and Hilbert transform. The original blood pressure oscillatory wave signal is decomposed into several intrinsic mode functions IMF (IMFs, Intrinsic Mode Functions) using empirical mode decomposition EMD. Each IMF represents a different frequency component in the signal; the decomposition steps are as follows: (1) Envelope fitting: Using local extreme points, the upper envelope and lower envelope of the signal are respectively constructed, usually obtained by spline interpolation; (2) Extract the initial IMF: Remove the mean value of the envelope from the original signal to obtain a candidate IMF, and iterate repeatedly until it meets the IMF conditions, that is: The number of zero-crossing points and extreme points of the IMF is equal or differs by no more than 1; In the whole signal, the local average value of the envelope curve of the IMF is zero; (3) Remove the IMF and repeat the steps: Remove the IMF from the original signal, and then continue to decompose the residual signal until all mode functions are extracted.

[0035] The result of empirical mode decomposition EMD is a set of IMFs, which are regarded as components of the signal at different frequency levels; Perform the Hilbert transform on each IMF, calculate its instantaneous frequency and instantaneous amplitude, and the formula for the Hilbert transform is: ; where: represents the Cauchy principal value, is the original oscillatory wave signal, is the integration variable.

[0036] Based on the time-frequency diagram, as Figure 6 shown, the following key features can be extracted: (1) Instantaneous frequency variation: Observe the dynamic variation of the frequency components in the blood pressure oscillatory wave signal over time. For example, high-frequency components are usually related to rapidly changing cardiovascular activities, while low-frequency components may reflect the long-term trend of blood pressure; (2) Energy distribution: The energy concentration regions in the time-frequency diagram correspond to important events of the signal. By analyzing these energy peak regions, the frequency-domain characteristics of key events (such as heartbeats or blood pressure fluctuations) can be identified; (3) Frequency bandwidth: By observing the variation of the frequency bandwidth, infer the changes in blood pressure fluctuations at different time periods; (4) Time-frequency clustering: Analyze the clustering of the signal in the time-frequency diagram to determine whether the blood pressure signal exhibits stable periodic characteristics; Step 4.4: Supervised feature selection based on the oscillatory wave database; For the preprocessed oscillatory wave signal, calculate the mutual information and F-value between the time-frequency features calculated using the blood pressure measurement values in the oscillatory wave database and SBP and DBP, and screen the top 20 features strongly related to blood pressure: , ; where, represents the mutual information, is and is the joint probability when and are the marginal probabilities of X and Y respectively, , are the means of feature j in the high- and low-blood pressure groups, , are the variances.

[0037] Step 5: Perform standardization processing and supervised feature fusion on the time domain, frequency domain, and time-frequency features of the oscillatory wave signal obtained in Step 4, as Figure 7 shown; Step 5.1: Standardize the oscillatory wave signal feature data; The multi-class features of the oscillatory wave signal have different dimensions and numerical ranges. Standardization eliminates the influence of dimensions by converting all features to the same scale, preventing the model training from being biased towards features with large numerical ranges. Standardizing the features also ensures that all features contribute approximately equally to the objective function, avoiding certain features from dominating the parameter updates. Additionally, standardization helps prevent model overfitting and improves the model's generalization ability to unseen data. The following standardization processes are performed on the time domain, frequency domain, and time-frequency features of the oscillatory wave signal: ; where \(x\) is the original feature value of the oscillatory wave signal, \(\mu\) is the mean of the feature, \(\sigma\) is the standard deviation of the feature, and \(x_1\) is the standardized feature value; Step 5.2: Calculate the covariance matrix for the standardized oscillatory wave signal feature data; perform eigenvalue decomposition on the covariance matrix to obtain a series of eigenvalues and corresponding eigenvectors; use the Fisher score to rank the features and retain the top 80% of the features; Step 5.2.1: Calculate the covariance matrix: Calculate the covariance matrix for the standardized oscillatory wave signal feature data; Step 5.2.2: Eigenvalue decomposition: Perform eigenvalue decomposition on the covariance matrix to obtain a series of eigenvalues and corresponding eigenvectors; Step 5.2.3: Feature selection: Use the Fisher score to rank the features and retain the top 80% of the features; For each feature j , the Fisher score F j is defined as: ; where K is the total number of classes; n k is the number of samples in class k ; μ k,j is the mean of the k -th feature in class j ; μ j is the j -th feature's overall mean (mean across all classes); σ k,j 2 is the variance of the k -th feature in class j ; Step 5.2.4: Data projection: Project the original oscillatory wave data onto the selected principal components to obtain the data with reduced dimensions of the oscillatory wave signal features; Step 5.3: Construct a supervised variational autoencoder, train the model, and obtain the fused features; Step 5.3.1: Construct a supervised variational autoencoder and use the blood pressure measurement values in the oscillatory wave database as the supervision signal to train the supervised variational autoencoder; the variational autoencoder consists of an encoder and a decoder; the encoder is responsible for compressing the input features into a low-dimensional space, and the decoder is responsible for reconstructing the original features from the low-dimensional space; The supervised variational autoencoder is a model that combines the variational autoencoder and supervised learning. While learning the latent representation of the data, it uses the supervision signal (here, the blood pressure data in the database) to optimize the model's parameters to better complete specific tasks such as blood pressure prediction. By using the blood pressure data in the database as the supervision signal, the model can learn the features and patterns related to blood pressure, thereby improving the prediction accuracy and reliability of blood pressure; Step 5.3.2: Training of the supervised variational autoencoder: Use the dimensionality-reduced feature data as the input of the supervised variational autoencoder, and use the mean square error as the loss function to train the supervised variational autoencoder to minimize the difference between the input features and the reconstructed features; Step 5.3.3: Feature fusion representation: After training is completed, the output of the decoder, that is, the feature representation in the low-dimensional space, is used as the fused feature; The joint loss function of the supervised variational autoencoder is: ; where is the mean square error of feature reconstruction, is the mean square error of blood pressure prediction of the blood pressure values in the database, and β is a hyperparameter used to balance the relative importance of the mean square error of feature reconstruction and the mean square error of blood pressure prediction of the blood pressure values in the database in the joint loss function L; Step 5.4: Feature weight calculation based on XGBoost; To more accurately measure the importance of each feature in non-invasive blood pressure measurement, the present invention uses the XGBoost algorithm to calculate the feature weights based on the blood pressure and waveform data in the database. The specific steps are as follows: Step 5.4.1: Use the oscillatory wave data after preprocessing and feature extraction as the input, and the corresponding blood pressure measurement values (systolic blood pressure SBP and diastolic blood pressure DBP) as the output to construct a training set; Step 5.4.2: Use the training set to train the XGBoost model; XGBoost is a gradient boosting tree algorithm that iteratively trains multiple decision trees to gradually optimize the prediction performance of the model. During the training process, the mean square error MSE is used as the loss function, and the formula is as follows: ; where n is the number of samples, is the true blood pressure value, is the blood pressure value predicted by the XGBoost model; Step 5.4.3: After training, the weights of each feature are obtained, and the weighted feature vector is obtained; The XGBoost model outputs the importance scores of each feature. After normalizing these scores, the weights of each feature are obtained ; These weights reflect the relative importance of each feature in predicting the blood pressure value.

[0038] Step 6: Perform feature weighted fusion to achieve non-invasive blood pressure measurement; as Figure 8 shown; Based on the fusion features extracted from the supervised variational autoencoder in Step 5.3, these features already contain the key information of the original data, but have a lower dimension and more concentrated information.

[0039] Based on the Fisher score, the feature importance is evaluated, and the change of the reconstruction error of the variational autoencoder after each fusion feature is removed is analyzed; The greater the contribution of the feature to the reconstruction error, the more important the feature is for the representation of the data, which is convenient for evaluating the importance of the fusion features; Step 6.1: Based on the Fisher score, the feature importance is evaluated; First, feature removal is performed. Each fused feature is sequentially removed from the input data, and the remaining features are used to reconstruct through the trained variational autoencoder; For the data after each feature removal, the reconstruction error is calculated, compared with the reconstruction error of the original model, and the importance of the feature is reflected according to the increase in the reconstruction error after feature removal; Step 6.1.1: Feature removal: Each fused feature is sequentially removed from the input data, and the remaining features are used to reconstruct through the trained supervised variational autoencoder; Step 6.1.2: Reconstruction error calculation: For the data after each feature removal, the reconstruction error is calculated and compared with the reconstruction error of the model without removing any fused features; Step 6.1.3: Importance evaluation: The importance of the feature is reflected according to the increase in the reconstruction error after feature removal; Step 6.2: According to the feature importance evaluation results, perform feature weighted fusion to obtain the weighted feature vector; The weight of the feature is proportional to its importance. Divide the importance score of each feature by the sum of the importance scores of all features to obtain the weight of the feature, so as to perform feature weight allocation; Based on the weight allocation, perform feature weighted fusion; Apply the calculated weight to each feature vector to obtain the weighted feature vector; Step 6.2.1: Feature weight allocation; The weight of a feature is directly proportional to its importance. Divide the importance score of each feature by the sum of the importance scores of all features to obtain the weight of that feature; Step 6.2.2: Perform feature weighted fusion based on weight assignment; Apply the calculated weight to each feature vector to obtain a weighted feature vector; Step 6.3: Perform non-invasive blood pressure measurement based on multi-index feature weighted fusion; Use the weighted feature vector obtained in Step 6.2 above and the feature weights of XGBoost obtained in Step 5.4 as the feature template of the current object to be detected; Adopt the weighted cosine similarity algorithm and the Mahalanobis distance algorithm to perform SBP, DBP, and MAP recognition and final output in the pre-stored oscillatory wave feature database; Step 6.3.1: Calculate the weighted cosine similarity and weighted Mahalanobis distance between the sample to be recognized and the SBP waveform, DBP waveform, and MAP waveform samples; Specifically: Use the weighted feature vector and combine the XGBoost weights to calculate the weighted cosine similarity and weighted Mahalanobis distance between the sample to be recognized and each known SBP waveform, DBP waveform, and MAP waveform sample; Step 6.3.2: Based on the weighted cosine similarity and weighted Mahalanobis distance between the sample to be recognized and each known SBP waveform, DBP waveform, and MAP waveform sample, assign corresponding weight values and calculate the weighted similarity; Among them, for the sample A to be recognized and the known waveform sample B, the weighted cosine similarity is as described by the following formula: ; where is the weighted cosine similarity between sample A and sample B; is the feature weight output by the XGBoost model trained based on existing data; By introducing the XGBoost weight, the importance of each feature in blood pressure measurement can be more accurately reflected, thereby improving the accuracy of similarity matching; m is the number of feature dimensions; represents the weighted dot product, which takes into account the weight of each feature; is the norm of the weighted vector A, is the norm of the weighted vector B; The weighted Mahalanobis distance is as described by the following formula: ; where is the weighted Mahalanobis distance between sample A and sample B; W is the weight matrix, which is a diagonal matrix in the form of: ; Step 6.3.3: Perform SBP, DBP, and MAP recognition based on the weighted similarity; By comparing the weighted similarity between the sample to be recognized and all known SBP waveform, DBP waveform, and MAP waveform samples, the lower the weighted similarity, the closer the waveform is to SBP / DBP / MAP, thus the higher the credibility of blood pressure measurement. Effectively utilize the information in the weighted feature vector, and select the SBP, DBP, and MAP corresponding to the sample to be recognized with the lowest weighted similarity as the final result.

[0040] By comparing the weighted similarity between the sample to be recognized and the waveform samples in all known databases, select the blood pressure value corresponding to the sample with the highest similarity as the final result, and ensure stability through the consistency verification of three consecutive cardiac cycles.

[0041] The above description is only a preferred embodiment of the present disclosure and an explanation of the applied technical principles. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present disclosure is not limited to the technical solutions formed by the specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above inventive concept. For example, the technical solutions formed by mutually replacing the above features with the (but not limited to) technical features with similar functions disclosed in the embodiments of the present disclosure.

Claims

1. A blood pressure measurement system based on oscillometric wave feature extraction and supervised fusion, characterized in that, Including: An oscillatory wave signal acquisition and available signal discrimination module, a preprocessing module, a multi-index feature extraction module, a multi-index feature normalization processing and supervised fusion module, and a non-invasive blood pressure measurement module; The oscillatory wave signal acquisition and available signal discrimination module is used to continuously collect the original oscillatory wave signal in real time, judge the oscillatory wave signal according to the cuff fitting and arm movement conditions, and discriminate the available oscillatory wave signal and the unavailable oscillatory wave signal. The unavailable oscillatory wave signal is a signal with distorted waveform or completely disappeared; classify the available oscillatory wave signal into signals with good quality and poor quality; among them, the quality of the available oscillatory wave signal is evaluated by fusing time-domain and frequency-domain features. The available oscillatory wave with the fused feature greater than the set threshold is called an oscillatory wave signal with good quality, and the remaining available oscillatory waves are called oscillatory wave signals with poor quality; The preprocessing module preprocesses the oscillatory wave signal with good quality, removes the noise in the oscillatory wave signal, and improves the quality of the oscillatory wave signal; The multi-index feature extraction module extracts the time-domain features of the oscillatory wave signal based on the Hamilton algorithm, extracts the frequency-domain features of the oscillatory wave signal based on continuous wavelet transform, and extracts the time-frequency features of the oscillatory wave signal based on the Hilbert-Huang transform HHT; The multi-index feature normalization processing and supervised fusion module performs normalization processing on the feature data of the oscillatory wave signal, applies independent component analysis ICA to reduce the dimension of the normalized oscillatory wave signal feature data, and uses a supervised variational autoencoder to perform feature learning and fusion on the features selected by the Fisher criterion; specifically: after the oscillatory wave signal feature extraction, first perform normalization processing on the extracted feature data, then apply the Fisher criterion to perform feature selection on the normalized feature data, calculate the discriminant ability between each feature and the target variable, screen out the most discriminative feature subset, and use the variational autoencoder to further perform feature learning on the selected features; The non-invasive blood pressure measurement module is based on the fused features extracted by the supervised variational autoencoder, evaluates the feature importance based on the reconstruction error contribution, performs multi-index feature weighted fusion based on the XGBoost weight assignment, and applies a pre-trained model for non-invasive blood pressure prediction to measure the non-invasive blood pressure, and finally completes the blood pressure measurement; specifically, combines the two measurement methods of cosine similarity and Mahalanobis distance, and combines the two by setting weight parameters. Among them, cosine similarity is used to measure the direction consistency of feature vectors, while Mahalanobis distance is used to measure the difference of feature vectors considering the data distribution; by optimizing the weight parameters, multi-angle evaluation of feature similarity is realized.

2. A blood pressure measurement method based on oscillometric wave feature extraction and supervised fusion, which is implemented by a blood pressure measurement system based on oscillometric wave feature extraction and supervised fusion described in claim 1, characterized in that, Including the following steps: Step 1: Construct an oscillatory wave database, which includes oscillatory wave signals and their corresponding blood pressure measurement values; Step 2: Obtain the original oscillatory wave signal of the sample to be recognized, and judge the available oscillatory wave signal and the unavailable oscillatory wave signal according to the sensor fitting and wire connection detection indicators, and evaluate the quality of the available oscillatory wave signal; Step 3: Preprocess the oscillatory wave signal with good quality, including noise removal, filtering and downsampling operations, to remove the noise in the oscillatory wave signal; Step 4: Extract time-domain, frequency-domain, and time-frequency features from the preprocessed oscillatory wave signals respectively to obtain multi-index features of the oscillatory wave signals; Step 5: Perform standardization processing and supervised feature fusion on the time-domain, frequency-domain, and time-frequency features of the oscillatory wave signals obtained in Step 4; Step 6: Perform feature weighted fusion to realize non-invasive blood pressure measurement.

3. The blood pressure measurement method based on oscillation wave feature extraction and supervised fusion according to claim 2, wherein The said Step 1 includes the following steps: Step 1.1: Collect the oscillatory wave signals of the subjects and their corresponding blood pressure measurement values as supervised indicators; Step 1.2: Preprocess the oscillatory wave signals in the oscillatory wave database; Step 1.3: Extract the time-domain, frequency-domain, and time-frequency features of the oscillatory wave signals in the oscillatory wave database, and select the supervised features related to blood pressure measurement; Step 1.4: Construct and train a supervised variational autoencoder; Step 1.5: Calculate the feature weights based on XGBoost and the oscillatory wave signals to be analyzed.

4. A blood pressure measurement method based on oscillation wave feature extraction and supervised fusion according to claim 2, characterized in that, The said Step 2 includes the following steps: Step 2.1: Detect errors in the collected oscillatory wave signals to distinguish available oscillatory wave signals from unavailable oscillatory wave signals; The unavailable oscillatory wave signals are oscillatory wave signals with distorted waveforms or completely disappeared, losing effective information; Step 2.2: For available oscillatory wave signals, comprehensively consider the indicators in both the time domain and the frequency domain to evaluate the quality of the signals; Step 2.2.1: Conduct time-domain analysis on the oscillatory wave signals to determine the signal quality evaluation indicators in the time domain; In the time-domain analysis, the amplitude change degree index, baseline drift degree index, and pulse rate consistency degree index are used to evaluate the quality of the oscillatory wave signals; Specifically: Perform variational mode decomposition and period segmentation on the oscillatory wave signals respectively to obtain the reconstructed signal and single-period signal representing the baseline drift situation of the signal; The amplitude change degree index is obtained by the amplitude difference of the signal waveforms in multiple cardiac cycles; the amplitude change degree index is obtained by the amplitude difference of the signal waveforms in multiple cardiac cycles; the amplitude change degree index within the i-th cycle of the oscillatory wave signal a ( i ) is shown by the following formula: ; wherein, and are respectively the maximum value and the minimum value of the waveform amplitude of the i th cycle of the oscillating wave signal after filtering; The baseline drift degree index is obtained by windowing the reconstructed signal; the baseline drift degree index is obtained by windowing the reconstructed signal; the reconstructed signal is windowed to obtain the baseline drift degree index within the i-th cycle of the oscillatory wave signal b ( i ), as shown in the following formula: ; wherein, bWave i is the reconstructed signal of the i-th cycle of the oscillatory wave signal, b is the baseline standard value of the oscillatory wave signal in the stationary state; The pulse rate consistency index is calculated by computing the consistency of the signal time lengths over multiple periods; the pulse rate consistency index is calculated by computing the consistency of the signal time lengths over multiple periods; the period length of the oscillatory wave signal is set to be l , and the pulse rate consistency index within the i-th period of the oscillatory wave signal is obtained , as shown in the following formula: ; wherein, is the standard deviation of the length of the i -th calculated signal cycle, ) is the mean value of the length of the i-th calculated signal cycle; Step 2.2.2: Conduct frequency-domain analysis on the oscillatory wave signals to determine the signal quality evaluation indicators in the frequency domain; In the frequency-domain analysis, the frequency component index is used to evaluate the quality of the oscillatory wave signals; Specifically: the effective frequency band of the oscillatory wave signal is selected through fast Fourier transform to calculate the power spectral density to obtain the frequency component index; the frequency component index within the i-th period of the oscillatory wave signal f ( i ) is as follows: ; wherein, P is the power of the oscillating wave signal, f represents the frequency, and the effective frequency band of the oscillating wave signal is selected to be 0.5 Hz to 10 Hz; Step 2.2.3: Calculate the entropy value based on the signal quality evaluation indicators in the time domain and the frequency domain to obtain the weight coefficient, and perform multi-index weighted fusion of the oscillatory wave signals in the time-frequency domain to obtain the fusion index; The fusion index SQI ( i ) is shown by the following formula: ; among them , , , are respectively the entropy weight values of the evaluation indicators , , ; Step 2.2.4: Perform threshold discrimination on the fusion index by setting a threshold, and classify the available oscillatory wave signals into signals with good quality and signals with poor quality; If the fusion index is greater than the set threshold, the available oscillatory wave signal is a signal with good quality, otherwise it is a signal with poor quality; for signals with good quality, execute Step 3, and further analyze in the time domain, frequency domain, and time-frequency domain after filtering, denoising, and baseline removal; for signals with poor quality, the oscillatory wave needs to be collected again.

5. A blood pressure measurement method based on oscillometric wave feature extraction and supervised fusion according to claim 2, characterized in that, The said Step 3 includes the following steps: Step 3.1: Remove noise from the oscillatory wave signals using a digital filter; specifically including low-pass filtering and median filtering; Step 3.2: Use moving window averaging and a high-pass filter to filter, smooth the signal, and remove baseline drift and low-frequency noise; Step 3.3: Use the decimation method for downsampling, and selectively retain key feature samples through interpolation and multi-stage decimation; Step 3.4: Construct a supervised feature selection unit, align the preprocessed oscillatory wave signal with the blood pressure measurement values in the oscillatory wave database, and label them as supervised samples.

6. The blood pressure measurement method based on oscillation wave feature extraction and supervised fusion according to claim 2, characterized in that The said Step 4 includes the following steps: Step 4.1: Extract the time-domain features of the oscillatory wave signal based on the Hamilton algorithm; Step 4.2: Extract the frequency-domain features of the oscillatory wave signal based on the continuous wavelet transform (CWT); By performing the continuous wavelet transform (CWT) on the signal, the signal is decomposed into wavelet basis functions of different scales; the mathematical formula of CWT is as follows: ; among them, t is a time variable, x(t) is the input oscillatory wave signal; ψ ( t ) is the mother wavelet function; a is the scale parameter; b is the translation parameter; CWTx ( a , b ) is the wavelet coefficient; Step 4.3: Extract the time-frequency features of the oscillatory wave signal based on the Hilbert-Huang transform (HHT); The said Hilbert-Huang transform (HHT) includes empirical mode decomposition (EMD) and Hilbert transform. The original blood pressure oscillatory wave signal is decomposed into several intrinsic mode functions (IMFs) using empirical mode decomposition (EMD); each IMF represents different frequency components in the signal; Perform the Hilbert transform on each IMF to calculate its instantaneous frequency and instantaneous amplitude. The formula of the Hilbert transform is: ; wherein: represents the Cauchy principal value, is the original oscillatory wave signal, is the integration variable; Step 4.4: Supervised feature selection based on the oscillatory wave database; For the preprocessed oscillatory wave signal, calculate the mutual information and F-value between the time-frequency features calculated using the blood pressure measurement values in the oscillatory wave database and SBP and DBP, and screen the top 20 features strongly correlated with blood pressure: , ; wherein, represents mutual information, is and the joint probability when and are the marginal probabilities of X and Y respectively, , are the means of feature j in the high and low blood pressure groups, , are the variances.

7. A blood pressure measurement method based on oscillation wave feature extraction and supervised fusion according to claim 2, characterized in that, The said Step 5 includes the following steps: Step 5.1: Standardize the oscillatory wave signal feature data; Perform the following standardization processing on the time-domain, frequency-domain, and time-frequency features of the oscillatory wave signal: ; where x is the original eigenvalue of the oscillatory wave signal, μ is the average value of the feature, σ is the standard deviation of the feature, and x1 is the normalized eigenvalue; Step 5.2: Calculate the covariance matrix for the standardized oscillatory wave signal feature data; perform eigenvalue decomposition on the covariance matrix to obtain a series of eigenvalues and corresponding eigenvectors; use the Fisher score to rank the features and retain the top 80% of the features; Step 5.2.1: Calculate the covariance matrix: Calculate the covariance matrix for the standardized oscillatory wave signal feature data; Step 5.2.2: Eigenvalue decomposition: Perform eigenvalue decomposition on the covariance matrix to obtain a series of eigenvalues and corresponding eigenvectors; Step 5.2.3: Feature selection: Use the Fisher score to rank the features and retain the top 80% of the features; For each feature j , the Fisher score F j is defined as: ; where, K is the total number of categories; n k is the number of samples in category k ; μ k,j is the mean of the k th j feature in category μ j is the overall mean of the j th σ k,j 2 is the variance of the k th j feature in category Step 5.2.4: Data projection: Project the original oscillatory wave data onto the selected principal components to obtain the data after dimensionality reduction of the oscillatory wave signal features; Step 5.3: Construct, train a supervised variational autoencoder, and obtain fused features; Step 5.3.1: Construct a supervised variational autoencoder and train the supervised variational autoencoder using the blood pressure measurement values in the oscillatory wave database as the supervised signal; the variational autoencoder consists of an encoder and a decoder; the encoder is responsible for compressing the input features into a low-dimensional space, and the decoder is responsible for reconstructing the original features from the low-dimensional space; Step 5.3.2: Training of the supervised variational autoencoder: Use the dimensionality-reduced feature data as the input of the supervised variational autoencoder, use the mean square error as the loss function, and train the supervised variational autoencoder to minimize the difference between the input features and the reconstructed features; Step 5.3.3: Feature Fusion Representation: After the training is completed, the output of the decoder, which is the feature representation in the low-dimensional space, is used as the fused feature; The joint loss function of the supervised variational autoencoder is as follows: ; where is the mean square error of feature reconstruction, is the mean square error of blood pressure prediction for blood pressure values in the database, and β is a hyperparameter used to balance the mean square error of feature reconstruction and the mean square error of blood pressure prediction for blood pressure values in the database in the relative importance in the joint loss function L; Step 5.4: Feature Weight Calculation Based on XGBoost; Step 5.4.1: Use the preprocessed and feature-extracted oscillatory wave data as the input and the corresponding blood pressure measurement values as the output to construct a training set; Step 5.4.2: Use the training set to train the XGBoost model to obtain a pre-trained model for non-invasive blood pressure prediction; During the training process, the mean squared error (MSE) is used as the loss function, and the formula is as follows: ; where n is the number of samples, is the true blood pressure value, is the blood pressure value predicted by the XGBoost model; After the training is completed, the weights of each feature are obtained to get the weighted feature vector; The XGBoost model outputs the importance scores of each feature. After normalizing these scores, the weights of each feature are obtained. .

8. A blood pressure measurement method based on oscillometric wave feature extraction and supervised fusion according to claim 2, characterized in that, Step 6 includes the following steps: Step 6.1: Feature Importance Evaluation Based on Fisher Score; Step 6.1.1: Feature Removal: Remove each fused feature in the input data in turn, and use the remaining features to reconstruct through the trained supervised variational autoencoder; Step 6.1.2: Reconstruction Error Calculation: For the data after each feature removal, calculate the reconstruction error and compare it with the reconstruction error of the model when no fused feature is removed; Step 6.1.3: Importance Evaluation: The importance of the feature is reflected by the increase in the reconstruction error after the feature is removed; Step 6.2: Feature Weighted Fusion According to the Feature Importance Evaluation Results to Obtain a Weighted Feature Vector; Step 6.2.1: Feature Weight Assignment; The weight of a feature is proportional to its importance. Divide the importance score of each feature by the sum of the importance scores of all features to obtain the weight of this feature; Step 6.2.2: Feature Weighted Fusion Based on Weight Assignment; Apply the calculated weight to each feature vector to get the weighted feature vector; Step 6.3: Non-invasive Blood Pressure Measurement Based on Multi-index Feature Weighted Fusion; Use the weighted feature vector obtained in Step 6.2 above and the feature weights of XGBoost obtained in Step 5.4 as the feature template of the current detected object; Adopt the weighted cosine similarity algorithm and the Mahalanobis distance algorithm to perform SBP, DBP, and MAP recognition and final output in the pre-stored oscillatory wave feature database; Step 6.3.1: Calculate the weighted cosine similarity and weighted Mahalanobis distance between the sample to be recognized and the SBP waveform, DBP waveform, and MAP waveform samples; Specifically: Use the weighted feature vector and combine the XGBoost weights to calculate the weighted cosine similarity and weighted Mahalanobis distance between the sample to be recognized and each known SBP waveform, DBP waveform, and MAP waveform sample; Step 6.3.2: Based on the weighted cosine similarity and weighted Mahalanobis distance between the sample to be recognized and each known SBP waveform, DBP waveform, and MAP waveform sample, assign corresponding weight values and calculate the weighted similarity; Among them, for the sample A to be recognized and the known waveform sample B, the weighted cosine similarity is as described in the following formula: ; where is the weighted cosine similarity between sample A and sample B; is the feature weight output by the XGBoost model trained based on existing data; m is the number of feature dimensions; represents the weighted dot product; is the norm of the weighted vector A, is the norm of the weighted vector B; The weighted Mahalanobis distance is as described in the following formula: ; wherein the weighted Mahalanobis distance between sample A and sample B; W is the weight matrix, which is a diagonal matrix and has the form: ； Step 6.3.3: Perform SBP, DBP, and MAP recognition based on the weighted similarity; By comparing the weighted similarities between the samples to be recognized and all known SBP waveform, DBP waveform, and MAP waveform samples, the SBP, DBP, and MAP corresponding to the sample to be recognized with the lowest weighted similarity are selected as the final results.

Images