Method and system for predicting injury condition of sick and wounded based on machine learning

Abstract

The invention provides a sick and wounded injury condition prediction method and system based on machine learning. The method comprises the following steps: collecting and integrating real-time information flow, and generating an initial data set of the sick and wounded; based on the initial data set of the sick and wounded, a random field algorithm is applied, complex relations and dynamic characteristics are captured, a time-dependent node embedding technology is adopted, a potential injury condition development trend and an abnormal mode are revealed, and an optimized sick and wounded state evolution graph is generated; based on the optimized state evolution graph of the sick and wounded, searching and positioning a historical case set highly similar to the current state of the sick and wounded by using a K-nearest neighbor algorithm, calculating the actual area covered by each target segmentation result by using a mask area filtering method, and generating an optimized state prediction model of the sick and wounded; and based on the optimized sick and wounded state prediction model, establishing a dynamic risk assessment and interactive feedback optimization mechanism, and generating a sick and wounded injury state prediction scheme. According to the technical scheme provided by the invention, the accuracy and reliability of injury prediction of the sick and wounded are comprehensively improved.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of medical health information technology, and in particular to a method and system for predicting the injury of an injured person based on machine learning. Background Art

[0002] With the rapid development of medical information technology and Internet of Things technology, the types and scale of medical data are exploding. Medical institutions need to collect information streams from diverse data sources in real time, including but not limited to electronic medical records, physiological monitoring equipment, wearable devices, environmental sensors, etc. Generating an initial data set for the wounded and sick through a multi-source data fusion method can provide comprehensive and accurate basic data for subsequent analysis. This application scenario places high demands on technology. It not only needs to process a large amount of heterogeneous data, but also needs to have strong real-time processing capabilities and efficient data integration capabilities to ensure the accurate construction of the evolution map of the wounded and sick status.

[0003] At present, the prediction of the injury status of the injured and sick mainly relies on traditional statistical analysis methods and simple machine learning models. For example, empirical rule matching based on historical cases, linear regression models, etc. In addition, some more advanced solutions have begun to introduce random field algorithms and time series analysis techniques to capture the complex relationships and dynamic characteristics in the evolution of the status of the injured and sick. However, most of these methods focus on static feature extraction and simple time series analysis, lack a deep understanding of time series characteristics, and fail to make full use of the latest deep learning and graph embedding technologies to enhance prediction capabilities.

[0004] Traditional methods often fail to effectively capture the complex relationships and dynamic characteristics in the evolution of the patient's status, resulting in insufficient accuracy in predicting the patient's condition. In addition, most existing methods ignore the application of time-dependent node embedding technology, making it difficult to reveal potential trends and abnormal patterns in the development of injuries. Especially when faced with large-scale, heterogeneous data, the data processing capabilities and real-time response speeds of existing solutions are also insufficient. Therefore, there is an urgent need for a more advanced and intelligent method that can comprehensively use multi-source data fusion, random field algorithms, time-dependent node embedding technology, and K-nearest neighbor algorithms to improve the accuracy and reliability of patient injury prediction, and establish a dynamic risk assessment and interactive feedback optimization mechanism to meet the needs of the modern medical environment. Summary of the invention

[0005] The embodiments of the present application provide a method and system for predicting the condition of the injured and sick based on machine learning, so as to solve the problem of insufficient accuracy in predicting the condition of the injured and sick in the prior art.

[0006] In a first aspect, an embodiment of the present application provides a method for predicting the injury of an injured person based on machine learning, comprising:

[0007] Collect and integrate real-time information streams from diverse data sources, and generate initial datasets of the injured and sick through multi-source data fusion methods;

[0008] Based on the initial data set of the injured and sick, the random field algorithm is used to process and analyze the time series data in the evolution map of the injured and sick states, capture the complex relationship and dynamic characteristics between the states of the injured and sick, and use the time-dependent node embedding technology to enhance the ability to understand the time series characteristics, reveal the potential development trend and abnormal pattern of the injury, and generate an optimized evolution map of the injured and sick states;

[0009] Based on the optimized sick and wounded status evolution map, the K nearest neighbor algorithm is used to perform similarity matching analysis, and a historical case set that is highly similar to the current sick and wounded status is searched and located to accurately adjust the weights. The mask area filtering method is used to calculate the actual area covered by each target segmentation result, and the results that are less than the preset threshold are filtered out to generate an optimized sick and wounded status prediction model.

[0010] Based on the optimized patient status prediction model, a dynamic risk assessment and interactive feedback optimization mechanism is established to generate a patient injury prediction plan.

[0011] Optionally, based on the initial data set of the injured and sick, a random field algorithm is used to process and analyze the time series data in the state evolution map of the injured and sick, capture the complex relationship and dynamic characteristics between the states of the injured and sick, adopt time-dependent node embedding technology to enhance the ability to understand the time series characteristics, reveal the potential development trend and abnormal pattern of the injury, and generate an optimized state evolution map of the injured and sick, including:

[0012] Based on the initial data set of the injured and sick, the time series data of the injured and sick are node-processed to generate the node timestamp of the status of the injured and sick;

[0013] Based on the timestamps of the patient status nodes, the random field algorithm is used to model the relationship between nodes, capture the complex relationship and dynamic characteristics between the patient status, identify the interdependence and time-varying nature between nodes in different statuses, reveal the potential development trend and abnormal pattern of the injury, and generate a random field model;

[0014] Based on the random field model, the time-dependent node embedding technology is used to deeply process and analyze the time series data in the state evolution map of the injured and sick, and each state node is embedded in the multidimensional space, so that the adjacent nodes are closer in space, and the time-dependent node embedding result is generated;

[0015] Based on the time-dependent node embedding results, key development trends and abnormal patterns are further annotated to generate an optimized evolution map of the status of the injured and sick.

[0016] Optionally, based on the timestamp of the patient status node, a random field algorithm is used to model the relationship between nodes, capture the complex relationship and dynamic characteristics between the patient status, identify the interdependence and time-varying nature between nodes in different statuses, reveal the potential injury development trend and abnormal pattern, and generate a random field model, including:

[0017] Based on the timestamps of the patient status nodes, a correlation analysis is performed on the patient status nodes corresponding to each timestamp to generate a patient status node relationship network;

[0018] Based on the node relationship network of the patient status, the random field algorithm is used to model the relationship between nodes, capture the complex relationship and dynamic characteristics between the patient status, identify the interdependence and time-varying nature between nodes of different statuses, reveal the potential development trend and abnormal pattern of the injury, and generate a node relationship model;

[0019] Based on the node relationship model, a statistical method is used to evaluate the stability and accuracy of the model to ensure that the real evolution of the status of the injured and sick is accurately reflected and an optimized node relationship model is generated;

[0020] Based on the optimized node relationship model, combined with the timestamp of the patient status node, the influence of external environmental factors is further introduced to generate a random field model.

[0021] Optionally, based on the random field model, the time-dependent node embedding technology is used to perform in-depth processing and analysis on the time series data in the state evolution map of the injured and sick, and each state node is embedded in a multidimensional space to make adjacent nodes closer in space, and generate a time-dependent node embedding result, including:

[0022] Based on the random field model, the time series data in the state evolution graph of the injured and sick is structured and converted into a format suitable for embedding operations to generate a structured state node set;

[0023] Based on the structured state node set, a time-dependent node embedding technique is used to map each state node into a multidimensional space, reflect the time dependency relationship and interaction between state nodes, ensure that adjacent nodes are closer in space, and generate a time-dependent node embedding vector;

[0024] Based on the time-dependent node embedding vector, the distance and position between nodes are optimized and adjusted to better capture the real dynamic characteristics of the evolution of the status of the injured and sick, and generate an optimized node embedding layout;

[0025] Based on the optimized node embedding layout, the path of the patient's state change and the time dependency are displayed through a visualization tool, and a time-dependent node embedding result is generated.

[0026] Optionally, based on the optimized sick and wounded status evolution map, a K-nearest neighbor algorithm is used to perform similarity matching analysis, a historical case set that is highly similar to the current sick and wounded status is searched and located, weights are accurately adjusted, a mask area filtering method is used, the actual area covered by each target segmentation result is calculated, results less than a preset threshold are filtered out, and an optimized sick and wounded status prediction model is generated, including:

[0027] Based on the optimized sick and wounded status evolution map, current sick and wounded status information is compared with data in the historical case library to generate a preliminary similar case set;

[0028] Based on the preliminary similar case set, the K nearest neighbor algorithm is used to perform accurate similarity matching analysis, by calculating the distance between the current sick and wounded status node and each case in the historical case set, selecting the closest K neighbor cases, searching for historical cases that are highly similar to the current situation, and generating a selected similar case set;

[0029] Based on the selected similar case set, a mask area filtering method is applied to calculate the actual area covered by each target segmentation result, and results smaller than a preset threshold are filtered out to generate case quality screening results;

[0030] Based on the case quality screening results, all high-quality case data are integrated to generate an optimized prediction model for the status of the injured and sick.

[0031] Optionally, based on the preliminary similar case set, a K nearest neighbor algorithm is used to perform an accurate similarity matching analysis, by calculating the distance between the current sick and wounded status node and each case in the historical case set, selecting the closest K neighbor cases, searching for historical cases that are highly similar to the current situation, and generating a selected similar case set, including:

[0032] Based on the preliminary similar case set, calculating the similarity between the time series feature and other related attributes to generate a preprocessed similar case set;

[0033] Based on the preprocessed similar case set, the K nearest neighbor algorithm is used to perform accurate similarity matching analysis, and the K closest neighbor cases are selected by calculating the distance between the current sick and wounded status node and each case in the historical case set to generate a selected neighbor case set;

[0034] Based on the selected neighbor case set, a detailed comparative analysis of the time series data of each selected case is performed to deeply evaluate the subtle differences between each neighbor case and the current status of the injured and sick, and generate a high-precision similar case assessment report;

[0035] Based on the high-precision similar case evaluation report, similarity confirmation and detailed comparative analysis are carried out to generate a selected similar case set.

[0036] Optionally, based on the optimized injury and patient status prediction model, a dynamic risk assessment and interactive feedback optimization mechanism is established to generate an injury prediction plan for the injured patients, including:

[0037] Based on the optimized injury and patient status prediction model, the evolution of the injured patients' status is monitored in real time, potential risk factors are identified and quantified, and a real-time risk factor report is generated;

[0038] Based on the real-time risk factor report, a dynamic risk assessment system is constructed. By combining the prediction model parameters with the real-time risk factor data, the probability of the injured patients' injury deterioration is obtained, and a dynamic risk assessment result is generated;

[0039] Based on the dynamic risk assessment result, an interactive feedback optimization mechanism is established. The post-event evaluations of medical experts are collected as incremental learning samples, and a feedback optimization record is generated;

[0040] Based on the feedback optimization record, the optimized injury and patient status prediction model is periodically fine-tuned to adapt to new injury situations and the evolution of medical practice, and an injury prediction plan for the injured patients is generated.

[0041] In a second aspect, an embodiment of the present application provides a machine learning-based injury prediction system for injured patients, including:

[0042] A collection module for collecting and integrating real-time information flows from diverse data sources, and generating an initial dataset of injured patients through a multi-source data fusion method;

[0043] An analysis module for processing and analyzing the time-series data in the injury and patient status evolution map based on the initial dataset of injured patients by using a random field algorithm, capturing the complex relationships and dynamic characteristics among the injured patients' statuses, adopting a time-dependent node embedding technique to enhance the understanding ability of temporal characteristics, revealing potential injury development trends and abnormal patterns, and generating an optimized injury and patient status evolution map;

[0044] An adjustment module for performing similarity matching analysis based on the optimized injury and patient status evolution map by using the K-nearest neighbor algorithm, searching and locating a historical case set highly similar to the current injured patient's condition to accurately adjust the weights, and using a masked area filtering method to calculate the actual area covered by each target segmentation result, filtering out results smaller than a preset threshold, and generating an optimized injury and patient status prediction model;

[0045] A generation module for establishing a dynamic risk assessment and interactive feedback optimization mechanism based on the optimized injury and patient status prediction model, and generating an injury prediction plan for the injured patients.

[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 predicting the injury of a patient based on machine learning 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 predicting the injury of a patient based on machine learning as described in the first aspect.

[0048] In the embodiment of the present application, real-time information streams from diverse data sources are collected and integrated, and an initial data set of the injured and sick is generated through a multi-source data fusion method; based on the initial data set of the injured and sick, a random field algorithm is used to process and analyze the time series data in the state evolution map of the injured and sick, so as to capture the complex relationship and dynamic characteristics between the states of the injured and sick, and a time-dependent node embedding technology is used to enhance the ability to understand the time series characteristics, reveal the potential development trend and abnormal patterns of injuries, and generate an optimized state evolution map of the injured and sick; based on the optimized state evolution map of the injured and sick, a K-nearest neighbor algorithm is used to perform similarity matching analysis, search and locate a historical case set that is highly similar to the current state of the injured and sick, so as to accurately adjust the weights, and a mask area filtering method is used to calculate the actual area covered by each target segmentation result, and the results less than the preset threshold are filtered out to generate an optimized state prediction model of the injured and sick; based on the optimized state prediction model of the injured and sick, a dynamic risk assessment and interactive feedback optimization mechanism is established to generate an injury prediction plan for the injured and sick. By integrating real-time information streams from diverse data sources, an initial dataset of patients is generated, and the patient status evolution map is processed using random field algorithms and time-dependent node embedding technology. This method can not only capture the complex relationships and dynamic characteristics between the patient status, but also reveal the potential development trends and abnormal patterns of injuries, and ultimately generate an optimized patient status evolution map. In addition, the accuracy of historical case selection and the quality of the prediction model are ensured by the K nearest neighbor algorithm and mask area filtering method. Finally, a dynamic risk assessment and interactive feedback optimization mechanism is established to improve the accuracy and timeliness of the patient injury prediction.

[0049] Furthermore, through node processing, random field algorithm modeling and time-dependent node embedding technology, this method can capture the time series characteristics of the evolution of the status of the wounded and sick more carefully. Specifically, the timestamps of the status nodes of the wounded and sick are generated, the random field model is constructed, and each status node is embedded in the multidimensional space, so that the adjacent nodes are closer in space. This process not only enhances the ability to understand the time series characteristics, but also marks the key development trends and abnormal patterns, and finally generates an optimized map of the evolution of the status of the wounded and sick. This makes subsequent analysis and prediction more accurate, providing a solid data foundation for medical decision-making.

[0050] Furthermore, a preliminary set of similar cases was generated through comparative processing, and the K nearest neighbor algorithm was used for accurate similarity matching analysis to select the closest K neighbor cases. Subsequently, the mask area filtering method was applied to filter out low-quality cases and generate high-quality case screening results. Finally, all high-quality case data were integrated to generate an optimized prediction model for the status of the injured and sick. This method ensures the accuracy of historical case selection, improves the reliability and practicality of the prediction model, provides a scientific basis for the prediction of the status of the injured and sick, and supports more accurate medical decision-making.

[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 predicting the injury of an injured or sick person based on machine learning provided in an embodiment of the present application;

[0054] Figure 2 A schematic diagram of the structure of a system for predicting the condition of an injured or sick person based on machine learning 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 predicting the injury of a patient based on machine learning is provided for an embodiment of the present application, such as Figure 1 As shown, the method includes:

[0060] 101. Collect and integrate real-time information streams from diverse data sources, and generate initial data sets of the injured and sick through multi-source data fusion methods;

[0061] In this step, diverse data sources include hospital information systems (such as electronic medical records, diagnostic records), wearable devices (such as heart rate monitors, smart bracelets), environmental sensors (such as air quality monitors, temperature and humidity meters), and social media platforms.

[0062] Real-time information flow refers to the continuous acquisition of the latest data from the above data sources to ensure that the system can respond to changes in the status of the injured and sick in a timely manner. This real-time performance is crucial for rapid decision-making in emergency situations.

[0063] Multi-source data fusion method refers to the technology of integrating and processing data from different sources. It not only includes simple data merging, but also involves pre-processing operations such as data cleaning, format conversion, missing value filling and outlier detection to ensure the quality and consistency of the data.

[0064] The initial dataset of the injured and sick is a unified data set generated after multi-source data fusion processing, which contains all relevant information of the injured and sick. This dataset provides basic support for subsequent analysis and prediction.

[0065] In an embodiment of the present application, relevant information of the injured and sick is first collected in real time from multiple data sources (such as hospital information systems, wearable devices, environmental sensors, and social media), and then these heterogeneous data are cleaned, format converted, and outliers detected through a multi-source data fusion method, and finally a high-quality initial data set of the injured and sick is generated, providing a solid foundation for subsequent analysis.

[0066] Suppose a hospital introduces a system for predicting the condition of the injured and sick based on machine learning; first, the system collects the basic information, diagnosis records and treatment plans of the injured and sick from the hospital's electronic medical record system in real time; second, it integrates physiological parameters such as heart rate, blood pressure, body temperature, etc. from wearable devices to provide more detailed health monitoring data; third, it combines external environmental factors such as air quality, temperature and humidity provided by environmental sensors to evaluate their impact on the condition of the injured and sick; finally, it uses social media platforms to obtain information on the psychological state and social support of the injured and sick, enriches the data dimension, and lays a solid foundation for subsequent analysis.

[0067] 102. Based on the initial data set of the injured and sick, the random field algorithm is used to process and analyze the time series data in the evolution map of the injured and sick states, capture the complex relationship and dynamic characteristics between the states of the injured and sick, and use time-dependent node embedding technology to enhance the ability to understand the time series characteristics, reveal the potential development trend and abnormal pattern of the injury, and generate an optimized evolution map of the states of the injured and sick;

[0068] In this step, the random field algorithm is a probabilistic model that is widely used in image processing, natural language processing and other fields. It captures the complex relationships and dynamic change characteristics between the injured and sick in different states by modeling time series data.

[0069] The patient status evolution map refers to a graph or chart that describes the changes in the patient's status over time. It shows the patient's status at different time points and helps identify their development trends and abnormal patterns.

[0070] Time series data is a series of data points arranged in chronological order, reflecting the continuous changes in the status of the injured and sick. This type of data is crucial to capturing the dynamic characteristics of the status of the injured and sick.

[0071] Time-dependent node embedding technology is a method that maps each state node into a multidimensional space, making adjacent nodes closer in space. This method enhances the understanding of temporal characteristics and helps to reveal potential trends and abnormal patterns in the evolution of the status of the injured and sick.

[0072] The optimized evolution map of the status of the injured and sick is the result of processing the original map through random field algorithm and time-dependent node embedding technology. It more accurately reflects the evolution process of the status of the injured and sick, and marks the key development trends and abnormal patterns.

[0073] In the embodiment of the present application, firstly, the system uses a random field algorithm to model the time series data in the state evolution map of the injured and sick based on the initial data set of the injured and sick, so as to capture the complex relationship and dynamic change characteristics between the injured and sick in different states; secondly, the time-dependent node embedding technology is applied to map each state node into a multidimensional space, so that adjacent nodes are closer in space; thirdly, the system automatically marks key development trends and abnormal patterns to help medical personnel quickly identify potential risks; finally, an optimized state evolution map of the injured and sick is generated to provide detailed data support for subsequent similarity matching analysis.

[0074] For example, continuing with the above example, assume that the system has generated an initial data set of injured and sick patients; first, the system uses a random field algorithm to model the time series data of the injured and sick patients to capture the interdependence between states; second, it applies time-dependent node embedding technology to embed each state node into a multidimensional space to enhance the understanding of time series characteristics; third, the system automatically labels key development trends and abnormal patterns to help medical personnel quickly identify potential risks; finally, it generates an optimized map of the evolution of the states of the injured and sick patients to provide detailed data support for subsequent similarity matching analysis and prediction model construction.

[0075] 103. Based on the optimized sick and wounded status evolution map, the K nearest neighbor algorithm is used to perform similarity matching analysis, and the historical case set that is highly similar to the current sick and wounded status is searched and located, so as to accurately adjust the weights, and the mask area filtering method is used to calculate the actual area covered by each target segmentation result, and the results less than the preset threshold are filtered out to generate an optimized sick and wounded status prediction model;

[0076] In this step, the K nearest neighbor algorithm is an instance-based learning method that selects the closest K neighbor cases by calculating the distance between the current patient status node and historical cases.

[0077] Similarity matching analysis refers to finding historical cases that are highly similar to the current conditions of the injured and sick by calculating and comparing the similarities between different cases. This method helps to improve the accuracy and reliability of predictions.

[0078] The historical case collection is a collection of data on the evolution of the status of the injured and sick in the past. By comparing with the current status of the injured and sick, historical cases with similar characteristics can be found, thus providing a reference.

[0079] The mask area filtering method is a technique used to evaluate the quality of segmentation results. It calculates the actual area covered by each target segmentation result and filters out results with an area smaller than a preset threshold to ensure that only high-quality prediction results are retained.

[0080] The optimized prediction model for the status of the injured and sick is a prediction model constructed by accurately adjusting weights and parameters based on the screening of high-quality historical cases. It can significantly improve the accuracy and reliability of the prediction of the injury status of the injured and sick.

[0081] In the embodiment of the present application, firstly, the system uses the K nearest neighbor algorithm to search and locate a set of historical cases that are highly similar to the current condition of the injured and sick based on the optimized evolution map of the injured and sick status; secondly, the distance between the current injured and sick status node and the historical cases is accurately calculated, and the K closest neighbor cases are selected; thirdly, the mask area filtering method is applied to calculate the actual area covered by each target segmentation result, and the results with an area less than a preset threshold are filtered out; finally, all high-quality case data are integrated to generate an optimized prediction model for the status of the injured and sick, providing a reliable basis for subsequent risk assessment.

[0082] For example, continuing with the above example, assume that the system has generated an optimized map of the evolution of the status of the injured and sick; first, based on the map, the system uses the K nearest neighbor algorithm to search and locate a set of historical cases that are highly similar to the current status of the injured and sick; secondly, accurately calculate the distance between the current status node of the injured and sick and the historical cases, and select the closest K neighbor cases; thirdly, apply the mask area filtering method to calculate the actual area covered by each target segmentation result, and filter out results with an area less than the preset threshold; finally, integrate all high-quality case data to generate an optimized prediction model for the status of the injured and sick, providing a reliable basis for subsequent risk assessment.

[0083] 104. Based on the optimized patient status prediction model, a dynamic risk assessment and interactive feedback optimization mechanism is established to generate a patient injury prediction plan.

[0084] In this step, dynamic risk assessment is a method of real-time monitoring of the evolution of the status of the injured and sick, and providing timely warnings and response suggestions by assessing the risks that may be faced in the future.

[0085] The interactive feedback optimization mechanism allows medical experts to adjust the parameters and weights of the prediction model according to actual conditions and provide personalized emergency treatment recommendations. This mechanism enhances the flexibility and practicality of the system and ensures that the prediction model can adapt to the ever-changing medical environment.

[0086] The injury prediction program for the wounded and sick is a specific treatment and management plan generated based on an optimized prediction model of the wounded and sick status and a dynamic risk assessment mechanism. It provides a scientific basis for medical decision-making and ensures that the wounded and sick receive the most appropriate treatment and support.

[0087] In the embodiment of the present application, firstly, the system establishes a dynamic risk assessment mechanism based on the optimized prediction model of the status of the injured and sick, monitors the evolution of the status of the injured and sick in real time, and assesses the risks they may face in the future; secondly, an interactive feedback optimization mechanism is introduced to allow medical experts to adjust the parameters and weights of the prediction model according to actual conditions; thirdly, the system provides personalized emergency treatment suggestions to ensure timely response to emergencies; finally, a prediction plan for the injury status of the injured and sick is generated to provide a scientific basis for medical decision-making and ensure that the injured and sick receive the most appropriate treatment.

[0088] For example, continuing with the above example, assume that the system has generated an optimized prediction model for the status of the sick and wounded; first, based on the model, the system establishes a dynamic risk assessment mechanism to monitor the evolution of the status of the sick and wounded in real time; second, an interactive feedback optimization mechanism is introduced to allow medical experts to adjust the parameters and weights of the prediction model according to actual conditions; third, the system provides personalized emergency treatment suggestions to ensure timely response to emergencies; finally, a prediction plan for the injury status of the sick and wounded is generated to provide a scientific basis for medical decision-making and ensure that the sick and wounded receive the most appropriate treatment.

[0089] In summary, steps 101 to 104 cover the complete process from data collection, preprocessing, feature extraction, similarity matching to model optimization and risk assessment, aiming to provide a comprehensive and intelligent solution for predicting the condition of the injured and sick, meeting the needs of accurate, real-time and personalized medical services in the modern medical environment. Through the application of multi-source data fusion, random field algorithm, time-dependent node embedding technology and K nearest neighbor algorithm, this method significantly improves the accuracy and reliability of the prediction of the condition of the injured and sick, providing strong support for medical decision-making.

[0090] In order to solve the problem of capturing complex relationships and dynamic characteristics in the state evolution map of the wounded and sick, in some embodiments, the processing and analysis of the time series data in the state evolution map of the wounded and sick described in step 102 includes: based on the initial data set of the wounded and sick, the time series data of the wounded and sick are node-processed to generate the state node timestamps of the wounded and sick; based on the state node timestamps of the wounded and sick, the random field algorithm is used to model the relationship between nodes, capture the complex relationship and dynamic characteristics between the states of the wounded and sick, identify the mutual dependence and time variation between different state nodes, reveal the potential development trend and abnormal pattern of the injury, and generate a random field model; based on the random field model, the time-dependent node embedding technology is used to deeply process and analyze the time series data in the state evolution map of the wounded and sick, embed each state node into a multidimensional space, make adjacent nodes closer in space, and generate a time-dependent node embedding result; based on the time-dependent node embedding result, further annotate the key development trends and abnormal patterns to generate an optimized state evolution map of the wounded and sick.

[0091] In this embodiment, node processing refers to converting the time series data of the patients into discrete time stamp nodes, each node representing the status of the patients at a time point, which is convenient for subsequent modeling and analysis. This method helps to capture the changes in the status of the patients at different time points.

[0092] The patient status node timestamp is a time series data point generated after node processing. Each timestamp contains the patient's status information at a specific time point, such as physiological parameters, diagnosis results, etc. These timestamps are used to build the basis for the patient status evolution map.

[0093] The random field model is a probabilistic model that captures the complex relationships and dynamic characteristics between the states of the injured and sick by modeling the relationships between nodes. It can identify the interdependence and time-varying nature of nodes in different states and reveal potential injury development trends and abnormal patterns.

[0094] The time-dependent node embedding result is generated by mapping each state node into a multidimensional space, making adjacent nodes closer in space. This method enhances the understanding of temporal characteristics and helps to reveal potential trends and abnormal patterns in the evolution of the status of the injured and sick.

[0095] The optimized map of the evolution of the status of the injured and sick is the result of processing the original map through random field algorithm and time-dependent node embedding technology. It more accurately reflects the evolution process of the status of the injured and sick, and marks the key development trends and abnormal patterns.

[0096] In the embodiment of the present application, first, the system performs node processing on the time series data of the injured and sick based on the initial data set of the injured and sick, and generates timestamps of the nodes of the injured and sick status; secondly, based on these timestamps, the system uses the random field algorithm to model the relationship between the nodes, capture the complex relationship and dynamic characteristics between the states of the injured and sick, identify the interdependence and time-varying properties between nodes of different states, reveal the potential development trends and abnormal patterns of injuries, and generate a random field model; thirdly, the system uses time-dependent node embedding technology to perform in-depth processing and analysis of the time series data in the evolution map of the state of the injured and sick, embed each state node into the multi-dimensional space, and make the adjacent nodes closer in space; finally, based on the time-dependent node embedding results, the system further marks the key development trends and abnormal patterns, and generates an optimized evolution map of the state of the injured and sick.

[0097] Here is a specific example:

[0098] For example, a machine learning-based injury prediction system for the injured and sick is used in a large emergency center. First, the system processes the time series data of the injured and sick into nodes and generates timestamps for the nodes of the injured and sick status to ensure that the status information at each time point is accurately recorded. Secondly, based on these timestamps, the system uses random field algorithms to model the relationship between nodes, capture the complex relationship and dynamic characteristics between the states of the injured and sick, identify the interdependence and time-varying properties between nodes of different states, reveal potential injury development trends and abnormal patterns, and generate random field models. Thirdly, the system uses time-dependent node embedding technology to perform in-depth processing and analysis of the time series data in the evolution map of the state of the injured and sick, embed each state node into multi-dimensional space, make adjacent nodes closer in space, and enhance the understanding of time series characteristics. Finally, based on the results of time-dependent node embedding, the system further annotates key development trends and abnormal patterns, generates an optimized evolution map of the state of the injured and sick, and provides detailed data support for subsequent similarity matching analysis.

[0099] In order to solve the problem of capturing complex relationships and dynamic characteristics in the wounded and sick status evolution map, in some embodiments, the modeling of the relationship between nodes described in step 102 includes: based on the timestamps of the wounded and sick status nodes, correlation analysis is performed on the wounded and sick status nodes corresponding to each timestamp to generate a wounded and sick status node relationship network; based on the wounded and sick status node relationship network, a random field algorithm is used to model the relationship between nodes to capture the complex relationships and dynamic characteristics between the wounded and sick status, identify the mutual dependence and time-varying properties between nodes in different states, reveal potential injury development trends and abnormal patterns, and generate a node relationship model; based on the node relationship model, a statistical method is used to evaluate the stability and accuracy of the model to ensure accurate reflection of the actual evolution of the wounded and sick status and generate an optimized node relationship model; based on the optimized node relationship model, combined with the timestamps of the wounded and sick status nodes, the influence of external environmental factors is further introduced to generate a random field model.

[0100] In this embodiment, the correlation analysis process is a quantitative analysis of the relationship between the patient status nodes corresponding to each timestamp, and by calculating the correlation coefficient between different nodes or using other statistical methods, it is possible to identify which status nodes have significant mutual influence.

[0101] The patient status node relationship network is a network structure generated by performing correlation analysis on the patient status nodes corresponding to each timestamp. The network shows the interdependence between nodes of different statuses, which helps to identify the complex patterns and time-varying characteristics in the evolution of the patient status.

[0102] Inter-node interdependence refers to the causal relationship or correlation between nodes in different states. This dependency can be direct (such as one state directly affecting another state) or indirect (such as the influence transmitted through a series of intermediate states).

[0103] Time variability refers to the changes in the relationships between nodes over time. Capturing time variability is crucial for accurately predicting the future evolution of the patient's status.

[0104] The node relationship model is the result of modeling the relationship between nodes through the random field algorithm. It not only captures the complex relationship and dynamic characteristics between the states of the injured and the sick, but also identifies the interdependence and time-varying nature of nodes in different states, revealing the potential development trends and abnormal patterns of injuries.

[0105] The optimized node relationship model is the result generated after evaluating the model stability and accuracy through statistical methods, ensuring that the model can accurately reflect the real evolution of the status of the injured and sick, and improving the reliability and accuracy of the prediction.

[0106] Statistical methods refer to a range of techniques used to evaluate model performance, including but not limited to mean square error (MSE), R 2 Values, cross validation, etc.

[0107] In the embodiment of the present application, first, based on the timestamp of the patient status node, the patient status node corresponding to each timestamp is subjected to correlation analysis and processing to generate a patient status node relationship network; secondly, based on the relationship network, the system uses a random field algorithm to model the relationship between nodes, capture the complex relationship and dynamic characteristics between the patient status, identify the interdependence and time-varying characteristics between nodes of different statuses, reveal the potential development trend and abnormal patterns of injuries, and generate a node relationship model; thirdly, the system uses statistical methods to evaluate the stability and accuracy of the node relationship model to ensure that it accurately reflects the real evolution of the patient status and generates an optimized node relationship model; finally, the system combines the optimized node relationship model with the timestamp of the patient status node, further introduces the influence of external environmental factors, and generates a random field model.

[0108] Here is a specific example:

[0109] For example, a machine learning-based injury prediction system for injured and sick patients was introduced into the medical support center of a large-scale sports event. First, based on the timestamp of the injured and sick patient status nodes, the system performs a correlation analysis on each injured and sick patient status node corresponding to the timestamp to generate an injured and sick patient status node relationship network, showing the interdependent relationships between different status nodes. Second, based on this relationship network, the system uses the random field algorithm to model the relationships between nodes, capture the complex relationships and dynamic characteristics between the statuses of injured and sick patients, identify the interdependence and time-varying nature between different status nodes, reveal the potential injury development trends and abnormal patterns, and generate a node relationship model. Third, the system uses statistical methods to evaluate the stability and accuracy of the node relationship model to ensure that it accurately reflects the real evolution of the status of injured and sick patients and generates an optimized node relationship model. Finally, the system combines the optimized node relationship model and the timestamp of the injured and sick patient status nodes, further introduces the influence of external environmental factors (such as stadium temperature, humidity, and spectator density), generates a random field model, provides a more comprehensive analysis of the evolution of the status of injured and sick patients, and supports emergency medical decision-making during the event.

[0110] To address the problem of capturing the complexity and dynamic characteristics of time series data in the injury and sick patient status evolution graph, in some embodiments, the in-depth processing and analysis of the time series data in the injury and sick patient status evolution graph described in step 102 includes: based on the random field model, performing structured processing on the time series data in the injury and sick patient status evolution graph, converting it into a format suitable for embedding operations, and generating a structured status node set; based on the structured status node set, using time-dependent node embedding technology to map each status node into a multi-dimensional space, reflecting the time-dependent relationships and interactions between status nodes, ensuring that adjacent nodes are closer in space, and generating time-dependent node embedding vectors; based on the time-dependent node embedding vectors, optimizing and adjusting the distances and positions between nodes to better capture the real dynamic characteristics of the injury and sick patient status evolution, and generating an optimized node embedding layout; based on the optimized node embedding layout, using a visualization tool to display the injury and sick patient status change path and time dependence, and generating a time-dependent node embedding result.

[0111] In this embodiment, structured processing refers to converting the time series data in the injury and sick patient status evolution graph into a format suitable for embedding operations. This process usually involves operations such as data cleaning, format conversion, and feature extraction to ensure that the data meets the requirements of subsequent embedding technologies.

[0112] The structured status node set is a set of time series data points after structured processing. Each node contains the status information of the injured and sick patient at a specific time point, such as physiological parameters, diagnosis results, etc.

[0113] Time-dependent node embedding technology is a method that maps each state node into a multidimensional space, making adjacent nodes closer in space. This method not only retains the sequential information of the original time series, but also increases the spatial proximity, making similar state nodes more closely clustered in space, thereby better reflecting the time dependency and interaction between state nodes.

[0114] The time-dependent node embedding vectors are generated by the time-dependent node embedding technology, which represent the position of each state node in the multidimensional space. These vectors not only reflect the time dependency of the state nodes, but also capture the interaction between nodes, which helps to reveal the true dynamic characteristics of the evolution of the state of the injured and sick.

[0115] The optimized node embedding layout is generated by adjusting the distance and position between nodes based on the time-dependent node embedding vector. This layout better captures the real dynamic characteristics of the evolution of the status of the injured and sick, and improves the accuracy and reliability of the prediction model.

[0116] Visualization tools are used to display the path and time dependency of the patient's condition changes. Through the graphical interface, medical personnel can intuitively observe the changing trends and potential risks of the patient's condition, and assist in decision-making.

[0117] In the embodiment of the present application, firstly, based on the random field model, the time series data in the state evolution graph of the wounded and sick are structured and converted into a format suitable for embedding operations to generate a structured state node set; secondly, based on the structured state node set, the system adopts the time-dependent node embedding technology to map each state node into a multidimensional space, reflecting the time dependency and interaction between state nodes, ensuring that adjacent nodes are closer in space, and generating a time-dependent node embedding vector; thirdly, based on the time-dependent node embedding vector, the system optimizes and adjusts the distance and position between nodes to better capture the real dynamic characteristics of the state evolution of the wounded and sick, and generates an optimized node embedding layout; finally, based on the optimized node embedding layout, the system displays the state change path and time dependency of the wounded and sick through visualization tools, and generates a time-dependent node embedding result.

[0118] Here is a specific example:

[0119] For example, a machine learning-based injury prediction system for patients was introduced in the intensive care unit (ICU) of a large hospital. First, based on the random field model, the system structured the time series data in the state evolution graph of patients in the ICU, converted it into a format suitable for embedding operations, and generated a structured state node set. Secondly, based on the structured state node set, the system uses time-dependent node embedding technology to map each state node into a multidimensional space, reflecting the time dependency and interaction between state nodes, ensuring that adjacent nodes are closer in space, and generating a time-dependent node embedding vector. Thirdly, based on the time-dependent node embedding vector, the system optimizes and adjusts the distance and position between nodes to better capture the real dynamic characteristics of the patient's state evolution and generate an optimized node embedding layout. Finally, based on the optimized node embedding layout, the system uses visualization tools to display the patient's state change path and time dependency, generates time-dependent node embedding results, and helps doctors monitor the patient's state in real time, adjust the treatment plan in time, and improve the efficiency of treatment.

[0120] In order to solve the problem of insufficient accuracy of historical case matching in the prediction of the status of the injured and sick, in some embodiments, the similarity matching analysis of the current status information of the injured and sick with the data in the historical case library described in step 103 includes: based on the optimized injured and sick status evolution map, comparing the current status information of the injured and sick with the data in the historical case library to generate a preliminary similar case set; based on the preliminary similar case set, using the K nearest neighbor algorithm to perform accurate similarity matching analysis, by calculating the distance between the current injured and sick status node and each case in the historical case set, selecting the closest K neighbor cases to search for historical cases that are highly similar to the current situation, and generating a selected similar case set; based on the selected similar case set, applying the mask area filtering method, calculating the actual area covered by each target segmentation result, filtering out results less than a preset threshold, and generating a case quality screening result; based on the case quality screening result, integrating all high-quality case data to generate an optimized injured and sick status prediction model.

[0121] In this embodiment, the preliminary similar case set is generated by comparing the current patient status information with the data in the historical case library. It contains historical cases that have certain similarities with the current patient status, providing a basis for further precise matching.

[0122] The selected similar case set is the result of precise similarity matching analysis based on the preliminary similar case set through the K nearest neighbor algorithm. It selects historical cases that are highly similar to the current conditions of the injured and sick to ensure the accuracy and relevance of the match.

[0123] The mask area filtering method is a technique used to evaluate the quality of segmentation results. It calculates the actual area covered by each target segmentation result and filters out results with an area smaller than a preset threshold to ensure that only high-quality prediction results are retained.

[0124] The case quality screening results are generated after the mask area filtering method, and contain high-quality historical case data. These cases provide a reliable basis for the subsequent prediction model construction.

[0125] The optimized prediction model for the status of the injured and sick is generated based on the integration of all high-quality case data. It not only improves the accuracy of the prediction, but also enhances the adaptability and generalization ability of the model, providing strong support for medical decision-making.

[0126] In the embodiment of the present application, firstly, based on the optimized evolution map of the status of the injured and sick, the current status information of the injured and sick is compared with the data in the historical case library to generate a preliminary similar case set; secondly, based on the preliminary similar case set, the system uses the K nearest neighbor algorithm to perform accurate similarity matching analysis, calculates the distance between the current status node of the injured and sick and each case in the historical case set, selects the closest K neighbor cases to search for historical cases that are highly similar to the current situation, and generates a selected similar case set; thirdly, based on the selected similar case set, the system applies the mask area filtering method to calculate the actual area covered by each target segmentation result, filters out the results that are less than the preset threshold, and generates a case quality screening result; finally, based on the case quality screening result, the system integrates all high-quality case data to generate an optimized prediction model for the status of the injured and sick.

[0127] Here is a specific example:

[0128] For example, a machine learning-based injury prediction system for the sick and wounded was introduced in a community hospital. First, based on the optimized sick and wounded status evolution map, the system compares the current sick and wounded status information with the data in the historical case library to generate a preliminary similar case set, ensuring that historical cases with similar characteristics are initially screened out. Second, based on the preliminary similar case set, the system uses the K nearest neighbor algorithm to perform precise similarity matching analysis. By calculating the distance between the current sick and wounded status node and each case in the historical case set, the closest K neighbor cases are selected to search for historical cases that are highly similar to the current situation, generate a selected similar case set, and ensure the accuracy of the match. Third, based on the selected similar case set, the system applies the mask area filtering method to calculate the actual area covered by each target segmentation result, filter out results that are less than the preset threshold, generate case quality screening results, and remove noise and irrelevant cases. Finally, based on the case quality screening results, the system integrates all high-quality case data to generate an optimized sick and wounded status prediction model, providing a scientific basis for doctors in community hospitals, assisting them in formulating more accurate treatment plans, and improving the quality of medical services.

[0129] In order to solve the problem of insufficient accuracy of historical case matching in the prediction of the status of the injured and sick, in some embodiments, the accurate similarity matching analysis of the current status information of the injured and sick with the historical case library data described in step 103 includes: based on the preliminary similar case set, calculating the similarity between the time series characteristics and other related attributes, and generating a preprocessed similar case set; based on the preprocessed similar case set, using the K nearest neighbor algorithm to perform accurate similarity matching analysis, by calculating the distance between the current injured and sick status node and each case in the historical case set, selecting the closest K neighbor cases, and generating a selected neighbor case set; based on the selected neighbor case set, performing a detailed comparative analysis of the time series data of each selected case, deeply evaluating the subtle differences between each neighbor case and the current status of the injured and sick, and generating a high-precision similar case evaluation report; based on the high-precision similar case evaluation report, performing similarity confirmation and detailed comparative analysis to generate a selected similar case set.

[0130] In this embodiment, the time series feature refers to the features that change over time in the patient's state evolution graph, such as the changing trends of physiological parameters such as heart rate, blood pressure, and body temperature.

[0131] Other related attributes include, but are not limited to, non-time series features such as basic information of the injured and sick (age, gender), diagnosis results, and treatment plans.

[0132] The preprocessed similar case set is the result generated after calculating the similarity between time series features and other related attributes. It selects historical cases that have high similarity with the current conditions of the injured and sick in multiple dimensions, providing an optimized basis for further precise matching.

[0133] The selected neighbor case set is the result generated after precise similarity matching analysis using the K nearest neighbor algorithm based on the preprocessed similar case set. It selects historical cases that are highly similar to the current conditions of the injured and sick to ensure the accuracy and relevance of the match.

[0134] The high-precision similar case assessment report is the result of a detailed comparative analysis of the time series data of each selected case. It deeply evaluates the subtle differences between each neighbor case and the current status of the injured and sick, ensuring that the final selected case has the highest similarity and reference value.

[0135] Similarity confirmation and detailed comparative analysis refers to further verification and refinement of cases in the high-precision similar case assessment report.

[0136] The selected similar case set is the final result generated after multiple rounds of screening and detailed comparative analysis. It contains historical cases that are highly similar to the current conditions of the injured and sick, providing a reliable foundation for the subsequent construction of the prediction model.

[0137] In the embodiment of the present application, first, the system calculates the similarity between the time series features and other related attributes based on the preliminary similar case set, and generates a preprocessed similar case set; secondly, based on the preprocessed similar case set, the system uses the K nearest neighbor algorithm to perform accurate similarity matching analysis, calculates the distance between the current sick and wounded status node and each case in the historical case set, selects the closest K neighbor cases, and generates a selected neighbor case set; thirdly, based on the selected neighbor case set, the system performs a detailed comparative analysis of the time series data of each selected case, deeply evaluates the subtle differences between each neighbor case and the current sick and wounded status, and generates a high-precision similar case evaluation report; finally, based on the high-precision similar case evaluation report, the system performs similarity confirmation and detailed comparative analysis to generate a selected similar case set.

[0138] Here is a specific example:

[0139] For example, a machine learning-based injury prediction system for the sick and wounded was introduced in a certain military medical unit. First, based on the preliminary similar case set, the system calculated the similarity between time series features (such as heart rate and blood pressure) and other related attributes (such as age, gender, and diagnosis results), and generated a pre-processed similar case set to ensure that historical cases with high similarity to the current sick and wounded conditions in multiple dimensions were initially screened out. Second, based on the pre-processed similar case set, the system used the K nearest neighbor algorithm to perform precise similarity matching analysis. By calculating the distance between the current sick and wounded status node and each case in the historical case set, the closest K neighbor cases were selected to generate a selected neighbor case set to ensure the accuracy of the match. Third, based on the selected neighbor case set, the system conducted a detailed comparative analysis of the time series data of each selected case, deeply evaluated the subtle differences between each neighbor case and the current sick and wounded status, and generated a high-precision similar case evaluation report to ensure the high similarity of each selected case. Finally, based on the high-precision similar case evaluation report, the system performed similarity confirmation and detailed comparative analysis to generate a selected similar case set to help military medical personnel quickly develop the best treatment plan and improve battlefield treatment efficiency.

[0140] In order to solve the problem of real-time risk monitoring and continuous model optimization in the prediction of the injury status of the wounded and sick, in some embodiments, the real-time monitoring of the evolution of the status of the wounded and sick and the generation of a prediction plan described in step 104 include: based on the optimized wounded and sick status prediction model, real-time monitoring of the evolution of the status of the wounded and sick, identifying and quantifying potential risk factors, and generating a real-time risk factor report; based on the real-time risk factor report, building a dynamic risk assessment system, combining the prediction model parameters with the real-time risk factor data, obtaining the probability of the wounded and sick's injury deterioration, and generating a dynamic risk assessment result; based on the dynamic risk assessment result, establishing an interactive feedback optimization mechanism, collecting medical experts' subsequent evaluations as incremental learning samples, and generating feedback optimization records; based on the feedback optimization records, periodically fine-tuning the optimized wounded and sick status prediction model to adapt to new injury conditions and the evolution of medical practice, and generating a wounded and sick injury prediction plan.

[0141] In this embodiment, the real-time risk factor report is generated through real-time monitoring of the evolution of the status of the injured and sick. It identifies and quantifies potential risk factors, such as abnormal physiological parameters, environmental changes, etc., and provides detailed data support for subsequent risk assessment.

[0142] The dynamic risk assessment system is a platform for real-time monitoring and evaluation of the evolution of the status of the injured and sick. It combines the prediction model parameters with real-time risk factor data to calculate the probability of the injured and sick's condition worsening and generate dynamic risk assessment results.

[0143] The probability of worsening of the patient's condition refers to the possibility that the patient's condition will worsen in the future, calculated by the dynamic risk assessment system. This probability value provides early warning information to medical personnel, helping them take preventive measures in a timely manner.

[0144] The interactive feedback optimization mechanism allows medical experts to adjust the parameters and weights of the prediction model according to actual conditions and provide personalized emergency treatment suggestions. This mechanism enhances the flexibility and practicality of the system and ensures that the prediction model can adapt to the ever-changing medical environment.

[0145] Incremental learning samples refer to data sets of post-evaluations by medical experts. These evaluations serve as new training samples to fine-tune the prediction model so that it more accurately reflects the changes and developments in actual medical practice.

[0146] The feedback optimization record is the result generated after collecting and organizing incremental learning samples. It records the specific content and effect of each adjustment and provides a basis for periodic fine-tuning of the model.

[0147] Periodic fine-tuning refers to regularly making small adjustments to the predictive model based on feedback optimization records. This approach ensures that the model can adapt to new injury conditions and evolution of medical practices, maintaining its accuracy and reliability.

[0148] In the embodiment of the present application, the system monitors the evolution of the status of the injured and sick in real time based on the optimized prediction model of the status of the injured and sick, identifies and quantifies potential risk factors, and generates a real-time risk factor report; secondly, the system builds a dynamic risk assessment system based on the real-time risk factor report, combines the prediction model parameters with the real-time risk factor data, obtains the probability of the injured and sick's condition worsening, and generates a dynamic risk assessment result; thirdly, based on the dynamic risk assessment result, the system establishes an interactive feedback optimization mechanism, collects the subsequent evaluation of medical experts as incremental learning samples, and generates feedback optimization records; finally, based on the feedback optimization records, the system periodically fine-tunes the optimized prediction model of the status of the injured and sick to adapt to new injury conditions and the evolution of medical practice, and generates an injury prediction plan for the injured and sick.

[0149] For example, a machine learning-based injury prediction system for injured and sick patients was introduced in a certain field medical rescue team. First, based on an optimized injury prediction model for injured and sick patients, the system monitors the evolution of the status of injured and sick patients at the rescue site in real time, identifies and quantifies potential risk factors, and generates a real-time risk factor report to ensure that any factors that may affect the status of injured and sick patients are captured in a timely manner. Second, based on the real-time risk factor report, the system constructs a dynamic risk assessment system, combines the prediction model parameters with the real-time risk factor data, obtains the probability of the deterioration of the injuries of the injured and sick patients, and generates a dynamic risk assessment result to help rescue personnel quickly assess the risk level of each injured and sick patient. Third, based on the dynamic risk assessment result, the system establishes an interactive feedback optimization mechanism, collects the post-event evaluations of medical experts as incremental learning samples, and generates feedback optimization records to ensure that each medical decision can provide valuable experience for model optimization. Finally, based on the feedback optimization records, the system periodically fine-tunes the optimized injury prediction model for injured and sick patients to adapt to new injury situations and the evolution of medical practices, generates an injury prediction plan for injured and sick patients, helps the rescue team formulate a more accurate treatment plan, and improves the efficiency and success rate of field rescue.

[0150] Considering that in order to solve the problem of insufficient prediction accuracy of the evolution of the status of injured and sick patients in the prior art, in some embodiments, a method for modeling the relationship between nodes based on the random field algorithm is proposed. In the prior art, since there are problems such as difficulty in capturing the complex relationships and dynamic characteristics between the statuses of injured and sick patients, the inventive embodiments propose this alternative solution to solve the above technical problems. Therefore, a new alternative solution is proposed, and this solution includes:

[0151] Based on the relationship network of the status nodes of the injured and sick patients, use the random field algorithm to model the relationship between nodes, capture the complex relationships and dynamic characteristics between the statuses of the injured and sick patients, identify the mutual dependence and time-variability between different status nodes, reveal the potential injury development trends and abnormal patterns, and generate a node relationship model, including:

[0152] Based on the relationship network of the status nodes of the injured and sick patients, through time series analysis and feature engineering, extract the time-dependent features of each node to enhance the understanding of the node status and generate conditional probabilities;

[0153] Calculate the conditional probability through the following formula:

[0154]

[0155] where P(s i ∣s -i , θ) is the conditional probability of node s i under the condition of given the statuses s -i of all other nodes and parameter θ; j is the index of the neighbor node of node i; N(i) is the neighbor set of node i; w ijis the weight between nodes i and j; f(s i ,s j ) is the nonlinear function of the relationship between nodes i and j; β is the influence coefficient of external features; g(s i ) is the external eigenvalue of node i; Z(θ) is the normalization factor to ensure the validity of the probability distribution;

[0156] Based on the conditional probability, nonlinear transformation and weight adjustment mechanism are introduced to quantify the complex relationship pattern between nodes, consider the influence of external characteristics, and build a framework reflecting the overall dynamic characteristics of the system to generate an energy function;

[0157] The energy function is calculated using the following formula:

[0158]

[0159] Where E(s; Θ) is the energy function of the entire system; s is the state vector of all nodes; Θ is the model parameter set, including θ, w ii , β i ,λ,μ,v;w ii is the weight of node i itself; f(s i ,s i ) is a nonlinear function of the node i’s own state; β i is the external characteristic influence coefficient of node i; g i (s i ) is the external eigenvalue of node i; λ is the adjustment term used to control the influence of the nonlinear part in the energy function; μ is the adjustment coefficient of the triple interaction term; φ(s i ,s -i ) is considered the node i and its neighbors s -i is a function of more complex nonlinear interactions between them; v is the adjustment coefficient of the time dependence; ψ(s i ,s -i ,t) is the node i and its neighbors s -i A nonlinear function of time-varying and interdependent properties at time t; i is the index of the node, from 1 to n; n is the total number of nodes;

[0160] Based on the energy function, the gradient descent optimization method is used to minimize the energy function, find the optimal parameter configuration, introduce a regularization term to prevent overfitting, ensure the generalization ability of the model, and generate a node relationship model.

[0161] This method aims to introduce random field algorithms and gradually generate a node relationship model that can accurately reflect the evolution of the status of the injured and sick through time series analysis, feature engineering, conditional probability calculation, energy function construction and optimization methods. This method not only improves the accuracy of prediction, but also enhances the adaptability and generalization ability of the model.

[0162] In the conditional probability, the normalization factor term Ensure the validity of the probability distribution, so that the sum of all conditional probabilities is equal to 1, and ensure the correctness of the probability distribution. This is because the unnormalized probability may not meet the basic requirements of probability theory; the exponential term exp(-∑ j∈N(i) w ij f(s i ,s j )+βg(s i )): It is used to capture the interdependence between nodes and the influence of external environmental factors. This item is designed to pass the nonlinear function f(s i ,s j ) and external eigenvalues ​​g(s i ) to enhance the model’s understanding of complex relationships, thereby improving prediction accuracy;

[0163] Among them, θ is obtained through historical data training, and these parameters reflect the adaptability of the model to the changes in the status of the injured and sick under different conditions; w ij It is calculated by the historical interaction frequency between node and j and is used to measure the strength of the relationship between the two nodes; f(s i ,s j ) is calculated from the nonlinear features extracted by feature engineering and is used to capture the complex interactions between nodes; γ is obtained from the environmental monitoring system and is used to represent the impact of external features on the status of the injured and sick; g(s i ) is directly read from the patient status record to reflect the patient's current status characteristics.

[0164] In the energy function, the state term It is used to measure the influence of the node's own state and its external characteristics, reflecting the stability of the node itself. This item is designed to capture the characteristics of the node itself and ensure that the model can accurately reflect the state of each node. The denominator normalization term Used to normalize the numerator term to ensure that the numerical range of the energy function is reasonable. This term is designed to avoid numerical instability and ensure that the calculation results of the energy function have practical significance; nonlinear interaction adjustment term It is used to adjust the triple interaction term, control the influence of the nonlinear part in the energy function, and prevent overfitting. This term is designed to adjust the degree of nonlinear interaction through hyperparameters λ and μ to improve the generalization ability of the model; the time dependency term vψ(s i ,s -i ,t): A nonlinear function used to consider the time-varying and interdependence of nodes and their neighbors in time. This item is designed to capture the time evolution characteristics of the status of the injured and sick, ensure that the model can adapt to time changes, and improve the accuracy of prediction;

[0165] Among them, Θ is obtained by gradient descent optimization. These parameters reflect the adaptability of the model to the changes in the status of the injured and sick under different conditions; w ii It is calculated through the historical records of the node's own status and is used to measure the stability of the node itself; f(s i ,s i ) is calculated through nonlinear features extracted by feature engineering and is used to capture the complex characteristics of the node itself; β i Obtained from the environmental monitoring system, used to indicate the impact of external features on the status of the injured or sick; g i (s i ) is directly read from the patient's status record to reflect the patient's current status characteristics; λ, μ, ν are obtained as hyperparameters through cross-validation tuning to adjust the complexity and adaptability of the model; φ(s i ,s -i ) is calculated by the nonlinear interaction function extracted by feature engineering and is used to capture the complex interactions between nodes; ψ(s i ,s -i ,t) is calculated through time series analysis,

[0166] Assume θ = [0.5, 0.3, 0.2]; w ij =[0.7,0.6,0.4]; f(s i ,s j )=[0.9,0.8,0.7]; β=[0.5,0.4,0.3]; g(s i ) = [0.6, 0.5, 0.4];

[0167]

[0168] Assume Θ = [0.5, 0.3, 0.2, 0.4, 0.6, 0.8]; w ii =[0.7,0.6,0.4]; f(s i ,s i )=[0.9,0.8,0.7];β i =[0.5,0.4,0.3]; g i (s i )=[0.6,0.5,0.4]; λ=0.5; μ=0.4; v=0.3; φ(s i ,s -i )=[0.8,0.7,0.6];ψ(s i ,s -i ,t)=[0.9,0.8,0.7];

[0169]

[0170] Assuming the thresholds are set to 0.9 and -1.0, since the conditional probability calculation result 0.92 is greater than the set threshold 0.9, and the energy function calculation result -0.75 is higher than the set energy threshold -1.0, it indicates that the current state of the patient is stable and the risk is low, and can continue to be observed without taking emergency measures immediately. This is because the higher conditional probability reflects the stability of the patient's state under the current state, and the energy function value close to or higher than the set energy threshold indicates that the system's internal stability and resistance to external interference are strong. Through the above steps, the accurate prediction of the evolution of the patient's state is ensured, the scientificity and timeliness of medical decision-making are improved, and the accuracy and response speed of the entire health monitoring system are enhanced.

[0171] In order to solve the problem of insufficient accuracy in predicting the evolution of the status of the injured and sick in the prior art, in some embodiments, a method for accurate similarity matching analysis based on the K nearest neighbor algorithm is proposed. Because it is difficult to capture the complex relationship and dynamic characteristics between the status of the injured and sick in the prior art, the embodiment of the invention proposes this optional solution to solve the above technical problem. Therefore, a new optional solution is proposed, which includes:

[0172] Based on the preprocessed similar case set, feature enhancement processing is performed, and the periodicity and trend components of the patient status are extracted through time series decomposition. The components are adjusted in combination with the event-driven model to generate the current patient status node s i With historical cases j The distance between

[0173] The current patient status node s is calculated using the following formula: i With historical cases j Distance between:

[0174]

[0175] Among them, d(s i ,h j ) is the current patient status node s i With historical cases j The distance between them; k is the index of the feature dimension, from 1 to m; m is the number of feature dimensions; s ik is the value of the kth feature of the current patient status node i; h jk is the value of historical case j on the kth feature; σ k is the standard deviation of the kth feature for normalization; α is the weight coefficient of the nonlinear term; γ is the adjustment coefficient of the nonlinear term; p is the number of external feature functions; f l (s i ) is the result of the lth external characteristic function applied to the current sick and wounded state node i; fl (h j ) is the result of applying the l-th external feature function to historical case j; l is the index of the external feature function;

[0176] Based on the current patient status node s i With historical cases j The distance between them is calculated by introducing a time decay factor and constructing a comprehensive scoring mechanism so that newer historical cases have higher weights. Each historical case is mapped into a low-dimensional space using graph embedding technology to generate a set of the closest K neighbor cases.

[0177] The set of cases closest to K neighbors is calculated using the following formula:

[0178]

[0179] Among them, S K is the set of the closest K neighbor cases selected; h j is the jth historical case; represents the selection of element j in the index set J that minimizes the objective function; J is a subset of size K selected from all historical case index sets {1, 2, …, n}; n is the total number of historical cases; d(s i ,h j ) is the current patient status node s i With historical cases j The distance between them; γ is the weight coefficient of the logarithmic term; δ is the adjustment coefficient of the exponential decay term; j is the index of the historical case;

[0180] Based on the K closest neighbor case sets, a domain expert rule engine is applied to filter out cases lacking clinical significance according to the medical knowledge base, an interactive feedback mechanism is introduced, and a selected neighbor case set is generated through multiple rounds of iterative optimization.

[0181] This method aims to introduce a method that can effectively utilize historical case data. It uses the K nearest neighbor algorithm and combines feature enhancement processing, time series decomposition, event-driven model adjustment, nonlinear terms and time decay factors to gradually generate a set of selected neighbor cases that can accurately reflect the current status of the injured and sick. This method not only improves the accuracy of prediction, but also enhances the efficiency and adaptability of the model to historical data.

[0182] In the current sick and injured status node s i With historical cases j In the distance between Measures the difference in each feature dimension, and normalizes it so that features of different magnitudes can be compared fairly; nonlinear terms Capture the nonlinear differences in the results of external characteristic functions and enhance the model's understanding of complex relationships;

[0183] Among them, m is the number of feature dimensions, which is determined according to the number of features in the data set; s ik and h jk are the values ​​of the current patient status node and the historical case on the kth feature, which are directly read from the data set; k is the standard deviation of the kth feature, which is used for normalization and is calculated by statistical methods; α and β are the weight coefficients and adjustment coefficients of the nonlinear term, which are obtained by cross-validation tuning; p is the number of external feature functions, which is set according to application requirements; f l (s i ) and f l (h j ) is the result of the lth external feature function applied to the current sick and wounded status node and the historical case, extracted through feature engineering;

[0184] In the case of the closest K neighbors, the reciprocal term The inverse of the distance is measured, and the closer the historical case is, the higher the score; the logarithmic term γlog(1+∑ j∈J exp(-δd(s i ,h j ))):Introduce an exponential decay factor to give newer historical cases higher weights and ensure the model's sensitivity to new information; the objective function minimizes Make sure to select the element j in the index set J that minimizes the objective function;

[0185] Where n is the total number of historical cases, which is determined according to the number of cases in the data set; K is the number of neighbor cases, which is set according to the application scenario; γ and δ are the weight coefficients of the logarithmic term and the adjustment coefficients of the exponential decay term, which are obtained through cross-validation tuning; d(s i ,h j ) is the distance between the current patient status node and the historical case, through the current patient status node s i With historical cases j The distance between

[0186] For example, the emergency department of a hospital developed a system to predict the injury of patients and wounded to improve the scientificity and timeliness of medical decision-making;

[0187] Assume m = 5; s ik =[0.6,0.7,0.8,0.9,0.5]; h jk =[0.5,0.6,0.7,0.8,0.4]; σ k=[0.1,0.1,0.1,0.1,0.1];α=0.5;β=0.3;p=3;f l (s i )=[0.9,0.8,0.7];f l (h j ) = [0.8, 0.7, 0.6];

[0188]

[0189] Assume n = 100; K = 5; γ = 0.4; δ = 0.2; d(s i ,h j )=0.75; Assuming the historical case h 1 ,h 2 ,h 3 ,h 4 ,h 5 The distances are 0.75, 0.80, 0.85, 0.90, 0.95 respectively; J = {1, 2, 3, 4, 5};

[0190]

[0191] Assuming that the threshold is set to 0.85, since the distance calculation result 0.75 is less than the set distance threshold 0.85, and the comprehensive scoring mechanism selects the 5 closest neighbor cases, it indicates that the current status of the injured and sick is highly similar to the historical cases, and can continue to be observed without taking immediate emergency measures. This is because the lower distance value reflects the high similarity between the current status of the injured and sick and the historical cases, and the comprehensive scoring mechanism ensures that the selected neighbor cases are the most relevant. Through the above steps, the accurate prediction of the evolution of the status of the injured and sick is ensured, the scientificity and timeliness of medical decision-making are improved, and the accuracy and response speed of the entire health monitoring system are enhanced.

[0192] Figure 2 A structural diagram of a system for predicting the condition of injured and sick persons based on machine learning is provided for an embodiment of the present application, such as Figure 2 As shown, the device comprises:

[0193] The collection module 21 is used to collect and integrate real-time information flows from various data sources, and generate an initial data set of the injured and sick through a multi-source data fusion method;

[0194] The analysis module 22 is used to process and analyze the time series data in the state evolution map of the injured and sick based on the initial data set of the injured and sick using the random field algorithm, capture the complex relationship and dynamic characteristics between the states of the injured and sick, use the time-dependent node embedding technology to enhance the ability to understand the time series characteristics, reveal the potential development trend and abnormal pattern of the injury, and generate an optimized state evolution map of the injured and sick;

[0195] The adjustment module 23 is used to perform similarity matching analysis based on the optimized sick and wounded status evolution map using the K nearest neighbor algorithm, search and locate historical case sets that are highly similar to the current sick and wounded status, accurately adjust the weights, use the mask area filtering method, calculate the actual area covered by each target segmentation result, filter out results that are less than a preset threshold, and generate an optimized sick and wounded status prediction model;

[0196] The generation module 24 is used to establish a dynamic risk assessment and interactive feedback optimization mechanism based on the optimized patient status prediction model to generate a patient injury prediction plan.

[0197] Figure 2 The machine learning-based injury prediction system can be implemented Figure 1 The implementation principle and technical effect of the method for predicting the condition of the injured and sick based on machine learning described in the embodiment shown are not repeated here. The specific way in which each module and unit performs operations in the system for predicting the condition of the injured and sick based on machine learning in the above embodiment has been described in detail in the embodiment of the method, and will not be elaborated here.

[0198] In one possible design, Figure 2 The machine learning-based injury prediction system of the illustrated embodiment 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;

[0199] 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 .

[0200] The processing component 32 is used to: collect and integrate real-time information flows from diverse data sources, and generate an initial data set of the injured and sick through a multi-source data fusion method; based on the initial data set of the injured and sick, use a random field algorithm to process and analyze the time series data in the state evolution map of the injured and sick, capture the complex relationship and dynamic characteristics between the states of the injured and sick, use time-dependent node embedding technology to enhance the ability to understand time series characteristics, reveal potential injury development trends and abnormal patterns, and generate an optimized state evolution map of the injured and sick; based on the optimized state evolution map of the injured and sick, use a K-nearest neighbor algorithm to perform similarity matching analysis, search and locate a historical case set that is highly similar to the current state of the injured and sick, accurately adjust the weight, use a mask area filtering method, calculate the actual area covered by each target segmentation result, filter out results less than a preset threshold, and generate an optimized state prediction model of the injured and sick; based on the optimized state prediction model of the injured and sick, establish a dynamic risk assessment and interactive feedback optimization mechanism to generate an injury prediction plan for the injured and sick.

[0201] 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.

[0202] 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.

[0203] Of course, the computing device may also include other components, such as input / output interfaces, display components, communication components, etc.

[0204] 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.

[0205] The communication component is configured to facilitate, among other things, wired or wireless communications between the computing device and other devices.

[0206] 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.

[0207] 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 predicting the injury status of the injured and sick based on machine learning.

[0208] 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.

[0209] 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.

[0210] 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.

[0211] 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 predicting the injury of the injured based on machine learning, characterized in that: include: Collect and integrate real-time information streams from diverse data sources, and generate initial datasets of the injured and sick through multi-source data fusion methods; Based on the initial data set of the injured and sick, the random field algorithm is used to process and analyze the time series data in the evolution map of the injured and sick states, capture the complex relationship and dynamic characteristics between the states of the injured and sick, and use the time-dependent node embedding technology to enhance the ability to understand the time series characteristics, reveal the potential development trend and abnormal pattern of the injury, and generate an optimized evolution map of the injured and sick states; Based on the optimized sick and wounded status evolution map, the K nearest neighbor algorithm is used to perform similarity matching analysis, and a historical case set that is highly similar to the current sick and wounded status is searched and located to accurately adjust the weights. The mask area filtering method is used to calculate the actual area covered by each target segmentation result, and the results that are less than the preset threshold are filtered out to generate an optimized sick and wounded status prediction model. Based on the optimized patient status prediction model, a dynamic risk assessment and interactive feedback optimization mechanism is established to generate a patient injury prediction plan.

2. The method according to claim 1, characterized in that Based on the initial data set of the injured and sick, the random field algorithm is used to process and analyze the time series data in the state evolution map of the injured and sick, capture the complex relationship and dynamic characteristics between the states of the injured and sick, adopt the time-dependent node embedding technology, enhance the ability to understand the time series characteristics, reveal the potential development trend and abnormal pattern of the injury, and generate an optimized state evolution map of the injured and sick, including: Based on the initial data set of the injured and sick, the time series data of the injured and sick are node-processed to generate the node timestamp of the status of the injured and sick; Based on the timestamps of the patient status nodes, the random field algorithm is used to model the relationship between nodes, capture the complex relationship and dynamic characteristics between the patient status, identify the interdependence and time-varying nature between nodes in different statuses, reveal the potential development trend and abnormal pattern of the injury, and generate a random field model; Based on the random field model, the time-dependent node embedding technology is used to deeply process and analyze the time series data in the state evolution map of the injured and sick, and each state node is embedded in the multidimensional space, so that the adjacent nodes are closer in space, and the time-dependent node embedding result is generated; Based on the time-dependent node embedding results, key development trends and abnormal patterns are further annotated to generate an optimized evolution map of the status of the injured and sick.

3. The method according to claim 2, characterized in that Based on the timestamp of the patient status node, the random field algorithm is used to model the relationship between nodes, capture the complex relationship and dynamic characteristics between the patient status, identify the interdependence and time-varying nature between nodes in different statuses, reveal the potential development trend and abnormal pattern of the injury, and generate a random field model, including: Based on the timestamps of the patient status nodes, a correlation analysis is performed on the patient status nodes corresponding to each timestamp to generate a patient status node relationship network; Based on the node relationship network of the patient status, the random field algorithm is used to model the relationship between nodes, capture the complex relationship and dynamic characteristics between the patient status, identify the interdependence and time-varying nature between nodes of different statuses, reveal the potential development trend and abnormal pattern of the injury, and generate a node relationship model; Based on the node relationship model, a statistical method is used to evaluate the stability and accuracy of the model to ensure that the real evolution of the status of the injured and sick is accurately reflected and an optimized node relationship model is generated; Based on the optimized node relationship model, combined with the timestamp of the patient status node, the influence of external environmental factors is further introduced to generate a random field model.

4. The method according to claim 2, characterized in that: Based on the random field model, the time-dependent node embedding technology is used to deeply process and analyze the time series data in the state evolution map of the injured and sick, embed each state node into a multidimensional space, make adjacent nodes closer in space, and generate a time-dependent node embedding result, including: Based on the random field model, the time series data in the state evolution graph of the injured and sick is structured and converted into a format suitable for embedding operations to generate a structured state node set; Based on the structured state node set, a time-dependent node embedding technique is used to map each state node into a multidimensional space, reflect the time dependency relationship and interaction between state nodes, ensure that adjacent nodes are closer in space, and generate a time-dependent node embedding vector; Based on the time-dependent node embedding vector, the distance and position between nodes are optimized and adjusted to better capture the real dynamic characteristics of the evolution of the status of the injured and sick, and generate an optimized node embedding layout; Based on the optimized node embedding layout, the path of the patient's state change and the time dependency are displayed through a visualization tool, and a time-dependent node embedding result is generated.

5. The method according to claim 1, characterized in that Based on the optimized sick and wounded status evolution map, the K nearest neighbor algorithm is used to perform similarity matching analysis, search and locate historical case sets that are highly similar to the current sick and wounded status, accurately adjust the weights, use the mask area filtering method, calculate the actual area covered by each target segmentation result, filter out results less than a preset threshold, and generate an optimized sick and wounded status prediction model, including: Based on the optimized sick and wounded status evolution map, current sick and wounded status information is compared with data in the historical case library to generate a preliminary similar case set; Based on the preliminary similar case set, the K nearest neighbor algorithm is used to perform accurate similarity matching analysis, by calculating the distance between the current sick and wounded status node and each case in the historical case set, selecting the closest K neighbor cases, searching for historical cases that are highly similar to the current situation, and generating a selected similar case set; Based on the selected similar case set, a mask area filtering method is applied to calculate the actual area covered by each target segmentation result, and results smaller than a preset threshold are filtered out to generate case quality screening results; Based on the case quality screening results, all high-quality case data are integrated to generate an optimized prediction model for the status of the injured and sick.

6. The method according to claim 5, characterized in that Based on the preliminary similar case set, the K nearest neighbor algorithm is used to perform accurate similarity matching analysis, by calculating the distance between the current sick and wounded status node and each case in the historical case set, selecting the closest K neighbor cases to search for historical cases that are highly similar to the current situation, and generating a selected similar case set, including: Based on the preliminary similar case set, calculating the similarity between the time series feature and other related attributes to generate a preprocessed similar case set; Based on the preprocessed similar case set, the K nearest neighbor algorithm is used to perform accurate similarity matching analysis, and the K closest neighbor cases are selected by calculating the distance between the current sick and wounded status node and each case in the historical case set to generate a selected neighbor case set; Based on the selected neighbor case set, a detailed comparative analysis of the time series data of each selected case is performed to deeply evaluate the subtle differences between each neighbor case and the current status of the injured and sick, and generate a high-precision similar case assessment report; Based on the high-precision similar case evaluation report, similarity confirmation and detailed comparative analysis are carried out to generate a selected similar case set.

7. The method according to claim 1, characterized in that The optimized patient status prediction model is used to establish a dynamic risk assessment and interactive feedback optimization mechanism to generate a patient injury prediction plan, including: Based on the optimized patient status prediction model, the patient status evolution is monitored in real time, potential risk factors are identified and quantified, and a real-time risk factor report is generated; Based on the real-time risk factor report, a dynamic risk assessment system is constructed, and the probability of the injury of the injured and sick worsening is obtained by combining the prediction model parameters with the real-time risk factor data, and a dynamic risk assessment result is generated; Based on the dynamic risk assessment results, an interactive feedback optimization mechanism is established to collect the subsequent evaluations of medical experts as incremental learning samples and generate feedback optimization records; Based on the feedback optimization records, the optimized patient status prediction model is periodically fine-tuned to adapt to new injury conditions and the evolution of medical practice, and a patient injury prediction plan is generated.

8. A system for predicting the condition of injured and sick patients based on machine learning, characterized in that: include: The collection module is used to collect and integrate real-time information flows from diverse data sources and generate an initial dataset of the injured and sick through a multi-source data fusion method; An analysis module is used to process and analyze the time series data in the state evolution map of the injured and sick based on the initial data set of the injured and sick using a random field algorithm, capture the complex relationship and dynamic characteristics between the states of the injured and sick, use time-dependent node embedding technology to enhance the ability to understand the time series characteristics, reveal potential injury development trends and abnormal patterns, and generate an optimized state evolution map of the injured and sick; An adjustment module is used to perform similarity matching analysis based on the optimized sick and wounded status evolution map using the K nearest neighbor algorithm, search and locate a historical case set that is highly similar to the current sick and wounded status, accurately adjust the weight, and use a mask area filtering method to calculate the actual area covered by each target segmentation result, filter out results that are less than a preset threshold, and generate an optimized sick and wounded status prediction model; A generation module is used to establish a dynamic risk assessment and interactive feedback optimization mechanism based on the optimized patient status prediction model to generate a patient injury prediction plan.

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 predicting the injury of a patient based on machine learning 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 predicting the injury condition of a patient based on machine learning as described in any one of claims 1 to 7 is implemented.

Images