Patient health monitoring method and system based on dynamic electrocardiogram analysis

Abstract

The invention relates to the technical field of health monitoring, and discloses a patient health monitoring method and system based on dynamic electrocardiogram analysis, and the method comprises the steps: synchronously obtaining multi-lead dynamic electrocardiogram data continuously collected by a patient within a preset duration, and related physiological parameters of exercise intensity, respiratory rate and body position change; performing dynamic self-adaptive preprocessing on the dynamic electrocardiogram data based on the associated physiological parameters to obtain standardized electrocardiosignals; extracting a multi-dimensional characteristic parameter set from the standardized electrocardiosignal, and sampling according to a preset time window to generate a characteristic parameter sequence with a timestamp; and inputting the characteristic parameter sequence into a dynamic optimization analysis model capable of iteratively updating parameters through real-time physiological parameter feedback, and performing graded evaluation on the health state of the patient to generate an evaluation result. The system corresponds to the method. By adopting the method and the system, the accuracy of dynamic electrocardiogram monitoring and the evaluation refinement level meet the dynamic monitoring requirement of cardiovascular health.

Description

Technical Field

[0001] This invention relates to the field of health monitoring technology, specifically a patient health monitoring method and system based on dynamic electrocardiogram analysis. Background Technology

[0002] Electrocardiogram (ECG) monitoring is a core technology for clinically assessing cardiovascular health and detecting diseases such as arrhythmias and myocardial ischemia at an early stage. Routine ECG and 24-hour Holter monitoring are the two most widely used basic monitoring methods in current medical settings. Routine ECG acquires a patient's ECG signal statically for a short period (usually a few minutes), quickly obtaining immediate ECG characteristics, suitable for routine outpatient screening. However, due to the limited monitoring time, it is difficult to capture intermittent abnormal signals such as paroxysmal arrhythmias and transient myocardial ischemia, easily leading to missed diagnoses. 24-hour Holter monitoring, on the other hand, continuously collects ECG data from patients during their daily activities using portable devices, significantly improving the detection rate of intermittent abnormal events and becoming an important tool for long-term monitoring of cardiovascular diseases.

[0003] However, existing patient health monitoring technologies based on routine and 24-hour Holter monitoring still have limitations in their overall technical system when addressing the needs of precise, personalized, and dynamic health management. At the signal processing level, existing technologies are insufficiently adaptable to the dynamically changing physiological scenarios of patients' daily activities. ECG signal preprocessing often uses fixed algorithms and parameters, making it difficult to effectively counteract interference factors in complex scenarios (such as movement, changes in body position, and respiratory fluctuations). This limits the accuracy of extracting key features from ECG signals, thus affecting the reliability of subsequent analysis results. At the health assessment level, existing analysis models are mostly qualitative, only outputting binary results of normal / abnormal, lacking quantitative assessment of health risks, and unable to dynamically adapt to individual patient differences (such as medical history and physiological characteristics), making it difficult to achieve early identification and tracing of potential health risks.

[0004] Therefore, existing monitoring technologies based on conventional and 24-hour Holter monitoring are insufficient to fully meet the clinical needs for comprehensive, accurate, and dynamic monitoring of patients' cardiovascular health. There is an urgent need to build a more adaptable, refined, and collaborative health monitoring technology system. Summary of the Invention

[0005] The purpose of this invention is to provide a patient health monitoring method and system based on dynamic electrocardiogram analysis to solve the technical problems mentioned in the background.

[0006] To achieve the above objectives, the present invention discloses the following technical solutions: In a first aspect, the present invention discloses a patient health monitoring method based on dynamic electrocardiogram analysis, the method comprising: S1: Simultaneously acquire the patient's dynamic electrocardiogram data and related physiological parameters. The dynamic electrocardiogram data includes multi-lead electrocardiogram signals continuously acquired within a preset time period. The related physiological parameters include exercise intensity, respiratory rate and body position change data. S2: Based on the associated physiological parameters, perform dynamic adaptive preprocessing on the dynamic electrocardiogram data to obtain a standardized electrocardiogram signal; S3: Extract a multidimensional feature parameter set from the standardized electrocardiogram signal, and periodically sample the multidimensional feature parameter set according to a preset time window to generate a feature parameter sequence with timestamps; wherein the multidimensional feature parameter set includes: heart rate variability parameters, waveform morphology features, and rhythm feature parameters; S4: Input the feature parameter sequence into the dynamic optimization analysis model, and perform a graded assessment of the patient's health status based on the dynamic optimization analysis model to generate an assessment result with a probability of health risk. The assessment result includes normal status, potential abnormal status, acute abnormal status and corresponding abnormality tracing information. The dynamic optimization analysis model iteratively updates its model parameters by receiving real-time feedback of related physiological parameters.

[0007] Preferably, the dynamic adaptive preprocessing includes: Based on the dynamic adjustment of the noise filtering threshold according to exercise intensity, adaptive filtering is performed on dynamic electrocardiogram data; Baseline drift correction is performed on dynamic electrocardiogram data based on respiratory rate characteristics; Based on body position change information, the dynamic electrocardiogram data is optimized by signal segmentation.

[0008] Preferably, the extraction of a multidimensional feature parameter set from the standardized electrocardiogram signal includes: Heart rate variability analysis is performed on the standardized electrocardiogram signal to extract time-domain features and frequency-domain features of energy distribution in different frequency bands, forming heart rate variability parameters. The time-domain features include the standard deviation of the normal sinus interval and the root mean square of the difference between adjacent normal intervals. The P wave, QRS complex and T wave in the standardized electrocardiogram signal are morphologically identified by a deep learning model, and the amplitude, width and slope of the waveform are extracted to form waveform morphological features. Time series analysis of continuous electrocardiogram signals is performed to capture the rhythmic variation patterns of the cardiac cycle and the frequency and duration of abnormal rhythms, thereby forming rhythmic characteristic parameters; The heart rate variability parameter, the waveform morphology feature, and the rhythm feature parameter are integrated to form a multidimensional feature parameter set.

[0009] Preferably, the integration of the heart rate variability parameter, the waveform morphological features, and the rhythm feature parameters includes: Normalization is performed on the heart rate variability parameter, the waveform morphology feature, and the rhythm feature parameter, respectively. Calculate the mutual information values ​​of the heart rate variability parameter, the waveform morphology feature, and the rhythm feature parameter with the historical health status label, retain feature parameters with mutual information values ​​greater than a preset threshold, and remove redundant and noisy features to form a core feature subset; Preset feature weight adjustment coefficients corresponding to different age groups and underlying disease types, and obtain an optimized feature parameter set that fits the individual characteristics of patients through weighted calculation. Use this optimized feature parameter set as the integrated multidimensional feature parameter set.

[0010] Preferably, the dynamic optimization analysis model includes: Basic evaluation layer: The multidimensional feature parameter set of each time window in the feature parameter sequence is initially classified using a random forest model; Dynamic correction layer: Based on the patient's historical health data, a bias correction factor is constructed to correct the bias in the preliminary classification results; Risk prediction layer: The feature parameter sequence is analyzed for time series trends using a long short-term memory network to generate health risk probability prediction values ​​for multiple future time windows; Results Output Layer: Based on the preliminary classification results after bias correction and the probability of health risks, the assessment results are output.

[0011] Preferably, the preliminary classification of the multidimensional feature parameter set for each time window in the feature parameter sequence using a random forest model includes the following steps: Based on the random forest model, for each time window in the feature parameter sequence, the preliminary classification probability of the health status corresponding to the time window is output, and the state with the highest probability is taken as the preliminary classification result of the time window; wherein, the training samples of each decision tree in the random forest model are obtained from the historical feature parameter sequence database through bootstrap sampling, and when each node splits, a portion of the multidimensional feature parameters are randomly selected to calculate the optimal split point.

[0012] Preferably, the method of constructing a bias correction factor based on the patient's historical health data to correct the bias in the preliminary classification results includes the following steps: Obtain the characteristic parameter sequences, clinical diagnostic records, and intervention feedback data corresponding to the patient's historical dynamic electrocardiograms; The deviation between the preliminary classification results of the random forest model and the clinical diagnosis results in historical data is calculated. Based on the deviation value, a correlation is established with the individual characteristics of the patient to generate a deviation correction factor with patient-specific attributes. The deviation correction factor is then embedded into a preset mathematical model to form a personalized correction function for the patient. The preliminary classification result is input into the personalized correction function. The bias correction factor integrated within the personalized correction function dynamically adjusts the preliminary classification result, and the corrected classification probability is output as the preliminary classification result after bias correction.

[0013] Preferably, the step of performing time-series trend analysis on the feature parameter sequence using a long short-term memory network to generate health risk probability prediction values ​​for multiple future time windows includes the following steps: A Long Short-Term Memory (LSTM) network is constructed, comprising an input layer, an output layer, and multiple hidden layers. The input layer receives feature parameter sequence segments from multiple consecutive time windows. The hidden layers process the temporal correlation of feature parameters within the segments through a gating mechanism to obtain temporal trend features. The output layer maps the temporal trend features to health risk probabilities based on a preset probability activation rule, generating and outputting predicted health risk probabilities for multiple future time windows. The constructed long short-term memory network was trained. During the training process, time series cross-validation was used, with the recent patient feature parameter sequence and the corresponding actual health risk records as the validation set. The deviation between the health risk probability prediction value output by the long short-term memory network based on the time series trend features and the actual health risk records was used as the optimization objective. The internal weights and bias parameters of the long short-term memory network were iteratively adjusted until the health risk probability prediction error met the preset requirements. The trained Long Short-Term Memory (LSTM) network extracts the frequency characteristic trend of the central rate variability parameter, the morphological change slope of the waveform characteristic parameter, and the cumulative change of the rhythm characteristic parameter through preset feature extraction rules. It also captures key nonlinear correlation information formed by the frequency characteristic trend, morphological change slope, and cumulative change in a continuous time window through a gating mechanism. This key nonlinear correlation information is transformed into a time-series trend feature that can reflect the dynamic change trend of the patient's health status, and the corresponding health risk probability prediction value is output by the output layer.

[0014] Preferably, the process of outputting the assessment result based on the preliminary classification result after bias correction and the health risk probability includes the following steps: Multiple preset threshold ranges for health risk probability are set, corresponding to normal state, potential abnormal state and acute abnormal state respectively; Based on the weighted fusion mechanism, the results of the preliminary classification after bias correction are fused with the health risk probability according to the preset weights. When the fusion result falls into a certain preset threshold range, the corresponding state evaluation result is output.

[0015] Secondly, this invention discloses a patient health monitoring system based on dynamic electrocardiogram analysis, which applies the patient health monitoring method based on dynamic electrocardiogram analysis as described above. The system includes: The data acquisition module is configured to: synchronously acquire the patient's dynamic electrocardiogram data and related physiological parameters, wherein the dynamic electrocardiogram data includes multi-lead electrocardiogram signals continuously collected within a preset time period, and the related physiological parameters include exercise intensity, respiratory rate and body position change data; The data processing module is configured to: perform dynamic adaptive preprocessing on the dynamic electrocardiogram data based on the associated physiological parameters to obtain a standardized electrocardiogram signal; The feature extraction module is configured to: extract a multidimensional feature parameter set from the standardized electrocardiogram signal, and periodically sample the multidimensional feature parameter set according to a preset time window to generate a feature parameter sequence with timestamps; wherein the multidimensional feature parameter set includes: heart rate variability parameters, waveform morphology features, and rhythm feature parameters; The health assessment module is configured to: input the feature parameter sequence into a dynamic optimization analysis model, and perform a graded assessment of the patient's health status based on the dynamic optimization analysis model, generating an assessment result with a probability of health risk. The assessment result includes a normal state, a potentially abnormal state, an acute abnormal state, and corresponding abnormality tracing information. The dynamic optimization analysis model iteratively updates its model parameters by receiving real-time feedback of related physiological parameters.

[0016] Beneficial Effects: The patient health monitoring method and system based on dynamic electrocardiogram (ECG) analysis of the present invention improves the adaptability and accuracy of ECG signal processing under complex physiological scenarios by simultaneously acquiring patient dynamic ECG data and associated physiological parameters, and performing dynamic adaptive preprocessing on the dynamic ECG data based on the associated physiological parameters. Simultaneously, by extracting a multi-dimensional feature parameter set from standardized ECG signals and generating a feature parameter sequence with timestamps, combined with a dynamic optimization analysis model that iteratively updates parameters through real-time physiological parameter feedback, the system can perform graded assessment of patient health status, outputting assessment results that include health risk probability and abnormality tracing information. This achieves refined assessment of patient health status and early identification of potential risks, thereby improving the practicality of dynamic ECG-based patient health monitoring technology and better meeting the needs for accurate and dynamic monitoring of cardiovascular health status in clinical and daily health management. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating a patient health monitoring method based on dynamic electrocardiogram analysis provided in an embodiment of the present invention. Detailed Implementation

[0019] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] In this document, the term "comprising" is intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0021] This embodiment aims to address the shortcomings of existing monitoring technologies based on conventional and 24-hour Holter monitoring in terms of signal processing adaptability, health assessment refinement, and closed-loop monitoring process, which make it difficult to meet the clinical and health management needs for accurate and dynamic monitoring of patients' cardiovascular health status.

[0022] Based on this, this embodiment provides a first aspect such as Figure 1 The patient health monitoring method based on dynamic electrocardiogram analysis shown herein includes: S1: Simultaneously acquire the patient's dynamic electrocardiogram data and related physiological parameters. The dynamic electrocardiogram data includes multi-lead electrocardiogram signals continuously collected within a preset time period. The related physiological parameters include exercise intensity, respiratory rate and body position change data. S2: Dynamic adaptive preprocessing of dynamic electrocardiogram data based on associated physiological parameters to obtain standardized electrocardiogram signals; S3: Extract a multidimensional feature parameter set from the standardized electrocardiogram signal, and periodically sample the multidimensional feature parameter set according to a preset time window to generate a feature parameter sequence with timestamps; the multidimensional feature parameter set includes: heart rate variability parameters, waveform morphology features, and rhythm feature parameters; S4: Input the feature parameter sequence into the dynamic optimization analysis model, and perform a graded assessment of the patient's health status based on the dynamic optimization analysis model to generate an assessment result with a probability of health risk. The assessment result includes normal status, potential abnormal status, acute abnormal status and corresponding abnormality tracing information; wherein, the dynamic optimization analysis model iteratively updates the model parameters through real-time feedback of related physiological parameters.

[0023] Based on the above, the patient health monitoring method based on dynamic electrocardiogram analysis in this embodiment simultaneously acquires multi-lead dynamic electrocardiogram data of patients and related physiological parameters such as exercise intensity, respiratory rate, and body position changes. It then performs dynamic adaptive preprocessing on the dynamic electrocardiogram data using these related physiological parameters, extracts a multi-dimensional feature parameter set from the standardized electrocardiogram signal, generates a feature parameter sequence with timestamps, and inputs it into a dynamic optimization analysis model that iteratively updates parameters based on real-time physiological parameter feedback. This model performs a graded assessment of the patient's health status and generates an assessment result containing health risk probability and abnormality tracing information. This solves the problems of poor signal preprocessing adaptability (unable to cope with interference from complex physiological scenarios), single feature extraction dimension (lacking temporal correlation), static assessment model (unable to dynamically adapt to patient status), and coarse assessment results (only binary judgment without risk quantification and tracing) in existing dynamic electrocardiogram monitoring technologies. It achieves precision in dynamic electrocardiogram signal processing, comprehensive feature information, refined health assessment (grading + risk quantification), and dynamic model adaptability, providing more reliable technical support for patient cardiovascular health monitoring.

[0024] In detail, before performing dynamic adaptive preprocessing, it is necessary to ensure the time synchronization between dynamic electrocardiogram data and associated physiological parameters. In this embodiment, a flexible wearable electrocardiogram monitoring device integrating multimodal sensors is used as the data acquisition terminal. In addition to a dry electrode array for acquiring dynamic electrocardiogram data of multiple leads (such as chest leads V1-V6 and limb leads II), the device also has built-in motion sensors (such as triaxial accelerometers), respiratory sensors (such as chest and abdominal strain gauges), and position sensors (such as gyroscopes). The sampling frequency of the three types of sensors and the electrocardiogram electrode array is uniformly 250Hz~500Hz, and each frame of data carries the same timestamp, ensuring that the data of exercise intensity, respiratory rate, and positional changes are accurately correlated with the dynamic electrocardiogram data at the corresponding time, providing a time reference for the subsequent "adjustment of preprocessing strategy based on associated parameters".

[0025] Furthermore, dynamic adaptive preprocessing includes: Adaptive filtering of dynamic electrocardiogram (ECG) data is performed by dynamically adjusting the noise filtering threshold based on exercise intensity. Specifically, this includes: 1. The patient's real-time acceleration data is collected using a triaxial accelerometer (integrated into a wearable ECG monitoring device) in the existing technology. By calculating the root mean square value of acceleration (RMS) per unit time, the exercise intensity is divided into three levels: low, medium and high (conventional exercise intensity grading logic).

[0026] 2. The Least Mean Square Error (LMS) algorithm or Recursive Least Squares (RLS) algorithm in the existing technology is used as the filtering framework, and the noise filtering threshold corresponding to the motion intensity level is preset; a higher threshold is set for low motion intensity (to reduce the loss of signal details), a medium threshold is set for medium motion intensity (to balance noise filtering and signal fidelity), and a lower threshold is set for high motion intensity (to enhance noise filtering).

[0027] 3. Synchronize the exercise intensity classification results with the dynamic electrocardiogram data according to the timestamp. The filtering algorithm calls the corresponding threshold according to the current exercise intensity and performs frame-by-frame filtering on the dynamic electrocardiogram data of the same time period, and outputs the preliminary electrocardiogram signal after filtering out exercise interference.

[0028] Baseline drift correction is performed on Holter ECG data based on respiratory rate characteristics. Specifically, this includes: 1. Using existing technologies such as chest and abdominal strain gauge sensors or piezoelectric respiratory sensors (integrated into wearable devices), the patient's chest expansion / contraction data are collected, the number of breaths per unit time is calculated to obtain the real-time respiratory rate, and the fundamental frequency and phase of the respiratory cycle are extracted using existing Fourier transform technology (distinguishing between the inspiratory and expiratory phases).

[0029] 2. The baseline components in the dynamic electrocardiogram data are extracted using existing wavelet transform algorithms or polynomial fitting algorithms. Based on the characteristic that "the respiratory cycle is consistent with the baseline drift cycle", the correction parameters that match the current respiratory phase (inspiratory phase / expiratory phase) are called to perform reverse compensation on the baseline components.

[0030] 3. The compensated baseline components are reconstructed with the high-frequency effective components (such as P waves and QRS complexes) of the original ECG signal to obtain an ECG signal that eliminates respiratory-related baseline drift.

[0031] Based on body position change information, signal segmentation optimization is performed on the dynamic electrocardiogram data. Specifically, this includes: 1. Use existing technologies such as gyroscopes or tilt sensors (integrated into wearable devices) to collect the patient's real-time posture angles (such as pitch angle and roll angle). When the change in posture angle per unit time exceeds the conventional judgment threshold (such as 30°), it is judged as a change in body position and the change timestamp is recorded.

[0032] 2. The basic segmentation of the dynamic electrocardiogram signal adopts the conventional rule of "using the peak value of the R wave as the reference" in the existing technology (such as each segment containing several consecutive cardiac cycles is a segment); when a change in body position is detected, the basic rule is paused and the segment boundary is adjusted to "the last R wave before the change is the endpoint and the first R wave after the change is the starting point" to ensure that the electrocardiogram signal in the same segment comes from the same body position.

[0033] 3. Add a "position association identifier" to adjacent segments before and after the positional change. The subsequent feature extraction process can use this identifier (existing data labeling technology) to perform routine verification of the feature parameters of these two signal segments (excluding non-pathological waveform differences caused by position).

[0034] Based on the above, the patient health monitoring method based on dynamic electrocardiogram analysis in this embodiment adaptively filters dynamic electrocardiogram data by dynamically adjusting the noise filtering threshold based on exercise intensity, processes dynamic electrocardiogram data using baseline drift correction technology combined with respiratory rate, and optimizes the segmentation of dynamic electrocardiogram data based on body position changes using signal segmentation optimization technology. This solves the problem in existing dynamic adaptive preprocessing methods that do not develop differentiated processing strategies for the interference characteristics of different associated physiological parameters, resulting in the inability to effectively eliminate signal segmentation deviations caused by exercise interference, respiratory baseline drift, and body position changes. It achieves targeted filtering and optimization of different types of physiological interference in dynamic electrocardiogram data, further improving the purity and segmentation rationality of standardized electrocardiogram signals, and laying a more solid foundation for the accuracy of subsequent feature extraction.

[0035] In this embodiment, a multidimensional feature parameter set is extracted from the self-standardized electrocardiogram signal, including: Heart rate variability analysis was performed on standardized electrocardiogram signals to extract time-domain features and frequency-domain features of energy distribution in different frequency bands, forming heart rate variability parameters. The time-domain features include the standard deviation of normal sinus interval (SDNN) and the root mean square of the difference between adjacent normal intervals (RMSSD). The P wave, QRS complex and T wave in the standardized electrocardiogram signal are morphologically identified by a deep learning model, and the amplitude, width and slope of the waveform are extracted to form waveform morphological features. Time series analysis of continuous electrocardiogram signals is performed to capture the rhythmic variation patterns of the cardiac cycle and the frequency and duration of abnormal rhythms, thereby forming rhythmic characteristic parameters; By integrating heart rate variability parameters, waveform morphology features, and rhythm feature parameters, a multidimensional feature parameter set is formed.

[0036] Based on the above, the patient health monitoring method based on dynamic electrocardiogram analysis in this embodiment extracts time-domain and frequency-domain features by performing heart rate variability analysis on standardized electrocardiogram signals, identifying P waves / QRS complexes / T waves using a deep learning model and extracting waveform morphology features, and capturing rhythm features by performing time series analysis on continuous electrocardiogram signals. These three types of features are then integrated into a multi-dimensional feature parameter set. This solves the problems of incomplete heart rate variability feature dimensions (lacking frequency domain information), low waveform feature recognition accuracy (relying on manual work or simple algorithms), lack of continuous correlation analysis of rhythm features, and fragmented information due to the lack of systematic integration of various features in existing feature extraction technologies. It expands electrocardiogram features from a single dimension to multiple dimensions (time domain-frequency domain-morphology-rhythm), improving the comprehensiveness and accuracy of feature extraction and providing richer and more reliable input information for dynamic optimization analysis models.

[0037] Furthermore, heart rate variability parameters, waveform morphology features, and rhythm characteristic parameters are integrated, including: Normalization was performed on heart rate variability parameters, waveform morphology features, and rhythm feature parameters respectively. For numerical heart rate variability parameters, Z-score standardization was used to eliminate dimensional differences. For morphological waveform morphology features, extreme value normalization was used to unify the feature amplitude range. For time-series rhythm feature parameters, sliding window normalization was used to align with the time dimension benchmark. The mutual information values ​​of heart rate variability parameters, waveform morphology features, and rhythm features with historical health status labels (such as clinically diagnosed health status) are calculated. Feature parameters with mutual information values ​​greater than a preset threshold are retained, while redundant features (such as time-domain parameters that repeatedly characterize heart rate fluctuations) and noise features (such as morphological parameters with a signal-to-noise ratio of less than 15dB) are removed to form a core feature subset.

[0038] Preset feature weight adjustment coefficients for different age groups (adolescents / middle-aged / elderly) and underlying disease types (coronary heart disease / hypertension / history of arrhythmia). For example, increase the weight of heart rate variability parameters for elderly patients (increase by 20%-30%), and increase the weight of waveform morphology features for patients with coronary heart disease (increase by 30%-40%). Calculate the optimized feature parameter set that fits the individual characteristics of the patients through weighted calculation. Use this optimized feature parameter set as the integrated multidimensional feature parameter set for subsequent health status assessment.

[0039] Mutual information (MI) measures the dependency between two random variables; a higher value indicates a stronger correlation. In this method, for discrete features (such as abnormal rhythm markers), the discretization method used in existing technologies is employed to calculate mutual information; for continuous features (such as heart rate fluctuation amplitude), the K-nearest neighbor method or kernel density estimation method used in existing technologies is employed. Efficient calculation is achieved by calling mature functions in existing open-source libraries (such as scikit-learn), and highly correlated features are then selected based on the calculation results.

[0040] For a combination of two types of variables, feature parameters X (such as SDNN in heart rate variability parameters) and historical health status labels Y (such as normal / abnormal status), when X is a discrete feature (such as abnormal rhythm label: 0 = no abnormal rhythm, 1 = occasional premature beats, 2 = frequent premature beats; or features that have been processed by existing discretization methods), and Y is a discrete health status label (such as... When mutual information is calculated, it is obtained by summing discrete probability distributions, and the formula is: , in, For all possible values ​​of discrete feature X (such as {0,1,2} for abnormal rhythm labeling); For the set of all possible values ​​of the health status label Y (e.g., {0, 1, 2}); The marginal probability of feature X taking a certain discrete value x is calculated as "X = number of samples of x in historical samples ÷ total number of historical samples"; The marginal probability of label Y taking a certain discrete value y is calculated as "the number of historical samples Y = y ÷ the total number of historical samples"; The joint probability of feature X=x and label Y=y is calculated as "the number of historical samples that simultaneously satisfy X=x and Y=y ÷ the total number of historical samples". It is a base-2 logarithm to ensure that the mutual information result is in bits (the larger the value, the stronger the correlation between X and Y).

[0041] When X is a continuous feature (such as the heart rate variability parameter SDNN, heart rate fluctuation amplitude) and Y is a discrete health status label, mutual information needs to be calculated by "integral of probability density function" (because the single-point probability of a continuous variable is 0, it cannot be directly summed using discrete methods), the formula is: , in, The actual range of values ​​for continuous feature X (e.g., the normal physiological range of SDNN is 20~200ms). The marginal probability density function of a continuous feature X describes the probability distribution density of X at the value x. It needs to be obtained from historical samples using existing techniques (such as K-nearest neighbor method and kernel density estimation method). Let X be the joint probability density function of feature X and label Y, describing the probability distribution density when X=x and Y=y, and it is also fitted by the K nearest neighbor method or kernel density estimation method. The marginal probability of label Y (same as the discrete formula, since Y is still a discrete variable); To integrate over all possible values ​​of a continuous feature X, replacing the summation in discrete scenarios, we can achieve mutual information quantization of continuous variables.

[0042] Regardless of whether the feature is discrete or continuous, the mutual information value is ultimately calculated using the corresponding formulas described above. : when The larger the value (e.g., >0.2), the stronger the correlation between the feature and the health status label, and the higher the discriminative power of the health assessment. when The smaller the value (e.g., <0.05), the weaker the correlation between the feature and the label. These features are redundant or noisy and need to be filtered out based on the calculation results. Finally, highly correlated features are retained for subsequent model analysis.

[0043] Based on the above, the patient health monitoring method based on dynamic electrocardiogram analysis in this embodiment performs normalization processing on heart rate variability parameters, waveform morphology features, and rhythm feature parameters respectively, calculates the mutual information value between features and health status labels to screen core features, and then generates an optimized feature parameter set by combining preset weight coefficients of patient age group and underlying disease type. This solves the problems in existing feature integration, such as feature weight imbalance caused by failure to eliminate dimensional differences, redundant / noisy features leading to redundant model calculations, and poor feature adaptability due to failure to consider individual patient differences. It achieves standardization (eliminating the influence of dimensions), simplification (reducing redundant calculations), and personalization (adapting to different patient features) of feature parameters, effectively improving the efficiency and pertinence of subsequent model analysis and reducing evaluation errors caused by feature quality issues.

[0044] In this embodiment, the dynamic optimization analysis model includes: Basic evaluation layer: The random forest model is used to perform preliminary classification of the multidimensional feature parameter set of each time window in the feature parameter sequence; Dynamic correction layer: Based on the patient's historical health data, a bias correction factor is constructed to correct the bias in the preliminary classification results; Risk prediction layer: The feature parameter sequence is analyzed over time using a long short-term memory network to generate a predicted value of health risk probability for multiple future time windows; Results Output Layer: Based on the preliminary classification results after bias correction and the probability of health risks, the assessment results are output.

[0045] Based on the above, the patient health monitoring method based on dynamic electrocardiogram analysis in this embodiment constructs a dynamic optimization analysis model containing a basic assessment layer (random forest preliminary classification), a dynamic correction layer (historical data to construct bias correction factors), a risk prediction layer (long short-term memory network temporal trend analysis), and a result output layer (fusion correction results and risk probabilities). This solves the problems of existing analysis models, such as a single structure (only single-level analysis), lack of individual bias correction (assessment results are easily affected by individual patient differences), lack of temporal trend analysis (unable to capture dynamic changes in health status), and lack of result output fusion mechanism (low reliability of single results). It realizes the upgrade of health assessment from "single analysis" to "multi-level collaborative analysis", improving the accuracy (bias correction), foresight (temporal trend prediction), and reliability (multi-result fusion) of assessment results, and meeting the needs of dynamic health status monitoring.

[0046] Furthermore, a preliminary classification of the multidimensional feature parameter sets for each time window in the feature parameter sequence is performed using a random forest model, including the following steps: Based on the random forest model, for each time window in the feature parameter sequence, the preliminary classification probability of the health status corresponding to the time window is output, and the state with the highest probability is taken as the preliminary classification result of the time window. In the random forest model, the training samples of each decision tree are obtained from the historical feature parameter sequence database through bootstrap sampling, and when each node splits, some multidimensional feature parameters are randomly selected to calculate the optimal split point.

[0047] Based on the above, the patient health monitoring method based on dynamic electrocardiogram analysis in this embodiment adopts a random forest model containing multiple decision trees. It obtains training samples based on bootstrap sampling and randomly selects some features to calculate the optimal split point when splitting decision tree nodes. It outputs the classification probability of the multi-dimensional feature parameter set of each time window in the feature parameter sequence and takes the highest probability as the preliminary classification result. It solves the problems of poor model generalization ability (easy to overfit), unbalanced feature utilization (dependence on some features leading to bias), and lack of probability support for classification results (only qualitative judgment) in existing preliminary classification techniques. It realizes the anti-overfitting ability of preliminary classification (multiple decision trees + bootstrap sampling), the balance of feature utilization (random feature selection for splitting), and the interpretability of classification results (probability quantification), providing more objective and stable basic data for subsequent bias correction.

[0048] Secondly, a bias correction factor is constructed based on the patient's historical health data to correct the bias in the preliminary classification results, including the following steps: Obtain the characteristic parameter sequences, clinical diagnostic records, and intervention feedback data corresponding to the patient's historical dynamic electrocardiograms; The deviation between the preliminary classification results of the random forest model and the clinical diagnosis results in historical data is calculated. Based on the deviation value, the correlation between the patient's individual characteristics (such as age, type of underlying disease, and medication history) is established to generate a deviation correction factor with patient-specific attributes. The deviation correction factor is then embedded into a preset mathematical model to form a personalized correction function for the patient. The preliminary classification results are input into a personalized correction function. The bias correction factor integrated within the personalized correction function dynamically adjusts the preliminary classification results, and the corrected classification probability is output as the preliminary classification result after bias correction.

[0049] Based on the above, the patient health monitoring method based on dynamic electrocardiogram analysis in this embodiment obtains the patient's historical dynamic electrocardiogram feature sequence, clinical diagnosis records, and intervention feedback data. It calculates the deviation between the historical preliminary classification results and the clinical diagnosis, establishes the correlation between the deviation and the patient's individual characteristics, generates personalized correction factors and correction functions, and then uses them to adjust the preliminary classification results. This solves the problem that existing models lack a personalized correction mechanism, resulting in large deviations in classification results for different patients (such as different medical histories and physiological characteristics), and failing to adapt to individual differences. It achieves "individualized calibration" of the preliminary classification results, significantly reduces the assessment deviation caused by individual patient differences, makes the corrected classification results more consistent with the patient's actual health status, and improves the accuracy of subsequent assessments.

[0050] Simultaneously, by performing time-series trend analysis on the feature parameter sequence using a long short-term memory network, predictive values ​​of health risk probability for multiple future time windows are generated, including the following steps: A Long Short-Term Memory (LSTM) network is constructed, which includes an input layer, an output layer, and multiple hidden layers. The input layer receives feature parameter sequence segments from multiple consecutive time windows. The hidden layers process the temporal correlation of feature parameters within the segments through a gating mechanism (including an input gate, a forget gate, and an output gate) to obtain temporal trend features. The output layer maps the temporal trend features to health risk probabilities based on a preset probability activation rule, and generates and outputs predicted health risk probabilities for multiple future time windows. The constructed long short-term memory network was trained. During the training process, time series cross-validation was used, with the recent patient feature parameter sequence and the corresponding actual health risk records as the validation set. The deviation between the health risk probability prediction value output by the long short-term memory network based on the time series trend features and the actual health risk records was used as the optimization objective. The internal weights and bias parameters of the long short-term memory network were iteratively adjusted until the health risk probability prediction error met the preset requirements. The trained Long Short-Term Memory (LSTM) network extracts the frequency trend of the central rate variability parameter, the morphological slope of the waveform feature parameter, and the cumulative change of the rhythm feature parameter through a pre-defined feature extraction rule. It then captures key nonlinear correlation information formed by the frequency trend, morphological slope, and cumulative change within a continuous time window (such as the synergistic change pattern between different parameters and the correlation between a sudden change in a parameter and other parameters) through a gating mechanism. This key nonlinear correlation information is then transformed into time-series trend features that reflect the dynamic changes in the patient's health status (such as the stable trend of health status corresponding to gradual parameter changes and the upward trend of health risk corresponding to parameter mutations). The corresponding predicted health risk probability value is then output by the output layer.

[0051] Based on the above, the patient health monitoring method based on dynamic electrocardiogram analysis in this embodiment constructs a long short-term memory network containing an input layer (receiving continuous time window feature sequence segments), a hidden layer (gating mechanism to process temporal correlations and generate temporal trend features), and an output layer (mapping temporal trend features into health risk probabilities). The network is trained using time-series cross-validation and its parameters are iteratively optimized. The trained network then extracts temporal correlation information from feature sequences and transforms it into temporal trend features. This solves the problems of existing time-series analysis techniques failing to effectively capture long-term dependencies in feature sequences (such as gradual trends in health status), lacking temporal feature support for risk probability prediction (low prediction accuracy), and not optimizing network training for time-series data (poor generalization ability). It achieves accurate capture of long-term temporal correlations in feature sequences, improves the foresight and accuracy of health risk probability prediction, provides reliable temporal dimension basis for health status grading assessment, and supports early identification of potential risks.

[0052] In addition, based on the results of the preliminary classification after bias correction and the probability of health risk, the assessment results are output, including the following steps: Multiple preset threshold ranges for health risk probability are set, corresponding to normal state, potential abnormal state and acute abnormal state respectively; Based on the weighted fusion mechanism, the results of the preliminary classification after bias correction are fused with the health risk probability according to the preset weights. When the fusion result falls into a certain preset threshold range, the corresponding state evaluation result is output.

[0053] Based on the above, the patient health monitoring method based on dynamic electrocardiogram analysis in this embodiment sets multiple health risk probability threshold intervals corresponding to different health states, and uses a weighted fusion mechanism to fuse the preliminary classification results after bias correction with the health risk probability for calculation. The assessment result is output according to the threshold interval into which the fusion result falls. This solves the problems of existing assessment results output lacking clear quantitative standards (subjective state judgment) and lacking multi-dimensional assessment basis (low reliability of single result). It realizes "standardized judgment" (threshold interval quantification) and "multi-basis synergy" (weighted fusion) in health status assessment, ensuring the objectivity and reliability of the assessment results, avoiding misjudgment of the state due to the bias of a single assessment basis, and making the output health status classification more accurate.

[0054] This embodiment provides a patient health monitoring system based on dynamic electrocardiogram analysis in a second aspect, applied to the patient health monitoring method based on dynamic electrocardiogram analysis as described above. The system includes: The data acquisition module is configured to: synchronously acquire the patient's dynamic electrocardiogram data and related physiological parameters. The dynamic electrocardiogram data includes multi-lead electrocardiogram signals continuously collected within a preset time period, and the related physiological parameters include exercise intensity, respiratory rate and body position change data. The data processing module is configured to perform dynamic adaptive preprocessing on dynamic electrocardiogram data based on associated physiological parameters to obtain standardized electrocardiogram signals. The feature extraction module is configured to: extract a multidimensional feature parameter set from the standardized electrocardiogram signal, and periodically sample the multidimensional feature parameter set according to a preset time window to generate a feature parameter sequence with timestamps; wherein the multidimensional feature parameter set includes: heart rate variability parameters, waveform morphological features and rhythm feature parameters; The health assessment module is configured to: input the feature parameter sequence into the dynamic optimization analysis model, and perform a graded assessment of the patient's health status based on the dynamic optimization analysis model, generating an assessment result with a probability of health risk. The assessment result includes normal status, potential abnormal status, acute abnormal status, and corresponding abnormality tracing information; wherein, the dynamic optimization analysis model iteratively updates the model parameters through real-time feedback of related physiological parameters.

[0055] Based on the above, the patient health monitoring system based on dynamic electrocardiogram analysis in this embodiment, by setting up a data acquisition module (synchronously collecting dynamic electrocardiograms and related physiological parameters), a data processing module (dynamic adaptive preprocessing), a feature extraction module (generating multi-dimensional feature parameter sets and feature parameter sequences), and a health assessment module (dynamically optimizing the analysis model for hierarchical assessment), collaboratively executes the aforementioned patient health monitoring method based on dynamic electrocardiogram analysis. This solves the problem in existing monitoring systems where data acquisition, processing, feature extraction, and assessment are fragmented, leading to low monitoring efficiency and the inability to form a complete technical closed loop. It integrates patient health monitoring from scattered processes into a unified closed loop, ensuring functional compatibility of each module, improving the efficiency and continuity of the monitoring process, providing a stable and reliable system platform for the implementation of the aforementioned patient health monitoring method based on dynamic electrocardiogram analysis, and guaranteeing the practicality and operability of the overall monitoring technology. It should be noted that the patient health monitoring system based on dynamic electrocardiogram analysis in this embodiment corresponds to the aforementioned patient health monitoring method based on dynamic electrocardiogram analysis. Therefore, the parts of the patient health monitoring system based on dynamic electrocardiogram analysis in this embodiment that are not described in detail (including but not limited to specific technical means and technical effects) can be referred to the relevant descriptions in the aforementioned patient health monitoring method based on dynamic electrocardiogram analysis, and will not be repeated here.

[0056] In the embodiments provided by this invention, it should be understood that the embodiments described herein can be implemented in hardware, software, firmware, middleware, code, or any suitable combination thereof. For hardware implementation, the processor can be implemented in one or more of the following: application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), processors, controllers, microcontrollers, microprocessors, other electronic units designed to implement the functions described herein, or combinations thereof. For software implementation, some or all of the processes of the embodiments can be performed by a computer program instructing the associated hardware. During implementation, the program can be stored in a computer-readable storage medium or transmitted as one or more instructions or code on a computer-readable storage medium. Computer-readable storage media include computer storage media and communication media, wherein communication media include any medium that facilitates the transmission of a computer program from one place to another. Storage media can be any available medium accessible to a computer. Computer-readable storage media can include, but are not limited to, RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code having the form of instructions or data structures and accessible to a computer.

[0057] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A patient health monitoring method based on dynamic electrocardiogram analysis, characterized in that, The method includes: S1: Simultaneously acquire the patient's dynamic electrocardiogram data and related physiological parameters. The dynamic electrocardiogram data includes multi-lead electrocardiogram signals continuously acquired within a preset time period. The related physiological parameters include exercise intensity, respiratory rate and body position change data. S2: Based on the associated physiological parameters, perform dynamic adaptive preprocessing on the dynamic electrocardiogram data to obtain a standardized electrocardiogram signal; S3: Extract a multidimensional feature parameter set from the standardized electrocardiogram signal, and periodically sample the multidimensional feature parameter set according to a preset time window to generate a feature parameter sequence with timestamps; wherein the multidimensional feature parameter set includes: heart rate variability parameters, waveform morphology features, and rhythm feature parameters; S4: Input the feature parameter sequence into the dynamic optimization analysis model, and perform a graded assessment of the patient's health status based on the dynamic optimization analysis model to generate an assessment result with a probability of health risk. The assessment result includes normal status, potential abnormal status, acute abnormal status and corresponding abnormality tracing information. The dynamic optimization analysis model iteratively updates its model parameters by receiving real-time feedback of related physiological parameters.

2. The patient health monitoring method based on dynamic electrocardiogram analysis according to claim 1, characterized in that, The dynamic adaptive preprocessing includes: Based on the dynamic adjustment of the noise filtering threshold according to exercise intensity, adaptive filtering is performed on dynamic electrocardiogram data; Baseline drift correction is performed on dynamic electrocardiogram data based on respiratory rate characteristics; Based on body position change information, the dynamic electrocardiogram data is optimized by signal segmentation.

3. The patient health monitoring method based on dynamic electrocardiogram analysis according to claim 1, characterized in that, The extraction of a multidimensional feature parameter set from the standardized electrocardiogram signal includes: Heart rate variability analysis is performed on the standardized electrocardiogram signal to extract time-domain features and frequency-domain features of energy distribution in different frequency bands, forming heart rate variability parameters. The time-domain features include the standard deviation of the normal sinus interval and the root mean square of the difference between adjacent normal intervals. The P wave, QRS complex and T wave in the standardized electrocardiogram signal are morphologically identified by a deep learning model, and the amplitude, width and slope of the waveform are extracted to form waveform morphological features. Time series analysis of continuous electrocardiogram signals is performed to capture the rhythmic variation patterns of the cardiac cycle and the frequency and duration of abnormal rhythms, thereby forming rhythmic characteristic parameters; The heart rate variability parameter, the waveform morphology feature, and the rhythm feature parameter are integrated to form a multidimensional feature parameter set.

4. The patient health monitoring method based on dynamic electrocardiogram analysis according to claim 3, characterized in that, The integration of the heart rate variability parameter, the waveform morphological features, and the rhythm feature parameters includes: Normalization is performed on the heart rate variability parameter, the waveform morphology feature, and the rhythm feature parameter, respectively. Calculate the mutual information values ​​of the heart rate variability parameter, the waveform morphology feature, and the rhythm feature parameter with the historical health status label, retain feature parameters with mutual information values ​​greater than a preset threshold, and remove redundant and noisy features to form a core feature subset; Preset feature weight adjustment coefficients corresponding to different age groups and underlying disease types, and obtain an optimized feature parameter set that fits the individual characteristics of patients through weighted calculation. Use this optimized feature parameter set as the integrated multidimensional feature parameter set.

5. The patient health monitoring method based on dynamic electrocardiogram analysis according to claim 1, characterized in that, The dynamic optimization analysis model includes: Basic evaluation layer: The multidimensional feature parameter set of each time window in the feature parameter sequence is initially classified using a random forest model; Dynamic correction layer: Based on the patient's historical health data, a bias correction factor is constructed to correct the bias in the preliminary classification results; Risk prediction layer: The feature parameter sequence is analyzed for time series trends using a long short-term memory network to generate health risk probability prediction values ​​for multiple future time windows; Results Output Layer: Based on the preliminary classification results after bias correction and the probability of health risks, the assessment results are output.

6. The patient health monitoring method based on dynamic electrocardiogram analysis according to claim 5, characterized in that, The preliminary classification of the multidimensional feature parameter set for each time window in the feature parameter sequence using a random forest model includes the following steps: Based on the random forest model, for each time window in the feature parameter sequence, the preliminary classification probability of the health status corresponding to the time window is output, and the state with the highest probability is taken as the preliminary classification result of the time window; wherein, the training samples of each decision tree in the random forest model are obtained from the historical feature parameter sequence database through bootstrap sampling, and when each node splits, a portion of the multidimensional feature parameters are randomly selected to calculate the optimal split point.

7. The patient health monitoring method based on dynamic electrocardiogram analysis according to claim 5, characterized in that, The method of constructing a bias correction factor based on patients' historical health data to correct biases in the preliminary classification results includes the following steps: Obtain the characteristic parameter sequences, clinical diagnostic records, and intervention feedback data corresponding to the patient's historical dynamic electrocardiograms; The deviation between the preliminary classification results of the random forest model and the clinical diagnosis results in historical data is calculated. Based on the deviation value, a correlation is established with the individual characteristics of the patient to generate a deviation correction factor with patient-specific attributes. The deviation correction factor is then embedded into a preset mathematical model to form a personalized correction function for the patient. The preliminary classification result is input into the personalized correction function. The bias correction factor integrated within the personalized correction function dynamically adjusts the preliminary classification result, and the corrected classification probability is output as the preliminary classification result after bias correction.

8. The patient health monitoring method based on dynamic electrocardiogram analysis according to claim 5, characterized in that, The method of performing time-series trend analysis on the feature parameter sequence using a long short-term memory network to generate predicted health risk probabilities for multiple future time windows includes the following steps: A Long Short-Term Memory (LSTM) network is constructed, comprising an input layer, an output layer, and multiple hidden layers. The input layer receives feature parameter sequence segments from multiple consecutive time windows. The hidden layers process the temporal correlation of feature parameters within the segments through a gating mechanism to obtain temporal trend features. The output layer maps the temporal trend features to health risk probabilities based on a preset probability activation rule, generating and outputting predicted health risk probabilities for multiple future time windows. The constructed long short-term memory network was trained. During the training process, time series cross-validation was used, with the recent patient feature parameter sequence and the corresponding actual health risk records as the validation set. The deviation between the health risk probability prediction value output by the long short-term memory network based on the time series trend features and the actual health risk records was used as the optimization objective. The internal weights and bias parameters of the long short-term memory network were iteratively adjusted until the health risk probability prediction error met the preset requirements. The trained Long Short-Term Memory (LSTM) network extracts the frequency characteristic trend of the central rate variability parameter, the morphological change slope of the waveform characteristic parameter, and the cumulative change of the rhythm characteristic parameter through preset feature extraction rules. It also captures key nonlinear correlation information formed by the frequency characteristic trend, morphological change slope, and cumulative change in a continuous time window through a gating mechanism. This key nonlinear correlation information is transformed into a time-series trend feature that can reflect the dynamic change trend of the patient's health status, and the corresponding health risk probability prediction value is output by the output layer.

9. The patient health monitoring method based on dynamic electrocardiogram analysis according to claim 5, characterized in that, The assessment results, based on the preliminary classification results after bias correction and the probability of health risk, are output, including the following steps: Multiple preset threshold ranges for health risk probability are set, corresponding to normal state, potential abnormal state and acute abnormal state respectively; Based on the weighted fusion mechanism, the results of the preliminary classification after bias correction are fused with the health risk probability according to the preset weights. When the fusion result falls into a certain preset threshold range, the corresponding state evaluation result is output.

10. A patient health monitoring system based on dynamic electrocardiogram analysis, employing the patient health monitoring method based on dynamic electrocardiogram analysis as described in any one of claims 1-9, characterized in that, The system includes: The data acquisition module is configured to: synchronously acquire the patient's dynamic electrocardiogram data and related physiological parameters, wherein the dynamic electrocardiogram data includes multi-lead electrocardiogram signals continuously collected within a preset time period, and the related physiological parameters include exercise intensity, respiratory rate and body position change data; The data processing module is configured to: perform dynamic adaptive preprocessing on the dynamic electrocardiogram data based on the associated physiological parameters to obtain a standardized electrocardiogram signal; The feature extraction module is configured to: extract a multidimensional feature parameter set from the standardized electrocardiogram signal, and periodically sample the multidimensional feature parameter set according to a preset time window to generate a feature parameter sequence with timestamps; wherein the multidimensional feature parameter set includes: heart rate variability parameters, waveform morphology features, and rhythm feature parameters; The health assessment module is configured to: input the feature parameter sequence into a dynamic optimization analysis model, and perform a graded assessment of the patient's health status based on the dynamic optimization analysis model, generating an assessment result with a probability of health risk. The assessment result includes a normal state, a potentially abnormal state, an acute abnormal state, and corresponding abnormality tracing information. The dynamic optimization analysis model iteratively updates its model parameters by receiving real-time feedback of related physiological parameters.

Images