Critical patient data index monitoring system and method

Abstract

The invention discloses a critical patient data index monitoring system and method, and the system comprises a data collection and fusion module which is used for accessing a multi-source medical device, and carrying out the data collection, protocol analysis, time sequence alignment and noise filtering; the dynamic association analysis module is used for constructing a risk conduction relation chain based on a knowledge graph and a dynamic association entropy weight algorithm and discovering a novel association relation in real time; the intelligent decision module is used for risk prediction, early warning generation and decision support; the interactive application module is used for displaying a risk conduction path and pushing early warning information; the dynamic association analysis module further comprises a knowledge graph library used for storing a pathological relation chain; the association chain generator is used for automatically generating a potential risk conduction path on the basis of the knowledge graph; the risk simulation engine is used for carrying out large-scale pathological state deduction and carrying out efficient calculation based on a GPU acceleration Bayesian network; the method has the characteristics of accuracy and comprehensiveness.

Description

Technical Field

[0001] The present invention relates to the field of medical information technology, and in particular to a system and method for monitoring data indicators of critically ill patients. Background Art

[0002] In the intensive care unit (ICU) environment, monitoring patients' vital signs is crucial. Traditional monitoring systems often face problems such as heterogeneous data interfaces from multiple medical devices, uneven data quality, and complex pathological relationships that are difficult to analyze in real time. For example, different medical devices use different data transmission protocols, which makes data collection and integration difficult. The time difference and noise interference between devices further affect the accuracy and reliability of the data. In addition, the condition of critically ill patients often involves the interaction of multiple physiological systems. The dynamic changes in pathological relationships and the complexity of the conduction pathways make risk prediction and early warning generation extremely challenging.

[0003] Traditional methods for monitoring critically ill patient data indicators rely heavily on static models and empirical rules, making it difficult to capture the dynamic changes in pathological relationships and the emergence of new correlations. Furthermore, the lack of intuitive risk pathway visualization and effective decision-making support mechanisms makes it difficult for medical staff to make accurate and prompt judgments when faced with complex conditions. Therefore, there is an urgent need for a system that can accurately and comprehensively monitor critically ill patient data indicators in real time, dynamically analyze pathological relationships, predict risks, generate early warnings, and provide decision support. Summary of the Invention

[0004] The purpose of the present invention is to provide a system and method for monitoring data indicators of critically ill patients to solve the problems raised in the above background technology.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions: a critically ill patient data indicator monitoring system, comprising a data acquisition and fusion module for accessing multi-source medical equipment to perform data acquisition, protocol analysis, timing alignment, and noise filtering;

[0006] Dynamic association analysis module, used to build risk transmission relationship chains based on knowledge graphs and dynamic association entropy weight algorithms, and discover new association relationships in real time;

[0007] Intelligent decision-making module for risk prediction, warning generation and decision support;

[0008] Interactive application module, used to display risk transmission paths and push early warning information;

[0009] The dynamic association analysis module further includes:

[0010] Knowledge graph library, used to store pathological relationship chains;

[0011] Association chain generator, used to automatically generate potential risk transmission paths based on the knowledge graph;

[0012] Risk simulation engine, used for large-scale pathological state deduction, based on GPU-accelerated Bayesian networks for efficient calculation;

[0013] An online learning unit is used to annotate false positive cases based on doctor feedback, dynamically update knowledge graph node weights, and update no less than 15% of association rules weekly;

[0014] The intelligent decision-making module further includes:

[0015] Cluster analysis unit, used to classify different types of pathological relationship chains;

[0016] An early warning generator, used to generate differentiated response instructions based on risk levels, including low-risk pop-up reminders and high-risk medical terminal strong reminders;

[0017] Decision support library, used to store historical case plans.

[0018] According to the above technical solution, the data acquisition and fusion module further includes:

[0019] Multi-source interface unit, used to be compatible with the data transmission protocols of different medical devices, and to achieve data access and format conversion;

[0020] Timing alignment unit, used to eliminate millisecond-level acquisition time differences between devices based on a dynamic time warping algorithm;

[0021] Noise filtering unit to improve data quality.

[0022] According to the above technical solution, the interactive application module further includes:

[0023] A three-dimensional visualization console is used to dynamically display the risk transmission pathways between organs and present the pathological spread in an intuitive manner;

[0024] The multi-terminal push interface is used to send warning information to different devices and realize multi-terminal synchronous notification.

[0025] A method for monitoring data indicators of critically ill patients, comprising:

[0026] Step S1: Connecting ICU equipment through a multi-source interface unit to achieve real-time collection of vital sign data, followed by data cleaning, protocol analysis, and time sequence alignment;

[0027] Step S2: Using knowledge graph drive, pathological relationships are extracted from the clinical guideline library to construct a medical knowledge graph;

[0028] Step S3: A dynamic association entropy weight algorithm is used to calculate the association weights between variables. When the rate of change of the entropy value exceeds a preset threshold, new associations are automatically marked. The risk simulation engine in the dynamic association analysis module is used to simulate the pathological diffusion path based on a GPU-accelerated Bayesian network.

[0029] Step S4: Establish a hierarchical response mechanism and take corresponding response actions according to the risk level;

[0030] Step S5: Use an online learning engine to automatically update at least 15% of the association rule base every week to optimize the prediction model. Through a closed-loop mechanism of doctor feedback, false positive cases are marked and the model weights are optimized.

[0031] According to the above technical solution, the implementation of step S1 specifically includes:

[0032] Establish stable connection with various medical devices in ICU through multi-source interface unit;

[0033] Real-time collection of vital sign data output by the above-mentioned equipment, including heart rate, blood pressure, respiratory rate, and blood oxygen saturation;

[0034] Identify and eliminate obviously erroneous or invalid data points to pre-process the collected raw data;

[0035] When a data point x satisfies |x-μ|>3σ', it is judged as abnormal, where μ is the sliding window mean and σ' is the corrected standard deviation. Abnormal data is filtered to ensure data quality so that the effective data rate is not lower than the preset standard;

[0036] Identify and parse the data transmission protocols used by different medical devices to achieve unified conversion of data formats for subsequent processing and analysis;

[0037] When collecting data and causing timing asynchrony due to differences in collection frequency or transmission delay between devices, a dynamic time warping algorithm is used to eliminate the millisecond-level collection time difference between devices and ensure the consistency of data in timing.

[0038] According to the above technical solution, step S3 further includes:

[0039] Based on the calculation results of dynamic correlation entropy weight, strongly correlated variable pairs with weight values ≥ 0.7 are extracted and automatically combined to generate correlation chains according to the dependency relationship of physiological systems;

[0040] The levels are divided according to the number of nodes contained in the association chain, among which,

[0041] When the chain length is between 3 and 5 nodes, it is classified as a short chain, which is used to characterize the pathological conduction within a single organ;

[0042] When the chain length is between 6 and 9 nodes, it is classified as a medium chain to characterize the risk spread between adjacent organs;

[0043] When the chain length is 10 nodes or more, it is classified as a long chain to characterize multi-system cascade reactions.

[0044] According to the above technical solution, step S3 specifically includes:

[0045] Short chain risk assessment: Using a real-time trend comparison algorithm, a primary warning is triggered when three consecutive variables in the short chain fluctuate by more than 20% of the baseline value;

[0046] Long-chain risk deduction: The cross-system transmission probability is calculated based on the organ state transition matrix. The formula is:

[0047]

[0048] Where ΔE i is the change in energy metabolism of organ i, E max Take 120% of the patient's energy metabolism value in a healthy state, W 解剖 is the anatomical adjacency weight coefficient;

[0049] Divide the risk level according to the transmission probability. 传导 <0.3: judged as low risk and marked in blue; when 0.3≤P 传导 <0.6: judged as medium risk and marked in yellow; P 传导 ≥0.6: Judged as high risk and marked in red.

[0050] According to the above technical solution, the implementation method of the hierarchical response mechanism in step S4 is:

[0051] Initiate local device-side alerts for short-chain primary warnings;

[0052] The risk warning in the long chain triggers the vibration alarm of the medical staff PDA;

[0053] The emergency call system will be activated simultaneously with the long-chain high-risk warning, and the corresponding case handling plan will be pre-loaded.

[0054] According to the above technical solution, step S3 further includes:

[0055] By improving the K-means algorithm, feature vectors were extracted with a 5-minute time window, and the cluster radius was set to Euclidean distance ≤ 0.35 to quickly cluster short chains;

[0056] Construct the disease variable similarity matrix and perform cluster analysis on the long-chain spectrum. The calculation expression is:

[0057]

[0058] Where σ is the Gaussian kernel width adjustment factor, and the value of σ is dynamically adjusted by the maximum likelihood estimation method, with the initial preset value being 0.8;

[0059] Finally, the abnormal chains are detected. When the silhouette coefficient of one cluster is less than 0.2, it is determined to be an abnormal association pattern and the new association discovery process is started.

[0060] According to the above technical solution, starting the new association discovery process specifically includes:

[0061] Extract the timestamp features and spatial distribution features of the anomaly chain;

[0062] Perform subgraph matching with the knowledge graph library, and generate candidate association pairs for the unmatched parts;

[0063] When the dynamic entropy weight of a candidate association pair increases by more than 15% for five consecutive collection cycles, it is automatically added to the knowledge graph.

[0064] Compared with the existing technology, the beneficial effects achieved by the present invention are as follows: the present invention realizes the spatiotemporal consistency fusion of cross-device data through protocol adaptive parsing and millisecond-level timing alignment technology of multi-source heterogeneous medical devices, and solves the monitoring lag problem caused by ICU equipment data islands. Secondly, a two-way driving mechanism of dynamic correlation entropy weight algorithm and knowledge graph is introduced. Under the existing medical knowledge framework, the cascade effect of risk transmission between organs is deduced through Bayesian network, and the online learning engine is used to capture new pathological associations, so that the system has the ability of continuous evolution of clinical cognition. At the same time, a hierarchical response mechanism and multimodal interactive design are adopted to transform the complex pathological transmission logic into a visual risk diffusion path, and dynamically evaluate the probability of multiple organ failure based on anatomical adjacency weights, so that clinical decision-making can be transformed from experience-driven to data-driven precise intervention. It effectively improves the real-time and predictive nature of intensive care. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0066] Figure 1 This is a schematic diagram of a critically ill patient data indicator monitoring system of the present invention.

[0067] Figure 2 This is a schematic diagram of a method for monitoring data indicators of critically ill patients according to the present invention. DETAILED DESCRIPTION

[0068] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0069] Example 1

[0070] See also Figure 1 This embodiment discloses a critically ill patient data indicator monitoring system, including:

[0071] The data acquisition and fusion module is used to access multi-source medical equipment to perform data acquisition, protocol analysis, timing alignment, and noise filtering. The data acquisition and fusion module further includes:

[0072] Multi-source interface unit, used to be compatible with data transmission protocols of different medical devices, realize data access and format conversion, support HL7 and DICOM protocols, and protocol parsing delay is less than 10ms;

[0073] The timing alignment unit is used to eliminate millisecond-level acquisition time differences between devices based on a dynamic time warping algorithm, ensuring that the alignment error does not exceed 50ms.

[0074] Noise filtering unit, used to improve data quality, removes abnormal data based on the improved 3σ criterion, and ensures that the effective data rate is not less than 98.5%;

[0075] Dynamic association analysis module, used to build risk transmission relationship chains based on knowledge graphs and dynamic association entropy weight algorithms, and discover new association relationships in real time;

[0076] The dynamic association analysis module further includes:

[0077] Knowledge graph library, used to store pathological relationship chains;

[0078] Association chain generator, used to automatically generate potential risk transmission paths based on the knowledge graph;

[0079] Risk simulation engine, used for large-scale pathological state deduction, based on GPU-accelerated Bayesian networks for efficient calculation;

[0080] An online learning unit is used to annotate false positive cases based on doctor feedback, dynamically update knowledge graph node weights, and update no less than 15% of association rules weekly;

[0081] An intelligent decision-making module is used for risk prediction, warning generation, and decision support; wherein the intelligent decision-making module further includes:

[0082] Cluster analysis unit, used to classify different types of pathological relationship chains, using the K-means algorithm to analyze short-chain risks and the spectral clustering algorithm to analyze long-chain complex pathological relationships;

[0083] An early warning generator, which generates differentiated response instructions based on risk levels, including low-risk pop-up reminders and high-risk medical terminal strong reminders. It also generates early warning information in JSON format, including risk score, risk level, and disposal suggestions.

[0084] Decision support library, used to store historical case plans and support rapid matching based on case similarity. The database contains more than 3,000 historical cases, with a similarity matching accuracy rate of no less than 92%;

[0085] Interactive application module, used to display risk transmission paths and push early warning information;

[0086] The interactive application module further includes:

[0087] A three-dimensional visualization console is used to dynamically display the risk transmission pathways between organs and present the pathological spread in an intuitive manner;

[0088] The multi-terminal push interface is used to send warning information to different devices and realize multi-terminal synchronous notification.

[0089] By designing a data acquisition and fusion module, the problem of heterogeneous protocols in traditional ICU equipment is cleverly solved, achieving unified access to multi-source medical equipment and efficient data cleaning and processing, laying a solid foundation for subsequent analysis. Secondly, by introducing a knowledge graph and a dynamic association entropy weight algorithm, it can not only automatically construct and update the medical knowledge graph, but also discover new associations in real time, greatly improving the accuracy and timeliness of pathological conduction pathway predictions. In addition, the intelligent decision-making module achieves accurate prediction and differentiated early warning of risks through cluster analysis and a hierarchical response mechanism, providing medical staff with timely and effective decision-making support. Finally, the interactive application module's three-dimensional visualization console and multi-terminal push interface display risk conduction pathways in an intuitive and convenient manner, and realizes multi-terminal synchronous notification of early warning information, greatly improving clinical work efficiency and patient safety.

[0090] Example 2

[0091] See also Figure 2 This embodiment discloses a method for monitoring data indicators of critically ill patients, including:

[0092] Step S1: Connect ICU equipment, including ECG monitors and ventilators, through a multi-source interface unit to collect vital sign data in real time, and then perform data cleaning, protocol analysis, and timing alignment.

[0093] The implementation of step S1 specifically includes:

[0094] Establish stable connection with various medical devices in ICU through multi-source interface unit;

[0095] Real-time collection of vital sign data output by the above-mentioned equipment, including heart rate, blood pressure, respiratory rate, and blood oxygen saturation;

[0096] Identify and eliminate obviously erroneous or invalid data points to pre-process the collected raw data;

[0097] When a data point x satisfies |x-μ|>3σ', it is judged as abnormal, where μ is the sliding window mean and σ' is the corrected standard deviation. By dynamically adjusting the standard deviation threshold, the traditional 3σ criterion is avoided from over-filtering non-normally distributed data, abnormal data is filtered out, and data quality is ensured so that the effective data rate is not lower than the preset standard.

[0098] Identify and parse the data transmission protocols used by different medical devices to achieve unified conversion of data formats for subsequent processing and analysis;

[0099] When collecting data and causing timing asynchrony due to differences in collection frequency or transmission delay between devices, a dynamic time warping algorithm is used to eliminate the millisecond-level collection time difference between devices and ensure the consistency of data in timing.

[0100] Step S2: Using knowledge graph drive, pathological relationships are extracted from the clinical guideline library to construct a medical knowledge graph;

[0101] Step S3: The dynamic association entropy weight algorithm is used to calculate the association weights between variables. When the entropy value change rate exceeds the preset threshold, the new association is automatically marked. The pathological diffusion path simulation is performed based on the GPU-accelerated Bayesian network through the risk simulation engine in the dynamic association analysis module.

[0102] Step S3 further comprises:

[0103] Based on the calculation results of dynamic correlation entropy weight, strongly correlated variable pairs with weight values ≥ 0.7 are extracted and automatically combined to generate correlation chains according to the dependency relationship of physiological systems;

[0104] The levels are divided according to the number of nodes contained in the association chain, among which,

[0105] When the chain length is between 3 and 5 nodes, it is classified as a short chain, which is used to characterize the pathological conduction within a single organ;

[0106] When the chain length is between 6 and 9 nodes, it is classified as a medium chain to characterize the risk spread between adjacent organs;

[0107] When the chain length is 10 nodes or more, it is classified as a long chain to characterize multi-system cascade reactions.

[0108] Among them, short chain risk assessment: using a real-time trend comparison algorithm, when three consecutive variables in the short chain fluctuate by more than 20% of the baseline value, a primary warning is triggered;

[0109] Long-chain risk deduction: The cross-system transmission probability is calculated based on the organ state transition matrix. The formula is:

[0110]

[0111] Where ΔE i is the change in energy metabolism of organ i, E max Take 120% of the patient's energy metabolism value in a healthy state, W 解剖 The anatomical adjacency weight coefficient is calculated based on organ spacing data from Gray's Anatomy atlas. The weight decreases by 0.1 for every 5 cm increase in spacing. This coefficient quantifies the cascade risk of multiple organ failure through the combined effects of metabolic changes and anatomical relationships. The product form reflects the assumption of event independence, and the 1-π structure indicates that deterioration of any organ may trigger conduction.

[0112] Divide the risk level according to the transmission probability. 传导 <0.3: judged as low risk and marked in blue; when 0.3≤P 传导 <0.6: judged as medium risk and marked in yellow; P 传导 ≥0.6: Judged as high risk and marked in red.

[0113] Step S3 further includes:

[0114] By improving the K-means algorithm, feature vectors were extracted with a 5-minute time window, and the cluster radius was set to Euclidean distance ≤ 0.35 to quickly cluster short chains;

[0115] Construct the disease variable similarity matrix and perform cluster analysis on the long-chain spectrum. The calculation expression is:

[0116]

[0117] Where σ is the Gaussian kernel width adjustment factor, and the value of σ is dynamically adjusted by the maximum likelihood estimation method, with an initial preset value of 0.8. In the formula, by integrating the temporal and spatial similarity with the correlation of pathological changes, abnormal association patterns can be more accurately identified;

[0118] Finally, the abnormal chains are detected. When the silhouette coefficient of one cluster is less than 0.2, it is determined to be an abnormal association pattern and the new association discovery process is started.

[0119] The process of initiating new association discovery specifically includes:

[0120] Extract the timestamp features and spatial distribution features of the anomaly chain;

[0121] Perform subgraph matching with the knowledge graph library, and generate candidate association pairs for the unmatched parts;

[0122] When the dynamic entropy weight of a candidate association pair increases by more than 15% for five consecutive collection cycles, it is automatically added to the knowledge graph.

[0123] Step S4: Establish a hierarchical response mechanism and take corresponding response actions according to the risk level; the implementation method of the hierarchical response mechanism in step S4 is as follows:

[0124] Initiate local device-side alerts for short-chain primary warnings;

[0125] The risk warning in the long chain triggers the vibration alarm of the medical staff PDA;

[0126] The emergency call system is activated simultaneously with long-chain high-risk warnings, and the corresponding case handling plan is pre-loaded;

[0127] Step S5: Using an online learning engine, automatically update at least 15% of the association rule base every week to optimize the prediction model. Through a closed-loop mechanism of doctor feedback, mark false positive cases and optimize the model weight. By building a complete data monitoring method process, covering the five core steps of data collection, knowledge graph construction, dynamic association analysis, hierarchical response and model optimization, it is possible to achieve unified access of medical devices through multi-source interface units, solve the problem of heterogeneous protocols of traditional ICU equipment, and use the dynamic association entropy weight algorithm combined with Bayesian network deduction to improve the prediction accuracy of pathological conduction pathways.

[0128] The purpose of the present invention is to construct an intensive care system that organically integrates dynamic knowledge evolution and intelligent decision-making, breaking through the limitations of the static rule base of the traditional monitoring system. First, through the protocol adaptive parsing and millisecond-level timing alignment technology of multi-source heterogeneous medical devices, the spatiotemporal consistency fusion of cross-device data is achieved, solving the monitoring lag problem caused by ICU equipment data islands. Secondly, a two-way driving mechanism of dynamic correlation entropy weight algorithm and knowledge graph is introduced. Under the existing medical knowledge framework, the cascade effect of risk transmission between organs is deduced through the Bayesian network, and the online learning engine is used to capture new pathological associations, so that the system has the ability to continuously evolve clinical cognition. At the same time, a hierarchical response mechanism and multimodal interactive design are adopted to transform the complex pathological transmission logic into a visual risk diffusion path, and dynamically evaluate the probability of multiple organ failure based on anatomical adjacency weights, so that clinical decision-making can be transformed from experience-driven to data-driven precise intervention. It effectively improves the real-time and predictive nature of intensive care.

[0129] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0130] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0131] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0132] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the purpose of the present invention and the claims, which are all protected by the present invention.

Claims

1. A critically ill patient data indicator monitoring system, characterized by include: Data acquisition and fusion module, used to access multi-source medical equipment for data acquisition, protocol analysis, timing alignment, and noise filtering; Dynamic association analysis module, used to build risk transmission relationship chains based on knowledge graphs and dynamic association entropy weight algorithms, and discover new association relationships in real time; Intelligent decision-making module for risk prediction, warning generation and decision support; Interactive application module, used to display risk transmission paths and push early warning information; The dynamic association analysis module further includes: Knowledge graph library, used to store pathological relationship chains; Association chain generator, used to automatically generate potential risk transmission paths based on the knowledge graph; Risk simulation engine, used for large-scale pathological state deduction, based on GPU-accelerated Bayesian networks for efficient calculation; An online learning unit is used to annotate false positive cases based on doctor feedback, dynamically update knowledge graph node weights, and update no less than 15% of association rules weekly; The intelligent decision-making module further includes: Cluster analysis unit, used to classify different types of pathological relationship chains; An early warning generator, used to generate differentiated response instructions based on risk levels, including low-risk pop-up reminders and high-risk medical terminal strong reminders; Decision support library, used to store historical case plans.

2. The critically ill patient data indicator monitoring system according to claim 1, characterized in that: The data acquisition and fusion module further includes: Multi-source interface unit, used to be compatible with the data transmission protocols of different medical devices, and to achieve data access and format conversion; Timing alignment unit, used to eliminate millisecond-level acquisition time differences between devices based on a dynamic time warping algorithm; Noise filtering unit to improve data quality.

3. The critically ill patient data indicator monitoring system according to claim 1, characterized in that: The interactive application module further includes: A three-dimensional visualization console is used to dynamically display the risk transmission pathways between organs and present the pathological spread in an intuitive manner; The multi-terminal push interface is used to send warning information to different devices and realize multi-terminal synchronous notification.

4. A method for monitoring data indicators of critically ill patients, characterized in that include: Step S1: Connecting ICU equipment through a multi-source interface unit to achieve real-time collection of vital sign data, followed by data cleaning, protocol analysis, and time sequence alignment; Step S2: Using knowledge graph drive, pathological relationships are extracted from the clinical guideline library to construct a medical knowledge graph; Step S3: A dynamic association entropy weight algorithm is used to calculate the association weights between variables. When the rate of change of the entropy value exceeds a preset threshold, new associations are automatically marked. The risk simulation engine in the dynamic association analysis module is used to simulate the pathological diffusion path based on a GPU-accelerated Bayesian network. Step S4: Establish a hierarchical response mechanism and take corresponding response actions according to the risk level; Step S5: Use an online learning engine to automatically update at least 15% of the association rule base every week to optimize the prediction model. Through a closed-loop mechanism of doctor feedback, false positive cases are marked and the model weights are optimized.

5. The method for monitoring data indicators of critically ill patients according to claim 4, characterized in that: The implementation of step S1 specifically includes: Establish stable connection with various medical devices in ICU through multi-source interface unit; Real-time collection of vital sign data output by the above-mentioned equipment, including heart rate, blood pressure, respiratory rate, and blood oxygen saturation; Identify and eliminate obviously erroneous or invalid data points to pre-process the collected raw data; When a data point x satisfies |x-μ|>3σ', it is judged as abnormal, where μ is the sliding window mean and σ' is the corrected standard deviation. Abnormal data is filtered to ensure data quality so that the effective data rate is not lower than the preset standard; Identify and parse the data transmission protocols used by different medical devices to achieve unified conversion of data formats for subsequent processing and analysis; When collecting data and causing timing asynchrony due to differences in collection frequency or transmission delay between devices, a dynamic time warping algorithm is used to eliminate the millisecond-level collection time difference between devices and ensure the consistency of data in timing.

6. The method for monitoring data indicators of critically ill patients according to claim 4, characterized in that: The step S3 further comprises: Based on the calculation results of dynamic correlation entropy weight, strongly correlated variable pairs with weight values ≥ 0.7 are extracted and automatically combined to generate correlation chains according to the dependency relationship of physiological systems; The levels are divided according to the number of nodes contained in the association chain, among which, When the chain length is between 3 and 5 nodes, it is classified as a short chain, which is used to characterize the pathological conduction within a single organ; When the chain length is between 6 and 9 nodes, it is classified as a medium chain to characterize the risk spread between adjacent organs; When the chain length is 10 nodes or more, it is classified as a long chain to characterize multi-system cascade reactions.

7. The method for monitoring data indicators of critically ill patients according to claim 6, characterized in that: The step S3 specifically includes: Short chain risk assessment: Using a real-time trend comparison algorithm, a primary warning is triggered when three consecutive variables in the short chain fluctuate by more than 20% of the baseline value; Long-chain risk deduction: The cross-system transmission probability is calculated based on the organ state transition matrix. The formula is: Where ΔE i is the change in energy metabolism of organ i, E max Take 120% of the patient's energy metabolism value in a healthy state, W 解剖 is the anatomical adjacency weight coefficient; Divide the risk level according to the transmission probability. 传导 <0.3: judged as low risk and marked in blue; when 0.3≤P 传导 <0.6: judged as medium risk and marked in yellow; P 传导 ≥0.6: Judged as high risk and marked in red.

8. The method for monitoring data indicators of critically ill patients according to claim 7, characterized in that: The implementation method of the hierarchical response mechanism in step S4 is: Initiate local device-side alerts for short-chain primary warnings; The risk warning in the long chain triggers the vibration alarm of the medical staff PDA; The emergency call system will be activated simultaneously with the long-chain high-risk warning, and the corresponding case handling plan will be pre-loaded.

9. The method for monitoring data indicators of critically ill patients according to claim 6, characterized in that: The step S3 further comprises: By improving the K-means algorithm, feature vectors were extracted with a 5-minute time window, and the cluster radius was set to Euclidean distance ≤ 0.35 to quickly cluster short chains; Construct the disease variable similarity matrix and perform cluster analysis on the long-chain spectrum. The calculation expression is: Where σ is the Gaussian kernel width adjustment factor, and the value of σ is dynamically adjusted by the maximum likelihood estimation method, with the initial preset value being 0.8; Finally, the abnormal chains are detected. When the silhouette coefficient of one cluster is less than 0.2, it is determined to be an abnormal association pattern and the new association discovery process is started.

10. The method for monitoring data indicators of critically ill patients according to claim 9, characterized in that: Initiating the new association discovery process specifically includes: Extract the timestamp features and spatial distribution features of the anomaly chain; Perform subgraph matching with the knowledge graph library, and generate candidate association pairs for the unmatched parts; When the dynamic entropy weight of a candidate association pair increases by more than 15% for five consecutive collection cycles, it is automatically added to the knowledge graph.

Images