Sick and wounded state evaluation method and system based on vital sign real-time monitoring

Abstract

The invention provides a sick and wounded state evaluation method and system based on vital sign real-time monitoring. The method comprises the steps that vital sign data streams of the sick and wounded are collected in real time through wearable equipment, preliminary standardization processing is carried out, and a standardized vital sign data sequence is generated; based on the standardized vital sign data sequence, a dynamic Bayesian network algorithm is applied to establish a time sequence dependency relationship model, a wavelet transform technology is adopted to decompose a signal spectrum, and refined abnormal mode data is generated; based on the refined abnormal mode data, constructing a multi-level decision tree set by using a deep forest algorithm, analyzing and capturing a vital sign change trend by adopting a time sequence, and generating a personalized state evaluation report; and based on the personalized state evaluation report, determining an intervention measure suggestion, converting the intervention measure suggestion into an executable operation guide, and generating a sick and wounded state response plan. According to the technical scheme, accurate intervention measures and suggestions are provided, and the timeliness and accuracy of state evaluation of the sick and wounded are remarkably improved.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of medical health monitoring technology, and in particular to a method and system for assessing the status of an injured or sick person based on real-time monitoring of vital signs. Background Art

[0002] With the development of medical technology and the Internet of Things, real-time monitoring of the vital signs of the wounded and sick has become an important part of the modern medical and health management system. In hospitals, emergency sites, long-term care institutions and other scenarios, accurate assessment and timely response to the status of the wounded and sick are crucial. Through wearable devices, real-time collection of vital signs data streams can be achieved, and preliminary standardization processing can be performed to generate standardized vital signs data sequences. This not only improves the quality and reliability of the data, but also lays the foundation for subsequent complex analysis. However, this real-time monitoring system needs to have efficient data processing capabilities, which can complete the cleaning, standardization and feature extraction of large amounts of data in a short period of time to ensure timeliness and accuracy.

[0003] At present, there are many schemes for real-time monitoring and analysis of vital signs data of the wounded and sick. Traditional monitoring methods mainly rely on manual recording and simple statistical analysis, which makes it difficult to capture complex temporal dependencies and potential abnormal patterns. In recent years, some machine learning-based methods have been introduced, such as the dynamic Bayesian network algorithm, which is used to establish a temporal dependency model between vital sign parameters. These methods can identify potential abnormal vital sign patterns, and use wavelet transform technology to decompose the signal spectrum, refine abnormal signal features, and generate refined abnormal pattern data. In addition, the deep forest algorithm has also been used to construct a multi-level decision tree set, combining historical health records with current environmental factors for comprehensive evaluation, thereby improving the generalization ability and prediction accuracy of the model.

[0004] Although the existing schemes have improved the accuracy and efficiency of the status assessment of the wounded and sick to a certain extent, there are still some shortcomings. First, the data processing capacity of traditional methods is limited and cannot meet the needs of large-scale real-time data, resulting in data delays and information loss, which in turn leads to insufficient timeliness of assessment; second, although the existing machine learning methods can capture some temporal dependencies, they still do not have a deep enough understanding of complex physiological processes, especially when faced with multivariate interactions; finally, the existing schemes are relatively lacking in personalized intervention measures and fail to make full use of historical health records and current environmental factors for comprehensive assessment, resulting in the generated status response plans lack of pertinence and flexibility; therefore, there is an urgent need for a more intelligent and efficient method for the status assessment of the wounded and sick to meet the above challenges and provide better medical services, especially to improve the timeliness of assessment and ensure that the wounded and sick can receive rapid and accurate medical response. Summary of the invention

[0005] The embodiments of the present application provide a method and system for assessing the status of a patient based on real-time monitoring of vital signs, so as to solve the problem of insufficient timeliness of assessment in the prior art.

[0006] In a first aspect, an embodiment of the present application provides a method for assessing the status of a patient based on real-time monitoring of vital signs, comprising:

[0007] Through wearable devices, the patient's vital signs data stream is collected in real time, and preliminary standardization processing is performed to generate a standardized vital signs data sequence;

[0008] Based on the standardized vital sign data sequence, a dynamic Bayesian network algorithm is used to establish a temporal dependency model between vital sign parameters, identify potential abnormal vital sign patterns, and use wavelet transform technology to decompose the signal spectrum, refine abnormal signal features, and generate refined abnormal pattern data;

[0009] Based on the refined abnormal pattern data, a multi-level decision tree set is constructed using the deep forest algorithm, a comprehensive evaluation is performed in combination with historical health records and current environmental factors, and time series analysis is used to capture the trend of vital signs changes to generate a personalized status assessment report;

[0010] Based on the personalized status assessment report, specific intervention measures are recommended, which are converted into executable operation guidelines through an intelligent recommendation system to generate a response plan for the status of the injured and sick.

[0011] Optionally, based on the standardized vital sign data sequence, a dynamic Bayesian network algorithm is used to establish a temporal dependency model between vital sign parameters, identify potential abnormal vital sign patterns, use wavelet transform technology to decompose signal spectrum, refine abnormal signal features, and generate refined abnormal pattern data, including:

[0012] Based on the standardized vital sign data sequence, denoising preprocessing and missing value filling are performed to generate a high-quality data set;

[0013] Based on the high-quality data set, a dynamic Bayesian network algorithm is used to learn the interactions and changing patterns between different vital signs, capture the complex temporal dependencies between vital sign parameters, and generate a vital sign temporal dependency model;

[0014] Based on the vital signs temporal dependency model, data points with large deviations from normal patterns are detected, potential abnormal vital signs patterns are identified, and signal spectrum is decomposed using wavelet transform technology to refine and highlight specific features of abnormal signals and generate abnormal signal feature sets;

[0015] Based on the abnormal signal feature set, combined with time information and signal strength key indicators, refined abnormal pattern data is generated.

[0016] Optionally, based on the high-quality data set, a dynamic Bayesian network algorithm is used to learn the interactions and changing rules between different vital signs, capture the complex temporal dependencies between vital sign parameters, and generate a vital sign temporal dependency model, including:

[0017] Based on the high-quality data set, the data is segmented into time series, divided into multiple time windows, and time series segmentation data is generated;

[0018] Based on the time series segmentation data, a dynamic Bayesian network algorithm is used to construct a network structure to integrate vital sign parameters and potential dependencies, preliminarily learn the interactions between vital sign parameters, and generate a preliminary dynamic Bayesian network structure;

[0019] Based on the preliminary dynamic Bayesian network structure, the connection weights and conditional probability distributions between nodes are adjusted by expectation maximization to accurately reflect the actual interactions between vital sign parameters and generate an optimized dynamic Bayesian network structure;

[0020] Based on the optimized dynamic Bayesian network structure, the network structure and parameters are continuously updated iteratively to converge the model and generate a vital sign temporal dependency model.

[0021] Optionally, based on the vital signs temporal dependency model, detecting data points with large deviations from normal patterns, identifying potential abnormal vital signs patterns, using wavelet transform technology to perform signal spectrum decomposition, refining and highlighting specific features of abnormal signals, and generating abnormal signal feature sets, includes:

[0022] Based on the vital sign time series dependency model, real-time monitoring and analysis of the vital sign data in the high-quality data set, detecting data points with large deviations from normal patterns, and generating potential abnormal data points;

[0023] Based on the potential abnormal data points, further analyzing time series characteristics, confirming and identifying potential abnormal vital sign patterns, and generating potential abnormal patterns;

[0024] Based on the potential abnormal pattern, wavelet transform technology is used to perform spectrum decomposition on the abnormal signal, extract signal features under different frequency components, refine and highlight the specific features of the abnormal signal, and generate detailed abnormal signal features;

[0025] Based on the detailed abnormal signal features, combined with time information and signal strength key indicators, an abnormal signal feature set is generated.

[0026] Optionally, based on the refined abnormal pattern data, a multi-level decision tree set is constructed using a deep forest algorithm, a comprehensive evaluation is performed in combination with historical health records and current environmental factors, and a time series analysis is used to capture the trend of vital signs changes to generate a personalized status assessment report, including:

[0027] Based on the refined abnormal pattern data, statistical features and time domain features are extracted to ensure data quality and generate high-quality abnormal pattern data;

[0028] Based on the high-quality abnormal pattern data, a deep forest algorithm is used to combine multiple random subspaces and random forests, traverse all levels of decision trees for recursive partitioning, capture data features at different levels, and generate a multi-level decision tree set;

[0029] Based on the multi-level decision tree set, a multi-level comprehensive evaluation is performed in combination with historical health records and current environmental factors to generate a multi-level comprehensive evaluation result;

[0030] Based on the multi-level comprehensive evaluation results, time series analysis is used to identify temporal dynamic characteristics, capture potential trends and periodic changes, and generate a personalized status assessment report.

[0031] Optionally, based on the high-quality abnormal pattern data, a deep forest algorithm is used to combine multiple random subspaces and random forests, traverse all hierarchical decision trees for recursive partitioning, capture data features at different levels, and generate a multi-level decision tree set, including:

[0032] Based on the high-quality abnormal pattern data, normalization processing is performed to unify the range of each vital sign parameter to the same scale, screen out the most influential features for abnormal pattern recognition, and generate normalized abnormal pattern data;

[0033] Based on the normalized abnormal pattern data, a deep forest algorithm is used to randomly divide multiple subspaces, reduce feature dimensions to improve the generalization ability of the model, and a random forest is constructed in all subspaces to generate an initial subspace data set;

[0034] Based on the initial subspace data set, randomly select features and samples to generate multiple decision trees, traverse all hierarchical decision trees for recursive partitioning, gradually refine data features, and generate a preliminary multi-level decision tree structure;

[0035] Based on the preliminary multi-level decision tree structure, data features at different levels are captured to generate a multi-level decision tree set.

[0036] Optionally, based on the personalized status assessment report, specific intervention measures are determined, converted into executable operation guidelines through an intelligent recommendation system, and a response plan for the status of the injured or sick is generated, including:

[0037] Based on the personalized status assessment report, identify key health issues and potential risks, and generate specific intervention recommendations;

[0038] Based on the specific intervention measures, the intelligent recommendation system is used to optimize the process, and personalized adjustments are made based on the injury history and health records to generate a preliminary intervention plan;

[0039] Based on the preliminary intervention plan, break it down into detailed steps and implementation instructions to generate executable operation guidelines;

[0040] Based on the executable operation guide, combined with the follow-up tracking plan and emergency response measures, a response plan for the status of the injured and sick is generated.

[0041] In a second aspect, the embodiment of the present application provides a system for assessing the status of a patient based on real-time monitoring of vital signs, including:

[0042] The acquisition module is used to collect the vital signs data stream of the injured and sick in real time through wearable devices, perform preliminary standardization processing, and generate a standardized vital signs data sequence;

[0043] An identification module is used to establish a temporal dependency model between vital sign parameters based on the standardized vital sign data sequence using a dynamic Bayesian network algorithm, identify potential abnormal vital sign patterns, decompose signal spectra using wavelet transform technology, refine abnormal signal features, and generate refined abnormal pattern data;

[0044] An analysis module is used to construct a multi-level decision tree set based on the refined abnormal pattern data using a deep forest algorithm, conduct a comprehensive evaluation based on historical health records and current environmental factors, use time series analysis to capture the trend of vital signs changes, and generate a personalized status assessment report;

[0045] A generation module is used to determine specific intervention measures based on the personalized status assessment report, convert it into an executable operation guide through an intelligent recommendation system, and generate a response plan for the status of the injured and sick.

[0046] In a third aspect, an embodiment of the present application provides a computing device, comprising a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement a method for assessing the status of a patient based on real-time monitoring of vital signs as described in the first aspect.

[0047] In a fourth aspect, an embodiment of the present application provides a computer storage medium storing a computer program. When the computer program is executed by a computer, it implements a method for assessing the status of a patient based on real-time monitoring of vital signs as described in the first aspect.

[0048] In an embodiment of the present application, a wearable device is used to collect the vital signs data stream of the injured and sick in real time, and a preliminary standardization process is performed to generate a standardized vital signs data sequence; based on the standardized vital signs data sequence, a dynamic Bayesian network algorithm is used to establish a temporal dependency model between vital signs parameters, identify potential abnormal vital signs patterns, and use wavelet transform technology to decompose the signal spectrum, refine the abnormal signal characteristics, and generate refined abnormal pattern data; based on the refined abnormal pattern data, a deep forest algorithm is used to construct a multi-level decision tree set, and a comprehensive evaluation is performed in combination with historical health records and current environmental factors, and time series analysis is used to capture the trend of vital signs changes, and a personalized status assessment report is generated; based on the personalized status assessment report, specific intervention measures are determined, and converted into executable operation guidelines through an intelligent recommendation system to generate a status response plan for the injured and sick. Real-time data collection through wearable devices ensures that the vital signs of the injured and sick can be monitored immediately, improving the timeliness and accuracy of status assessment; the collected data is initially standardized to generate a standardized vital signs data sequence, reducing data noise and deviation, and improving the reliability of subsequent analysis; the dynamic Bayesian network algorithm is used to establish a temporal dependency model to identify potential abnormal patterns, capture the complex dependencies between vital sign parameters, and enhance the expressive power of the model; the wavelet transform technology is used to decompose the signal spectrum, refine the abnormal signal characteristics, generate refined abnormal pattern data, and improve the accuracy of anomaly detection; combined with historical health records and current environmental factors, a personalized status assessment report is generated, and converted into an executable operation guide through an intelligent recommendation system, providing personalized intervention measures and improving the scientific nature and pertinence of medical decision-making.

[0049] Furthermore, by performing denoising preprocessing and missing value filling on the standardized vital signs data series, a high-quality data set is generated, which ensures the quality of the input data and improves the effect of model training; the dynamic Bayesian network algorithm is used to learn the interactions and changing patterns between different vital signs, capture the complex temporal dependencies between vital sign parameters, and enhance the model's understanding of complex physiological processes; based on the generated vital sign temporal dependency model, data points with large deviations from the normal pattern are detected, and wavelet transform technology is used to further refine the abnormal signal characteristics, thereby improving the accuracy and sensitivity of abnormal pattern recognition; combining time information with key indicators of signal strength, refined abnormal pattern data is generated, providing rich feature information for subsequent analysis and supporting more in-depth medical research and clinical applications.

[0050] Furthermore, by extracting statistical features and time domain features, high-quality abnormal pattern data is generated, which ensures the quality and integrity of the input data and provides a solid foundation for subsequent analysis; the deep forest algorithm is used to combine multiple random subspaces and random forests, and all levels of decision trees are traversed for recursive division to capture data features at different levels, thereby enhancing the generalization and robustness of the model; historical health records and current environmental factors are combined for multi-level comprehensive evaluation to generate multi-level comprehensive evaluation results, providing a comprehensive and detailed status assessment and supporting more scientific medical decision-making; time series analysis is used to identify time dynamic characteristics, capture potential trends and cyclical changes, and generate personalized status assessment reports, which provide strong support for the status prediction and management of the wounded and sick and improve the level of personalization of medical services.

[0051] These and other aspects of the present application will become more clearly understood in the description of the following embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0053] Figure 1 A flowchart of a method for assessing the status of a patient based on real-time monitoring of vital signs provided in an embodiment of the present application;

[0054] Figure 2 A schematic diagram of the structure of a patient status assessment system based on real-time monitoring of vital signs provided in an embodiment of the present application;

[0055] Figure 3 A schematic diagram of the structure of a computing device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0056] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0057] In some of the processes described in the specification and claims of this application and the above-mentioned figures, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this article or executed in parallel. The serial numbers of the operations, such as 101, 102, etc., are only used to distinguish between different operations, and the serial numbers themselves do not represent any execution order. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this article are used to distinguish different messages, devices, modules, etc., do not represent the order of precedence, and do not limit the "first" and "second" to be different types.

[0058] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of this application.

[0059] Figure 1 A flowchart of a method for assessing the status of a patient based on real-time monitoring of vital signs is provided for an embodiment of the present application, such as Figure 1 As shown, the method includes:

[0060] 101. Through wearable devices, the vital signs data stream of the injured and sick is collected in real time, and preliminary standardization processing is performed to generate a standardized vital signs data sequence;

[0061] In this step, the wearable device is a smart bracelet or patch that integrates multiple sensors, which is used to collect the vital signs data stream of the injured and sick in real time. These devices include but are not limited to heart rate monitors, blood pressure monitors, blood oxygen saturation monitors and body temperature sensors, etc., which are used to continuously collect multiple vital sign parameters such as heart rate, respiratory rate, blood pressure, blood oxygen saturation and body temperature.

[0062] The initial standardization process is to preprocess the raw data to eliminate noise, fill in missing values, and unify data from different units to the same scale to generate a standardized vital sign data series. This process ensures the data quality and consistency of subsequent analysis, and specifically includes denoising, missing value filling, and normalization.

[0063] The standardized vital sign data sequence is a set of data generated after the above processing. These data are collected and stored in a secure data environment, providing a basis for subsequent processing. This sequence not only improves the quality and reliability of the data, but also lays the foundation for complex analysis.

[0064] In an embodiment of the present application, assuming a busy emergency room environment, first, medical staff will wear a set of smart bracelets with integrated multiple sensors for each newly admitted patient; the bracelet can collect data such as heart rate, respiratory rate, blood pressure, and blood oxygen saturation once a minute; the system automatically denoises the received data to remove abnormal fluctuations caused by movement or poor contact of the equipment; for occasional missing values, the system uses linear interpolation to fill them; all data are converted to standard units and normalized so that subsequent algorithms can process them more efficiently.

[0065] 102. Based on the standardized vital sign data sequence, a dynamic Bayesian network algorithm is used to establish a temporal dependency model between vital sign parameters, identify potential abnormal vital sign patterns, use wavelet transform technology to decompose the signal spectrum, refine abnormal signal features, and generate refined abnormal pattern data;

[0066] In this step, the dynamic Bayesian network algorithm is a probabilistic graphical model used to establish a temporal dependency model between vital sign parameters. It not only considers the vital sign status at the current moment, but also combines data from multiple time points in the past to capture the complex interactions and changing patterns between different parameters.

[0067] The temporal dependency model between vital sign parameters is a model constructed through a dynamic Bayesian network algorithm, which aims to identify the historical associations and future predictions between different vital sign parameters. This model helps to understand the changing trends of vital signs and their mutual influence, thereby improving the accuracy and timeliness of abnormality detection.

[0068] Potentially abnormal vital sign pattern recognition is to detect data points that are significantly different from the normal pattern. These anomalies may indicate potential health problems. This process analyzes the model output, marks the data points that deviate from the normal pattern, and further refines the abnormal signal characteristics.

[0069] Wavelet transform technology is a signal processing method used to decompose signal spectrum and refine abnormal signal characteristics. This technology can capture local features in the signal and is particularly suitable for the analysis of non-stationary signals, thereby improving the accuracy of anomaly detection.

[0070] The refined abnormal pattern data are the abnormal signal feature data processed by wavelet transform technology. These data describe the specific characteristics of the abnormal signal in more detail and provide high-quality input for further analysis.

[0071] In an embodiment of the present application, it is assumed that an accident occurs during a large-scale outdoor activity; first, the emergency personnel receive standardized vital signs data from the smart bracelet and use the dynamic Bayesian network algorithm to build a dynamic model containing multiple time steps; secondly, the model learns the historical change trend of each parameter and its correlation with other parameters; thirdly, when a parameter deviates from its historical pattern, the system will automatically mark it as a potential anomaly and apply wavelet transform to further analyze the frequency domain characteristics of the abnormal signal; finally, a detailed abnormal pattern data set is generated for the next step of analysis.

[0072] Optionally, the method in step 102 is based on the standardized vital sign data sequence, uses a dynamic Bayesian network algorithm to establish a temporal dependency model between vital sign parameters, identifies potential abnormal vital sign patterns, uses wavelet transform technology to decompose signal spectrum, refines abnormal signal features, and generates refined abnormal pattern data, including: based on the standardized vital sign data sequence, performs denoising preprocessing and missing value filling to generate a high-quality data set; based on the high-quality data set, uses a dynamic Bayesian network algorithm to learn the interactions and change patterns between different vital signs, captures the complex temporal dependencies between vital sign parameters, and generates a vital sign temporal dependency model; based on the vital sign temporal dependency model, detects data points with large deviations from normal patterns, identifies potential abnormal vital sign patterns, uses wavelet transform technology to decompose signal spectrum, refines and highlights specific features of abnormal signals, and generates an abnormal signal feature set; based on the abnormal signal feature set, combines time information and signal strength key indicators to generate refined abnormal pattern data.

[0073] In this step, denoising preprocessing is a data cleaning method used to remove noise from the standardized vital signs data sequence to ensure the accuracy and reliability of the data.

[0074] Missing value filling is to fill the missing values ​​in the data set through interpolation or other statistical methods to ensure the integrity of the data.

[0075] The high-quality dataset is a data set generated after the above processing. These data have high quality and consistency, providing a solid foundation for subsequent analysis.

[0076] Potentially abnormal vital sign pattern recognition detects data points that are significantly different from normal patterns. These anomalies may indicate potential health problems.

[0077] The abnormal signal feature set is the abnormal signal feature data processed by wavelet transform technology. These data describe the specific characteristics of the abnormal signal in more detail and provide high-quality input for further analysis.

[0078] Refined anomaly pattern data is generated by combining time information and key indicators of signal strength. These data not only include the time characteristics of abnormal signals, but also important information such as signal strength, providing detailed information support for subsequent personalized evaluation.

[0079] In the embodiments of the present application, firstly, denoising preprocessing and missing value filling are performed based on the standardized vital signs data sequence to ensure the quality and integrity of the data; secondly, a high-quality data set is used to apply a dynamic Bayesian network algorithm to learn the interactions and changing patterns between different vital signs, and to capture the complex temporal dependencies between vital sign parameters; thirdly, based on the established vital sign temporal dependency model, data points with large deviations from the normal pattern are detected, potential abnormal vital sign patterns are identified, and wavelet transform technology is used to decompose the signal spectrum to refine and highlight the specific characteristics of the abnormal signal; finally, time information and signal strength key indicators are combined to generate refined abnormal pattern data to provide detailed data support for subsequent analysis.

[0080] Assume that in the emergency department of a large hospital, first, the system performs denoising preprocessing on the received standardized vital signs data to remove abnormal fluctuations caused by movement or poor equipment contact, and uses linear interpolation to fill in occasional missing values ​​to generate a high-quality data set; secondly, the system uses the high-quality data set to start the dynamic Bayesian network algorithm to learn the historical change trend of each vital sign parameter and its correlation with other parameters, and construct a vital sign time series dependency model; thirdly, based on the model, the system automatically marks data points that deviate from the normal pattern, identifies potential abnormal patterns, and applies wavelet transform to further analyze the frequency domain characteristics of abnormal signals to generate a detailed abnormal signal feature set; finally, the system combines time information with key indicators of signal strength to generate refined abnormal pattern data for the next step of analysis to ensure the accuracy and timeliness of anomaly detection.

[0081] Optionally, based on the high-quality data set, a dynamic Bayesian network algorithm is used to learn the interactions and changing patterns between different vital signs, capture the complex temporal dependencies between vital sign parameters, and generate a vital sign temporal dependency model, including: based on the high-quality data set, time series segmentation of the data is performed, divided into multiple time windows, and time series segmentation data is generated; based on the time series segmentation data, a dynamic Bayesian network algorithm is used to construct a network structure to integrate vital sign parameters and potential dependencies, preliminarily learn the interactions between vital sign parameters, and generate a preliminary dynamic Bayesian network structure; based on the preliminary dynamic Bayesian network structure, the connection weights and conditional probability distributions between nodes are adjusted by expectation maximization to accurately reflect the actual interactions between vital sign parameters, and an optimized dynamic Bayesian network structure is generated; based on the optimized dynamic Bayesian network structure, the network structure and parameters are continuously iteratively updated to converge the model, and a vital sign temporal dependency model is generated.

[0082] The method of detecting data points with large deviations from normal patterns based on the vital signs temporal dependency model, identifying potential abnormal vital signs patterns, using wavelet transform technology to perform signal spectrum decomposition, refining and highlighting specific features of abnormal signals, and generating an abnormal signal feature set includes: based on the vital signs temporal dependency model, real-time monitoring and analyzing vital signs data in the high-quality data set, detecting data points with large deviations from normal patterns, and generating potential abnormal data points; based on the potential abnormal data points, further analyzing time series characteristics, confirming and identifying potential abnormal vital signs patterns, and generating potential abnormal patterns; based on the potential abnormal patterns, using wavelet transform technology to perform spectrum decomposition of abnormal signals, extracting signal features under different frequency components, refining and highlighting specific features of abnormal signals, and generating detailed abnormal signal features; based on the detailed abnormal signal features, combining time information with key indicators of signal strength, generating an abnormal signal feature set.

[0083] In this step, time series segmentation is to divide the high-quality data set into multiple time windows of fixed length in chronological order. The data in each window represents the changes in vital signs in a time period, which helps to capture the changing trends of vital sign parameters over time and generate time series segmentation data for subsequent analysis.

[0084] The preliminary dynamic Bayesian network structure is the initial network structure constructed based on time series segmentation data. It integrates vital sign parameters and potential dependencies, preliminarily reflects the interaction between these parameters, and provides a basis for subsequent optimization.

[0085] Expectation maximization is an iterative optimization method used to adjust the connection weights and conditional probability distribution between nodes so that the model can more accurately reflect the actual interactions between vital sign parameters, and ultimately generate an optimized dynamic Bayesian network structure. Through continuous iteration, the model is ensured to be stable and can accurately predict the changing trends of vital signs.

[0086] The optimized dynamic Bayesian network structure is a network structure adjusted by expectation maximization, which more accurately reflects the actual interaction between vital sign parameters and enhances the model's expressive power and prediction accuracy.

[0087] Potentially abnormal data points are data points that deviate significantly from normal patterns and are detected through real-time monitoring and analysis of vital sign data in high-quality data sets, indicating possible health problems.

[0088] Potential abnormal patterns are abnormal vital sign patterns confirmed and identified by further analyzing the time series characteristics. Potential abnormal patterns are generated to provide a basis for subsequent processing. These patterns help identify specific health risks.

[0089] Detailed abnormal signal characteristics are abnormal signal characteristics processed by wavelet transform technology. They combine time information and key indicators of signal strength to generate detailed abnormal signal characteristics, providing high-quality data support for subsequent evaluation.

[0090] The abnormal signal feature set is data generated by combining time information and key indicators of signal strength. This data not only includes the time characteristics of the abnormal signal, but also covers important information such as signal strength, providing detailed information support for subsequent personalized evaluation.

[0091] In the embodiment of the present application, firstly, based on the high-quality data set, the data is segmented into time series, divided into multiple time windows, and time series segmentation data is generated; secondly, using these time series segmentation data, a preliminary dynamic Bayesian network structure is constructed, the vital sign parameters and potential dependencies are integrated, and the interaction between the parameters is preliminarily learned; again, the connection weights and conditional probability distribution between nodes are adjusted by expectation maximization, the network structure is optimized to accurately reflect the actual interaction between the vital sign parameters, and the network structure and parameters are continuously updated iteratively to make the model converge and generate a stable vital sign time series dependency model. At the same time, based on the generated vital sign time series dependency model, the vital sign data in the high-quality data set is monitored and analyzed in real time, and the data points with large deviations from the normal mode are detected, and the time series characteristics are further analyzed to confirm and identify the potential abnormal vital sign mode, and the wavelet transform technology is used to perform spectral decomposition of the abnormal signal, extract the signal features under different frequency components, and refine and highlight the specific features of the abnormal signal; finally, combining the time information and the key indicators of signal strength, an abnormal signal feature set is generated to provide detailed data support for subsequent evaluation.

[0092] Assuming that on a remote medical rescue platform, first, the system performs time series segmentation on the received high-quality data set, divides it into multiple time windows, generates time series segmentation data, and starts the construction of a preliminary dynamic Bayesian network structure, integrates vital sign parameters and potential dependencies, and preliminarily learns the interaction between parameters; secondly, the system adjusts the connection weights and conditional probability distribution between nodes by expectation maximization, optimizes the network structure, ensures that the model can accurately reflect the actual interaction between vital sign parameters, and monitors and analyzes the vital sign data in the high-quality data set in real time, detects data points with large deviations from the normal mode, and generates potential abnormal data points; thirdly, the system further analyzes the time series characteristics, confirms and identifies potential abnormal vital sign patterns, generates potential abnormal patterns, and uses wavelet transform technology to perform spectral decomposition on abnormal signals, extracts signal features under different frequency components, and refines and highlights the specific features of abnormal signals; finally, the system continuously iterates and updates the network structure and parameters to converge the model, generate a stable vital sign time series dependency model, and combines time information with key indicators of signal strength to generate an abnormal signal feature set for the next step of analysis, ensuring the accuracy and timeliness of abnormal detection.

[0093] The present application takes into account that in the prior art, due to the problems of difficulty in capturing potential dependencies between vital sign parameters, low data quality and model overfitting, the invention embodiment proposes this optional solution, which aims to build a network structure through a dynamic Bayesian network algorithm to integrate vital sign parameters and potential dependencies, preliminarily learn the interactions between vital sign parameters, and solve the technical problem of improving model prediction accuracy and generalization ability.

[0094] Optionally, based on the time series segmentation data, a dynamic Bayesian network algorithm is used to construct a network structure to integrate vital sign parameters and potential dependencies, preliminarily learn the interactions between vital sign parameters, and generate a preliminary dynamic Bayesian network structure, including:

[0095] Based on the time series segmentation data, multiple interpolation methods are used to fill missing values, and smoothing is performed to improve data quality, and the most representative vital sign parameters are screened out using stepwise regression analysis to generate a vital sign parameter vector;

[0096] The vital sign parameter vector P(X t ∣X t-1 ,Θ):

[0097]

[0098] Among them, X t is the vital sign parameter vector at time t; X t-1is the vital sign parameter vector at time t-1; Θ is the model parameter vector including network structure parameters and conditional probability distribution parameters; Σ is the covariance matrix describing the dependency relationship between parameters; f(X t-1 ,Θ) is a nonlinear function that represents the impact of the vital sign parameters at the previous moment on the current moment; d is the dimension of the vital sign parameters; η is a nonlinear adjustment coefficient used to enhance the nonlinear expression ability of the model; δ i is the weight coefficient of the i-th nonlinear adjustment item; i is the index of the nonlinear adjustment, ranging from 1 to k; k is the number of nonlinear adjustment items;

[0099] Based on the vital sign parameter vector, the posterior probability of the latent variable is estimated, the model parameters are updated to maximize the likelihood function, and the network structure parameters and the conditional probability distribution parameters are gradually adjusted through multiple iterations to generate an updated model parameter vector;

[0100] The updated model parameter vector is calculated using the following formula:

[0101]

[0102] Among them, Θ new is the updated model parameter vector; T is the length of the time series; λ is the latent variable regularization coefficient used to balance the influence of explicit observation data and latent variables; P(X t ∣X t-1 ,Θ) is the conditional probability distribution in the first formula; Z i is the ith latent variable; P(Z i ∣X t-1 ,Θ) is the conditional probability distribution of latent variables; k is the number of latent variables; μ is the regularization coefficient of high-order latent variables used to balance the influence of high-order latent variables; W j is the jth high-order latent variable; P(E j ∣X t-1 ,Θ,Z) is the conditional probability distribution of high-order latent variables; m is the number of high-order latent variables; γ is the L2 regularization coefficient used to prevent overfitting; ∥Θ∥ 2 is the L2 norm of the model parameter vector; α is the high-dimensional latent variable regularization coefficient used to balance the impact of high-dimensional latent variables; β l is the weight coefficient of the lth high-dimensional latent variable; V l is a high-dimensional latent variable vector; h(X t-1 ,Θ) is a nonlinear function of high-dimensional latent variables, which represents the influence of the vital sign parameters and model parameters at the previous moment on the high-dimensional latent variables; Λ is the covariance matrix of the high-dimensional latent variables; p is the dimension of the high-dimensional latent variables; n is the number of high-dimensional latent variables; ρ is the periodic adjustment coefficient used to capture periodic changes; ζ s is the weight coefficient of the sth periodic adjustment item; ωs is the frequency coefficient of the sth periodic adjustment item; q is the number of periodic adjustment items; t is the index of the time point, from 1 to T; T is the length of the time series; i is the index of the latent variable, from 1 to k; k is the number of latent variables; j is the index of the high-order latent variable, from 1 to m; m is the number of high-order latent variables; l is the index of the high-dimensional latent variable, from 1 to n; n is the number of high-dimensional latent variables; s is the index of the periodic adjustment item, from 1 to q; q is the number of periodic adjustment items; max Θ represents the maximization operation of the model parameter vector Θ, the purpose of which is to find the optimal parameter configuration that maximizes the objective function (such as likelihood function or loss function); X t is the normalized abnormal pattern data vector at time t, representing the vital sign parameters at the current moment; X t-1 is the normalized abnormal pattern data vector at time t-1, representing the vital sign parameters at the previous moment; Θ is the model parameter vector, including network structure parameters and conditional probability distribution parameters, which is used to describe the relationship and dependency within the model; Z is the latent variable vector, which captures the state or factors that are not directly observed, and helps to improve the expression and generalization ability of the model; f(X t-1 ,Θ) is a nonlinear function, representing the vital sign parameter X at the previous moment t-1 The influence of the model parameters Θ on the current moment is used to predict the vital sign parameters at the current moment;

[0103] Based on the updated model parameter vector, the training set and the test set are divided by random sampling to obtain the test set performance index, so as to adjust the regularization coefficient and other hyperparameters, verify the model effectiveness and generalization ability, and generate a preliminary dynamic Bayesian network structure.

[0104] This method aims to segment data based on time series, fill missing values ​​and smooth processing using multiple interpolation methods, use stepwise regression analysis to screen the most representative vital sign parameters, and generate a vital sign parameter vector; then calculate the vital sign parameter vector through conditional probability distribution, estimate the posterior probability of latent variables, update the model parameters to maximize the likelihood function, and finally generate an updated model parameter vector. This design not only enhances the understanding of the complex dependencies between vital sign parameters, but also improves the predictive ability and generalization performance of the model in the medical field.

[0105] In the vital sign parameter vector, the Gaussian distribution term The conditional probability distribution of vital sign parameters based on the Gaussian distribution assumption describes the vital sign parameters X at the current moment. t At a given previous moment X t-1 and the probability under the model parameter Θ; nonlinear adjustment term By introducing the nonlinear activation function tanh and adjusting the coefficient η and weight δ i ,Enhance the nonlinear expression ability of the model and capture the complex dependencies among vital sign parameters;

[0106] Among them, X t and X t-1 Directly obtained from the time series segmentation data; Θ contains the network structure parameters and conditional probability distribution parameters, which are optimized by the back propagation algorithm during the training process; Σ is the covariance matrix, which describes the dependency between parameters and is estimated by statistical methods or during the training process; f(X t-1 ,Θ) is a nonlinear function, which represents the influence of the vital sign parameters at the previous moment on the current moment, and is realized through neural networks or other nonlinear models; d is the dimension of the vital sign parameters, which is directly determined according to the input data; η is the nonlinear adjustment coefficient, which is obtained through hyperparameter tuning; δ i is the weight coefficient of the i-th nonlinear adjustment item, which is obtained by optimizing the back propagation algorithm during the training process; k is the number of nonlinear adjustment items, which is set according to the complexity of the model;

[0107] In the updated model parameter vector, the conditional probability distribution term Calculate the conditional probability distribution of the observed data to ensure that the model can fully explain the observed data; latent variable regularization term Introduce hidden variables to capture states or factors that are not directly observed, and balance the influence of explicit observation data and hidden variables; high-order hidden variable regularization term Introduce high-order latent variables to balance the influence of high-order latent variables; L2 regularization term γ∥Θ∥ 2 : Prevent model overfitting by introducing regularization terms; high-dimensional latent variable regularization terms Introduce high-dimensional latent variables to balance the impact of high-dimensional latent variables; periodic adjustment term Capture cyclical changes in vital sign parameters;

[0108] Among them, Θ contains the network structure parameters and conditional probability distribution parameters, which are optimized by the back propagation algorithm during the training process; T is the length of the time series, which is determined according to the data set; λ is the latent variable regularization coefficient, which is obtained by hyperparameter tuning; Z i is the i-th latent variable, estimated through model training; P(Z i ∣X t-1 ,Θ) is the conditional probability distribution of latent variables, estimated through model training; k is the number of latent variables, set according to the complexity of the model; μ is the regularization coefficient of high-order latent variables, obtained through hyperparameter tuning; W j is the jth high-order latent variable, estimated through model training; P(W j ∣X t-1,Θ,Z) is the conditional probability distribution of high-order latent variables, estimated through model training; m is the number of high-order latent variables, set according to the complexity of the model; γ is the L2 regularization coefficient, obtained through hyperparameter tuning; α is the high-dimensional latent variable regularization coefficient, obtained through hyperparameter tuning; β l is the weight coefficient of the lth high-dimensional latent variable, obtained through hyperparameter tuning; V l is the lth high-dimensional latent variable vector, estimated through model training; h(X t-1 ,Θ) is a nonlinear function of high-dimensional latent variables, which is realized by neural networks or other nonlinear models; Λ is the covariance matrix of high-dimensional latent variables, which is estimated by statistical methods or training process; p is the dimension of high-dimensional latent variables, which is determined according to input data; n is the number of high-dimensional latent variables, which is set according to model complexity; ρ is the periodic adjustment coefficient, which is obtained by hyperparameter tuning; ζ s is the weight coefficient of the sth periodic adjustment term, obtained through hyperparameter tuning; ω s is the frequency coefficient of the sth periodic adjustment term, obtained through hyperparameter tuning; q is the number of periodic adjustment terms, set according to the model complexity;

[0109] Suppose there is a medical system that needs to monitor patients’ vital signs in real time and predict potential health risks;

[0110] Assume X t =[0.82,0.72,0.62],X t-1 =[0.78,0.68,0.58]; model parameter vector Θ =[0.9,0.8,0.7]; covariance matrix Nonlinear function f(X t-1 ,Θ)=[0.8,0.7,0.6]; nonlinear adjustment coefficient η=0.4,δ1=0.3; number of nonlinear adjustment items k=1; latent variable regularization coefficient λ=0.1; high-order latent variable regularization coefficient μ=0.05; L2 regularization coefficient γ=0.01; high-dimensional latent variable regularization coefficient α=0.2; periodic adjustment coefficient ρ=0.05;

[0111]

[0113] Assume that the threshold is set to θ = [0.85, 0.75, 0.65], since the updated model parameter vector Θ new= Each element of [0.88, 0.78, 0.68] is greater than the corresponding set threshold, which shows that the model has successfully adjusted the network structure parameters and conditional probability distribution parameters after multiple iterations, making the model more adaptable to actual data and improving prediction accuracy. Through the above steps, the system can effectively process time series vital sign data and extract high-quality multi-semantic information hidden state representation, thereby improving the accuracy and reliability of medical monitoring.

[0114] 103. Based on the refined abnormal pattern data, a multi-level decision tree set is constructed using a deep forest algorithm, a comprehensive evaluation is performed in combination with historical health records and current environmental factors, and time series analysis is used to capture the trend of vital signs changes to generate a personalized status assessment report;

[0115] In this step, the deep forest algorithm is a tree-structured machine learning method that captures data features at different levels by combining multiple random subspaces and random forests to enhance the generalization ability and robustness of the model.

[0116] The multi-level decision tree ensemble is a series of decision tree models built using the Deep Forest algorithm. Each tree is split based on a randomly selected feature subset to increase model diversity. This multi-level structure helps capture data features at different levels and improves the expressiveness of the model.

[0117] Historical health records are files containing information such as the patient's medical history, living habits, and medication use records. They are used to comprehensively evaluate the patient's health status. These files provide an important reference for personalized assessments and help to more comprehensively understand the patient's historical health status.

[0118] Current environmental factors refer to the external conditions in which the patient is located, such as weather, geographical location, indoor temperature and humidity, etc. These factors may affect vital signs. Considering these factors will help to more accurately assess the patient's condition.

[0119] Time series analysis is a statistical method used to capture the changing trends of vital signs and identify potential trends and cyclical changes. This method predicts future health status by analyzing historical data and provides a basis for personalized assessment.

[0120] The personalized status assessment report is a detailed report generated based on the above analysis, including predictions of the health status of the injured and sick and warnings of possible risks in the future. The report provides a scientific basis for formulating intervention measures and supports more accurate medical decision-making.

[0121] In an embodiment of the present application, suppose that in a long-term care facility, an elderly patient suddenly develops symptoms of discomfort. First, the system starts the deep forest algorithm to construct a multi-level decision tree set based on the refined abnormal pattern data generated in the previous step; the system integrates the patient's past medical history (such as hypertension, diabetes) and current environmental factors (such as indoor temperature and humidity); identifies long-term trends and short-term fluctuations in vital signs through time series analysis; and finally, generates a detailed status assessment report, which includes a prediction of the patient's health status and possible risk warnings in the future.

[0122] Optionally, in step 103, based on the refined abnormal pattern data, a deep forest algorithm is used to construct a multi-level decision tree set, a comprehensive evaluation is performed in combination with historical health records and current environmental factors, and time series analysis is used to capture the trend of changes in vital signs to generate a personalized status assessment report, including: based on the refined abnormal pattern data, statistical features and time domain features are extracted to ensure data quality, and high-quality abnormal pattern data are generated; based on the high-quality abnormal pattern data, a deep forest algorithm is used to combine multiple random subspaces and random forests, all hierarchical decision trees are traversed for recursive partitioning, data features at different levels are captured, and a multi-level decision tree set is generated; based on the multi-level decision tree set, a multi-level comprehensive evaluation is performed in combination with historical health records and current environmental factors to generate a multi-level comprehensive evaluation result; based on the multi-level comprehensive evaluation result, time series analysis is used to identify time dynamic characteristics, capture potential trends and periodic changes, and generate a personalized status assessment report.

[0123] Among them, based on the high-quality abnormal pattern data, the deep forest algorithm is used to combine multiple random subspaces and random forests, traverse all hierarchical decision trees for recursive division, capture data features at different levels, and generate a multi-level decision tree set, including: based on the high-quality abnormal pattern data, normalization processing is performed to unify the range of each vital sign parameter to the same scale, screen out the most influential features for abnormal pattern recognition, and generate normalized abnormal pattern data; based on the normalized abnormal pattern data, the deep forest algorithm is used to randomly divide multiple subspaces, reduce feature dimensions to improve model generalization ability, construct random forests in all subspaces, and generate an initial subspace data set; based on the initial subspace data set, features and samples are randomly selected to generate multiple decision trees, all hierarchical decision trees are traversed for recursive division, data features are gradually refined, and a preliminary multi-level decision tree structure is generated; based on the preliminary multi-level decision tree structure, data features at different levels are captured to generate a multi-level decision tree set.

[0124] In this step, statistical features and time domain features include statistics such as mean, variance, peak, and characteristics such as periodicity, trend, and mutation points in time series, which are used to evaluate data quality and ensure the reliability of subsequent analysis.

[0125] High-quality abnormal pattern data is a data set generated after statistical features and time domain features are extracted, which ensures the quality and consistency of the data and provides reliable input for the deep forest algorithm.

[0126] Historical health records are files that contain information such as the patient's medical history, living habits, and medication use records, and are used to comprehensively evaluate the patient's health status.

[0127] Current environmental factors refer to the external conditions in which the injured or sick are located, such as weather, geographical location, indoor temperature and humidity, etc. These factors may affect vital signs.

[0128] The multi-level comprehensive evaluation results are generated after a multi-level comprehensive evaluation combining historical health records with current environmental factors. They provide a comprehensive and detailed status assessment and support more scientific medical decision-making.

[0129] Time series analysis is a statistical method used to capture the changing trends of vital signs and identify potential trends and cyclical changes.

[0130] The personalized status assessment report is a detailed report generated based on the above analysis, including a prediction of the health status of the injured or sick and warnings of possible risks in the future.

[0131] This application takes into account that, first, based on the refined abnormal pattern data, statistical features and time domain features are extracted to ensure data quality and generate high-quality abnormal pattern data; secondly, based on the high-quality abnormal pattern data, the deep forest algorithm is used to combine multiple random subspaces and random forests, randomly divide multiple subspaces, reduce feature dimensions to improve the generalization ability of the model, and construct random forests in all subspaces to generate an initial subspace data set; thirdly, based on the initial subspace data set, features and samples are randomly selected to generate multiple decision trees, all levels of decision trees are traversed for recursive division, data features are gradually refined, and a preliminary multi-level decision tree structure is generated; finally, based on the preliminary multi-level decision tree structure, data features at different levels are captured, a multi-level decision tree set is generated, and a multi-level comprehensive evaluation is performed in combination with historical health records and current environmental factors to generate a multi-level comprehensive evaluation result, and time series analysis is used to identify time dynamic characteristics, capture potential trends and periodic changes, and finally generate a personalized status assessment report.

[0132] Assume that in an intensive care unit environment; first, the system extracts statistical features and time domain features from the received refined abnormal pattern data to generate high-quality abnormal pattern data to ensure data quality and consistency; second, the system performs normalization processing based on the high-quality abnormal pattern data, unifies the range of each vital sign parameter to the same scale, screens out the most influential features for abnormal pattern recognition, generates normalized abnormal pattern data, and uses the deep forest algorithm to randomly divide multiple subspaces, reduce feature dimensions to improve the generalization ability of the model, and constructs random forests in all subspaces to generate an initial subspace data set; third, the system is based on In the initial subspace data set, features and samples are randomly selected to generate multiple decision trees, and all levels of decision trees are traversed for recursive division, and the data features are gradually refined to generate a preliminary multi-level decision tree structure; finally, based on the preliminary multi-level decision tree structure, the system captures the data features of different levels, generates a multi-level decision tree set, and combines historical health records with current environmental factors for a multi-level comprehensive evaluation to generate a multi-level comprehensive evaluation result. Time series analysis is used to identify time dynamic characteristics, capture potential trends and cyclical changes, and finally generate a personalized status assessment report for reference by medical staff to ensure the accuracy and timeliness of the assessment.

[0133] The present application takes into account the problems in the prior art such as insufficient model generalization ability, feature redundancy and excessive model complexity due to high feature dimensions, so the invention embodiment proposes this optional solution, which aims to randomly divide multiple subspaces through a deep forest algorithm, reduce feature dimensions to improve the model generalization ability, and construct random forests in all subspaces to generate an initial subspace data set, thereby solving the problem of improving the model prediction accuracy and stability.

[0134] Optionally, based on the normalized abnormal pattern data, a deep forest algorithm is used to randomly divide multiple subspaces, reduce feature dimensions to improve model generalization ability, and construct a random forest in all subspaces to generate an initial subspace data set, including:

[0135] Based on the normalized abnormal pattern data, key features are identified in combination with the correlation between the features and the target variable;

[0136] Dimensionality reduction is performed through principal component analysis to reduce redundancy among key features and retain the main information to generate subspaces;

[0137] The subspace is calculated using the following formula:

[0138]

[0139] Among them, S i is the i-th subspace; X i,jis the jth feature in the normalized abnormal pattern data; rand(1,d,k) is a function that randomly selects k non-repeating integers from 1 to d; d is the feature dimension of the normalized abnormal pattern data; k is the number of features selected for each subspace; η is the nonlinear adjustment coefficient used to enhance the nonlinear expression ability of the model; ω l is the weight coefficient of the Ith nonlinear adjustment term; μ j is the mean of the jth feature; l is the index of the nonlinear adjustment item, from 1 to L; L is the number of nonlinear adjustment items;

[0140] Based on the subspace, multiple sample subsets are randomly selected, decision trees are constructed based on all sample subsets, only some features are used for node splitting to increase model diversity, and multiple decision trees are gradually constructed through multiple iterations to generate a random forest;

[0141] The random forest is calculated using the following formula:

[0142]

[0143] Among them, F i is the random forest of the ith subspace; m is the index of the decision tree in the random forest; M is the number of decision trees in the random forest; Σ m The covariance matrix of the mth decision tree describes the dependency between parameters; f(X i,j ,Θ m ) is a nonlinear function that represents the impact of the current feature on the decision tree; Θ m is the parameter vector of the mth decision tree, including the network structure parameters and conditional probability distribution parameters; λ is the latent variable regularization coefficient used to balance the influence of explicit observation data and latent variables; P(Z n ∣X i,j ,Θ m ) is the conditional probability distribution of the latent variable; Z n is the nth latent variable; n is the index of the latent variable, from 1 to N; N is the number of latent variables; k is the feature dimension;

[0144] Based on the random forest, a cross-validation method is used to evaluate the overall prediction performance, and a grid search is used to systematically traverse the preset parameter combinations to obtain the optimal parameter configuration, improve the model accuracy and stability, and generate an initial subspace data set.

[0145] The method aims to identify key features based on normalized abnormal pattern data and the correlation between features and target variables; perform dimensionality reduction through principal component analysis to reduce redundancy between key features and retain main information to generate subspaces; then construct a random forest in each subspace, using only some features for node splitting to increase model diversity; finally, use cross-validation to evaluate overall prediction performance, and systematically traverse preset parameter combinations through grid search to obtain the optimal parameter configuration. This design not only enhances the understanding of the complex dependencies between vital sign parameters, but also improves the model's predictive ability and generalization performance in the medical field.

[0146] In the subspace, the random feature selection term {X i,j |j∈rand(1,d,k)}: randomly select k non-repeated features from the original feature set to reduce the feature dimension to improve the generalization ability of the model; nonlinear adjustment term Introduce nonlinear activation function tanh and adjustment coefficient η and weight ω l , enhance the nonlinear expression ability of the model and capture the complex dependencies between features;

[0147] Among them, S i is the i-th subspace, calculated by the subspace formula; X i,j is the jth feature in the normalized abnormal pattern data, which is directly obtained from the adaptive input data; rand(1, d, k) is a function that randomly selects k non-repeating integers from 1 to d, obtained by random sampling; d is the feature dimension of the normalized abnormal pattern data, which is directly determined according to the input data; k is the number of features selected for each subspace, which is set according to the model complexity; η is the nonlinear adjustment coefficient, which is obtained by hyperparameter tuning; ω l is the weight coefficient of the lth nonlinear adjustment term, which is optimized by the back propagation algorithm during the training process; μ j is the mean of the jth feature, estimated by statistical methods or during training; l is the index of the nonlinear adjustment term, from 1 to L; L is the number of nonlinear adjustment terms, set according to the model complexity;

[0148] In random forests, the conditional probability distribution term Calculate the conditional probability distribution of the observed data to ensure that the model can fully explain the observed data; latent variable regularization term Introduce latent variables to capture states or factors that are not directly observed and balance the influence of explicit observation data and latent variables;

[0149] Among them, F iis the random forest of the ith subspace, calculated by the random forest formula; m is the index of the decision tree in the random forest, ranging from 1 to M; M is the number of decision trees in the random forest, set according to the model complexity; Σ m is the covariance matrix of the mth decision tree, describing the dependency between parameters, which is estimated through statistical methods or training process; f(X i,j ,Θ m ) is a nonlinear function that represents the impact of the current feature on the decision tree, which is implemented through a neural network or other nonlinear models; Θ m is the parameter vector of the mth decision tree, including the network structure parameters and conditional probability distribution parameters, which are optimized by the back propagation algorithm during the training process; λ is the latent variable regularization coefficient, which is obtained by hyperparameter tuning; P(Z n ∣X i,j ,Θ m ) is the conditional probability distribution of the latent variable, estimated through model training; Z n is the nth latent variable, estimated through model training; n is the index of the latent variable, from 1 to N; N is the number of latent variables, set according to the complexity of the model; k is the feature dimension, determined according to the input data;

[0150] Suppose there is an intelligent patient vital signs monitoring system, the system needs to evaluate vital sign parameters in real time;

[0151] Assume that the feature dimension d = 5; the number of features selected in each subspace k = 3; the nonlinear adjustment coefficient η = 0.5, ω1 = 0.3; the number of nonlinear adjustment items L = 1; the random feature selection result rand(1,5,3) = [1,3,5]; the feature mean μ1 = 0.8, μ3 = 0.7, μ5 = 0.6; the feature X in the normalized abnormal pattern data i,1 =0.82,X i,3 =0.72,X i,5 =0.62;

[0152] S i ={0.82,0.72,0.62}×(1+0.5·tanh(0.3·(0.82-0.8+0.72-0.7+0.62-0.6)))=[0.84,0.74,0.64];

[0153]

[0154] Assuming that the threshold is set to θ = 0.9, since the calculated result 0.92 is greater than the set threshold, it shows that the random forest is successfully constructed. After multiple iterations, the model has successfully adjusted the network structure parameters and conditional probability distribution parameters, improving the prediction accuracy. Through the above steps, the system can effectively process time series vital sign data and extract high-quality multi-semantic information hidden state representation, thereby improving the accuracy and reliability of medical monitoring.

[0155] 104. Based on the personalized status assessment report, specific intervention measures are recommended, which are converted into executable operation guidelines through an intelligent recommendation system to generate a response plan for the status of the injured or sick.

[0156] In this step, the personalized status assessment report is a detailed report generated based on the above analysis, covering the health status prediction of the injured and sick and possible risk warnings in the future, providing a scientific basis for formulating specific intervention measures and supporting more accurate medical decision-making.

[0157] Specific intervention measures recommended are targeted medical advice or lifestyle adjustment plans based on the assessment results, aimed at improving the health status of the injured and sick, including medication adjustments, dietary advice, exercise guidance, etc., to help the injured and sick better manage their health.

[0158] The intelligent recommendation system is an automated tool that converts the above suggestions into executable operation guidelines to help medical staff or patients take specific actions. The system can automatically generate detailed implementation steps based on the specific circumstances of the injured and patients to ensure the effective implementation of the suggestions.

[0159] The patient status response plan is a clear time and operational procedures formulated based on the intervention measures to ensure that all recommendations can be implemented in a timely and effective manner. The plan lists in detail the specific implementation time and steps for each recommendation to ensure the effectiveness and adaptability of the intervention measures.

[0160] Regular tracking and feedback adjustment enable the platform to regularly monitor the execution of the injured and sick, and adjust the plan based on feedback to ensure the effectiveness and adaptability of the intervention measures, thus ensuring the dynamic optimization of the entire intervention process and improving the treatment effect and quality of life.

[0161] In an embodiment of the present application, it is assumed that on a telemedicine service platform, a patient with a chronic disease needs continuous monitoring and management. First, the system tailors a set of intervention measures recommendations for the patient based on the personalized status assessment report, covering medication adjustments, dietary recommendations, and exercise guidance; the intelligent recommendation system converts these recommendations into easy-to-understand operating guidelines, such as daily medication reminders, nutritional meal matching plans, etc.; the system generates a response plan for the status of the injured and sick, listing in detail the specific execution time and steps of each recommendation; the platform regularly tracks the patient's execution status, and adjusts the plan based on feedback to ensure the effectiveness and adaptability of the intervention measures.

[0162] Optionally, the step 104 determines specific intervention measures based on the personalized status assessment report, converts them into executable operation guidelines through an intelligent recommendation system, and generates a sick and injured status response plan, including: based on the personalized status assessment report, identifying key health issues and potential risks, and generating specific intervention measures; based on the specific intervention measures, optimizing the recommendations through an intelligent recommendation system, making personalized adjustments based on the injury history health records, and generating a preliminary intervention plan; based on the preliminary intervention plan, breaking it down into detailed steps and execution instructions to generate an executable operation guide; based on the executable operation guide, combining subsequent tracking plans and emergency response measures, generating a sick and injured status response plan.

[0163] In this step, the personalized status assessment report is a detailed report generated based on the above analysis, including the health status prediction of the injured and sick and possible risk warnings in the future.

[0164] Key health issues and potential risks are major health challenges and possible risk factors identified by analyzing the personalized status assessment report.

[0165] Specific intervention recommendations are targeted medical advice or lifestyle adjustments based on the assessment results, aimed at improving the health status of the injured or sick.

[0166] The intelligent recommendation system is an automated tool that converts the above suggestions into executable operation guidelines to help medical staff or patients take specific actions.

[0167] The preliminary intervention plan is an initial intervention plan generated after personalized adjustment based on the patient's historical health records, taking into account the patient's medical history, living habits and other information to ensure that the intervention measures are more in line with individual needs.

[0168] The executable operational guide breaks down the initial intervention plan into detailed steps and implementation instructions to ensure that each recommendation can be implemented in a specific manner.

[0169] The follow-up plan is a regular monitoring plan for the implementation of intervention measures for the injured and sick, to ensure the effectiveness of the intervention and make timely adjustments based on feedback.

[0170] Emergency response measures are emergency plans developed for possible emergencies to ensure that effective actions can be taken quickly in emergency situations to protect the safety of the injured and sick.

[0171] In the embodiment of the present application, firstly, based on the personalized status assessment report, key health problems and potential risks are identified, and specific intervention measures are recommended; secondly, based on the specific intervention measures, optimization processing is performed through the intelligent recommendation system, and personalized adjustments are made in combination with the historical health records of the injured and sick to generate a preliminary intervention plan; thirdly, based on the preliminary intervention plan, it is broken down into detailed steps and execution instructions to generate an executable operation guide; finally, based on the executable operation guide, combined with the follow-up tracking plan and emergency response measures, a response plan for the status of the injured and sick is generated to ensure the effectiveness and adaptability of the intervention measures.

[0172] Assume the scenario of a remote rehabilitation center; first, based on the personalized status assessment report, the system identifies the key health problems and potential risks of the patient, such as hypertension fluctuations and heart discomfort, and generates specific intervention measures, such as adjusting the dosage of medicines and increasing the monitoring frequency; secondly, the system optimizes these suggestions through the intelligent recommendation system, and makes personalized adjustments based on the patient's historical health records (such as previous history of heart disease), and generates a preliminary intervention plan, including daily medication reminders and weekly heart rate monitoring arrangements; thirdly, the system breaks down the preliminary intervention plan into detailed steps and execution instructions, such as taking specific medicines at 7 am every day and measuring heart rate at 3 pm, to generate an executable operation guide; finally, the system combines follow-up tracking plans with emergency response measures, such as monthly doctor video consultations and calling emergency numbers in emergencies, to generate a complete patient status response plan to ensure that all suggestions can be implemented in a timely and effective manner to improve rehabilitation effects and safety.

[0173] In summary, steps 101 to 104 cover the complete process from processing normalized abnormal pattern data to building a random forest, aiming to provide an efficient and accurate feature selection and model training framework to meet the needs of improving model generalization ability and prediction accuracy in a high-dimensional data environment.

[0174] Figure 2 A schematic diagram of a system for assessing the status of a patient based on real-time monitoring of vital signs is provided for the present application. Figure 2 As shown, the device comprises:

[0175] The collection module 21 is used to collect the vital signs data stream of the injured and sick in real time through the wearable device, perform preliminary standardization processing, and generate a standardized vital signs data sequence;

[0176] The identification module 22 is used to establish a temporal dependency model between vital sign parameters based on the standardized vital sign data sequence using a dynamic Bayesian network algorithm, identify potential abnormal vital sign patterns, decompose the signal spectrum using wavelet transform technology, refine abnormal signal features, and generate refined abnormal pattern data;

[0177] An analysis module 23 is used to construct a multi-level decision tree set based on the refined abnormal pattern data using a deep forest algorithm, conduct a comprehensive evaluation in combination with historical health records and current environmental factors, use time series analysis to capture the trend of vital sign changes, and generate a personalized status assessment report;

[0178] The generation module 24 is used to determine specific intervention measures based on the personalized status assessment report, convert it into an executable operation guide through the intelligent recommendation system, and generate a response plan for the status of the injured and sick.

[0179] Figure 2 The patient status assessment system based on real-time monitoring of vital signs can be implemented Figure 1 The implementation principle and technical effect of the patient status assessment method based on real-time monitoring of vital signs described in the embodiment shown are not repeated here. The specific way in which each module and unit performs operations in the patient status assessment system based on real-time monitoring of vital signs in the above embodiment has been described in detail in the embodiment of the method, and will not be elaborated here.

[0180] In one possible design, Figure 2 The patient status assessment system based on real-time monitoring of vital signs of the embodiment shown can be implemented as a computing device, such as Figure 3 As shown, the computing device may include a storage component 31 and a processing component 32;

[0181] The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are called and executed by the processing component 32 .

[0182] The processing component 32 is used to: collect the vital signs data stream of the injured and sick in real time through wearable devices, perform preliminary standardization processing, and generate a standardized vital signs data sequence; based on the standardized vital signs data sequence, use a dynamic Bayesian network algorithm to establish a temporal dependency model between vital signs parameters, identify potential abnormal vital signs patterns, use wavelet transform technology to decompose the signal spectrum, refine abnormal signal characteristics, and generate refined abnormal pattern data; based on the refined abnormal pattern data, use a deep forest algorithm to construct a multi-level decision tree set, combine historical health records with current environmental factors for comprehensive evaluation, use time series analysis to capture vital signs change trends, and generate a personalized status assessment report; based on the personalized status assessment report, determine specific intervention measures and recommendations, convert them into executable operation guidelines through an intelligent recommendation system, and generate a status response plan for the injured and sick.

[0183] The processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above method. Of course, the processing component may also be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors or other electronic components to perform the above method.

[0184] The storage component 31 is configured to store various types of data to support operations at the terminal. The storage component can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0185] Of course, the computing device may also include other components, such as input / output interfaces, display components, communication components, etc.

[0186] The input / output interface provides an interface between the processing component and the peripheral interface module, which may be an output device, an input device, etc.

[0187] The communication component is configured to facilitate, among other things, wired or wireless communications between the computing device and other devices.

[0188] Among them, the computing device can be a physical device or an elastic computing host provided by a cloud computing platform, etc. In this case, the computing device can refer to a cloud server, and the above-mentioned processing components, storage components, etc. can be basic server resources rented or purchased from the cloud computing platform.

[0189] The present application also provides a computer storage medium storing a computer program, wherein the computer program can achieve the above-mentioned Figure 1 The illustrated embodiment is a method for assessing the status of a patient based on real-time monitoring of vital signs.

[0190] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0191] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0192] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0193] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for assessing the status of patients based on real-time monitoring of vital signs, characterized in that: include: Through wearable devices, the patient's vital signs data stream is collected in real time, and preliminary standardization processing is performed to generate a standardized vital signs data sequence; Based on the standardized vital sign data sequence, a dynamic Bayesian network algorithm is used to establish a temporal dependency model between vital sign parameters, identify potential abnormal vital sign patterns, and use wavelet transform technology to decompose the signal spectrum, refine abnormal signal features, and generate refined abnormal pattern data; Based on the refined abnormal pattern data, a multi-level decision tree set is constructed using the deep forest algorithm, a comprehensive evaluation is performed in combination with historical health records and current environmental factors, and time series analysis is used to capture the trend of vital signs changes to generate a personalized status assessment report; Based on the personalized status assessment report, specific intervention measures are recommended, which are converted into executable operation guidelines through an intelligent recommendation system to generate a response plan for the status of the injured and sick.

2. The method according to claim 1, characterized in that Based on the standardized vital sign data sequence, a dynamic Bayesian network algorithm is used to establish a temporal dependency model between vital sign parameters, identify potential abnormal vital sign patterns, use wavelet transform technology to decompose signal spectrum, refine abnormal signal features, and generate refined abnormal pattern data, including: Based on the standardized vital sign data sequence, denoising preprocessing and missing value filling are performed to generate a high-quality data set; Based on the high-quality data set, a dynamic Bayesian network algorithm is used to learn the interactions and changing patterns between different vital signs, capture the complex temporal dependencies between vital sign parameters, and generate a vital sign temporal dependency model; Based on the vital signs temporal dependency model, data points with large deviations from normal patterns are detected, potential abnormal vital signs patterns are identified, and signal spectrum is decomposed using wavelet transform technology to refine and highlight specific features of abnormal signals and generate abnormal signal feature sets; Based on the abnormal signal feature set, combined with time information and signal strength key indicators, refined abnormal pattern data is generated.

3. The method according to claim 2, characterized in that Based on the high-quality data set, the dynamic Bayesian network algorithm is used to learn the interactions and change rules between different vital signs, capture the complex temporal dependencies between vital sign parameters, and generate a vital sign temporal dependency model, including: Based on the high-quality data set, the data is segmented into time series, divided into multiple time windows, and time series segmentation data is generated; Based on the time series segmentation data, a dynamic Bayesian network algorithm is used to construct a network structure to integrate vital sign parameters and potential dependencies, preliminarily learn the interactions between vital sign parameters, and generate a preliminary dynamic Bayesian network structure; Based on the preliminary dynamic Bayesian network structure, the connection weights and conditional probability distributions between nodes are adjusted by expectation maximization to accurately reflect the actual interactions between vital sign parameters and generate an optimized dynamic Bayesian network structure; Based on the optimized dynamic Bayesian network structure, the network structure and parameters are continuously updated iteratively to converge the model and generate a vital sign temporal dependency model.

4. The method according to claim 2, characterized in that: Based on the vital signs temporal dependency model, data points with large deviations from normal patterns are detected, potential abnormal vital signs patterns are identified, and signal spectrum decomposition is performed using wavelet transform technology to refine and highlight specific features of abnormal signals, and generate abnormal signal feature sets, including: Based on the vital sign time series dependency model, real-time monitoring and analysis of the vital sign data in the high-quality data set, detecting data points with large deviations from normal patterns, and generating potential abnormal data points; Based on the potential abnormal data points, further analyzing time series characteristics, confirming and identifying potential abnormal vital sign patterns, and generating potential abnormal patterns; Based on the potential abnormal pattern, wavelet transform technology is used to perform spectrum decomposition on the abnormal signal, extract signal features under different frequency components, refine and highlight the specific features of the abnormal signal, and generate detailed abnormal signal features; Based on the detailed abnormal signal features, combined with time information and signal strength key indicators, an abnormal signal feature set is generated.

5. The method according to claim 1, characterized in that Based on the refined abnormal pattern data, a multi-level decision tree set is constructed using the deep forest algorithm, a comprehensive evaluation is performed in combination with historical health records and current environmental factors, and time series analysis is used to capture the trend of vital signs changes to generate a personalized status assessment report, including: Based on the refined abnormal pattern data, statistical features and time domain features are extracted to ensure data quality and generate high-quality abnormal pattern data; Based on the high-quality abnormal pattern data, a deep forest algorithm is used to combine multiple random subspaces and random forests, traverse all levels of decision trees for recursive partitioning, capture data features at different levels, and generate a multi-level decision tree set; Based on the multi-level decision tree set, a multi-level comprehensive evaluation is performed in combination with historical health records and current environmental factors to generate a multi-level comprehensive evaluation result; Based on the multi-level comprehensive evaluation results, time series analysis is used to identify temporal dynamic characteristics, capture potential trends and periodic changes, and generate a personalized status assessment report.

6. The method according to claim 5, characterized in that Based on the high-quality abnormal pattern data, the deep forest algorithm is used to combine multiple random subspaces and random forests, traverse all levels of decision trees for recursive partitioning, capture data features at different levels, and generate a multi-level decision tree set, including: Based on the high-quality abnormal pattern data, normalization processing is performed to unify the range of each vital sign parameter to the same scale, screen out the most influential features for abnormal pattern recognition, and generate normalized abnormal pattern data; Based on the normalized abnormal pattern data, a deep forest algorithm is used to randomly divide multiple subspaces, reduce feature dimensions to improve the generalization ability of the model, and a random forest is constructed in all subspaces to generate an initial subspace data set; Based on the initial subspace data set, randomly select features and samples to generate multiple decision trees, traverse all hierarchical decision trees for recursive partitioning, gradually refine data features, and generate a preliminary multi-level decision tree structure; Based on the preliminary multi-level decision tree structure, data features at different levels are captured to generate a multi-level decision tree set.

7. The method according to claim 1, characterized in that Based on the personalized status assessment report, specific intervention measures are determined, converted into executable operation guidelines through the intelligent recommendation system, and a response plan for the status of the injured and sick is generated, including: Based on the personalized status assessment report, identify key health issues and potential risks, and generate specific intervention recommendations; Based on the specific intervention measures, the intelligent recommendation system is used to optimize the process, and personalized adjustments are made based on the injury history and health records to generate a preliminary intervention plan; Based on the preliminary intervention plan, break it down into detailed steps and implementation instructions to generate executable operation guidelines; Based on the executable operation guide, combined with the follow-up tracking plan and emergency response measures, a response plan for the status of the injured and sick is generated.

8. A patient status assessment system based on real-time monitoring of vital signs, characterized in that: include: The acquisition module is used to collect the vital signs data stream of the injured and sick in real time through wearable devices, perform preliminary standardization processing, and generate a standardized vital signs data sequence; An identification module is used to establish a temporal dependency model between vital sign parameters based on the standardized vital sign data sequence using a dynamic Bayesian network algorithm, identify potential abnormal vital sign patterns, decompose signal spectra using wavelet transform technology, refine abnormal signal features, and generate refined abnormal pattern data; An analysis module is used to construct a multi-level decision tree set based on the refined abnormal pattern data using a deep forest algorithm, conduct a comprehensive evaluation based on historical health records and current environmental factors, use time series analysis to capture the trend of vital signs changes, and generate a personalized status assessment report; A generation module is used to determine specific intervention measures based on the personalized status assessment report, convert it into an executable operation guide through an intelligent recommendation system, and generate a response plan for the status of the injured and sick.

9. A computing device, characterized in that It comprises a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement a method for assessing the status of a patient based on real-time monitoring of vital signs as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that: A computer program is stored, and when the computer program is executed by a computer, a method for assessing the status of a patient based on real-time monitoring of vital signs as described in any one of claims 1 to 7 is implemented.

Images