Noninvasive continuous dynamic blood pressure detector data analysis method

Abstract

The invention discloses a non-invasive continuous dynamic blood pressure detector data analysis method, and relates to the technical field of cardiovascular and cerebrovascular diseases, skin bio-electricity signals, heart sound signals and micro-motion signals are collected, the signals collected in the step S1 are preprocessed and sampled and aligned, a deep neural network is utilized to extract time sequence features, a convolutional neural network is utilized to extract spatial features, and the time sequence features are extracted; according to the method, blood pressure related features are extracted through a multi-source signal acquisition mechanism in combination with a deep neural network and a convolutional neural network, the blood pressure related features are fused into a high-precision blood pressure prediction model, and continuous dynamic blood pressure prediction can be realized without depending on a cuff and an invasive sensor; the system is suitable for home health management, chronic disease follow-up visit and remote medical treatment, and has wide clinical application and industrialization potential.

Description

Technical Field

[0001] This invention relates to the field of cardiovascular and cerebrovascular disease technology, and in particular to a data analysis method for a non-invasive continuous dynamic blood pressure monitor. Background Technology

[0002] With the continuous rise in the incidence of chronic diseases such as hypertension and cardiovascular diseases, the demand for dynamic and accurate monitoring of blood pressure is increasing. Existing blood pressure monitoring devices are mostly based on cuff inflation or single optical sensor signals, typically employing intermittent detection methods that cannot achieve continuous monitoring. Furthermore, they are significantly affected by the measurement environment, wearing posture, and individual differences, resulting in insufficient data stability and predictive accuracy. Some non-invasive blood pressure monitoring technologies have attempted to incorporate pulse wave propagation time, optical volume pulse wave, or single-modal signal modeling methods for estimation. However, due to the lack of multi-source physiological signal fusion capabilities and personalized data modeling mechanisms, they often fail to meet the long-term, stable, and personalized monitoring needs of different populations, limiting their application value in home monitoring, telemedicine, and high-risk population early warning management. Therefore, there is an urgent need for a non-invasive blood pressure data analysis method that can fuse multimodal signals, possess continuous predictive capabilities, and combine individual profiles for risk identification and dynamic output, in order to improve the real-time performance, accuracy, and intervention capabilities of monitoring.

[0003] Currently, Chinese patent application number CN201810042467.0 discloses a method and system for generating ambulatory blood pressure monitoring reports with reminder function. The method uses an ambulatory blood pressure monitoring device to measure the blood pressure of the subject and uploads the measured blood pressure data to a data analysis platform. The data analysis platform receives and stores the test data from the ambulatory blood pressure monitoring device and performs index data analysis to obtain the values ​​of various blood pressure indicators. After comparing the measured blood pressure values ​​with normal reference values, the method automatically generates an ambulatory blood pressure monitoring report and provides a prompt for abnormal blood pressure indicators in the ambulatory blood pressure monitoring report.

[0004] The relevant technologies are insufficient to achieve non-invasive, continuous, and highly accurate dynamic monitoring and personalized health analysis of blood pressure. Summary of the Invention

[0005] The technical problem solved by this invention is that existing technologies are unable to achieve non-invasive, continuous, and highly accurate dynamic monitoring and personalized health analysis of blood pressure.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] A method for analyzing data from a non-invasive continuous ambulatory blood pressure monitor includes the following steps:

[0008] Step S1: Collect skin bioelectrical signals, heart sound signals, and micro-motion signals;

[0009] Step S2: Preprocess and align the signals acquired in step S1.

[0010] Step S3: Use a deep neural network to extract temporal features and a convolutional neural network to extract spatial features;

[0011] Step S4: Fuse features and perform regression prediction to generate blood pressure prediction results;

[0012] Step S5: Determine blood pressure status and mark it as normal or abnormal;

[0013] Step S6: Output blood pressure status information.

[0014] Preferably, step S1 includes the following sub-steps:

[0015] Step S101: Collect skin bioelectric signals from an electrophysiological sensor placed on the skin surface, wherein the skin bioelectric signals include potential change signals;

[0016] Step S102: Collect heart sound signals from an acoustic sensor deployed on the chest of the human body. The heart sound signals include a first heart sound and a second heart sound.

[0017] Step S103: Acquire micro-motion signals from an accelerometer located on the surface of a limb or torso. The micro-motion signals include periodic vibration components and transient displacement components.

[0018] Preferably, step S2 includes the following sub-steps:

[0019] Step S201: Preprocess the skin bioelectric signal. The preprocessing includes noise filtering based on a fixed frequency band filter. The noise includes fundamental frequency interference noise from the power supply system and power frequency coupling noise from the epidermal contact surface.

[0020] Step S202: The heart sound signal is subjected to amplitude normalization processing, wherein the amplitude normalization processing is based on the maximum absolute amplitude within the sampling interval and the amplitude is scaled.

[0021] Step S203: The skin bioelectric signal, heart sound signal and micro-motion signal are sampled and aligned respectively. The sampling and alignment process includes resampling all signals according to a uniform time step. The resampled signals are represented as an equally spaced time sequence.

[0022] Preferably, step S3 includes the following sub-steps:

[0023] Step S301: Use a deep neural network to learn and recognize the temporal features of the heart sound signal and the micro-motion signal:

[0024] After amplitude normalization and sampling alignment, the heart sound signal and micro-motion signal are passed as input data to the deep neural network. The deep neural network includes multiple fully connected layers, and the output of each fully connected layer is passed to the previous layer. The network is trained by layer-by-layer calculation and backpropagation algorithm to learn the temporal characteristics of the heart sound signal and micro-motion signal.

[0025] The logic for learning the temporal characteristics of heart sound signals and micro-motion signals is as follows:

[0026] Extract the periodic fluctuations, abrupt changes, and trend changes of heart sound signals and micro-motion signals over time;

[0027] Identify features associated with blood pressure changes based on signal patterns in historical data;

[0028] The deep neural network is trained using the backpropagation algorithm, and the network weights are gradually adjusted until the error between the output data of the deep neural network and the actual blood pressure data reaches the minimum value.

[0029] Step S302: Use a convolutional neural network to extract spatial features from the signal:

[0030] The heart sound signal and micro-motion signal, after amplitude normalization and sampling alignment processing, are passed as input data to a convolutional neural network. The convolutional neural network extracts the spatial features of the heart sound signal and micro-motion signal through convolution and pooling operations, and learns important local patterns in the signal through convolution kernels. The specific process includes:

[0031] Convolutional neural networks use multiple convolutional kernels to convolve heart sound signals and micro-motion signals to extract local features from them. Each convolutional kernel slides on the input signal and identifies frequency patterns, abrupt changes, and waveform features in the heart sound signals and micro-motion signals through convolution operations, outputting a feature map.

[0032] Pooling layers downsample the feature maps after convolution. After convolution and pooling operations, the convolutional neural network maps the extracted spatial features to a high-dimensional space through fully connected layers.

[0033] Preferably, step S4 includes the following sub-steps:

[0034] Step S401: The skin bioelectric signal, heart sound signal and micro-motion signal processed in step S3 are input into a deep neural network model. The deep neural network model includes a multi-layer fully connected structure. The fully connected structure is trained based on the backpropagation mechanism and outputs a blood pressure time-series feature vector.

[0035] Step S402: The skin bioelectric signal, heart sound signal and micro-motion signal processed in step S3 are input into the convolutional neural network model. The convolutional neural network model includes at least one set of convolutional layers, activation layers and pooling layers. The convolution operation is used to extract local spatial structure features in the signal, and the pooling operation is used to downsample local region features. The output is a blood pressure spatial feature vector.

[0036] Step S403: The blood pressure time-series feature vector and the blood pressure spatial feature vector are concatenated and fused to form a fused feature vector. The fused feature vector is input into the fully connected prediction layer for regression calculation, and the output is the blood pressure prediction result, which includes the predicted systolic blood pressure value and the predicted diastolic blood pressure value.

[0037] Step S404: Perform threshold comparison processing on the blood pressure prediction results;

[0038] If any blood pressure prediction value is higher than the first upper limit threshold or lower than the first lower limit threshold, it is marked as an abnormal blood pressure state.

[0039] If the predicted blood pressure value is between the first upper limit threshold and the first lower limit threshold, it is marked as normal blood pressure.

[0040] Step S405: Output the abnormal blood pressure status and the normal blood pressure status as blood pressure status information.

[0041] Preferably, the processing logic for the fused feature vector includes:

[0042] The blood pressure time-series feature vector and the blood pressure spatial feature vector are concatenated. The concatenation process involves connecting the two feature vectors along the feature dimension, and the output is a fused feature vector.

[0043] The fused feature vector is input into a multilayer perceptron network structure, which includes at least two fully connected layers and one output regression layer. The fully connected layers are used for nonlinear mapping and feature combination. The output regression layer is used for continuous numerical prediction of the fused features, and the output is the blood pressure prediction result.

[0044] The blood pressure prediction results include predicted systolic blood pressure and predicted diastolic blood pressure.

[0045] Preferably, step S5 includes the following sub-steps:

[0046] Step S501: Obtain the user's historical health data and personalized parameters. The historical health data includes past blood pressure measurement records, heart rate records, body mass index, exercise habits and sleep duration. The personalized parameters include the user's age, gender, past disease information and medication status. Construct a health status feature set based on the historical health data and personalized parameters. The health status feature set is used to generate a user health profile model.

[0047] Step S502: Perform correlation analysis between the blood pressure prediction results and the health profile model. The correlation analysis includes comparing the current systolic blood pressure prediction value with the user's historical systolic blood pressure distribution range, classifying the diastolic blood pressure prediction value with the target threshold range, and generating personalized health advice information by combining historical fluctuation trends and the current abnormal labeling status. The personalized health advice information includes exercise advice level, dietary control points, and recommended rest duration.

[0048] Step S503: Generate a blood pressure change trend curve based on the blood pressure prediction result sequence within a continuous time window, calculate the slope change and rate of change between adjacent prediction points, determine whether there is a rapid upward or downward trend, and generate blood pressure warning information if any continuous prediction point meets the set rate of change threshold. The blood pressure warning information includes warning level, associated risk indicators and suggested handling methods.

[0049] Preferably, step S6 includes the following sub-steps:

[0050] Step S601: Generate a health management report based on the predicted systolic blood pressure and predicted diastolic blood pressure.

[0051] Step S602: Transmit the abnormal marker records and blood pressure change trend data from the health management report to the client application. The client application includes a mobile terminal APP and a wearable device interface. The interface provides a blood pressure historical trend display module, an abnormal warning module, and a health advice module. The modules are used to present the blood pressure fluctuation status and response measures suggestions to the user in real time.

[0052] Step S603: When the number of abnormal blood pressure markers reaches a preset frequency threshold, a remote monitoring mechanism is triggered. The remote monitoring mechanism includes automatically pushing blood pressure prediction results and health management reports to a preset health service platform through a network interface. The health service platform is used to store, analyze, and forward abnormal data to the remote doctor's account.

[0053] Preferably, the deep neural network is used to process the temporal features of the heart sound signal. The temporal features include the periodic rhythm features of the heart sound signal, the time interval features between the first heart sound and the second heart sound, and the transient peak change features. The deep neural network model extracts the temporal features through multiple sets of one-dimensional convolutional layers and then connects them to a max pooling layer for feature compression. The output is a temporal feature vector of the heart sound signal after flattening.

[0054] The convolutional neural network is used to process the spatial features in the micro-motion signal. The spatial features include the local amplitude distribution, envelope morphology, and position-related response patterns of the micro-motion signal in different sampling segments. The convolutional neural network performs nonlinear mapping and combination of the micro-motion signal features through a multi-layer fully connected structure, and outputs a spatial feature vector of the micro-motion signal.

[0055] Preferably, the health management report includes:

[0056] A blood pressure change trend graph, which is generated based on blood pressure prediction results within a continuous time window, shows the rising, falling and stable sections of blood pressure values;

[0057] The health risk assessment results are calculated based on the degree of deviation of systolic and diastolic blood pressure from the user's individual baseline range, the duration of abnormality, and the predicted rate of change.

[0058] The improvement plan is generated based on the current blood pressure status and health profile, including dietary adjustment suggestions, exercise frequency suggestions, and sleep duration suggestions;

[0059] User behavior adjustment suggestions are provided, which are based on historical behavior patterns and blood pressure response trends, and output as text data of daily intervention measures.

[0060] The beneficial effects of this invention are as follows: This invention uses a multi-source signal acquisition mechanism, combined with deep neural networks and convolutional neural networks to extract blood pressure-related features, and integrates them into a high-precision blood pressure prediction model. It does not rely on cuffs or invasive sensors, and can achieve continuous dynamic blood pressure prediction, output health management reports and early warning information. It is suitable for home health management, chronic disease follow-up and telemedicine, and has broad clinical application and industrialization potential. Attached Figure Description

[0061] Figure 1 The flowchart illustrates the steps of a non-invasive continuous dynamic blood pressure monitoring data analysis method according to an embodiment of the present invention. Detailed Implementation

[0062] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0063] Example, refer to Figure 1 A method for analyzing data from a non-invasive continuous ambulatory blood pressure monitor is provided, comprising the following steps:

[0064] Step S1: Collect skin bioelectric signals, heart sound signals, and micro-motion signals.

[0065] Step S2 involves preprocessing and aligning the signals acquired in step S1.

[0066] Step S3: Use a deep neural network to extract temporal features and a convolutional neural network to extract spatial features.

[0067] Step S4: Fuse features and perform regression prediction to generate blood pressure prediction results.

[0068] Step S5: Determine the blood pressure status and mark it as normal or abnormal.

[0069] Step S6: Output blood pressure status information.

[0070] This invention utilizes a multi-source signal acquisition mechanism, combining deep neural networks and convolutional neural networks to extract blood pressure-related features, and integrates them into a high-precision blood pressure prediction model. It does not rely on cuffs or invasive sensors and can achieve continuous dynamic blood pressure prediction. Furthermore, it introduces a personalized health profile model and trend monitoring mechanism to conduct risk assessment based on historical data and individual characteristics, and outputs health management reports and early warning information. It is suitable for home health management, chronic disease follow-up, and telemedicine, and has broad clinical application and industrialization potential.

[0071] Step S1 includes the following sub-steps:

[0072] Step S101: Collect skin bioelectric signals from an electrophysiological sensor placed on the skin surface. The skin bioelectric signals include potential change signals, which are used to characterize the electrophysiological activity of the vascular region.

[0073] Step S101 uses skin bioelectric signals collected by skin electrophysiology sensors to monitor the electrophysiological activity of blood vessels in real time. In particular, during the process of vasoconstriction and vasodilation, the potential change signal provides key evidence for the dynamic changes in blood pressure.

[0074] Step S102: Acquire heart sound signals from acoustic sensors deployed on the chest of the human body. The heart sound signals include a first heart sound and a second heart sound, which are used to reflect the working state of the heart during systole and diastole.

[0075] Step S102, by collecting heart sound signals, especially the first and second heart sounds, can reflect the working state of the heart's contraction and relaxation, providing direct cardiac function information for blood pressure prediction and enhancing the accuracy and reliability of blood pressure prediction.

[0076] Step S103: Acquire micro-motion signals from an accelerometer located on the surface of a limb or torso. The micro-motion signals include periodic vibration components and transient displacement components, which are used to reflect the mechanical response characteristics of the skin surface caused by blood flow.

[0077] Step S103 uses the micro-motion signal collected by the accelerometer to capture the tiny mechanical response of blood flow to the skin surface, reflecting information such as blood flow velocity and vascular tension, which helps to predict blood pressure change trends and further improve the blood pressure prediction model.

[0078] Step S1 involves acquiring skin bioelectrical signals, heart sound signals, and micro-motion signals, combined with multimodal signal sources, to comprehensively obtain physiological characteristic data related to blood pressure. These signals can accurately reflect the state of blood vessels, heart, and blood flow, providing high-quality input data for subsequent blood pressure prediction.

[0079] Step S2 includes the following sub-steps:

[0080] Step S201: Preprocess the skin bioelectric signal. The preprocessing includes noise filtering based on a fixed frequency band filter. The noise includes fundamental frequency interference noise from the power supply system and power frequency coupling noise from the epidermal contact surface.

[0081] Step S201 filters out noise from the skin's bioelectrical signals, eliminating fundamental frequency interference noise from the power system and power frequency coupling noise from the epidermal contact surface, ensuring signal purity, enhancing signal stability and reliability, and thus improving the accuracy of subsequent analysis results.

[0082] Step S202: The heart sound signal is subjected to amplitude normalization processing, which is amplitude scaling based on the maximum absolute amplitude within the sampling interval.

[0083] Step S202 performs amplitude normalization on the heart sound signal, eliminating inconsistencies in signal intensity across different sampling intervals and ensuring data consistency under different signal acquisition conditions. Amplitude scaling based on the maximum absolute amplitude helps standardize the signal, making subsequent analysis more accurate.

[0084] Step S203: The skin bioelectric signal, heart sound signal and micro-motion signal are sampled and aligned respectively. The sampling and alignment process includes resampling all signals according to a uniform time step. The resampled signals are represented as an equally spaced time sequence.

[0085] Step S203 resamples all signals at a uniform time step, ensuring that the time series of skin bioelectrical signals, heart sound signals, and micro-motion signals are consistent. This operation ensures the synchronization of each signal in the time dimension, providing accurate data input for subsequent feature extraction and model training.

[0086] Deep neural networks are used to process the temporal features of heart sound signals. These temporal features include the periodic rhythm features of the heart sound signals, the time interval features between the first and second heart sounds, and the transient peak variation features. The deep neural network model extracts the temporal features through multiple sets of one-dimensional convolutional layers, then connects them to a max pooling layer for feature compression, and outputs the temporal feature vector of the heart sound signal through flattening.

[0087] Convolutional neural networks are used to process the spatial features of micro-motion signals. These spatial features include the local amplitude distribution, envelope morphology, and position-related response patterns of the micro-motion signals in different sampling segments. The convolutional neural network performs nonlinear mapping and combination of the micro-motion signal features through a multi-layer fully connected structure, and outputs a spatial feature vector of the micro-motion signal.

[0088] Step S2 preprocesses and aligns the multimodal signals to eliminate noise interference and normalize the signals, making subsequent feature extraction more accurate and reliable. This step effectively optimizes signal quality and ensures data consistency and accuracy in subsequent analysis processes.

[0089] Step S3 includes the following sub-steps:

[0090] Step S301: Use a deep neural network to learn and recognize the temporal features of the heart sound signal and the micro-motion signal:

[0091] After amplitude normalization and sampling alignment, the heart sound signal and micro-motion signal are passed as input data to the deep neural network. The deep neural network consists of multiple fully connected layers, and the output of each fully connected layer is passed to the previous layer. The network is trained by layer-by-layer calculation and backpropagation algorithm to learn the temporal characteristics of the heart sound signal and micro-motion signal.

[0092] The logic for learning the temporal characteristics of heart sound signals and micromotion signals is as follows:

[0093] Extract the periodic fluctuations, abrupt changes, and trend changes of heart sound signals and micro-motion signals over time.

[0094] Based on signal patterns in historical data, identify features associated with blood pressure changes.

[0095] The deep neural network is trained using the backpropagation algorithm, and the network weights are gradually adjusted until the error between the output data of the deep neural network and the actual blood pressure data reaches the minimum value.

[0096] Step S301 uses a deep neural network to extract the temporal features of heart sound signals and micro-motion signals, identify the periodic fluctuations, abrupt changes and trend changes of the signals in the time dimension, establish the mapping relationship between signal features and blood pressure changes, and obtain a temporal feature vector for regression prediction.

[0097] Step S302: Use a convolutional neural network to extract spatial features from the signal:

[0098] After amplitude normalization and sampling alignment, the heart sound signal and micro-motion signal are fed as input data into a convolutional neural network. The convolutional neural network extracts the spatial features of the heart sound signal and micro-motion signal through convolution and pooling operations, and learns important local patterns in the signal through convolution kernels. The specific process includes:

[0099] Convolutional neural networks use multiple convolutional kernels to convolve heart sound signals and micro-motion signals to extract local features from them. Each convolutional kernel slides on the input signal and identifies frequency patterns, abrupt changes, and waveform features in the heart sound signals and micro-motion signals through convolution operations, outputting a feature map.

[0100] Pooling layers downsample the feature maps after convolution. After convolution and pooling operations, the convolutional neural network maps the extracted spatial features to a high-dimensional space through fully connected layers.

[0101] Step S302 extracts the spatial features of heart sound signals and micro-motion signals through a convolutional neural network, identifies local amplitude distribution, waveform structure and frequency patterns, retains key features on the basis of dimensionality reduction and compression, and outputs spatial feature representation to enhance the structural representation capability of blood pressure modeling.

[0102] Step S3 extracts temporal and spatial features from heart sound signals and micro-motion signals using deep neural networks and convolutional neural networks, respectively, to model and represent the deep structure of multimodal signals. This provides high-dimensional and highly correlated input features for subsequent blood pressure prediction models, improving prediction accuracy and model generalization ability.

[0103] Step S4 includes the following sub-steps:

[0104] Step S401: The skin bioelectric signal, heart sound signal and micro-motion signal processed in step S3 are input into the deep neural network model. The deep neural network model includes a multi-layer fully connected structure. The fully connected structure is trained based on the backpropagation mechanism and the output is a blood pressure time-series feature vector.

[0105] Step S401 inputs the processed multi-source signal into a deep neural network model and outputs a blood pressure time-series feature vector that reflects the dynamic change trend of the signal, providing a core basis for capturing the time pattern of blood pressure changes.

[0106] Step S402: The skin bioelectric signal, heart sound signal and micro-motion signal processed in step S3 are input into the convolutional neural network model. The convolutional neural network model includes at least one set of convolutional layers, activation layers and pooling layers. The convolution operation is used to extract local spatial structure features in the signal, and the pooling operation is used to downsample local region features. The output is a blood pressure spatial feature vector.

[0107] Step S402 transmits the same input signal to the convolutional neural network model, extracts local structural and morphological features, and outputs a blood pressure spatial feature vector to enhance the ability to recognize static morphological changes in the signal.

[0108] Step S403: The time-series feature vector of blood pressure and the spatial feature vector of blood pressure are concatenated and fused to form a fused feature vector. The fused feature vector is input into the fully connected prediction layer for regression calculation, and the output is the blood pressure prediction result, which includes the predicted value of systolic blood pressure and the predicted value of diastolic blood pressure.

[0109] Step S403 concatenates the time-series feature vector and spatial feature vector of blood pressure, inputs them into the regression prediction structure, and outputs the predicted values ​​of systolic and diastolic blood pressure to achieve quantitative prediction of blood pressure levels, which serves as the basic data for the final analysis results.

[0110] Step S404: Perform threshold comparison processing on the blood pressure prediction results.

[0111] If any predicted blood pressure value is higher than the first upper limit threshold or lower than the first lower limit threshold, it is marked as an abnormal blood pressure state.

[0112] If the predicted blood pressure value is between the first upper limit threshold and the first lower limit threshold, it is marked as normal blood pressure.

[0113] Step S404 compares the upper and lower limits of the predicted blood pressure value to automatically determine whether the blood pressure is in an abnormal state, ensuring that the system has the ability to identify risks.

[0114] Step S405: Output the abnormal blood pressure status and the normal blood pressure status as blood pressure status information.

[0115] The processing logic for fusing feature vectors includes:

[0116] The blood pressure time-series feature vector and the blood pressure spatial feature vector are concatenated. The concatenation process involves connecting the two feature vectors along the feature dimension, and the output is a fused feature vector.

[0117] The fused feature vector is input into a multilayer perceptron network structure, which includes at least two fully connected layers and one output regression layer. The fully connected layers are used for nonlinear mapping and feature combination, and the output regression layer is used for continuous numerical prediction of the fused features. The output is the blood pressure prediction result.

[0118] Blood pressure prediction results include predicted systolic blood pressure and predicted diastolic blood pressure.

[0119] Step S405 outputs the threshold comparison result as blood pressure status information, providing clear and structured input labels for the subsequent health management module and supporting the establishment of a closed-loop processing mechanism for the system.

[0120] Step S4 establishes a complete multimodal blood pressure prediction model by fusing the temporal and spatial features of blood pressure extracted by the neural network, enabling continuous and refined prediction of systolic and diastolic blood pressure, and judging blood pressure status by threshold comparison, effectively improving the accuracy and discrimination ability of blood pressure analysis.

[0121] Step S5 includes the following sub-steps:

[0122] Step S501: Obtain the user's historical health data and personalized parameters. The historical health data includes past blood pressure measurement records, heart rate records, body mass index, exercise habits and sleep duration. The personalized parameters include the user's age, gender, past disease information and medication status. Based on the historical health data and personalized parameters, construct a health status feature set. The health status feature set is used to generate a user health profile model.

[0123] Step S501 constructs a set of health status features based on the user's historical health data and personalized physiological characteristics, and generates a health profile model that conforms to individual characteristics, providing a personalized basis for subsequent analysis and recommendations.

[0124] Step S502: Perform correlation analysis between the blood pressure prediction results and the health profile model. The correlation analysis includes comparing the current systolic blood pressure prediction value with the user's historical systolic blood pressure distribution range, classifying the diastolic blood pressure prediction value with the target threshold range, and generating personalized health advice information by combining historical fluctuation trends and the current abnormal labeling status. The personalized health advice information includes exercise advice level, dietary control points, and recommended rest duration.

[0125] Step S502 matches and compares the blood pressure prediction results with the health profile model. Through the correlation analysis between individual historical characteristics and current measurement results, personalized health advice information is generated to enhance the operability and user guidance value of the analysis results.

[0126] Step S503: Generate a blood pressure change trend curve based on the blood pressure prediction result sequence within a continuous time window, calculate the slope change and rate of change between adjacent prediction points, determine whether there is a rapid upward or downward trend, and generate blood pressure warning information if any continuous prediction point meets the set rate of change threshold. The blood pressure warning information includes the warning level, associated risk indicators, and suggested handling methods.

[0127] Step S503 calculates the blood pressure change trend based on the blood pressure prediction results over a continuous time period, identifies abrupt fluctuations, generates structured blood pressure early warning information, and supports the deployment of early intervention and risk control strategies.

[0128] Step S5 introduces the user's historical health data and personalized parameters, combines them with blood pressure prediction results to conduct correlation analysis and trend modeling, and outputs personalized health advice and blood pressure warning information, thereby realizing intelligent interpretation of blood pressure prediction results and extension of personalized health management.

[0129] Step S6 includes the following sub-steps:

[0130] Step S601: Generate a health management report based on the predicted systolic blood pressure and the predicted diastolic blood pressure.

[0131] Step S601 generates a health management report containing predicted systolic and diastolic blood pressure values, providing users with a visual representation of blood pressure trends, supporting a comprehensive understanding of their blood pressure status, and enhancing their initiative in health management.

[0132] Step S602: Transmit the abnormal marker records and blood pressure change trend data from the health management report to the client application. The client application includes a mobile terminal APP and a wearable device interface. The interface provides a blood pressure historical trend display module, an abnormal warning module, and a health advice module. The modules are used to present the blood pressure fluctuation status and response measures suggestions to the user in real time.

[0133] Step S602 transmits the abnormal marker records and blood pressure change trend data from the health management report to the client application in real time, ensuring that users can obtain blood pressure fluctuations, abnormal warnings and health advice through mobile terminal APP or wearable devices at any time, thereby improving users' health awareness and responsiveness.

[0134] Step S603: When the number of abnormal blood pressure markers reaches a preset frequency threshold, a remote monitoring mechanism is triggered. The remote monitoring mechanism includes automatically pushing blood pressure prediction results and health management reports to a preset health service platform via a network interface. The health service platform is used to store, analyze, and forward abnormal data to the remote doctor's account.

[0135] The health management report includes:

[0136] The blood pressure trend chart is generated based on blood pressure prediction results within a continuous time window, showing the rising, falling, and stable sections of blood pressure values.

[0137] The health risk assessment results are calculated based on the degree of deviation of systolic and diastolic blood pressure from the individual user's baseline range, the duration of abnormality, and the predicted rate of change.

[0138] Improvement plan suggestions are generated based on the current blood pressure status and health profile, including dietary adjustment suggestions, exercise frequency suggestions, and sleep duration suggestions.

[0139] User behavior adjustment suggestions are based on historical behavior patterns and blood pressure response trends, and the output is text data of daily intervention measures.

[0140] Step S603 When the number of abnormal blood pressure markers reaches the set threshold, the remote monitoring mechanism is triggered, and the blood pressure prediction results and health reports are automatically pushed to the health service platform through the network interface. This supports doctors or telemedicine teams to analyze and intervene in abnormal data in real time, and provides timely medical responses and treatment suggestions.

[0141] Step S6 generates a health management report and transmits it to the client application to achieve real-time feedback and visualization of blood pressure status, supports remote monitoring and health early warning functions, helps users, medical personnel and health service platforms to track and respond to blood pressure abnormalities in real time, and provides personalized health intervention plans.

[0142] This invention provides a non-invasive continuous dynamic blood pressure monitoring data analysis method. By constructing a multi-source signal acquisition mechanism based on skin bioelectric signals, heart sound signals and micro-motion signals, and combining deep neural networks and convolutional neural networks, it achieves comprehensive extraction of blood pressure-related temporal and spatial features, and integrates them into a high-precision blood pressure prediction model. This enables continuous dynamic prediction of blood pressure status without relying on cuffs and invasive sensors.

[0143] This method further introduces a personalized health profile model and trend monitoring mechanism, which can perform risk modeling based on historical blood pressure fluctuations, individual characteristics and behavioral habits, and output health management reports, early warning information and intervention suggestions. It improves the intelligence, precision and pre-intervention capabilities of blood pressure management, and is suitable for home health management, chronic disease follow-up, telemedicine and continuous status assessment of high-risk groups. It has good clinical application and industrial promotion value.

[0144] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0145] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for analyzing data from a non-invasive continuous ambulatory blood pressure monitor, characterized in that, Includes the following steps: Step S1: Collect skin bioelectrical signals, heart sound signals, and micro-motion signals; Step S2: Preprocess and align the signals acquired in step S1. Step S3: Use a deep neural network to extract temporal features and a convolutional neural network to extract spatial features; Step S4: Fuse features and perform regression prediction to generate blood pressure prediction results; Step S5: Determine blood pressure status and mark it as normal or abnormal; Step S6: Output blood pressure status information.

2. The data analysis method for a non-invasive continuous ambulatory blood pressure monitor as described in claim 1, characterized in that, Step S1 includes the following sub-steps: Step S101: Collect skin bioelectric signals from an electrophysiological sensor placed on the skin surface. The skin bioelectric signals include potential change signals, which are used to characterize the electrophysiological activity of the vascular region. Step S102: Collect heart sound signals from an acoustic sensor deployed on the chest of the human body. The heart sound signals include a first heart sound and a second heart sound. The first heart sound and the second heart sound are used to reflect the working state of the heart during systole and diastole. Step S103: Acquire micro-motion signals from an accelerometer located on the surface of a limb or torso. The micro-motion signals include periodic vibration components and transient displacement components, which are used to reflect the mechanical response characteristics of the skin surface caused by blood flow.

3. The data analysis method for a non-invasive continuous ambulatory blood pressure monitor as described in claim 2, characterized in that, Step S2 includes the following sub-steps: Step S201: Preprocess the skin bioelectric signal. The preprocessing includes noise filtering based on a fixed frequency band filter. The noise includes fundamental frequency interference noise from the power supply system and power frequency coupling noise from the epidermal contact surface. Step S202: The heart sound signal is subjected to amplitude normalization processing, wherein the amplitude normalization processing is based on the maximum absolute amplitude within the sampling interval and the amplitude is scaled. Step S203: The skin bioelectric signal, heart sound signal and micro-motion signal are sampled and aligned respectively. The sampling and alignment process includes resampling all signals according to a uniform time step. The resampled signals are represented as an equally spaced time sequence.

4. The data analysis method for a non-invasive continuous ambulatory blood pressure monitor as described in claim 3, characterized in that, Step S3 includes the following sub-steps: Step S301: Use a deep neural network to learn and recognize the temporal features of the heart sound signal and the micro-motion signal: After amplitude normalization and sampling alignment, the heart sound signal and micro-motion signal are passed as input data to the deep neural network. The deep neural network includes multiple fully connected layers, and the output of each fully connected layer is passed to the previous layer. The network is trained by layer-by-layer calculation and backpropagation algorithm to learn the temporal characteristics of the heart sound signal and micro-motion signal. The logic for learning the temporal characteristics of heart sound signals and micro-motion signals is as follows: Extract the periodic fluctuations, abrupt changes, and trend changes of heart sound signals and micro-motion signals over time; Identify features associated with blood pressure changes based on signal patterns in historical data; The deep neural network is trained using the backpropagation algorithm, and the network weights are gradually adjusted until the error between the output data of the deep neural network and the actual blood pressure data reaches the minimum value. Step S302: Use a convolutional neural network to extract spatial features from the signal: The heart sound signal and micro-motion signal, after amplitude normalization and sampling alignment processing, are passed as input data to a convolutional neural network. The convolutional neural network extracts the spatial features of the heart sound signal and micro-motion signal through convolution and pooling operations, and learns important local patterns in the signal through convolution kernels. The specific process includes: Convolutional neural networks use multiple convolutional kernels to convolve heart sound signals and micro-motion signals to extract local features from them. Each convolutional kernel slides on the input signal and identifies frequency patterns, abrupt changes, and waveform features in the heart sound signals and micro-motion signals through convolution operations, outputting a feature map. Pooling layers downsample the feature maps after convolution. After convolution and pooling operations, the convolutional neural network maps the extracted spatial features to a high-dimensional space through fully connected layers.

5. The data analysis method for a non-invasive continuous ambulatory blood pressure monitor as described in claim 4, characterized in that, Step S4 includes the following sub-steps: Step S401: The skin bioelectric signal, heart sound signal and micro-motion signal processed in step S3 are input into a deep neural network model. The deep neural network model includes a multi-layer fully connected structure. The fully connected structure is trained based on the backpropagation mechanism and outputs a blood pressure time-series feature vector. Step S402: The skin bioelectric signal, heart sound signal and micro-motion signal processed in step S3 are input into the convolutional neural network model. The convolutional neural network model includes at least one set of convolutional layers, activation layers and pooling layers. The convolution operation is used to extract local spatial structure features in the signal, and the pooling operation is used to downsample local region features. The output is a blood pressure spatial feature vector. Step S403: The blood pressure time-series feature vector and the blood pressure spatial feature vector are concatenated and fused to form a fused feature vector. The fused feature vector is input into the fully connected prediction layer for regression calculation, and the output is the blood pressure prediction result, which includes the predicted systolic blood pressure value and the predicted diastolic blood pressure value. Step S404: Perform threshold comparison processing on the blood pressure prediction results; If any blood pressure prediction value is higher than the first upper limit threshold or lower than the first lower limit threshold, it is marked as an abnormal blood pressure state. If the predicted blood pressure value is between the first upper limit threshold and the first lower limit threshold, it is marked as normal blood pressure. Step S405: Output the abnormal blood pressure status and the normal blood pressure status as blood pressure status information.

6. The data analysis method for a non-invasive continuous ambulatory blood pressure monitor as described in claim 5, characterized in that, The processing logic for the fused feature vectors includes: The blood pressure time-series feature vector and the blood pressure spatial feature vector are concatenated. The concatenation process involves connecting the two feature vectors along the feature dimension, and the output is a fused feature vector. The fused feature vector is input into a multilayer perceptron network structure, which includes at least two fully connected layers and one output regression layer. The fully connected layers are used for nonlinear mapping and feature combination. The output regression layer is used for continuous numerical prediction of the fused features, and the output is the blood pressure prediction result. The blood pressure prediction results include predicted systolic blood pressure and predicted diastolic blood pressure.

7. The data analysis method for a non-invasive continuous ambulatory blood pressure monitor as described in claim 6, characterized in that, Step S5 includes the following sub-steps: Step S501: Obtain the user's historical health data and personalized parameters. The historical health data includes past blood pressure measurement records, heart rate records, body mass index, exercise habits and sleep duration. The personalized parameters include the user's age, gender, past disease information and medication status. Construct a health status feature set based on the historical health data and personalized parameters. The health status feature set is used to generate a user health profile model. Step S502: Perform correlation analysis between the blood pressure prediction results and the health profile model. The correlation analysis includes comparing the current systolic blood pressure prediction value with the user's historical systolic blood pressure distribution range, classifying the diastolic blood pressure prediction value with the target threshold range, and generating personalized health advice information by combining historical fluctuation trends and the current abnormal labeling status. The personalized health advice information includes exercise advice level, dietary control points, and recommended rest duration. Step S503: Generate a blood pressure change trend curve based on the blood pressure prediction result sequence within a continuous time window, calculate the slope change and rate of change between adjacent prediction points, determine whether there is a rapid upward or downward trend, and generate blood pressure warning information if any continuous prediction point meets the set rate of change threshold. The blood pressure warning information includes warning level, associated risk indicators and suggested handling methods.

8. The data analysis method for a non-invasive continuous ambulatory blood pressure monitor as described in claim 7, characterized in that, Step S6 includes the following sub-steps: Step S601: Generate a health management report based on the predicted systolic blood pressure and the predicted diastolic blood pressure. Step S602: Transmit the abnormal marker records and blood pressure change trend data from the health management report to the client application. The client application includes a mobile terminal APP and a wearable device interface. The interface provides a blood pressure historical trend display module, an abnormal warning module, and a health advice module. The modules are used to present the blood pressure fluctuation status and response measures suggestions to the user in real time. Step S603: When the number of abnormal blood pressure markers reaches a preset frequency threshold, a remote monitoring mechanism is triggered. The remote monitoring mechanism includes automatically pushing blood pressure prediction results and health management reports to a preset health service platform through a network interface. The health service platform is used to store, analyze, and forward abnormal data to the remote doctor's account.

9. The data analysis method for a non-invasive continuous ambulatory blood pressure monitor as described in claim 8, characterized in that, The deep neural network is used to process the temporal features of the heart sound signal. The temporal features include the periodic rhythm features of the heart sound signal, the time interval features between the first and second heart sounds, and the transient peak change features. The deep neural network model extracts the temporal features through multiple sets of one-dimensional convolutional layers and then connects them to a max pooling layer for feature compression. The output is a temporal feature vector of the heart sound signal after flattening. The convolutional neural network is used to process the spatial features in the micro-motion signal. The spatial features include the local amplitude distribution, envelope morphology, and position-related response patterns of the micro-motion signal in different sampling segments. The convolutional neural network performs nonlinear mapping and combination of the micro-motion signal features through a multi-layer fully connected structure, and outputs a spatial feature vector of the micro-motion signal.

10. The data analysis method for a non-invasive continuous ambulatory blood pressure monitor as described in claim 9, characterized in that, The health management report includes: A blood pressure change trend graph, which is generated based on blood pressure prediction results within a continuous time window, shows the rising, falling and stable sections of blood pressure values; The health risk assessment results are calculated based on the degree of deviation of systolic and diastolic blood pressure from the user's individual baseline range, the duration of abnormality, and the predicted rate of change. The improvement plan is generated based on the current blood pressure status and health profile, including dietary adjustment suggestions, exercise frequency suggestions, and sleep duration suggestions; User behavior adjustment suggestions are provided, which are based on historical behavior patterns and blood pressure response trends, and output as text data of daily intervention measures.

Images