Method and System for Predicting Probability of Cardiovascular Events Based on Ambulatory Blood Pressure Monitoring

Abstract

The present invention discloses a method and system for predicting the occurrence probability of cardiovascular events based on ambulatory blood pressure monitoring, which relates to the field of artificial intelligence technology. In this method, a feature data set is constructed according to the ambulatory blood pressure sequence features and personal information features; the feature data set is input into a trained ambulatory blood pressure encoder to output a hidden feature matrix; according to a preset time, the hidden feature matrix is divided into a first-half-day hidden feature and a second-half-day hidden feature; based on the first-half-day hidden feature and the second-half-day hidden feature, a morning-time hidden feature and a night-time hidden feature are obtained; the morning-time hidden feature and the night-time hidden feature are input into a feature fusion and occurrence probability output module to obtain the occurrence probabilities of cardiovascular events corresponding to the morning time and the night time, wherein the feature fusion and occurrence probability output module is constructed based on a cross-attention mechanism. Implementing the technical solution of this application can accurately obtain the occurrence probabilities of cardiovascular events corresponding to the morning time and the night time.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and particularly to a method and system for predicting the occurrence probability of cardiovascular events based on ambulatory blood pressure monitoring. Background Art

[0002] Cardiovascular events (CVEs) refer to serious adverse events caused by heart and blood vessel diseases, such as heart attacks, strokes, sudden cardiac deaths, etc. The harm of cardiovascular events is very great. It is one of the leading causes of death globally, causing approximately 18 million deaths each year, and one-third of these deaths occur in people under 70 years old. In addition, cardiovascular events can also cause disabilities and severely reduce the quality of life of patients.

[0003] In research related to cardiovascular events, there is a lack of direct early warning of the occurrence probability of cardiovascular events, especially a lack of early warning of the occurrence probability of cardiovascular events at specific times such as early morning and night. This also makes it impossible for high-risk groups of cardiovascular events to obtain more comprehensive occurrence probability prompts and timely rescue measures.

[0004] Therefore, how to accurately obtain the occurrence probability of cardiovascular events corresponding to the early morning period and the night period has become a technical problem that needs to be solved urgently. Summary of the Invention

[0005] This application provides a method and system for predicting the occurrence probability of cardiovascular events based on ambulatory blood pressure monitoring, which can accurately obtain the occurrence probability of cardiovascular events corresponding to the early morning period and the night period.

[0006] In a first aspect, this application provides a method for predicting the occurrence probability of cardiovascular events based on ambulatory blood pressure monitoring. The method includes: obtaining the ambulatory blood pressure sequence features and personal information features of a target person, and constructing a feature data set according to the ambulatory blood pressure sequence features and personal information features;

[0007] Inputting the feature data set into a trained ambulatory blood pressure encoder to output a hidden feature matrix;

[0008] Dividing the hidden feature matrix into a first-half-day hidden feature and a second-half-day hidden feature according to a preset time;

[0009] Based on the first-half-day hidden feature and the second-half-day hidden feature, obtaining the early morning period hidden feature and the night period hidden feature corresponding to the target person;

[0010] Input the hidden features in the early morning period and the hidden features in the night period into a preset feature fusion and occurrence probability output module to obtain the occurrence probabilities of cardiovascular events corresponding to the target person in the early morning period and the night period, where the preset feature fusion and occurrence probability output module is constructed based on a cross-attention mechanism.

[0011] Optionally, before inputting the hidden features in the early morning period and the hidden features in the night period into a preset feature fusion and occurrence probability output module to obtain the occurrence probabilities of cardiovascular events corresponding to the target person in the early morning period and the night period, the method further includes:

[0012] Obtain the historical dynamic blood pressure sequence features and historical personnel information features of historical target persons, and construct a training feature data set according to the historical dynamic blood pressure sequence features and historical personnel information features;

[0013] Construct a dynamic blood pressure encoder based on a time series model or a non-time series model;

[0014] Input the training feature data set into a occurrence probability prediction model framework for training to obtain an initial occurrence probability prediction model. The occurrence probability prediction model framework includes the dynamic blood pressure encoder, a preset early morning encoder, a preset night encoder, and a preset feature fusion and occurrence probability output module;

[0015] Evaluate the initial occurrence probability prediction model by a cross-validation method to obtain the performance indicators of the initial occurrence probability prediction model; the performance indicators include cross-validation results;

[0016] Based on the cross-validation results, use a hyperparameter tuning method to adjust the hyperparameters of the initial occurrence probability prediction model to obtain the preset occurrence probability prediction model.

[0017] Optionally, the occurrence probabilities of cardiovascular events include the first cardiovascular event occurrence probability and the second cardiovascular event occurrence probability; inputting the hidden features in the early morning period and the hidden features in the night period into a preset feature fusion and occurrence probability output module to obtain the occurrence probabilities of cardiovascular events corresponding to the target person in the early morning period and the night period specifically includes:

[0018] Input the hidden features in the early morning period and the hidden features in the night period into the preset feature fusion and occurrence probability output module to obtain a first concatenated feature and a second concatenated feature;

[0019] Based on the first concatenated feature, obtain the first cardiovascular event occurrence probability corresponding to the target person in the early morning period;

[0020] Based on the second splicing feature, obtain the second probability of the target person having a cardiovascular event during the night period.

[0021] Optionally, the preset feature fusion and occurrence probability output module includes a first encoder, a second encoder, a first fully connected layer, a second fully connected layer, a first cross-attention module, and a second cross-attention module; the step of inputting the early morning period hidden feature and the night period hidden feature into the preset feature fusion and occurrence probability output module to obtain a first splicing feature and a second splicing feature specifically includes:

[0022] Input the early morning period hidden feature into the first encoder to obtain a first feature;

[0023] Input the night period hidden feature into the second encoder to obtain a second feature;

[0024] Input the first feature into the first fully connected layer to obtain a first intermediate feature;

[0025] Input the second feature into the second fully connected layer to obtain a second intermediate feature;

[0026] Input the first intermediate feature and the second feature into the first cross-attention module to obtain a first attention feature;

[0027] Input the second intermediate feature and the first feature into the second cross-attention module to obtain a second attention feature;

[0028] Splice the first feature and the first attention feature to obtain a first splicing feature;

[0029] Splice the second feature and the second attention feature to obtain a second splicing feature.

[0030] Optionally, the step of obtaining the early morning period hidden feature and the night period hidden feature of the target person based on the first half-day hidden feature and the second half-day hidden feature specifically includes:

[0031] Input the first half-day hidden feature into a preset early morning encoder to obtain the awakening time of the target person, and input the second half-day hidden feature into a preset night encoder to obtain the falling asleep time of the target person;

[0032] Extract the early morning period hidden feature from the first half-day hidden feature according to the awakening time, and extract the night period hidden feature from the second half-day hidden feature according to the falling asleep time.

[0033] Optionally, extracting the early morning period hidden feature from the first half day hidden feature according to the awakening moment, and extracting the night period hidden feature from the second half day hidden feature according to the falling asleep moment specifically includes:

[0034] Based on the awakening moment, determining a first time window corresponding to the awakening moment, where the awakening moment is the central moment of the first time window;

[0035] Extracting a first half day intermediate feature corresponding to the first time window from the first half day hidden feature;

[0036] Taking the first half day intermediate feature as the early morning period hidden feature;

[0037] Based on the falling asleep moment, determining a second time window corresponding to the falling asleep moment, where the falling asleep moment is the central moment of the second time window;

[0038] Extracting a second half day intermediate feature corresponding to the second time window from the second half day hidden feature;

[0039] Taking the second half day intermediate feature as the night period hidden feature.

[0040] Optionally, the dynamic blood pressure encoder is any one of a time series model or a non-time series model.

[0041] Optionally, constructing the feature data set according to the dynamic blood pressure sequence feature and the personnel information feature specifically includes:

[0042] Based on a preset evaluation criterion, determining a first disease label and a second disease label corresponding to the target person;

[0043] Constructing a basic data set based on the dynamic blood pressure sequence feature, the personnel information feature, the first disease label, and the second disease label;

[0044] Performing null value filling on the basic data set to obtain the feature data set.

[0045] Optionally, after obtaining the cardiovascular event occurrence probabilities corresponding to the target person in the early morning period and the night period, the method further includes:

[0046] When the cardiovascular event occurrence probability is greater than a preset disease occurrence probability, outputting a warning prompt message to the physician terminal.

[0047] In a second aspect of the present application, a cardiovascular event occurrence probability prediction system based on dynamic blood pressure monitoring is provided. The system includes: a data acquisition module, a feature acquisition module, and a feature fusion and output module;

[0048] The data acquisition module is used to obtain the dynamic blood pressure sequence features and personnel information features of the target person, and construct a feature dataset according to the dynamic blood pressure sequence features and personnel information features;

[0049] The feature acquisition module is used to input the feature dataset into the trained dynamic blood pressure encoder, and output a hidden feature matrix; according to a preset time, divide the hidden feature matrix into a first-half-day hidden feature and a second-half-day hidden feature; based on the first-half-day hidden feature and the second-half-day hidden feature, obtain the early morning period hidden feature and the night period hidden feature corresponding to the target person;

[0050] The feature fusion and output module is used to input the early morning period hidden feature and the night period hidden feature into a preset feature fusion and occurrence probability output module to obtain the cardiovascular event occurrence probabilities corresponding to the target person in the early morning period and the night period, wherein the preset feature fusion and occurrence probability output module is constructed based on a cross-attention mechanism.

[0051] Optionally, the system further includes: an early warning module;

[0052] The early warning module is used to output a warning prompt message to the physician terminal when the cardiovascular event occurrence probability is greater than a preset disease occurrence probability.

[0053] In a third aspect of the present application, an electronic device is provided, including a processor, a memory, a user interface, and a network interface. The memory is used to store instructions, the user interface and the network interface are used to communicate with other devices, and the processor is used to execute the instructions stored in the memory so that the electronic device executes the method according to any one of the first aspects of the present application.

[0054] In a fourth aspect of the present application, a computer-readable storage medium is provided. The computer-readable storage medium stores a computer program that can be loaded and executed by a processor to execute the method according to any one of the first aspects of the present application.

[0055] In summary, one or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages:

[0056] The present application provides a method for warning of the occurrence probability of cardiovascular events at specific times such as early morning and night based on dynamic blood pressure, which can predict whether a patient has a risk of early morning and nighttime cardiovascular adverse events, and prompt doctors of the blood pressure abnormalities of the patient in the early morning and night periods. Description of the Drawings

[0057] Figure 1It is one of the schematic flowcharts of a method for predicting the probability of cardiovascular events based on ambulatory blood pressure monitoring provided by an embodiment of the present application;

[0058] Figure 2 It is a schematic macro-frame diagram for predicting the probability of cardiovascular events provided by an embodiment of the present application;

[0059] Figure 3 It is the second of the schematic flowcharts of a method for predicting the probability of cardiovascular events based on ambulatory blood pressure monitoring provided by an embodiment of the present application;

[0060] Figure 4 It is the third of the schematic flowcharts of a method for predicting the probability of cardiovascular events based on ambulatory blood pressure monitoring provided by an embodiment of the present application;

[0061] Figure 5 It is the fourth of the schematic flowcharts of a method for predicting the probability of cardiovascular events based on ambulatory blood pressure monitoring provided by an embodiment of the present application;

[0062] Figure 6 It is the fifth of the schematic flowcharts of a method for predicting the probability of cardiovascular events based on ambulatory blood pressure monitoring provided by an embodiment of the present application;

[0063] Figure 7 It is a schematic macro-frame diagram of a feature fusion and probability of occurrence output module provided by an embodiment of the present application;

[0064] Figure 8 It is a schematic structural diagram of a system for predicting the probability of cardiovascular events based on ambulatory blood pressure monitoring provided by an embodiment of the present application;

[0065] Figure 9 It is a schematic structural diagram of an electronic device disclosed by an embodiment of the present application.

[0066] Explanation of reference numerals: 1. Data acquisition module; 2. Feature acquisition module; 3. Feature fusion and output module; 4. Probability of occurrence prediction model construction module; 5. Early warning module; 900. Electronic device; 901. Processor; 902. Communication bus; 903. User interface; 904. Network interface; 905. Memory. Detailed implementation manners

[0067] In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of this specification. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments.

[0068] In the description of the embodiments of the present application, words such as "for example" or "for illustration" are used to give examples, illustrations or explanations. Any embodiment or design solution described as "for example" or "for illustration" in the embodiments of the present application should not be construed as being more preferred or having more advantages than other embodiments or design solutions. Rather, the use of words such as "for example" or "for illustration" is intended to present relevant concepts in a specific manner.

[0069] In the description of the embodiments of the present application, the term "a plurality of" means two or more. For example, a plurality of systems means two or more systems, and a plurality of screen terminals means two or more screen terminals. In addition, the terms "first" and "second" are only used for descriptive purposes and cannot be construed as indicating or implying relative importance or implicitly specifying the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include one or more of such features. The terms "comprise", "include", "have" and their variants all mean "including but not limited to", unless otherwise specifically emphasized in other ways.

[0070] The present application provides a method for predicting the probability of cardiovascular events based on ambulatory blood pressure monitoring. Referring to Figure 1 , which shows one of the schematic flowcharts of a method for predicting the probability of cardiovascular events based on ambulatory blood pressure monitoring provided by the embodiments of the present application. The method includes steps S1 - S5, and the above steps are as follows:

[0071] Based on the descriptions of steps S1 - S5, for the convenience of understanding the overall solution, referring to Figure 2 , which shows a schematic diagram of the macroscopic framework for predicting the probability of cardiovascular events provided by the embodiments of the present application.

[0072] Step S1: Obtain the ambulatory blood pressure sequence features and personal information features of the target person, and construct a feature data set according to the ambulatory blood pressure sequence features and personal information features.

[0073] Specifically, in the process of implementing step S1, first determine the ambulatory blood pressure sequence features and personal information features of the target person as the basic data. The feature data set includes the ambulatory blood pressure sequence features and personal information features of the target person. The ambulatory blood pressure sequence features include blood pressure and heart rate, and the personal information features include age, gender, weight, and medical history; these data include physiological parameters such as blood pressure and heart rate and basic information such as age, gender, weight, and medical history, which are key factors for evaluating the probability of cardiovascular disease occurrence.

[0074] By implementing step S1, a comprehensive, accurate and complete feature data set can be constructed, which provides a solid foundation for subsequent prediction of the probability of cardiovascular events.

[0075] It should be noted that the specific implementation of step S1 will be described in detail in the subsequent embodiments.

[0076] Step S2: Input the feature dataset into the trained ambulatory blood pressure encoder to output a hidden feature matrix.

[0077] Specifically, in step S2, the preprocessed feature dataset is input into the already trained ambulatory blood pressure encoder. It should be noted that the ambulatory blood pressure encoder can be any one of a time series model or a non-time series model.

[0078] The purpose of this step is to convert the original physiological and personal information data into a hidden feature matrix that is more conducive to subsequent analysis. The selection of the ambulatory blood pressure encoder for this conversion is based on its ability to capture complex patterns and correlations in time series data.

[0079] When implementing this step, it is first necessary to determine the specific type of the ambulatory blood pressure encoder. Time series models such as long short-term memory networks (LSTM) or recurrent neural networks (RNN) are particularly suitable for processing time series data because they can capture the time dependence and dynamic changes in the data. Non-time series models, such as multi-layer perceptrons (MLP) or Transformers, although they do not directly handle the time series dependence, can still capture time series features through appropriate data preprocessing (such as the integration of time windows). This selection will be determined according to the specific application scenario and data characteristics.

[0080] The ambulatory blood pressure encoder outputs a hidden feature matrix by learning the patterns in the dataset. These hidden features comprehensively reflect the dynamic changes in the patient's blood pressure and its association with personal basic information. In this way, the original data is transformed into a form that is more suitable for machine learning analysis and probability of occurrence assessment. The output hidden feature matrix provides detailed and encoded information for the subsequent steps, which is the basis for accurately predicting the probability of cardiovascular events.

[0081] The effect of this implementation is that it can greatly improve the expressive ability of the data, converting simple physiological and personal information into a complex data form that can reflect the deep changes in the patient's health status. This not only improves the accuracy of the model in predicting the probability of future cardiovascular events but also makes it possible for a more detailed probability of occurrence assessment, thus supporting medical decision-making and the implementation of early intervention measures.

[0082] Step S3: Divide the hidden feature matrix into a morning hidden feature and an afternoon hidden feature according to the preset time.

[0083] Specifically, in step S3, the hidden feature matrix output by the ambulatory blood pressure encoder is segmented into a first-half-day hidden feature and a second-half-day hidden feature according to a preset time. This step is to divide the continuous 24-hour data into key time periods related to biological rhythms for more accurate health condition analysis and probability prediction of occurrence.

[0084] The necessity of step S3 lies in that since the physiological state of the human body has significant periodic changes during a day, and these changes are closely related to the probability of occurrence of cardiovascular events. For example, the incidence rate of cardiovascular events is usually higher in the early morning, which is related to the change of hormone levels in the body when waking up in the morning. Therefore, dividing the hidden feature matrix into first-half-day and second-half-day features helps to capture the changes in the probability of occurrence caused by these biological rhythms respectively.

[0085] Therefore, when specifically implementing step S3, it is first necessary to determine the middle time, usually the 12-hour point, to divide the 24-hour data into a first-half-day hidden feature from midnight to noon and a second-half-day hidden feature from noon to the next midnight. Such a division enables the data to better correspond to the patient's daily activities and physiological cycle in model analysis.

[0086] Step S4: Based on the first-half-day hidden feature and the second-half-day hidden feature, obtain the early morning period hidden feature and the night period hidden feature corresponding to the target person.

[0087] Specifically, in step S4, the segmented first-half-day hidden feature and second-half-day hidden feature are processed by a preset early morning encoder and a night encoder to further determine the personalized waking time and falling asleep time of the target person, and accordingly extract the corresponding early morning period hidden feature and night period hidden feature. The implementation of this step is crucial for accurately evaluating the probability of occurrence of cardiovascular events, because it allows the model to pay special attention to the peak periods of individual physiological activities, and these periods are often closely related to the increased probability of occurrence of cardiovascular events.

[0088] The necessity of step S4 lies in that generally, the probability of occurrence of cardiovascular events such as myocardial infarction increases significantly within a few hours after waking up in the morning, and the period from midnight to a few hours after falling asleep is also a critical period. By accurately identifying and analyzing the physiological data of these specific time periods, these potential probability periods can be predicted and prevented more effectively.

[0089] It should be noted that the specific implementation manner of step S4 will be described in detail in the subsequent embodiments.

[0090] Step S5: Input the hidden features in the early morning period and the hidden features in the night period into a preset feature fusion and occurrence probability output module to obtain the occurrence probabilities of cardiovascular events corresponding to the target person in the early morning period and the night period, where the preset feature fusion and occurrence probability output module is constructed based on the cross-attention mechanism.

[0091] Specifically, in Step S5, the hidden features in the early morning period and the hidden features in the night period obtained from the previous steps are input into a preset feature fusion and occurrence probability output module, which is constructed based on the cross-attention mechanism, with the aim of obtaining the occurrence probabilities of cardiovascular events of the target person in the early morning period and the night period. This step is the core of the entire warning system because it directly outputs the final prediction results, which are used to evaluate and possibly trigger preventive measures.

[0092] The necessity of Step S5 lies in that the prediction of the occurrence probability of cardiovascular events needs to consider temporal correlation and individual differences, especially at different stages of daily life, such as early morning awakening and night sleep. Through data analysis of specific time periods, high-occurrence probability periods can be more accurately identified, thereby achieving more effective prevention and intervention. The cross-attention mechanism plays a key role here. By strengthening the model's understanding of the differences and correlations between early morning and night data, it improves the accuracy and relevance of the prediction.

[0093] It should be noted that the specific implementation manner of Step S5 will be described in detail in the subsequent embodiments.

[0094] In a possible implementation manner, before Step S5, the method further includes the following steps:

[0095] Obtain the historical dynamic blood pressure sequence features and historical personnel information features of historical target persons, and construct a training feature data set based on the historical dynamic blood pressure sequence features and historical personnel information features.

[0096] Specifically, in this step, the task is to obtain the historical dynamic blood pressure sequence features and historical personnel information features of historical target persons, and construct a training feature data set based on these historical data. This step is to ensure that the cardiovascular event warning system can effectively learn and model construction based on historical data, thereby improving the accuracy and reliability of the prediction model.

[0097] The specific method for obtaining the historical dynamic blood pressure sequence features and historical personnel information features of historical target persons and constructing a training feature data set based on the historical dynamic blood pressure sequence features and historical personnel information features is the same as the implementation method of Step S1 in the foregoing embodiments, so it will not be elaborated here too much.

[0098] Construct a dynamic blood pressure encoder based on a time series model or a non-time series model.

[0099] Specifically, in this step, the task is to construct a dynamic blood pressure encoder based on a time-series model or a non-time-series model. The purpose of this step is to create a tool that can effectively extract features from dynamic blood pressure data and related health information to support subsequent analysis and prediction of the probability of cardiovascular events.

[0100] In specific operations, first, a suitable model architecture is selected according to the characteristics of the collected historical dataset. If the data shows obvious time dependence, such as the periodic changes of blood pressure and heart rate over time, it is more appropriate to use a time-series model such as LSTM or RNN; if the focus is on extracting features from data at each time point, MLP or CNN may be selected. Then, the model is designed and trained, including setting the number of network layers, neurons, activation functions, and optimization algorithms. The design and training of the above models are determined according to the actual application scenario requirements and are prior art. During the model training process, the historical feature dataset is used to train the encoder, and the parameters are adjusted to achieve the best learning effect, and a trained dynamic blood pressure encoder is obtained.

[0101] The training feature dataset is input into the probability prediction model framework for training to obtain an initial probability prediction model. The probability prediction model framework includes a dynamic blood pressure encoder, a preset early morning encoder, a preset night encoder, a preset feature fusion, and a probability output module.

[0102] Specifically, in this step, the task is to input the training feature dataset into the probability prediction model framework for training. This framework includes a dynamic blood pressure encoder, a preset early morning encoder, a preset night encoder, and a preset feature fusion and probability output module. The purpose of this step is to create a comprehensive prediction model that can accurately predict the probability of cardiovascular events based on the data features obtained from different encoders and modules.

[0103] In specific operations, first, it is necessary to ensure that all encoders and modules have been designed and optimized according to the requirements of pre-training. Then, the training feature dataset integrating dynamic blood pressure features, personal basic information features, and other relevant health indicators is input into the model framework. This step involves all aspects of model training, including selecting a suitable loss function and optimization algorithm, setting a reasonable training period, and making necessary hyperparameter adjustments to ensure that the model can effectively learn and capture the key patterns and relationships in the data.

[0104] Through in-depth learning of the training feature dataset, the prediction model can establish complex data associations and pattern recognition, which are indispensable for probability assessment and prediction. In addition, by comprehensively utilizing the outputs of multiple encoders and feature fusion modules, the model can provide a comprehensive prediction of the probability of cardiovascular events, and the accuracy of this prediction is much higher than that of a single data source or simple analysis methods.

[0105] The initial probability prediction model of occurrence is evaluated using the cross-validation method to obtain the performance metrics of the initial probability prediction model of occurrence; the performance metrics include cross-validation results.

[0106] Specifically, in the fields of machine learning and data science, evaluating the performance of a model and validating its predictive ability are crucial steps. Through cross-validation, the model is trained and tested on multiple data subsets, which can effectively evaluate its stability and accuracy, and avoid data contingency and overfitting. Cross-validation can provide a comprehensive evaluation to ensure that the model can achieve the expected effect in practical applications.

[0107] Therefore, in this step, the task is to evaluate the initial probability prediction model of occurrence using the cross-validation method to obtain the performance metrics of the initial probability prediction model of occurrence. This step is to ensure that the model has good generalization ability on different data samples and to verify its effect of predicting the probability of cardiovascular events.

[0108] In specific operations, first, an appropriate cross-validation scheme is set, such as k-fold cross-validation, where the dataset is divided into k smaller subsets. The model is trained on k - 1 subsets and tested on the remaining one subset. This process is repeated k times, and each time a different subset is selected as the test set to evaluate the performance of the model. After each test, the performance metrics of the model, such as accuracy, recall rate, area under the ROC curve (AUC), etc., are recorded.

[0109] Based on the cross-validation results, the hyperparameters of the initial probability prediction model of occurrence are adjusted using the hyperparameter tuning method to obtain the preset probability prediction model of occurrence.

[0110] Specifically, hyperparameter tuning is an important step in the development process of machine learning models. By adjusting the hyperparameters of the model, such as learning rate, number of neural network layers, number of nodes, batch size, etc., it can significantly affect the learning ability and final performance of the model. Appropriate hyperparameter settings can help the model learn more effectively from the data, avoid overfitting or underfitting, and thus improve the performance and stability of the model in practical applications.

[0111] Therefore, in this step, the task is to adjust the initial occurrence probability prediction model using a hyperparameter tuning method based on the cross-validation results to obtain an optimized occurrence probability prediction model. This step is to ensure that the model can execute its task with the best configuration and improve the accuracy and efficiency of prediction.

[0112] In specific operations, first, evaluate the current configuration of the model according to the performance metrics obtained during the cross-validation process. Then, select an appropriate hyperparameter tuning strategy, such as grid search, random search, or a method based on Bayesian optimization, to systematically explore the effects of different hyperparameter combinations. These methods search within a predefined hyperparameter space to find the parameter combination that can maximize the model performance metrics (such as accuracy, AUC, etc.).

[0113] After hyperparameter tuning, the model usually needs to be retrained on the same training dataset to verify the effect of the new parameter settings. The optimized model is evaluated again for performance to ensure that the adjustments made have indeed improved the model's prediction ability.

[0114] In a possible implementation, the cardiovascular event occurrence probability includes the first cardiovascular event occurrence probability and the second cardiovascular event occurrence probability. Refer to Figure 5 , which shows the fourth flowchart of a method for predicting the occurrence probability of cardiovascular events based on ambulatory blood pressure monitoring provided by an embodiment of the present application. Step S5 specifically includes steps S51 - S53:

[0115] Step S51: Input the hidden features in the early morning period and the hidden features in the night period into a preset feature fusion and occurrence probability output module to obtain a first concatenated feature and a second concatenated feature.

[0116] Specifically, in step S51, the task is to input the hidden features in the early morning period and the hidden features in the night period into a preset feature fusion and occurrence probability output module, and then obtain a first concatenated feature and a second concatenated feature. This step is to effectively integrate and analyze the key features extracted from different time periods to predict the occurrence probability of cardiovascular events of the target person in the early morning and night periods.

[0117] By synthesizing and strengthening the features extracted from the early morning and night periods, the model can more comprehensively evaluate the occurrence probability of cardiovascular events of the patient during these critical periods. This method not only improves the overall accuracy of occurrence probability prediction but also makes the prediction results more detailed.

[0118] It should be noted that the specific implementation manner of step S51 will be described in detail in the subsequent embodiments.

[0119] Step S52: Based on the first concatenated feature, obtain the first cardiovascular event occurrence probability corresponding to the target person in the early morning period.

[0120] Specifically, in the specific operation, first, the first spliced feature obtained through feature fusion and processing is input into a preset occurrence probability evaluation model. This model usually includes machine learning algorithms such as logistic regression, support vector machine, or deep learning network, and these algorithms can predict the occurrence probability of the first cardiovascular event according to the input features. The model will analyze the information in the first spliced feature, such as the patterns and trends extracted from physiological data, and the important physiological signals obtained from the cross-attention mechanism, to calculate the occurrence probability of cardiovascular events during the early morning period.

[0121] Step S53: Based on the second spliced feature, obtain the occurrence probability of the second cardiovascular event corresponding to the target person during the night period.

[0122] Specifically, first, use the second spliced feature processed by the feature fusion and occurrence probability output module. These features integrate the information extracted and enhanced from the night encoder, related fully connected layers, and cross-attention module. Then, input these features into a pre-trained prediction model, which may use machine learning algorithms such as random forest, gradient boosting machine, or deep learning network to predict the occurrence probability of the second cardiovascular event. In this process, the model will analyze the patterns and correlations in these features, such as heart rate variability, blood pressure fluctuations, etc., to predict the possibility of cardiovascular events at night.

[0123] In a possible implementation manner, the preset feature fusion and occurrence probability output module includes a first encoder, a second encoder, a first fully connected layer, a second fully connected layer, a first cross-attention module, and a second cross-attention module. Refer to Figure 6 , which shows the fifth schematic diagram of the process of a method for predicting the occurrence probability of cardiovascular events based on ambulatory blood pressure monitoring provided by an embodiment of the present application. Step S51 specifically includes steps S511-S518:

[0124] Based on the descriptions of steps S511-S518, for a better understanding of the overall solution, refer to Figure 7 , which shows a schematic diagram of the macroscopic framework of a feature fusion and occurrence probability output module provided by an embodiment of the present application.

[0125] Step S511: Input the hidden feature during the early morning period into the first encoder to obtain the first feature.

[0126] Specifically, in step S511, the task is to input the hidden feature during the early morning period into the first encoder to obtain the first feature. The implementation of this step is to extract deeper and more refined information through advanced encoding processing, which is crucial for accurately predicting the occurrence probability of cardiovascular events during the early morning period.

[0127] In specific operations, it is first necessary to ensure that the hidden features in the early morning accurately reflect the physiological data collected within two hours before and after waking up, such as the time series data of heart rate and blood pressure. These data are input into the first encoder, which usually adopts an advanced neural network structure such as a deep learning model, including but not limited to a convolutional neural network (CNN) or a recurrent neural network (RNN). These models are good at processing time series data and extracting complex features, so that the first encoder can refine more detailed and key information from the original hidden features to form the first feature.

[0128] The first encoder can generate the first feature containing deeper information by processing the input hidden features through advanced algorithms. These features more precisely reflect the physiological state and potential occurrence probability during the early morning period. For example, by identifying specific heart rate change patterns or abnormal increases in blood pressure, the model can more accurately predict the occurrence probability of cardiovascular events that may occur in the early morning.

[0129] Step S512: Input the hidden features in the night period into the second encoder to obtain the second feature.

[0130] Specifically, in step S512, the task is to input the hidden features in the night period into the second encoder to obtain the second feature. The purpose of this step is to further process and refine the hidden features in the night period using advanced encoding techniques to enhance the model's ability to identify and predict the factors of the occurrence probability of cardiovascular events at night.

[0131] In specific operations, it is first necessary to ensure that the hidden features in the night period accurately reflect the relevant physiological data collected within two hours before and after falling asleep. These data are input into the second encoder, which usually adopts an advanced neural network structure such as a deep learning model, including but not limited to a convolutional neural network (CNN) or a recurrent neural network (RNN). These models are good at processing time series data and extracting complex features, so that the second encoder can refine more detailed and key information from the original hidden features to form the second feature.

[0132] The second encoder not only improves the depth of data processing but also enhances the model's ability to identify the potential factors of the occurrence probability of cardiovascular events at night. For example, by identifying specific heart rate slowdown patterns or abnormal decreases in blood pressure, the second feature can help the model more accurately predict the occurrence probability of cardiovascular events that may occur at night.

[0133] Step S513: Input the first feature into the first fully connected layer to obtain the first intermediate feature.

[0134] Specifically, in step S513, the task is to input the first feature obtained from the first encoder into the first fully connected layer to obtain the first intermediate feature. The purpose of this step is to further process and refine the features in the early morning period, enhance the model's understanding and processing ability of these data, and thus improve the accuracy of predicting the probability of cardiovascular events.

[0135] In specific operations, first, the first feature, that is, the feature set processed by the first encoder, is input into the first fully connected layer. This fully connected layer consists of multiple neurons, and each neuron can learn different aspects and correlations in the data and optimize the output by adjusting weights and biases. Appropriate activation functions (such as ReLU or Sigmoid) are set to ensure non-linear feature transformation and better learning effects.

[0136] The processing of the first fully connected layer not only enhances the feature expression ability but also improves the fineness and accuracy of the prediction model when analyzing and evaluating the probability of cardiovascular events in the early morning. For example, it can help the model identify specific physiological change patterns, such as the feature of a rapid increase in blood pressure when waking up, which are key factors in judging the probability of cardiovascular events. Further feature processing through the fully connected layer significantly improves the prediction ability of the model, enabling it to more accurately identify which patients are at a higher risk of cardiovascular events in the early morning period.

[0137] Step S514: Input the second feature into the second fully connected layer to obtain the second intermediate feature.

[0138] Specifically, in step S514, the task is to input the second feature obtained from the second encoder into the second fully connected layer to obtain the second intermediate feature. The purpose of this step is to further process and refine the features in the night period, and improve the prediction accuracy of the probability of cardiovascular events at night by enhancing the model's processing ability of these data.

[0139] In specific operations, first, the second feature, that is, the feature set of the night period processed by the second encoder, is input into the second fully connected layer. This layer consists of a series of neurons, and each neuron can learn different features and patterns in the data. In this process, choosing appropriate activation functions (such as ReLU or Sigmoid) is crucial, and these functions help achieve non-linear feature transformation and optimize the learning effect.

[0140] The second fully connected layer not only enhances the expressiveness of features but also improves the meticulousness and accuracy of the prediction model in dealing with the probability of nocturnal cardiovascular events. For example, this layer can help the model better understand and identify patterns such as specific heart rate slowdowns or blood pressure drops that are unique to the night, which may be omens of impending cardiovascular events. Through the efficient feature processing of this layer, the model's prediction ability for the probability of cardiovascular events during the night period is significantly improved.

[0141] Step S515: Input the first intermediate feature and the second feature into the first cross-attention module to obtain the first attention feature.

[0142] Specifically, in step S515, the task is to input the first intermediate feature and the second feature into the first cross-attention module to obtain the first attention feature. The execution of this step is crucial because it uses the cross-attention mechanism to enhance the model's ability to integrate and analyze features from different time periods (such as early morning and night), which is essential for improving the overall accuracy and meticulousness of predicting the probability of cardiovascular events.

[0143] In the specific operation, first, the first intermediate feature (the output after processing the data of the early morning period) and the second feature (the output directly from the night encoder) are input into the first cross-attention module. The first cross-attention module automatically learns which feature combinations are most informative for the prediction model by calculating the relationship weights between features. In this process, attention mechanisms such as those in Transformer may be used, which adjust and optimize the flow and processing of information by calculating the relationships between Query, Key, and Value.

[0144] The first cross-attention module can more accurately map the complex physiological patterns that affect the probability of cardiovascular events by finely tuning the interaction between features. For example, this module can identify the impact of a rapid increase in heart rate in the early morning on blood pressure fluctuations at night, or vice versa, thus providing key physiological clues for early warning of cardiovascular events. In addition, through this fine feature integration and analysis, the model's prediction ability is significantly improved.

[0145] Step S516: Input the second intermediate feature and the first feature into the second cross-attention module to obtain the second attention feature.

[0146] Specifically, in step S516, the task is to input the second intermediate feature and the first feature into the second cross-attention module to obtain the second attention feature. This step is executed to utilize the cross-attention mechanism to enhance the model's ability to analyze and integrate features from different time periods, especially to combine night data with early morning data, thereby improving the comprehensiveness and accuracy of cardiovascular event probability prediction.

[0147] In the specific operation, first, the second intermediate feature (the output after processing night-time data) and the first feature (the output from the early morning encoder) are input into the second cross-attention module. Similarly, the second cross-attention module automatically learns which feature combinations are most informative for the prediction model by calculating the relationship weights between features. In this process, an attention mechanism such as that in Transformer may be used, which adjusts and optimizes the flow and processing of information by calculating the relationships between Query, Key, and Value.

[0148] The second cross-attention module enables the model to more meticulously explore and utilize the potential connections between night and early morning data, enhancing the understanding of factors related to cardiovascular event probability, such as the relationship between sleep quality and morning blood pressure peaks. This in-depth feature integration significantly improves the prediction accuracy of the model.

[0149] Step S517: Concatenate the first feature and the first attention feature to obtain the first concatenated feature.

[0150] Specifically, in step S517, the task is to concatenate the first feature and the first attention feature to obtain the first concatenated feature. The implementation of this step is to comprehensively utilize the finely processed early morning period data and the relevant features enhanced by the cross-attention module, thereby improving the prediction accuracy of cardiovascular events in the early morning.

[0151] In cardiovascular event prediction, integrating features from different processing stages is a key strategy to improve the accuracy of the prediction model. Concatenating the first feature and the first attention feature can fuse the information directly extracted from the original data and the information emphasized by the cross-attention mechanism. This method can more comprehensively reflect the cardiovascular health status during the early morning period, especially those subtle physiological changes that may indicate the probability of occurrence.

[0152] In the specific operation, first, the first feature (directly generated by the first encoder, reflecting the processing result based on the original data) and the first attention feature (generated by the first cross-attention module, highlighting the correlation and importance with the features of other time periods) are input into a concatenation operation in parallel. This operation simply combines the two sets of feature vectors element-wise into a longer vector, that is, the first concatenated feature, thus ensuring that the information extracted from the two different processing flows can be fully utilized in subsequent model training and prediction.

[0153] By concatenating the first feature and the first attention feature, the formed first concatenated feature provides a more abundant and detailed data basis for the model, enhancing the model's ability to capture the factors of the probability of cardiovascular events in the early morning period. This feature fusion strategy enables the prediction model to more accurately identify the high-probability occurrence states in the early morning, such as abnormal blood pressure increase or heart rate acceleration, etc., thereby providing more effective early warnings.

[0154] Step S518: Concatenate the second feature and the second attention feature to obtain the second concatenated feature.

[0155] Specifically, in step S518, the task is to concatenate the second feature and the second attention feature to generate the second concatenated feature. The implementation of this step is to effectively fuse the direct feature of the night time period and the correlation feature strengthened by cross-attention, improving the accuracy and comprehensiveness of the model's prediction of the probability of cardiovascular events at night.

[0156] In the prediction of cardiovascular events, integrating the features of different processing stages is a key strategy to improve the accuracy of the prediction model. Concatenating the second feature and the second attention feature can fuse the information directly extracted from the original data and the information emphasized by the cross-attention mechanism. This method can more comprehensively reflect the cardiovascular health status in the night time period, especially those subtle physiological changes that may indicate the probability of occurrence.

[0157] In the specific operation, similar to step S517, first, the second feature (generated by the second encoder, reflecting the processing result of the basic physiological data in the night time period) and the second attention feature (generated by the second cross-attention module, emphasizing the important connections and differences between the night and early morning features) are input into a concatenation operation in parallel. This operation simply combines the two sets of feature vectors element-wise into a longer vector, that is, the second concatenated feature, thus ensuring that the information extracted from the two different processing flows can be fully utilized in subsequent model training and prediction.

[0158] By splicing the second feature and the second attention feature, the second spliced feature provides a richer and more detailed data basis for the model, enhancing the model's ability to capture the probability factors of cardiovascular disease during the night. This feature fusion strategy enables the prediction model to more accurately identify the high probability of occurrence at night, such as abnormal heart rate or sudden changes in blood pressure during sleep, thereby providing more effective warnings.

[0159] In one possible implementation, refer to Figure 3 , which shows a second flow chart of a method for predicting the probability of occurrence of cardiovascular events based on dynamic blood pressure monitoring provided in an embodiment of the present application. Step S4 specifically includes steps S41-S42:

[0160] Step S41: input the latent features of the first half of the day into a preset morning encoder to obtain the awakening time of the target person, and input the latent features of the second half of the day into a preset night encoder to obtain the sleeping time of the target person.

[0161] Specifically, studies have shown that the occurrence of cardiovascular events such as myocardial infarction often peaks within a few hours after waking up in the morning, and the probability of occurrence also increases within a few hours after falling asleep. By accurately determining the moment of waking up and falling asleep, we can more effectively focus on these key periods and conduct more in-depth physiological data analysis, thereby effectively predicting and managing the probability of occurrence.

[0162] In step S41, the first half of the day latent features are input into a preset morning encoder to determine the target person's awakening time, and the second half of the day latent features are input into a preset night encoder to determine the target person's sleeping time. The purpose of this step is to accurately identify the key transition points in the individual's biorhythm, that is, the transition from the sleeping state to the awakening state and the transition from the awakening state to the sleeping state, which are closely related to the change in the probability of cardiovascular health.

[0163] In specific operations, we first need to use specially designed encoders to process the latent features of the first half of the day and the second half of the day. These encoders may be time series models based on deep learning, such as recurrent neural networks (RNNs) or long short-term memory networks (LSTMs), which can learn time-dependent patterns and features from continuous time series data. The morning encoder is specially adjusted to recognize physiological activity patterns in the morning period, such as changes in blood pressure and heart rate, while the night encoder focuses on changes in physiological activities in the evening.

[0164] Through the analysis of these encoders, the moment of awakening and the moment of falling asleep can be accurately marked, which is usually done by identifying significant points of changes in physiological parameters, such as increased blood pressure or increased heart rate. After determining these moments, the physiological data before and after these moments can be further analyzed to assess the probability of cardiovascular events.

[0165] Step S42: Extract the early morning period hidden features from the first half-day hidden features according to the awakening time, and extract the night period hidden features from the second half-day hidden features according to the falling asleep time.

[0166] Specifically, accurately identifying and utilizing the physiological data related to the awakening and falling asleep times can significantly improve the accuracy of cardiovascular disease prediction. Cardiovascular events, especially myocardial infarction and sudden cardiac death, are often significantly correlated with the daily physiological rhythm, especially the physiological changes when the human body transitions from the sleep state to the awakening state.

[0167] In step S42, by performing refined analysis and feature extraction to deeply mine the first half-day and second half-day hidden features, accurately locate and utilize the information of the awakening time and the falling asleep time, and then extract the corresponding early morning period hidden features and night period hidden features from the hidden feature matrix according to these critical moments. This step is a crucial part of the entire early warning system for the assessment of individual occurrence probabilities, because it allows the model to more specifically analyze the time periods most relevant to the occurrence probability of cardiovascular events.

[0168] In a possible implementation manner, referring to Figure 4 , which shows the third schematic flow diagram of a method for predicting the occurrence probability of cardiovascular events based on ambulatory blood pressure monitoring provided by an embodiment of the present application. Step S42 specifically includes steps S421 - S426:

[0169] Step S421: Based on the awakening time, determine the first time window corresponding to the awakening time, where the awakening time is the central time of the first time window.

[0170] Specifically, when the human body is at the awakening time, especially during the transition from the sleep state to the awakening state, it will experience a series of significant physiological changes, such as an increase in heart rate and blood pressure. These changes are closely related to the increased occurrence probability of cardiovascular events such as myocardial infarction.

[0171] In specific operations, first, the awakening time determined by the early morning encoder needs to be used as the center point. For example, if the encoder analyzes and identifies that the patient's awakening time is 6 am, then the first time window will be set from 4 am to 8 am. The selection of this time window is based on the significant physiological changes before and after awakening, which are key factors for evaluating the cardiovascular occurrence probability.

[0172] The effect of step S421 is to provide a highly targeted analysis window, enabling the model to focus on the key physiological data before and after the awakening time. By deeply analyzing the data within this time period, physiological markers related to a high cardiovascular occurrence probability, such as a sharp change in blood pressure or an irregular pattern of heart rate, can be accurately identified.

[0173] Step S422: extracting the intermediate features of the first half of the day corresponding to the first time window from the latent features of the first half of the day.

[0174] Specifically, in step S422, the task is to extract features corresponding to the first time window (the awakening moment and the two hours before and after) from the latent features of the first half of the day, and these features are defined as the latent features of the early morning period. The implementation of this step is crucial for in-depth analysis and utilization of physiological data related to awakening, because these data usually cover the physiological changes that are most relevant to the increased probability of cardiovascular events.

[0175] In the specific operation, the awakening time is first determined, for example, 6 am, and the first time window is determined from 4 am to 8 am. Then, the data in this time window is accurately selected from the latent feature data of the first half of the day processed by the dynamic blood pressure encoder. This includes cutting out the part corresponding to 4 am to 8 am from the latent feature matrix to ensure that this part of the data contains all the physiological changes related to awakening.

[0176] Step S423: Use the intermediate features of the first half of the day as latent features of the early morning period.

[0177] Specifically, in step S423, the first half-day intermediate features extracted from step S422 are set as the early morning period latent features. This step is implemented to transform the specific data analysis into a specific feature set that can be used for further prediction of the probability of occurrence, which is a key link in achieving accurate probability assessment in the cardiovascular event early warning system.

[0178] In the specific operation, we first ensure that the data segments before and after awakening have been correctly extracted from the latent features of the first half of the day, which usually involves identifying and segmenting the period from 4 am to 8 am in the time series data. These data are marked and stored as latent features of the early morning period, which contain the dynamic changes of physiological parameters such as heart rate and blood pressure during the awakening process. Such processing not only ensures the specificity and relevance of the data, but also provides the model with highly concentrated occurrence probability prediction factors.

[0179] Step S424: Based on the sleep onset time, determine a second time window corresponding to the sleep onset time, wherein the sleep onset time is the center time of the second time window.

[0180] Specifically, research has shown that the human body also undergoes a series of important physiological changes during the night's sleep onset phase, and these changes may be closely related to cardiovascular health. For example, a decrease in heart rate and blood pressure usually occurs during deep sleep, but any abnormal change pattern, such as nocturnal tachycardia or blood pressure fluctuations, may indicate an increased probability of cardiovascular events. By precisely defining and monitoring the data in this critical time window, it is possible to better identify early warning signs of nocturnal cardiovascular events.

[0181] In step S424, based on the sleep onset moment determined from the night encoder, a second time window corresponding to the sleep onset moment is set. Usually, this time window is centered around the sleep onset moment and includes data for two hours before and after the sleep onset. This step is to precisely capture the physiological data related to sleep onset, which is crucial in the assessment of the probability of cardiovascular events, especially when assessing the potential probability of nocturnal cardiovascular events.

[0182] In specific operations, once the night encoder identifies the patient's sleep onset moment, for example, 10 pm, the second time window is set to 8 pm to 12 midnight. Within this time window, relevant physiological data segments will be precisely extracted from the second half-day hidden features. This involves screening the time series data to ensure that the selected data segments can precisely cover the physiological activities before and after sleep onset.

[0183] The effect of step S424 is that by focusing on analyzing the key physiological changes before and after sleep onset, it is possible to effectively predict and manage the probability of nocturnal cardiovascular events. This method allows the model to gain in-depth understanding of the changes in nocturnal cardiovascular status, providing a scientific basis for medical providers to implement targeted intervention measures, especially for those patients who may have a high probability of nocturnal cardiovascular events.

[0184] Step S425: Extract the second half-day intermediate features corresponding to the second time window from the second half-day hidden features.

[0185] Specifically, in step S425, the task is to extract the features corresponding to the second time window (the sleep onset moment and two hours before and after it) from the second half-day hidden features, and these features are defined as the hidden features of the night period. The execution of this step is to ensure that the physiological changes related to sleep onset can be accurately captured and analyzed, which are crucial in the assessment of the probability of nocturnal cardiovascular events.

[0186] In specific operations, once the patient's bedtime is determined through the night encoder, for example, 10 p.m., the second time window is set from 8 p.m. to 12 a.m. During this time window, relevant physiological data segments are precisely selected and extracted from the second-half-day hidden features. This involves screening data within a specific time range from the time series to ensure that the selected data segments can accurately cover the physiological activities before and after falling asleep, with particular attention to key indicators such as heart rate decline and blood pressure changes.

[0187] Step S426: Use the second-half-day intermediate features as the hidden features for the night period.

[0188] In step S426, the task is to set the second-half-day intermediate features extracted from step S425 as the hidden features for the night period. This step is to use a specific set of physiological data for further analysis and prediction of the probability of nocturnal cardiovascular events. The determination and application of the hidden features for the night period are crucial because they concentrate the key physiological indicators that may affect cardiovascular health at night.

[0189] In specific operations, step S425 has ensured the precise extraction of the data segment from 8 p.m. to 12 a.m. from the second-half-day hidden features. This data segment is now officially defined as the hidden features for the night period, which includes the changing data of heart rate, blood pressure, etc. before and after falling asleep. This process ensures the concentration and relevance of the data, providing a scientific and accurate data basis for the assessment of the probability of nocturnal cardiovascular events.

[0190] In a possible implementation manner, step S1 specifically includes the following steps:

[0191] Based on a preset assessment criterion, determine the first disease label and the second disease label corresponding to the target person.

[0192] Specifically, the accurate definition of disease labels is fundamental and core to any medical prediction model because these labels are directly related to the accuracy of model training and the reliability of the final prediction. By clarifying the disease labels, it is ensured that precise predictions can be made for specific cardiovascular events during model development and subsequent analysis.

[0193] In specific operations, first, it is necessary to set the assessment criteria for cardiovascular events by collaborating with medical experts or referring to the latest clinical guidelines. These criteria are usually based on a series of clinical parameters such as electrocardiogram changes, blood pressure levels, cholesterol levels, past medical history, etc. For each participating patient, a comprehensive health examination and data collection are carried out according to these preset assessment criteria.

[0194] Next, based on the collected data, such as electrocardiogram results, blood test results, etc., one or more disease labels are assigned to each patient. For example, if a patient's electrocardiogram shows typical signals of myocardial infarction, then he / she will be labeled as an individual with a high probability of myocardial infarction (the first disease label). Similarly, if a patient has a history of sudden cardiac death or has factors that contribute to the occurrence probability of the disease, they will also be labeled accordingly.

[0195] The effect of this step is to provide an accurate target variable, i.e., the disease label, for subsequent data processing and model training, which helps the model learn and predict the occurrence probability of cardiovascular events more accurately. The disease label determined by this method makes the entire early warning system more targeted.

[0196] Construct a basic data set based on dynamic blood pressure sequence features, personal information features, the first disease label, and the second disease label.

[0197] Specifically, combine dynamic blood pressure sequence features, personal information features, and disease labels to construct a basic data set. This data integration provides comprehensive input for the dynamic blood pressure encoder, which helps to accurately extract hidden features in subsequent steps. The establishment of the basic data set is to ensure the integrity and multi-dimensionality of the input data, enabling the occurrence probability prediction model to capture the interrelationships between complex physiological and personal information.

[0198] This integration method can ensure that the mutual relevance and complexity of the data are fully utilized, providing sufficient information during the model training process to learn how to distinguish different health conditions and predict future cardiovascular events.

[0199] In specific operations, first integrate the dynamic blood pressure sequence features and personal information features of each collected patient into a data framework. The dynamic blood pressure sequence features may include, but are not limited to, data on blood pressure and heart rate changes within 24 hours, while the personal information features include basic and medical background information such as age, gender, weight, and medical history. After these data are integrated, the first disease label and the second disease label are added to each patient. Next, perform necessary data preprocessing on the integrated data set, such as filling in missing values, standardization, and regularization, to ensure the consistency and applicability of the data. The construction of this basic data set not only involves data collection and integration but also includes cleaning and formatting the data to make it suitable for advanced analysis and machine learning processing.

[0200] The effect of this step is to provide a rich, structured, and high-quality input data source for the cardiovascular disease prediction model. In this way, the prediction model can more effectively identify the key factors affecting cardiovascular events. In addition, this comprehensive data processing method enhances the representativeness of the data and the generalization ability of the prediction model, enabling the model to not only be applicable to the current research population but also provide reliable predictions in a wider population.

[0201] Fill in the null values in the basic data set to obtain the feature data set.

[0202] Specifically, during the data collection process, the problem of missing data is often encountered. Missing data may be caused by various reasons, such as measurement errors, data entry omissions, or sensor failures. Untreated null values will affect the quality and accuracy of data analysis, possibly causing the model to learn incorrect information and thus affecting the final prediction results. Therefore, appropriately handling these null values is crucial for ensuring the effectiveness of data analysis and subsequent model training.

[0203] In the specific operation, first evaluate the null values in the basic data set to identify which features have null values and the distribution of the null values. Select appropriate filling methods according to different types of features:

[0204] For the null values in the ambulatory blood pressure sequence features, consider using model fitting imputation methods, such as time series prediction, interpolation techniques, or filling based on the average value of adjacent time points, which helps to maintain the continuity and consistency of the data in the time series.

[0205] For the null values in the personal information features, such as age, weight, etc., the median or mode can be used for filling. This method is simple and effective and can reduce the impact of outliers on the overall distribution of the data set.

[0206] After performing these filling processes, the integrity of the data set will be significantly improved, creating a solid foundation for subsequent data analysis and model training. Through such data preprocessing, it is ensured that the data of all patients are complete, enabling the model to learn on the basis of complete data and improving the accuracy and robustness of the prediction model.

[0207] In a possible implementation manner, after step S5, the method further includes the following steps:

[0208] When the probability of the cardiovascular event occurring is greater than the preset disease occurrence probability, output a warning prompt message to the physician terminal.

[0209] Specifically, in this step, the task is to output a warning message to the physician terminal when it is determined that the probability of a cardiovascular event occurring in the target person is greater than a preset threshold. The implementation of this step is to ensure that any potential probability of cardiovascular events can be promptly identified and handled by medical professionals, so as to take appropriate intervention measures to reduce the probability of health problems for patients.

[0210] In specific operations, first, based on the trained probability prediction model, the probabilities of cardiovascular events occurring in the early morning and at night for each patient are evaluated. These probabilities are obtained from the final output of the model, including the first and second probabilities of cardiovascular events calculated through steps S51 to S53. Then, these probabilities are compared with the preset probability threshold. If any probability exceeds the threshold, the system will automatically generate a warning message and send the warning to the physician terminal through the medical information system, such as the hospital's electronic health record system or mobile medical application.

[0211] Refer to Figure 8 , which shows a schematic structural diagram of a cardiovascular event probability prediction system based on ambulatory blood pressure monitoring provided by an embodiment of the present application. The system includes: a data acquisition module 1, a feature acquisition module 2, and a feature fusion and probability output module 3;

[0212] The data acquisition module 1 is used to obtain the ambulatory blood pressure sequence features and personal information features of the target person, and construct a feature data set based on the ambulatory blood pressure sequence features and personal information features;

[0213] The feature acquisition module 2 is used to input the feature data set into the trained ambulatory blood pressure encoder and output a hidden feature matrix; according to the preset time, divide the hidden feature matrix into a first-half-day hidden feature and a second-half-day hidden feature; based on the first-half-day hidden feature and the second-half-day hidden feature, obtain the hidden features corresponding to the early morning period and the night period of the target person;

[0214] The feature fusion and output module 3 is used to input the hidden features corresponding to the early morning period and the night period into the preset feature fusion and probability output module to obtain the probabilities of cardiovascular events corresponding to the target person in the early morning period and the night period, where the preset feature fusion and output module 3 is constructed based on the cross-attention mechanism.

[0215] In a possible implementation, the system further includes: an occurrence probability prediction model construction module 4; the occurrence probability prediction model construction module 4 is used to obtain the historical dynamic blood pressure sequence features and historical personnel information features of historical target personnel, and construct a training feature data set according to the historical dynamic blood pressure sequence features and historical personnel information features; based on a time series model or a non-time series model, construct a dynamic blood pressure encoder; input the training feature data set into the occurrence probability prediction model framework for training to obtain an initial occurrence probability prediction model. The occurrence probability prediction model framework includes a dynamic blood pressure encoder, a preset early morning encoder, a preset night encoder, a preset feature fusion and occurrence probability output module; use the cross-validation method to evaluate the initial occurrence probability prediction model to obtain the performance indicators of the initial occurrence probability prediction model; the performance indicators include the cross-validation result; based on the cross-validation result, use the hyperparameter tuning method to adjust the hyperparameters of the initial occurrence probability prediction model to obtain a preset occurrence probability prediction model.

[0216] In a possible implementation, the feature fusion and output module 3 is further used to input the early morning period hidden features and the night period hidden features into the preset feature fusion and occurrence probability output module to obtain a first concatenated feature and a second concatenated feature; based on the first concatenated feature, obtain the first cardiovascular event occurrence probability corresponding to the target person during the early morning period; based on the second concatenated feature, obtain the second cardiovascular event occurrence probability corresponding to the target person during the night period.

[0217] In a possible implementation, the feature fusion and output module 3 is further used to input the early morning period hidden features into the first encoder to obtain a first feature; input the night period hidden features into the second encoder to obtain a second feature; input the first feature into the first fully connected layer to obtain a first intermediate feature; input the second feature into the second fully connected layer to obtain a second intermediate feature; input the first intermediate feature and the second feature into the first cross-attention module to obtain a first attention feature; input the second intermediate feature and the first feature into the second cross-attention module to obtain a second attention feature; concatenate the first feature and the first attention feature to obtain a first concatenated feature; concatenate the second feature and the second attention feature to obtain a second concatenated feature.

[0218] In a possible implementation, the feature acquisition module 2 is further used to input the first half-day hidden features into the preset early morning encoder to obtain the awakening time of the target person, and input the second half-day hidden features into the preset night encoder to obtain the sleep time of the target person; according to the awakening time, extract the early morning period hidden features from the first half-day hidden features, and according to the sleep time, extract the night period hidden features from the second half-day hidden features.

[0219] In a possible implementation manner, the feature acquisition module 2 is further configured to determine a first time window corresponding to the awakening moment based on the awakening moment, where the awakening moment is the central moment of the first time window; extract a first half-day intermediate feature corresponding to the first time window from the first half-day hidden features; use the first half-day intermediate feature as the early morning period hidden feature; determine a second time window corresponding to the falling asleep moment based on the falling asleep moment, where the falling asleep moment is the central moment of the second time window; extract a second half-day intermediate feature corresponding to the second time window from the second half-day hidden features; use the second half-day intermediate feature as the night period hidden feature

[0220] In a possible implementation manner, the data acquisition module 1 is further configured to determine a first disease label and a second disease label corresponding to the target person based on a preset evaluation criterion; construct a basic data set based on the dynamic blood pressure sequence feature, the person information feature, the first disease label, and the second disease label; perform null value filling on the basic data set to obtain a feature data set.

[0221] In a possible implementation manner, the system further includes: a warning module 5; the warning module 5 is configured to output a warning prompt message to the physician terminal when the probability of a cardiovascular event is greater than a preset disease occurrence probability.

[0222] It should be noted that: when the device provided in the above embodiment realizes its functions, only the above-mentioned division of each functional module is used for illustration. In practical applications, the above functions can be allocated to different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the device and method embodiments provided in the above embodiments belong to the same concept, and the specific implementation process is detailed in the method embodiment, which will not be repeated here.

[0223] This application also discloses an electronic device. Refer to Figure 9 , Figure 9 is a schematic structural diagram of an electronic device disclosed in an embodiment of this application. The electronic device 900 may include: at least one processor 901, at least one network interface 904, a user interface 903, a memory 905, and at least one communication bus 902.

[0224] Among them, the communication bus 902 is used to realize the connection and communication between these components.

[0225] Among them, the user interface 903 may include a display screen (Display) and a camera (Camera). Optionally, the user interface 903 may further include a standard wired interface and a wireless interface.

[0226] Among them, the network interface 904 may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface).

[0227] Among them, the processor 901 may include one or more processing cores. The processor 901 connects various parts within the entire server through various interfaces and lines. By running or executing instructions, programs, code sets, or instruction sets stored in the memory 905, and by calling the data stored in the memory 905, it executes various functions of the server and processes data. Optionally, the processor 901 may be implemented in at least one hardware form of digital signal processing (DSP), field-programmable gate array (FPGA), or programmable logic array (PLA). The processor 901 may integrate a combination of one or several of a central processing unit (CPU), a graphics processing unit (GPU), and a modem, etc. Among them, the CPU mainly processes the operating system, user interface, application programs, etc.; the GPU is responsible for rendering and drawing the content to be displayed on the display screen; the modem is used to process wireless communication. It can be understood that the above-mentioned modem may not be integrated into the processor 901 and may be implemented separately by a single chip.

[0228] Among them, the memory 905 may include random access memory (RAM) and may also include read-only memory. Optionally, the memory 905 includes a non-transitory computer-readable storage medium. The memory 905 can be used to store instructions, programs, code, code sets, or instruction sets. The memory 905 may include a program storage area and a data storage area. Among them, the program storage area may store instructions for implementing the operating system, instructions for at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the above-mentioned various method embodiments, etc.; the data storage area may store the data involved in the above-mentioned various method embodiments. Optionally, the memory 905 may also be at least one storage device located far from the aforementioned processor 901. Refer to Figure 9 , the memory 905, as a computer-readable storage medium, may include an operating system, a network communication module, a user interface module, and an application program.

[0229] In Figure 9In the electronic device 900 shown, the user interface 903 is mainly used to provide an interface for the user to input and obtain the data input by the user; and the processor 901 can be used to call an application program stored in the memory 905. When executed by one or more processors 901, the electronic device 900 is caused to execute the method(s) of one or more of the above-described embodiments. It should be noted that, for the foregoing method embodiments, for the sake of simple description, they are all expressed as a series of action combinations. However, those skilled in the art should know that this application is not limited by the described order of actions, because according to this application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0230] In the above embodiments, the descriptions of the respective embodiments have their own emphases. For the parts not detailed in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0231] In several implementation manners provided by this application, it should be understood that the disclosed device can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling or direct coupling or communication connection between each other can be through some service interfaces. The indirect coupling or communication connection of the device or unit can be in an electrical or other form.

[0232] The unit described as a separated component may or may not be physically separated, and the component displayed as a unit may or may not be a physical unit, that is, it may be located in one place, or may be distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0233] In addition, in each embodiment of this application, the functional units can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above integrated unit can be implemented in the form of hardware or in the form of a software functional unit.

[0234] When an integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable memory. Based on this understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present application. The aforementioned memory includes various media that can store program codes, such as USB flash drives, mobile hard disks, magnetic disks, or optical discs.

[0235] The above are only exemplary embodiments of the present disclosure, and the scope of the present disclosure cannot be limited thereby. That is, any equivalent changes and modifications made in accordance with the teachings of the present disclosure still fall within the scope covered by the present disclosure. After considering the specification and the disclosure of the practical truth, those skilled in the art will easily think of other implementation manners of the present disclosure.

[0236] The present application aims to cover any variations, uses, or adaptive changes of the present disclosure. These variations, uses, or adaptive changes follow the general principles of the present disclosure and include the common general knowledge or conventional technical means in the technical field not recorded in the present disclosure. The specification and the embodiments are only regarded as exemplary, and the scope and spirit of the present disclosure are defined by the claims.

Claims

1. A method for predicting the probability of cardiovascular events based on ambulatory blood pressure monitoring, characterized in that, The method includes: Obtaining the dynamic blood pressure sequence features and personal information features of the target person, and constructing a feature dataset according to the dynamic blood pressure sequence features and personal information features; Inputting the feature dataset into the trained dynamic blood pressure encoder to output a hidden feature matrix; Dividing the hidden feature matrix into a first-half-day hidden feature and a second-half-day hidden feature according to a preset time; Based on the first-half-day hidden feature and the second-half-day hidden feature, obtaining the early morning period hidden feature and the night period hidden feature corresponding to the target person; Inputting the early morning period hidden feature and the night period hidden feature into a preset feature fusion and occurrence probability output module to obtain the cardiovascular event occurrence probabilities corresponding to the target person in the early morning period and the night period, wherein the preset feature fusion and occurrence probability output module is constructed based on a cross-attention mechanism, and the preset feature fusion and occurrence probability output module includes a first encoder, a second encoder, a first fully connected layer, a second fully connected layer, a first cross-attention module, and a second cross-attention module; The obtaining the early morning period hidden feature and the night period hidden feature corresponding to the target person based on the first-half-day hidden feature and the second-half-day hidden feature specifically includes: Inputting the first-half-day hidden feature into a preset early morning encoder to obtain the awakening time of the target person, and inputting the second-half-day hidden feature into a preset night encoder to obtain the falling asleep time of the target person; Extracting the early morning period hidden feature from the first-half-day hidden feature according to the awakening time, and extracting the night period hidden feature from the second-half-day hidden feature according to the falling asleep time.

2. The method according to claim 1, wherein Before inputting the early morning period hidden feature and the night period hidden feature into a preset feature fusion and occurrence probability output module to obtain the cardiovascular event occurrence probabilities corresponding to the target person in the early morning period and the night period, the method further includes: Obtaining the historical dynamic blood pressure sequence features and historical personal information features of historical target persons, and constructing a training feature dataset according to the historical dynamic blood pressure sequence features and historical personal information features; Constructing a dynamic blood pressure encoder based on a time series model or a non-time series model; Inputting the training feature dataset into a occurrence probability prediction model framework for training to obtain an initial occurrence probability prediction model, where the occurrence probability prediction model framework includes the dynamic blood pressure encoder, a preset early morning encoder, a preset night encoder, and a preset feature fusion and occurrence probability output module; Evaluating the initial occurrence probability prediction model by a cross-validation method to obtain the performance index of the initial occurrence probability prediction model; the performance index includes the cross-validation result; Based on the cross-validation result, adjusting the hyperparameters of the initial occurrence probability prediction model by a hyperparameter tuning method to obtain a preset occurrence probability prediction model.

3. The method according to claim 1, wherein The probability of cardiovascular events includes the probability of the first cardiovascular event and the probability of the second cardiovascular event; inputting the hidden features in the early morning period and the hidden features in the night period into a preset feature fusion and probability output module to obtain the probability of cardiovascular events corresponding to the target person in the early morning period and the night period specifically includes: Inputting the hidden features in the early morning period and the hidden features in the night period into the preset feature fusion and probability output module to obtain a first spliced feature and a second spliced feature; Based on the first spliced feature, obtaining the probability of the first cardiovascular event corresponding to the target person in the early morning period; Based on the second spliced feature, obtaining the probability of the second cardiovascular event corresponding to the target person in the night period.

4. The method according to claim 3, characterized in that, The preset feature fusion and probability output module includes a first encoder, a second encoder, a first fully connected layer, a second fully connected layer, a first cross-attention module, and a second cross-attention module; inputting the hidden features in the early morning period and the hidden features in the night period into the preset feature fusion and probability output module to obtain a first spliced feature and a second spliced feature specifically includes: Inputting the hidden features in the early morning period into the first encoder to obtain a first feature; Inputting the hidden features in the night period into the second encoder to obtain a second feature; Inputting the first feature into the first fully connected layer to obtain a first intermediate feature; Inputting the second feature into the second fully connected layer to obtain a second intermediate feature; Inputting the first intermediate feature and the second feature into the first cross-attention module to obtain a first attention feature; Inputting the second intermediate feature and the first feature into the second cross-attention module to obtain a second attention feature; Splicing the first feature and the first attention feature to obtain a first spliced feature; Splicing the second feature and the second attention feature to obtain a second spliced feature.

5. The method according to claim 1, wherein Extracting the hidden features in the early morning period from the hidden features in the first half of the day according to the awakening time, and extracting the hidden features in the night period from the hidden features in the second half of the day specifically includes: Based on the awakening time, determining a first time window corresponding to the awakening time, where the awakening time is the central time of the first time window; Extracting a first-half-day intermediate feature corresponding to the first time window from the hidden features in the first half of the day; Using the first-half-day intermediate feature as the hidden features in the early morning period; Based on the falling asleep time, determining a second time window corresponding to the falling asleep time, where the falling asleep time is the central time of the second time window; Extracting a second-half-day intermediate feature corresponding to the second time window from the hidden features in the second half of the day; Using the second-half-day intermediate feature as the hidden features in the night period.

6. The method according to claim 1, wherein The dynamic blood pressure encoder is any one of a time series model or a non-time series model.

7. The method according to claim 1, characterized in that, Constructing a feature dataset according to the dynamic blood pressure sequence feature and the personnel information feature specifically includes: Based on a preset evaluation criterion, determine the first disease label and the second disease label corresponding to the target person; Construct a basic data set based on the dynamic blood pressure sequence features, the personal information features, the first disease label, and the second disease label; Fill in the null values in the basic data set to obtain the feature data set.

8. The method according to claim 1, wherein After obtaining the probabilities of cardiovascular events corresponding to the target person in the early morning period and the night period, the method further includes: When the probability of cardiovascular events is greater than the preset disease occurrence probability, output a warning prompt message to the physician terminal.

9. An auxiliary warning system for cardiovascular event warning, characterized in that, The system includes: a data acquisition module, a feature acquisition module, and a feature fusion and output module; the data acquisition module is used to acquire the dynamic blood pressure sequence features and personal information features of the target person, and construct a feature data set according to the dynamic blood pressure sequence features and personal information features; The feature acquisition module is used to input the feature data set into a trained dynamic blood pressure encoder, and output a hidden feature matrix; according to a preset time, divide the hidden feature matrix into a first half-day hidden feature and a second half-day hidden feature; based on the first half-day hidden feature and the second half-day hidden feature, obtain the early morning period hidden feature and the night period hidden feature corresponding to the target person; The feature fusion and output module is used to input the early morning period hidden feature and the night period hidden feature into a preset feature fusion and occurrence probability output module to obtain the probabilities of cardiovascular events corresponding to the target person in the early morning period and the night period, wherein the preset feature fusion and occurrence probability output module is constructed based on a cross-attention mechanism, and the preset feature fusion and occurrence probability output module includes a first encoder, a second encoder, a first fully connected layer, a second fully connected layer, a first cross-attention module, and a second cross-attention module; The obtaining the early morning period hidden feature and the night period hidden feature corresponding to the target person based on the first half-day hidden feature and the second half-day hidden feature specifically includes: Input the first half-day hidden feature into a preset early morning encoder to obtain the awakening time of the target person, and input the second half-day hidden feature into a preset night encoder to obtain the falling asleep time of the target person; According to the awakening time, extract the early morning period hidden feature from the first half-day hidden feature, and according to the falling asleep time, extract the night period hidden feature from the second half-day hidden feature.

10. The system according to claim 9, wherein The system further includes: a warning module; The warning module is used to output a warning prompt message to the physician terminal when the probability of cardiovascular events is greater than the preset disease occurrence probability.

11. An electronic device, characterized in that, It includes a processor (901), a memory (905), a user interface (903), and a network interface (904). The memory (905) is used to store instructions. The user interface (903) and the network interface (904) are used to communicate with other devices. The processor (901) is used to execute the instructions stored in the memory (905) so that the electronic device (900) executes the method according to any one of claims 1-8.

12. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions that, when executed, perform the method according to any one of claims 1-8.

Images