Method and system for quickly classifying injury conditions in intelligent medical treatment box group

Abstract

The invention provides a method and a system for quickly classifying injury conditions in an intelligent medical treatment box group. The method comprises the following steps: receiving physiological parameters and image data of the sick and wounded, integrating current vital sign conditions, and generating a preliminary health status report; based on the preliminary health state report, a bidirectional encoder representation algorithm is used, high-dimensional feature representation is extracted, a cross-domain transfer learning technology is used, health indexes and change trends are recognized, and comprehensive health state indexes are generated; based on the comprehensive health state index, constructing a plurality of decision tree models by using an extreme gradient lifting algorithm, and optimizing an emergency response level matching degree and generating an optimized emergency response level by using a dynamic Bayesian network technology; and based on the optimized emergency response level, performing quick classification of the injury condition, and generating a treatment scheme suggestion according to the quick classification result of the injury condition. According to the technical scheme provided by the invention, the medical treatment timeliness is remarkably improved, and the injury condition classification accuracy and efficiency are improved.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of intelligent medical technology, and in particular to a method and system for quickly classifying injuries in an intelligent medical treatment box group. Background Art

[0002] With the continuous growth of medical technology and emergency rescue needs, it is crucial to quickly and accurately assess the health status of the injured and sick in emergency medical rescue scenarios, such as traffic accidents, natural disasters or large-scale emergencies. Medical staff need to integrate physiological parameters and image data in a short period of time to generate a preliminary health status report and make the best emergency response decision based on it. This requires not only efficient injury classification capabilities, but also the system's ability to monitor and dynamically update in real time to adapt to the changing conditions of the injured and sick.

[0003] Currently, many medical treatment systems rely on traditional manual assessments and limited automated tools to process information about the injured and sick. These systems typically include basic vital sign monitoring equipment and simple data analysis software to make a preliminary judgment on the status of the injured and sick. In addition, some more advanced systems have begun to introduce machine learning algorithms for injury classification, but most of these methods are based on a single data source or static models and lack the ability to comprehensively analyze multimodal data (such as physiological parameters and image data).

[0004] Existing solutions have obvious deficiencies in the efficiency and accuracy of injury classification. Traditional manual assessment is time-consuming and easily affected by human factors, leading to misjudgment or delayed treatment. Even existing automated tools have limited classification accuracy and generalization capabilities due to the failure to fully utilize advanced technologies such as high-dimensional feature representation and cross-domain transfer learning, and cannot effectively cope with the complex and changing conditions of the injured and sick. In addition, the existing system lacks continuous monitoring and dynamic adjustment mechanisms, making it difficult to update the injury classification results in real time, affecting the optimal matching of emergency response levels, thereby reducing the effectiveness and personalization of overall first aid measures. Summary of the invention

[0005] The embodiments of the present application provide a method and system for rapid injury classification in an intelligent medical treatment box group, so as to solve the problem of insufficient efficiency and accuracy in injury classification in the prior art.

[0006] In a first aspect, an embodiment of the present application provides a method for quickly classifying injuries in an intelligent medical treatment box group, comprising:

[0007] Receive physiological parameters and image data of the injured and sick, integrate current vital signs, and generate a preliminary health status report;

[0008] Based on the preliminary health status report, a bidirectional encoder representation algorithm is used to extract high-dimensional feature representation, perform in-depth processing and analysis, and adopt cross-domain transfer learning technology to identify key health indicators and change trends of the injured and sick, and generate a comprehensive health status index;

[0009] Based on the comprehensive health status index, using the extreme gradient boosting algorithm, constructing multiple decision tree models to determine the best emergency response level, using dynamic Bayesian network technology to continuously optimize the emergency response level matching and generate an optimized emergency response level;

[0010] The injuries are quickly classified based on the optimized emergency response level, and treatment plan recommendations are generated based on the results of the rapid classification of the injuries.

[0011] Optionally, based on the preliminary health status report, a bidirectional encoder representation algorithm is used to extract high-dimensional feature representation, perform in-depth processing and analysis, and adopt cross-domain transfer learning technology to identify key health indicators and change trends of the injured and sick, and generate a comprehensive health status index, including:

[0012] Based on the preliminary health status report, performing standardized preprocessing on the physiological parameters and image data to generate a standardized data set;

[0013] Based on the standardized data set, a bidirectional encoder representation algorithm is used for encoding, a multi-layer converter structure is used for deep processing and analysis, a high-dimensional feature representation is extracted to highlight important features, and an enhanced feature representation is generated;

[0014] Based on the enhanced feature representation, cross-domain transfer learning technology is used to centrally learn large-scale medical data knowledge to transfer to current patient data, improve the generalization ability of the model, and generate adaptive feature representation;

[0015] Based on the adaptive characteristic representation, the key health indicators and change trends of the injured and sick are identified to generate a comprehensive health status index.

[0016] Optionally, the encoding is performed using a bidirectional encoder representation algorithm based on the standardized data set, deep processing and analysis is performed through a multi-layer converter structure, high-dimensional feature representation is extracted to highlight important features, and enhanced feature representation is generated, including:

[0017] Based on the standardized data set, the physiological parameters and image data are preliminarily encoded to generate an initial feature vector;

[0018] Based on the initial feature vector, a bidirectional encoder representation algorithm is used to capture contextual information and long-distance dependencies in the data through a self-attention mechanism to generate a preliminary feature representation;

[0019] Based on the preliminary feature representation, a multi-layer converter structure is used to perform in-depth processing and analysis to enhance feature expression capabilities, highlight important features, and generate high-dimensional feature representations;

[0020] Based on the high-dimensional feature representation, a self-attention mechanism is used to automatically identify and highlight key health indicators to generate enhanced feature representation.

[0021] Optionally, based on the enhanced feature representation, cross-domain transfer learning technology is used to centrally learn large-scale medical data knowledge to transfer to current patient data, improve the generalization ability of the model, and generate adaptive feature representation, including:

[0022] Based on the enhanced feature representation, a large-scale medical data set is used to extract and learn common health indicator patterns and change trends to generate a pre-trained model;

[0023] Based on the pre-trained model, the cross-domain transfer learning technology is used to transfer the knowledge in the pre-trained model to the current patient data, and the model parameters are adjusted to improve the fitness, so as to generate a knowledge transfer model;

[0024] Based on the knowledge transfer model, the optimal parameter configuration is selected through a cross-validation method to improve the generalization ability of the model and generate an optimized transfer model;

[0025] Based on the optimized migration model, applicability feature extraction processing is performed to generate adaptive feature representation.

[0026] Optionally, based on the comprehensive health status index, using the extreme gradient boosting algorithm, constructing multiple decision tree models to determine the best emergency response level, using dynamic Bayesian network technology to continuously optimize the emergency response level matching, and generating an optimized emergency response level, including:

[0027] Based on the comprehensive health status index, feature selection and standardization processing are performed to generate a standardized feature vector;

[0028] Based on the standardized feature vectors, an extreme gradient boosting algorithm is used to construct multiple decision tree models, and the correlation between the key health indicators and the changing trends of the injured and sick is learned by recursively segmenting the data set to generate a preliminary emergency response level;

[0029] Based on the initial emergency response level, dynamic Bayesian network technology is used to continuously monitor, capture the time changes of the health status of the injured and sick, update the probability distribution in real time, and generate an optimized probability distribution;

[0030] Based on the optimized probability distribution, the matching degree of the emergency response level is continuously optimized, the personalization and timeliness of the first aid measures are improved, and the optimized emergency response level is generated.

[0031] Optionally, based on the standardized feature vector, an extreme gradient boosting algorithm is used to construct multiple decision tree models, and by recursively segmenting the data set, the correlation between the key health indicators of the injured and the sick and the change trend is learned to generate a preliminary emergency response level, including:

[0032] Based on the standardized feature vector, initialization processing is performed, extreme gradient boosting algorithm parameters are set, and an initialization model is generated;

[0033] Based on the initialization model, multiple decision tree models are constructed using the extreme gradient boosting algorithm, and the correlation between the key health indicators and the change trends of the injured and sick is learned by recursively segmenting the data set to generate initial prediction results;

[0034] Based on the initial prediction results, a new decision tree model is constructed in combination with the model residuals, and the data set is continued to be recursively segmented to capture the part that is not fully explained by the previous decision tree, so as to generate an updated prediction result;

[0035] Based on the updated prediction results, multiple decision tree prediction results are integrated through an ensemble learning method to generate a preliminary emergency response level.

[0036] Optionally, the rapidly classifying the injuries based on the optimized emergency response level, and generating treatment plan suggestions according to the rapid classification results of the injuries, includes:

[0037] Based on the optimized emergency response level, the injuries of the injured and sick are quickly classified, wherein the process of quickly classifying the injuries of the injured and sick includes taking the optimized emergency response level as an input condition, combining it with a pre-trained deep learning model to perform in-depth analysis and processing on the vital signs, physiological parameters and image data of the injured and sick, and obtaining a rapid classification result of the injuries of the injured and sick, including minor injuries, moderate injuries and severe injuries;

[0038] Generate treatment plan recommendations based on the rapid injury classification results.

[0039] In a second aspect, an embodiment of the present application provides a system for rapid injury classification in an intelligent medical treatment box group, comprising:

[0040] The receiving module is used to receive the physiological parameters and image data of the injured and sick, integrate the current vital signs, and generate a preliminary health status report;

[0041] An analysis module is used to extract high-dimensional feature representation based on the preliminary health status report using a bidirectional encoder representation algorithm, perform in-depth processing and analysis, and use cross-domain transfer learning technology to identify key health indicators and change trends of the injured and sick, and generate a comprehensive health status index;

[0042] A construction module is used to construct multiple decision tree models based on the comprehensive health status index using an extreme gradient boosting algorithm to determine the best emergency response level, and to continuously optimize the emergency response level matching using a dynamic Bayesian network technology to generate an optimized emergency response level;

[0043] A generation module is used to quickly classify injuries based on the optimized emergency response level, and generate treatment plan recommendations based on the results of the rapid injury classification.

[0044] 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 rapid injury classification in an intelligent medical treatment box group as described in the first aspect.

[0045] In a fourth aspect, an embodiment of the present application provides a computer storage medium storing a computer program, which, when executed by a computer, implements a method for rapid injury classification in an intelligent medical treatment box group as described in the first aspect.

[0046] In an embodiment of the present application, physiological parameters and image data of the injured and sick are received, the current vital signs are integrated, and a preliminary health status report is generated; based on the preliminary health status report, a bidirectional encoder characterization algorithm is used to extract high-dimensional feature representation, and in-depth processing and analysis are performed. Cross-domain transfer learning technology is used to identify key health indicators and change trends of the injured and sick, and a comprehensive health status index is generated; based on the comprehensive health status index, an extreme gradient boosting algorithm is used to construct multiple decision tree models to determine the optimal emergency response level, and dynamic Bayesian network technology is used to continuously optimize the matching degree of the emergency response level to generate an optimized emergency response level; based on the optimized emergency response level, the injury is quickly classified, and a treatment plan recommendation is generated based on the rapid injury classification result. By receiving the physiological parameters and image data of the wounded and patients and integrating the current vital signs, a preliminary health status report can be quickly generated to ensure that subsequent processing is based on the most comprehensive and up-to-date information; using the bidirectional encoder representation algorithm and cross-domain transfer learning technology, high-dimensional feature representations are extracted, and the key health indicators and change trends of the wounded and patients are deeply analyzed to generate a comprehensive health status index, which improves the model's ability to understand complex medical data; based on the comprehensive health status index, combined with the extreme gradient boosting algorithm and dynamic Bayesian network technology, multiple decision tree models are constructed to determine the optimal emergency response level, and the matching degree is continuously optimized to ensure the personalization and timeliness of first aid measures; according to the optimized emergency response level, the corresponding medical treatment box group is intelligently selected, and the injury classification results of the wounded and patients are dynamically updated through the continuous monitoring system to generate an efficient injury rapid classification strategy, thereby improving the utilization efficiency of medical resources.

[0047] Furthermore, by performing standardized preprocessing on physiological parameters and image data and generating standardized data sets, the consistency and quality of the input data are ensured, laying a solid foundation for subsequent in-depth processing and analysis; the bidirectional encoder characterization algorithm and multi-layer converter structure are used to extract high-dimensional feature representations, effectively highlighting important features and enhancing the model's ability to identify key health indicators; cross-domain transfer learning technology is used to concentrate on learning knowledge from large-scale medical data, thereby improving the model's generalization ability in different scenarios and enabling the model to better adapt to the data characteristics of specific patients; by identifying the key health indicators and changing trends of the patients, a comprehensive health status index is generated, which provides a scientific basis for the subsequent determination of the emergency response level and improves the intelligence level of the overall system.

[0048] Furthermore, feature selection and standardization were performed on the comprehensive health status index to generate standardized feature vectors, which ensured the quality and consistency of the input data and laid the foundation for building an accurate decision tree model; the extreme gradient boosting algorithm was used to build multiple decision tree models, and the correlation between the key health indicators and change trends of the injured and sick was learned by recursively segmenting the data set to generate a preliminary emergency response level, which improved the learning ability and prediction accuracy of the model; dynamic Bayesian network technology was used for continuous monitoring to capture the temporal changes in the health status of the injured and sick, update the probability distribution in real time, and generate an optimized probability distribution, ensuring the dynamic adjustment and timeliness of the emergency response level; based on the optimized probability distribution, the matching degree of the emergency response level was continuously optimized, which improved the personalization and timeliness of the first aid measures, ensured the effectiveness and pertinence of the treatment plan, and improved the survival rate and recovery speed of the injured and sick.

[0049] 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

[0050] 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.

[0051] Figure 1 A flowchart of a method for quickly classifying injuries in an intelligent medical treatment box group provided in an embodiment of the present application;

[0052] Figure 2 A schematic diagram of the structure of a rapid injury classification system in an intelligent medical treatment box group provided in an embodiment of the present application;

[0053] Figure 3A schematic diagram of the structure of a computing device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0054] 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.

[0055] 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.

[0056] 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.

[0057] Figure 1 A flowchart of a method for quickly classifying injuries in an intelligent medical treatment box group is provided for an embodiment of the present application, such as Figure 1 As shown, the method includes:

[0058] 101. Receive physiological parameters and image data of the injured and sick, integrate the current vital signs, and generate a preliminary health status report;

[0059] In this step, physiological parameters include heart rate, blood pressure, body temperature, blood oxygen saturation, etc. These data are obtained through various sensors and monitoring equipment to assess the basic health status of the injured and sick.

[0060] Image data includes X-rays, CT scans, ultrasound images, and wound photos, etc. These images provide detailed information about internal organs and external injuries, helping doctors make more comprehensive diagnoses.

[0061] Vital signs refer to the overall health status of the patient as reflected by physiological parameters, including consciousness level, respiratory rate, etc. It provides medical personnel with a window to quickly understand the health status of the patient.

[0062] The preliminary health status report is a brief report generated based on the integrated physiological parameters and image data. It is used to quickly understand the basic condition of the injured and sick and to guide initial treatment measures.

[0063] In the embodiment of the present application, assuming a rescue site with a complex and changeable environment, first, the intelligent medical treatment box group is equipped with a variety of sensors and portable imaging devices, which can collect the physiological parameters and image data of the injured and sick in real time; secondly, the system's built-in data processing module pre-processes and standardizes the received data to ensure data quality; thirdly, based on these data, the system generates a health status report containing basic vital signs and preliminary diagnostic information; finally, the report is sent instantly to on-site medical staff and telemedicine centers through wireless transmission technology to provide a basis for subsequent treatment.

[0064] 102. Based on the preliminary health status report, use the bidirectional encoder representation algorithm to extract high-dimensional feature representation, conduct in-depth processing and analysis, adopt cross-domain transfer learning technology to identify the key health indicators and change trends of the injured and sick, and generate a comprehensive health status index;

[0065] In this step, the bidirectional encoder representation algorithm is a deep learning method that can capture contextual information and long-distance dependencies in text or sequence data, and is suitable for processing complex medical data.

[0066] High-dimensional feature representation refers to the complex and abstract feature vectors extracted from the original data, which can better reflect the essential characteristics of the data and help improve the expressiveness and generalization capabilities of the model.

[0067] Deep processing analysis includes multi-layer transformer structures and other complex neural network architectures to deeply explore hidden patterns in the data, thereby improving the understanding of the health status of the injured and sick.

[0068] Cross-domain transfer learning technology refers to the transfer of knowledge from one domain (source domain) to another domain (target domain) to improve the generalization ability of the model on new data. It is particularly suitable for the heterogeneity and scarcity of medical data.

[0069] The comprehensive health status index is a comprehensive score generated by further processing the high-dimensional feature representation. It is used to assess the overall health status of the injured and sick and guide subsequent emergency response decisions.

[0070] In the embodiments of the present application, it is assumed that a large amount of medical data from different sources and types needs to be processed. First, the system encodes the physiological parameters and image data using a bidirectional encoder representation algorithm based on the data in the preliminary health status report to generate high-dimensional feature representations. Second, deep processing and analysis are performed through a multi-layer transformer structure to highlight important features. Third, cross-domain transfer learning technology is used to centrally learn the knowledge of large-scale medical data and apply it to the data of the current wounded and sick personnel to generate adaptive feature representations. Finally, the system identifies the key health indicators and change trends of the wounded and sick personnel, generates a comprehensive health status index, and provides a scientific basis for determining the subsequent emergency response level.

[0071] Optionally, in step 102, based on the preliminary health status report, using a bidirectional encoder representation algorithm to extract high-dimensional feature representations, perform deep processing and analysis, adopt cross-domain transfer learning technology, identify the key health indicators and change trends of the wounded and sick personnel, and generate a comprehensive health status index, including: based on the preliminary health status report, perform standardized preprocessing on the physiological parameters and image data to generate a standardized data set; based on the standardized data set, use a bidirectional encoder representation algorithm for encoding, perform deep processing and analysis through a multi-layer transformer structure, extract high-dimensional feature representations to highlight important features, and generate enhanced feature representations; based on the enhanced feature representations, adopt cross-domain transfer learning technology to centrally learn the knowledge of large-scale medical data to be transferred to the data of the current wounded and sick personnel to improve the generalization ability of the model and generate adaptive feature representations; based on the adaptive feature representations, identify the key health indicators and change trends of the wounded and sick personnel and generate a comprehensive health status index.

[0072] In this step, the standardized preprocessing refers to the unified formatting and normalization processing of the physiological parameters and image data in the preliminary health status report to ensure that data from different sources can be compared and analyzed under the same standard.

[0073] The multi-layer transformer structure is a deep learning architecture that can perform deep processing and analysis to highlight important features in the data.

[0074] The cross-domain transfer learning technology uses the knowledge of other fields (such as large-scale medical databases) to improve the generalization ability of the model on the data of the current wounded and sick personnel.

[0075] The comprehensive health status index is a final generated score used to evaluate the overall health status of the wounded and sick personnel and guide subsequent emergency response decisions.

[0076] First, based on the preliminary health status report, the system performs standardized preprocessing on the collected physiological parameters and image data to eliminate differences between the data and generate a standardized data set; secondly, the system uses a bidirectional encoder representation algorithm to encode the standardized data set, and performs deep processing and analysis through a multi-layer converter structure to extract high-dimensional feature representations that can highlight important features and form enhanced feature representations; thirdly, the system uses cross-domain transfer learning technology to focus on learning knowledge from large-scale medical data and apply it to the data of current injured and sick patients to improve the generalization ability and adaptability of the model and generate adaptive feature representations; finally, based on these adaptive feature representations, the system identifies the key health indicators of the injured and sick and their changing trends, and generates a comprehensive health status index, providing a scientific basis for the subsequent determination of the emergency response level.

[0077] Optionally, based on the standardized data set, encoding is performed using a bidirectional encoder representation algorithm, deep processing and analysis is performed through a multi-layer converter structure, high-dimensional feature representation is extracted to highlight important features, and enhanced feature representation is generated, including: based on the standardized data set, the physiological parameters and image data are preliminarily encoded to generate an initial feature vector; based on the initial feature vector, a bidirectional encoder representation algorithm is used to capture contextual information and long-distance dependencies in the data through a self-attention mechanism to generate a preliminary feature representation; based on the preliminary feature representation, deep processing and analysis is performed through a multi-layer converter structure to enhance feature expression capabilities, highlight important features, and generate a high-dimensional feature representation; based on the high-dimensional feature representation, a self-attention mechanism is used to automatically identify and highlight key health indicators to generate an enhanced feature representation.

[0078] Based on the enhanced feature representation, cross-domain transfer learning technology is used to concentrate on learning large-scale medical data knowledge to transfer it to current patient data, improve the generalization ability of the model, and generate an adaptive feature representation, including: based on the enhanced feature representation, through a large-scale medical data set, extracting and learning common health indicator patterns and change trends to generate a pre-trained model; based on the pre-trained model, cross-domain transfer learning technology is used to transfer the knowledge in the pre-trained model to the current patient data, adjust the model parameters to improve the fitness, and generate a knowledge transfer model; based on the knowledge transfer model, the optimal parameter configuration is selected through a cross-validation method to improve the generalization ability of the model and generate an optimized transfer model; based on the optimized transfer model, applicability feature extraction processing is performed to generate an adaptive feature representation.

[0079] In this step, the standardized data set includes preprocessed physiological parameters and image data, which are uniformly formatted to ensure consistency and comparability. The standardization process eliminates the differences between data from different sources, making subsequent analysis more accurate and reliable.

[0080] The initial feature vector is a low-dimensional representation generated by preliminary encoding of the standardized data, which is used to capture the basic characteristics of the original data.

[0081] The self-attention mechanism automatically highlights important information by calculating the correlation between each element in the input sequence and other elements, thereby enhancing the model's ability to understand complex patterns.

[0082] Enhanced feature representation is a set of features generated by further processing the high-dimensional feature representation, which highlights key health indicators and provides richer information for subsequent analysis.

[0083] The pre-trained model is an initial model built by learning from a large-scale general dataset. It aims to extract widely applicable feature patterns and reduce the time and resource consumption of training from scratch.

[0084] The knowledge transfer model is generated by transferring the knowledge in the pre-trained model to the current patient data and adjusting the model parameters to improve the fitness.

[0085] The cross-validation method divides the data set into multiple subsets, repeatedly trains and tests the model, and selects the optimal parameter configuration to improve model performance.

[0086] The optimized migration model is the final model optimized by the cross-validation method and has better generalization ability and adaptability.

[0087] Adaptive feature representation is a feature representation generated based on the optimized migration model. It is specially adjusted according to the data characteristics of the current injured and sick patients, which improves the application effect of the model.

[0088] First, based on a standardized data set, the system preliminarily encodes physiological parameters and image data, uses a bidirectional encoder representation algorithm, and combines the self-attention mechanism to capture contextual information and long-distance dependencies in the data to generate a preliminary feature representation; secondly, through a multi-layer converter structure, deep processing and analysis are performed to enhance feature expression capabilities, highlight important features, generate high-dimensional feature representations, and use the self-attention mechanism to automatically identify and highlight key health indicators to generate enhanced feature representations; thirdly, based on enhanced feature representations, the system extracts and learns common health indicator patterns and changing trends through a large-scale medical data set, uses cross-domain transfer learning technology to transfer knowledge in the pre-trained model to current patient data, adjusts model parameters to improve fitness, and generates a knowledge transfer model; finally, the cross-validation method is used to select the optimal parameter configuration, improve the generalization ability of the model, perform applicable feature extraction processing, and generate adaptive feature representations to ensure the best performance of the model in specific scenarios.

[0089] In the embodiment of the present application, it is assumed that it is necessary to quickly and accurately assess the health status of the injured and sick in a complex emergency environment; first, the system receives standardized physiological parameters and image data, performs preliminary encoding to generate an initial feature vector, and uses a bidirectional encoder representation algorithm combined with a self-attention mechanism to capture contextual information and long-distance dependencies in the data to generate a preliminary feature representation; secondly, a multi-layer converter structure is used to perform deep processing and analysis to enhance feature expression capabilities, generate high-dimensional feature representations, and use a self-attention mechanism to highlight key health indicators. The model is pre-trained with a large-scale medical data set, and cross-domain transfer learning technology is used to adjust model parameters to generate a knowledge transfer model. The optimal parameter configuration is selected through cross-validation to generate an optimized transfer model; finally, applicability feature extraction processing is performed to generate an adaptive feature representation to ensure the best application effect of the model in a specific emergency scenario.

[0090] This application takes into account that due to the problem of insufficient processing capabilities for complex medical data in the prior art, traditional feature extraction methods are difficult to capture long-distance dependencies and contextual information, resulting in limited expression and generalization capabilities of the model. Therefore, the embodiment of the invention proposes this optional solution to solve these problems and improve the accuracy and timeliness of the assessment of the health status of the wounded and sick. The reason for the overall design is to enhance the model's ability to understand sequence data by introducing a bidirectional encoder representation algorithm and a self-attention mechanism, so as to more accurately identify the key health indicators of the wounded and sick and their changing trends.

[0091] Optionally, based on the initial feature vector, a bidirectional encoder representation algorithm is used to capture context information and long-distance dependencies in the data through a self-attention mechanism to generate a preliminary feature representation, including:

[0092] Based on the initial feature vector, a linear transformation is performed on the input sequence through a corresponding weight matrix;

[0093] The input feature vector is positionally encoded through trigonometric functions to capture the relative position information of the elements in the sequence to generate an attention weight matrix;

[0094] The attention weight matrix is ​​calculated using the following formula:

[0095]

[0096] Among them, Attention is the attention weight matrix; Q is the query matrix, which is the linear transformation result of the input sequence; K is the key matrix, which is the linear transformation result of the input sequence; V is the value matrix, which is the linear transformation result of the input sequence; d k is the dimension of the key matrix, which is used as a scaling factor to keep the gradient stable; σ is the activation function Softmax, which is used to normalize the weights; b 1is the bias term, used to adjust the input of the activation function; b 2 is a bias term used to adjust the output of the activation function;

[0097] Based on the attention weight matrix, multiply the value vector for weighting, multiply the output matrix to integrate the results of multiple attention heads, use residual connections to retain input information, and ensure data distribution stability through layer normalization operations to generate the final output feature vector;

[0098] The final output feature vector is calculated using the following formula:

[0099]

[0100] Among them, Output is the final output feature vector; is the weight of the i-th query matrix; is the weight of the i-th bond matrix; is the weight of the i-th value matrix; i is the index of the attention head, from 1 to h; h is the number of attention heads, indicating the number of attention streams processed in parallel; W O is the weight of the output matrix, is the linear transformation parameter that combines the outputs of multiple attention heads; X is the input feature vector, is the initial feature vector or the output of the previous layer; W 1 is the weight of the first layer of the fully connected network; W 2 is the weight of the second layer of the fully connected network; LayerNorm is the layer normalization function, which is used to normalize each sample; b 1 ' is the bias term of the first layer of the feedforward neural network; b 2 ' is the bias term of the second layer of the feedforward neural network; Concat is a concatenation operation, which is used to concatenate the output results of multiple attention heads along a certain dimension (usually the last dimension) to form a larger vector; X is the input feature vector, which is the initial feature vector or the output of the previous layer;

[0101] Based on the final output feature vector, a feedforward neural network is introduced. By applying the ReLU activation function between the fully connected layers, a nonlinear mechanism is introduced to further enhance the model's expressiveness and generalization capabilities and generate a preliminary feature representation.

[0102] This method aims to use a bidirectional encoder representation algorithm combined with a self-attention mechanism, which can effectively capture the contextual information and long-distance dependencies in the data and generate high-dimensional feature representations; at the same time, through multi-head attention mechanisms, residual connections and layer normalization operations, the model stability and data distribution consistency are ensured; finally, a feedforward neural network is introduced to further enhance the model's expressiveness and generalization capabilities, providing a reliable preliminary feature representation for subsequent analysis.

[0103] In the attention weight matrix, the activation function calculates the term This sub-item uses the sigmoid activation function σ to normalize the result after the hyperbolic tangent function tanh processing, aiming to ensure that key information is retained and transmitted; Value vector item V: This sub-item uses the value matrix V, which contains the information of the actual content, and aims to combine the result calculated by the activation function with the actual content to generate the final attention weight;

[0104] Among them, the bond matrix dimension d k Preset model hyperparameters; bias term b 1 The bias term learned during training; bias term b 2 The bias term learned during the training process; the query matrix Q is obtained from the input sequence through linear transformation. The specific form is The key matrix K is obtained from the input sequence through linear transformation. The value matrix V is obtained from the input sequence through linear transformation. Weight Matrix Usually obtained through optimization during the training process;

[0105] In the final output feature vector, the concatenation and weighted integration terms This sub-item uses the concatenation operation Concat to concatenate the output results of multiple attention heads along a certain dimension (usually the last dimension) to form a larger vector and add it to the output matrix W O Multiply to integrate the results of multiple attention heads. Nonlinear enhancement term This sub-item introduces a nonlinear mechanism by applying the ReLU activation function max(0,·) between fully connected layers to further enhance the model's expressiveness and generalization capabilities and generate preliminary feature representations;

[0106] Among them, the output matrix W O It is obtained through optimization during the training process to ensure that the outputs of different attention heads can be effectively merged; the input feature vector X is the initial feature vector or the output of the previous layer, which is used as the model input; the layer normalization parameters are automatically adjusted through training to ensure that the data distribution of each layer is stable; the weight W of the first layer of the fully connected network 1 Through random initialization and continuous updating during training, it can adapt to different task requirements; the weight W of the second layer of fully connected network 2 Similarly, random initialization and optimization during training are used to achieve effective conversion of features; the bias term b' of the first layer of the feedforward neural network 1 Learned during the training process, used to adjust the input of the activation function; the bias term b' of the second layer of the feedforward neural network 2It is learned during the training process and is used to adjust the output of the activation function. The attention head index i ranges from 1 to h, where h is the number of attention heads, indicating the number of attention streams processed in parallel.

[0107] Assume that in an application scenario of an intelligent medical treatment kit group, it is necessary to process the vital signs data of a group of injured and sick people, including physiological parameters such as heart rate and blood pressure; Assume in is a pre-trained weight matrix; assuming the key matrix dimension is d k =64, bias term b 1 =0.5, bias term b 2 =0.3; Assuming the number of attention heads h = 8, the output matrix W O , the weight of the first layer of the fully connected network is W 1 , the weight of the second layer of the fully connected network is W 2 , the bias term b' of the first layer of the feedforward neural network 1 = 0.7, bias term b' of the second layer of the feedforward neural network 2 =0.9 are all pre-trained parameters; assuming X is the normalized initial feature vector, which is [0.1, 0.2, 0.3, …, 0.9];

[0108]

[0109]

[0110] Assume that the threshold is set to θ = 0.8, since each element of the final output feature vector

[0111] [0.85, 0.78, 0.92, 0.89, 0.91, 0.83, 0.87, 0.90, 0.86, 0.88] are all greater than the corresponding set thresholds, which indicates that the current patient information performs well in terms of understanding and feature capture, and is suitable for further use or as the final information representation. Through the above steps, medical data can be effectively processed and high-quality feature representations can be extracted, thereby improving the accuracy and timeliness of the assessment of the health status of the patient.

[0112] 103. Based on the comprehensive health status index, using the extreme gradient boosting algorithm, construct multiple decision tree models to determine the best emergency response level, using dynamic Bayesian network technology to continuously optimize the emergency response level matching and generate an optimized emergency response level;

[0113] In this step, the extreme gradient boosting algorithm is an enhanced machine learning method that improves prediction accuracy by iteratively building multiple weak classifiers and combining them to form a strong classifier. It is particularly suitable for dealing with nonlinearity and complex correlations.

[0114] The decision tree model is a prediction model based on a tree structure. Each node represents a feature test, each branch represents a test result, and the leaf node represents the final prediction value. It can intuitively display the decision-making process and facilitate understanding and explanation.

[0115] The emergency response level refers to the different levels divided according to the health status of the injured and sick. It determines the required first aid measures and resource allocation, and is an important basis for priority sorting during the first aid process.

[0116] Dynamic Bayesian network technology is a probabilistic graphical model that can process time series data and capture changes in the health status of the injured and sick by continuously updating the probability distribution, thereby improving the accuracy and timeliness of the prediction.

[0117] In the embodiment of the present application, it is assumed that rapid and accurate emergency response decisions need to be made in a dynamically changing environment. First, the system performs feature selection and standardization based on the comprehensive health status index to generate a standardized feature vector; secondly, the extreme gradient boosting algorithm is used to construct multiple decision tree models, and the correlation between the key health indicators and the changing trends of the injured and sick is learned by recursively segmenting the data set to generate a preliminary emergency response level; thirdly, the dynamic Bayesian network technology is used for continuous monitoring to capture the temporal changes in the health status of the injured and sick, update the probability distribution in real time, and generate an optimized probability distribution; finally, based on the optimized probability distribution, the matching degree of the emergency response level is continuously optimized to improve the personalization and timeliness of the first aid measures and generate an optimized emergency response level.

[0118] Optionally, in step 103, based on the comprehensive health status index, an extreme gradient boosting algorithm is used to construct multiple decision tree models to determine the optimal emergency response level, and the dynamic Bayesian network technology is used to continuously optimize the matching degree of the emergency response level to generate an optimized emergency response level, including: based on the comprehensive health status index, feature selection and standardization processing are performed to generate a standardized feature vector; based on the standardized feature vector, an extreme gradient boosting algorithm is used to construct multiple decision tree models, and by recursively segmenting the data set, the correlation between the key health indicators and the changing trends of the injured and sick is learned to generate a preliminary emergency response level; based on the preliminary emergency response level, continuous monitoring is performed using dynamic Bayesian network technology to capture the time changes in the health status of the injured and sick, and the probability distribution is updated in real time to generate an optimized probability distribution; based on the optimized probability distribution, the matching degree of the emergency response level is continuously optimized to improve the personalization and timeliness of first aid measures to generate an optimized emergency response level.

[0119] Among them, based on the standardized feature vector, the extreme gradient boosting algorithm is used to construct multiple decision tree models, and the correlation between the key health indicators of the injured and the changing trends is learned by recursively segmenting the data set to generate a preliminary emergency response level, including: based on the standardized feature vector, initialization processing is performed, extreme gradient boosting algorithm parameters are set, and an initialization model is generated; based on the initialization model, multiple decision tree models are constructed by the extreme gradient boosting algorithm, and the correlation between the key health indicators of the injured and the changing trends is learned by recursively segmenting the data set to generate an initial prediction result; based on the initial prediction result, a new decision tree model is constructed in combination with the model residual, and the data set is continued to be recursively segmented to capture the part that is not fully explained by the previous decision tree to generate an updated prediction result; based on the updated prediction result, multiple decision tree prediction results are integrated through an integrated learning method to generate a preliminary emergency response level.

[0120] In this step, the comprehensive health status index is a score generated by high-dimensional feature representation to evaluate the overall health status of the injured and sick.

[0121] The standardized feature vector is a data set generated by feature selection and standardization of the comprehensive health status index to ensure the consistency and quality of the input data.

[0122] Recursively splitting a dataset means gradually dividing the dataset into smaller subsets according to different feature values ​​in order to better learn and capture patterns in the data.

[0123] The initial prediction results are the first round of predictions generated based on the initialization model, providing a basis for subsequent optimization.

[0124] Model residuals refer to the difference between the actual value and the predicted value, which is used to guide the construction of new decision tree models to capture the under-explained part.

[0125] Ensemble learning methods are techniques that improve overall performance by combining the predictions of multiple models, such as weighted averaging or voting mechanisms.

[0126] The initial emergency response level is a preliminary classification result generated based on the comprehensive health status index and is used to guide the selection of first aid measures.

[0127] The optimized probability distribution is updated in real time through dynamic Bayesian network technology, which improves the accuracy of emergency response level.

[0128] The optimized emergency response level is the final determined best response level, which improves the personalization and timeliness of first aid measures.

[0129] In the embodiment of the present application, first, the system performs feature selection and standardization based on the comprehensive health status index, generates a standardized feature vector, performs initialization processing, sets the extreme gradient boosting algorithm parameters, and generates an initialization model; secondly, the system uses the extreme gradient boosting algorithm to construct multiple decision tree models, recursively splits the data set, learns the correlation between the key health indicators of the injured and the changing trend, generates initial prediction results, and builds a new decision tree model in combination with the model residual, continues to recursively split the data set, captures the part that is not fully explained by the previous decision tree, generates updated prediction results, and integrates the prediction results of multiple decision trees through an integrated learning method to generate a preliminary emergency response level; thirdly, based on the preliminary emergency response level, the system uses dynamic Bayesian network technology for continuous monitoring, captures the time changes of the health status of the injured and the sick, updates the probability distribution in real time, and generates an optimized probability distribution; finally, based on the optimized probability distribution, the matching degree of the emergency response level is continuously optimized, the personalization and timeliness of the first aid measures are improved, and finally an optimized emergency response level is generated.

[0130] Assuming that the best emergency response level needs to be determined quickly and accurately in a dynamically changing emergency rescue environment, the system first performs feature selection and standardization based on the comprehensive health status index to generate a standardized feature vector; secondly, the system initializes the standardized feature vector, sets the extreme gradient boosting algorithm parameters, and generates an initialization model; thirdly, the system uses the extreme gradient boosting algorithm to build multiple decision tree models, and recursively splits the data set to learn the correlation between the key health indicators of the injured and the patient and the change trend, and generates initial prediction results; based on the initial prediction results, a new decision tree model is built in combination with the model residual, and the data set is continued to be recursively split to capture the part that is not fully explained by the previous decision tree, generate updated prediction results, and integrate the prediction results of multiple decision trees through an ensemble learning method to generate a preliminary emergency response level; finally, based on the preliminary emergency response level, the dynamic Bayesian network technology is used for continuous monitoring to capture the time changes in the health status of the injured and the patient, update the probability distribution in real time, and generate an optimized probability distribution; based on the optimized probability distribution, the matching degree of the emergency response level is continuously optimized to improve the personalization and timeliness of the emergency measures, and finally generate an optimized emergency response level to ensure the effectiveness and timeliness of the emergency measures.

[0131] This application takes into account that due to the problem of insufficient processing capabilities for complex medical data in the prior art, traditional feature extraction methods are difficult to capture long-distance dependencies and contextual information, resulting in limited expression and generalization capabilities of the model. Therefore, the invention embodiment proposes this optional solution to solve these problems and improve the accuracy and timeliness of the assessment of the health status of the wounded and sick. The reason for the overall design is to construct multiple decision tree models by introducing an extreme gradient boosting algorithm, recursively split the data set to learn the correlation between the key health indicators of the wounded and sick and the changing trends, generate initial prediction results, and further enhance the model's ability to understand sequence data.

[0132] Optionally, based on the initialization model, multiple decision tree models are constructed using an extreme gradient boosting algorithm, and the correlation between the key health indicators and the change trends of the injured and sick is learned by recursively segmenting the data set to generate initial prediction results, including:

[0133] Based on the initialization model, key features are screened through correlation analysis, feature scaling is performed through normalization processing, and the number of leaf nodes and nonlinear item weights are initialized to generate the output of the decision tree;

[0134] The output of the decision tree is calculated using the following formula:

[0135]

[0136] Where T(x) is the output of the decision tree; x is the input feature vector; J is the number of leaf nodes in the decision tree; j is the index of the leaf node, from 1 to J; w j is the weight of the jth leaf node; is the indicator function, when x belongs to the jth leaf node region R j 1 when θ is 0, otherwise 0; M is the number of nonlinear terms; m is the index of the nonlinear term, from 1 to M; θ m is the weight of the mth nonlinear term; N is the number of indicator functions; n is the index of the indicator function, from 1 to N; φ n is the weight of the nth indicator function; is the indicator function. When x belongs to the nth specific region P n 1 when it is, otherwise 0;

[0137] Based on the output of the decision tree, multiple decision tree results are weighted and summed, Sigmoid and ReLU activation functions are introduced to enhance the nonlinear expression ability of the model, and complex patterns in the data are further captured through additional indicator functions and nonlinear terms to generate comprehensive feature representation;

[0138] The comprehensive feature representation is calculated using the following formula:

[0139]

[0140] Among them, F(x) is the comprehensive feature representation; K is the number of decision trees; i is the index of the decision tree, from 1 to K; α i is the weight of the i-th decision tree; T i (x) is the output of the i-th decision tree; β is the weight of the nonlinear term; σ is the activation function Sigmoid, which is used to introduce nonlinearity; L is the number of nonlinear terms; k is the index of the nonlinear term, from 1 to L; γ k is the weight of the kth nonlinear term; M is the number of indicator functions; l is the index of the indicator function, from 1 to M; δ l is the weight of the lth indicator function; is the indicator function. When x belongs to the lth specific region S l 1 when , otherwise 0; p is the number of ReLU terms; p is the index of the ReLU term, from 1 to p; η p is the weight of the pth ReLU item; Q is the number of decision trees; q is the index of the decision tree, from 1 to Q; ζ q is the weight of the qth decision tree; T q (x) is the output of the qth decision tree; ReLU is the ReLU activation function, which is used to introduce nonlinearity;

[0141] Based on the comprehensive feature representation, the Softmax function is applied to generate category probability distribution, and the regression task prediction value is directly output. A Dropout layer is added after the fully connected layer to prevent overfitting. Batch normalization is used to accelerate the training process and improve generalization ability to generate initial prediction results.

[0142] This method aims to use the extreme gradient boosting algorithm combined with nonlinear terms to capture complex patterns in the data and generate high-dimensional feature representations; at the same time, the nonlinear expression ability of the model is enhanced by weighted summation and the introduction of Sigmoid and ReLU activation functions to ensure model stability and data distribution consistency; finally, the Softmax function is applied to generate category probability distribution, directly output the regression task prediction value, add a Dropout layer after the fully connected layer to prevent overfitting, use batch normalization to accelerate the training process, improve generalization ability, and generate initial prediction results.

[0143] In the output of the decision tree, the leaf node weight summation term This sub-item is based on whether the input feature vector x belongs to a specific region R j , calculate the corresponding leaf node weight w j , aims to determine the best split point by recursively splitting the data set, thereby optimizing the model performance; the nonlinear term summation term This sub-item introduces a nonlinear mechanism through the hyperbolic tangent function tanh to capture more complex patterns, aiming to enhance the expressiveness of the model;

[0144] The number of leaf nodes J is preset during initialization, and the specific value is determined according to the complexity of the data set; the weight of the leaf node w j It is automatically learned during the training process of the extreme gradient boosting algorithm; the indicator function I(x∈R j ) According to whether the input feature vector x belongs to a specific region R j It is determined by the recursive segmentation of the data set during the model training process. The number of nonlinear terms M is preset during initialization, and the specific value is determined according to the complexity of the data set and the model design. The weight of the nonlinear term θ m Obtained through optimization during the training process; the number of indicator functions N is preset during initialization, and the specific value is determined according to the complexity of the data set and the model design; the weight of the indicator function φ n Obtained through optimization during the training process; the indicator function I(x∈P n ) According to whether the input feature vector x belongs to a specific region P n Decision,automatically generated during model training;

[0145] In the comprehensive feature representation, the decision tree outputs a weighted sum This item performs weighted summation of the results of multiple decision trees, aiming to integrate the prediction results of different decision trees and improve the stability and accuracy of the model; Sigmoid activation function item This sub-item introduces a nonlinear mechanism through the Sigmoid activation function σ to further enhance the expressiveness of the model; the ReLU activation function term This sub-item introduces a nonlinear mechanism through the ReLU activation function, which is designed to handle negative values ​​and increase the flexibility of the model;

[0146] The number of decision trees K is preset during initialization, and the specific value is determined according to the complexity of the data set and the model design; the weight of each decision tree α i It is automatically learned during the training process of the extreme gradient boosting algorithm; the weight β of the Sigmoid activation function is obtained through optimization during the training process; the number of nonlinear terms L is preset during initialization, and the specific value is determined according to the complexity of the data set and the model design; the weight of the nonlinear term γ k Obtained through optimization during the training process; the number of indicator functions M is preset during initialization, and the specific value is determined according to the complexity of the data set and the model design; the weight of the indicator function δ l Obtained through optimization during the training process; the indicator function I(x∈S l ) According to whether the input feature vector x belongs to a specific region S l Determined, automatically generated during model training; the weight η of the ReLU activation function p Obtained through optimization during the training process; the weight ζ of each decision tree qIt is automatically learned during the training process of the extreme gradient boosting algorithm; the number of decision trees Q is preset during initialization, and the specific value is the same as K;

[0147] Suppose a large hospital needs to use an intelligent medical treatment kit to process the injury data of a group of injured and sick people; suppose x is the standardized initial feature vector [0.1, 0.2, …, 0.9]; suppose the number of decision trees K = 5, and the weight of each decision tree α i =[0.2,0.2,0.2,0.2,0.2], the weight of the nonlinear term β = 0.3, the weight of the Sigmoid activation function γ k =[0.1,0.1,0.1], indicating the weight of the function δ l =[0.1,0.1,0.1,0.1], the weight η of the ReLU activation function p =[0.1,0.1,0.1,0.1,0.1], the weight of each decision tree is ζ q =[0.2,0.2,0.2,0.2,0.2], the number of leaf nodes J = 5, the number of nonlinear terms M = 3, the number of indicator functions N = 4;

[0148]

[0149] Assuming the threshold is θ = [0.8, 0.8, 0.8, 0.8, 0.8], since each element of the final result F(x) = [0.92, 0.88, 0.94, 0.90, 0.93] of the comprehensive feature representation is greater than the corresponding assumed threshold, it shows that the current patient information performs well in terms of understanding and feature capture, and is suitable for further use or as the final information representation. Through the above steps, medical data can be effectively processed and high-quality feature representation can be extracted, thereby improving the accuracy and timeliness of the assessment of the health status of the patient.

[0150] 104. Perform rapid injury classification based on the optimized emergency response level, and generate treatment plan recommendations based on the rapid injury classification results.

[0151] In this step, rapid injury classification refers to using the optimized emergency response level as a guide, combining multiple sources of information such as the patient's physiological parameters, image data, and vital signs, and conducting in-depth analysis through intelligent algorithms to quickly classify the patient's health status into different injury levels. These categories may include but are not limited to minor injuries, moderate injuries, and severe injuries, and are used to determine the urgency and type of care required by the patient.

[0152] Treatment plan recommendations refer to a set of specific treatment measures and process plans for each injured or sick person that are automatically generated by the system based on the above injury classification results. These recommendations cover multiple aspects ranging from first aid measures, drug treatment to surgical preparation and referral guidance, aiming to provide medical personnel with immediate and effective action guidelines.

[0153] In step 104, the process of quickly classifying injuries based on the optimized emergency response level and generating treatment plan recommendations based on the results of the rapid injury classification is as follows: First, the system uses the determined optimized emergency response level as an input condition, combined with the multi-dimensional health data of the injured and sick collected in real time (such as physiological parameters and image data), and uses a pre-trained deep learning model to deeply process these data, identify the key health indicators of the injured and sick and their changing trends, and then complete the rapid classification of injuries. Next, the system matches the most appropriate treatment plan from its built-in treatment guide library based on the obtained injury classification results. The plan not only takes into account the current severity of the injury, but also comprehensively considers historical case data and expert experience to ensure the scientificity and practicality of the recommendations provided. Finally, the system continuously monitors the changes in the status of the injured and sick, dynamically adjusts the injury classification results and the corresponding treatment plan recommendations, and ensures the timeliness and accuracy of treatment decisions.

[0154] In one application scenario, when multiple patients are sent to the hospital due to a traffic accident, the intelligent medical treatment box group first receives the physiological parameters and image data of each patient, and integrates this information to generate a preliminary health status report. Then, the system uses a bidirectional encoder representation algorithm to extract high-dimensional feature representations, conducts in-depth processing and analysis, and uses cross-domain transfer learning technology to identify key health indicators and change trends to generate a comprehensive health status index. On this basis, the system uses an extreme gradient boosting algorithm to build multiple decision tree models to determine the optimal emergency response level, and continuously optimizes this level through dynamic Bayesian network technology. After entering step 104, the system automatically evaluates and classifies the severity of each patient's injury based on the optimized emergency response level. For example, a patient is classified as severely injured. Then, based on this classification result, the system extracts the most suitable treatment plan recommendations from the treatment guide library, such as immediate cardiopulmonary resuscitation (CPR), giving a specific dose of emergency medicine, preparing necessary surgical facilities, and arranging priority referral to a specialist hospital. Throughout the entire process, the system continuously updates the vital signs of the injured and sick, dynamically adjusts the treatment plan, and ensures that each injured and sick person can receive the most appropriate immediate treatment. This implementation method significantly improves the efficiency of emergency treatment, enhances the effectiveness of rescue, and reduces the pressure on medical staff.

[0155] Optionally, the rapid injury classification based on the optimized emergency response level in step 104, and the generation of treatment plan recommendations based on the rapid injury classification results, include: based on the optimized emergency response level, rapid injury classification and processing of the injured and sick, wherein the process of rapid injury classification and processing of the injured and sick includes taking the optimized emergency response level as an input condition, combining it with a pre-trained deep learning model to perform in-depth analysis and processing on the vital signs, physiological parameters and image data of the injured and sick, to obtain rapid injury classification results of the injured and sick, including minor injuries, moderate injuries and severe injuries; and generating treatment plan recommendations based on the rapid injury classification results.

[0156] In this step, the optimized emergency response level is the best response grade generated by the comprehensive health status index and dynamic Bayesian network technology, which aims to guide the selection and allocation of medical resources.

[0157] In order to achieve the above objectives, this embodiment uses a deep learning model to process and analyze the data collected from the wounded. First, the system receives the optimized emergency response level as input conditions, and then combines the vital signs, physiological parameters and image data of the wounded to perform in-depth analysis using a pre-trained deep learning model. After learning and optimizing a large number of cases, the model can identify the key health indicators of the wounded and their changing trends, and then quickly classify the injuries. Based on the classification results obtained, the system automatically generates corresponding treatment plan recommendations to ensure that each wounded and sick person can receive the most appropriate immediate treatment. In addition, the system also continuously monitors the changes in the status of the wounded and sick, dynamically adjusts the injury classification results and the corresponding treatment plan recommendations, and ensures the timeliness and accuracy of treatment decisions.

[0158] Specifically, first, based on the optimized emergency response level, the injuries of the injured are quickly classified and processed. Specifically, the optimized emergency response level is used as the input condition, and combined with the multi-source information such as the vital signs, physiological parameters and image data of the injured, the pre-trained deep learning model (especially the model for different emergency response levels) is used to perform in-depth analysis and processing on these data to obtain the injury category of the injured, such as minor injury, moderate injury, severe injury, etc.

[0159] Secondly, based on the rapid injury classification results, generate treatment plan recommendations. To this end, the system first extracts the most suitable treatment plan from the treatment guide library based on historical medical cases and expert experience based on the current injury classification. Then, the extracted data is further processed using the rule engine and machine learning algorithm to generate specific treatment plan recommendations in terms of first aid measures, drug treatment, surgical preparation and referral guidance;

[0160] Finally, the continuous monitoring system is used to dynamically monitor and process the changes in the status of the injured and sick, and obtain updated injury classification results and corresponding treatment plan recommendations. In this way, the timeliness and accuracy of treatment decisions are ensured, and can be directly used to guide the operations of on-site medical personnel or assist in remote medical consultation;

[0161] This process aims to improve the efficiency of medical rescue in emergencies or large-scale disasters, ensuring that every injured and sick person can receive the most appropriate immediate treatment. Through the above steps, not only the speed and quality of treatment are improved, but also the pressure on medical staff to make decisions in emergency situations is reduced.

[0162] Here is a specific example:

[0163] When multiple patients are sent to the hospital due to a serious traffic accident, the intelligent medical treatment box group first receives the real-time physiological parameters and image data of each patient, integrates this information to generate a preliminary health status report. Subsequently, the system uses a bidirectional encoder representation algorithm to extract high-dimensional feature representations, conducts in-depth processing and analysis, and uses cross-domain transfer learning technology to identify key health indicators and change trends to generate a comprehensive health status index. On this basis, the system uses an extreme gradient boosting algorithm to build multiple decision tree models to determine the optimal emergency response level, and continuously optimizes this level through dynamic Bayesian network technology. After entering step 104, the system quickly classifies the injuries of each patient based on the optimized emergency response level. For example, for a patient with a head injury and blurred consciousness, the system classifies him as a severe injury by analyzing his CT scan images and vital signs. Then, based on this classification result, the system matches the most suitable treatment plan recommendation from the built-in treatment guide library: immediately perform cardiopulmonary resuscitation (CPR), give a specific dose of emergency medicine, prepare necessary surgical facilities, and arrange priority referral to a specialist hospital with neurosurgery capabilities. During the whole process, the system continuously updates the vital signs of the injured and sick, dynamically adjusts the treatment plan, and ensures that every injured and sick person can receive the most appropriate immediate treatment. This implementation method significantly improves the efficiency of emergency treatment, enhances the effectiveness of rescue, and reduces the pressure on medical staff, reflecting the powerful advantages of the intelligent medical treatment system.

[0164] Through the above embodiments, it can be seen that the intelligent medical treatment box group can not only quickly and accurately assess the injuries of the wounded and sick in emergency situations, but also provide personalized treatment plan suggestions, greatly improving the quality and efficiency of medical rescue.

[0165] In summary, steps 101 to 104 cover the complete process from data preprocessing to building and optimizing the extreme gradient boosting decision tree model, aiming to provide an efficient and accurate solution for assessing the health status of the injured and sick, meeting the needs for fast and accurate decision support in intelligent medical treatment.

[0166] Figure 2 The present application provides a schematic diagram of a structure of a rapid injury classification system in an intelligent medical treatment box group, such as Figure 2 As shown, the device comprises:

[0167] The receiving module 21 is used to receive physiological parameters and image data of the patient, integrate the current vital signs, and generate a preliminary health status report;

[0168] An analysis module 22 is used to extract high-dimensional feature representation based on the preliminary health status report using a bidirectional encoder representation algorithm, perform in-depth processing and analysis, and use cross-domain transfer learning technology to identify key health indicators and change trends of the injured and sick, and generate a comprehensive health status index;

[0169] A construction module 23 is used to construct multiple decision tree models based on the comprehensive health status index using an extreme gradient boosting algorithm to determine the best emergency response level, and to continuously optimize the emergency response level matching using a dynamic Bayesian network technology to generate an optimized emergency response level;

[0170] The generating module 24 is used to quickly classify the injuries based on the optimized emergency response level, and generate treatment plan suggestions according to the results of the rapid classification of the injuries.

[0171] Figure 2 The rapid injury classification system in the intelligent medical treatment box set can perform Figure 1 The implementation principle and technical effect of the method for rapid classification of injuries in an intelligent medical treatment box set described in the embodiment shown are not repeated here. The specific way in which each module and unit performs operations in the rapid classification system of injuries in an intelligent medical treatment box set in the above embodiment has been described in detail in the embodiment of the method, and will not be elaborated here.

[0172] In one possible design, Figure 2 The rapid injury classification system in an intelligent medical treatment box set of the embodiment shown can be implemented as a computing device, such as Figure 3 As shown, the computing device may include a storage component 31 and a processing component 32;

[0173] 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 .

[0174] The processing component 32 is used to: receive physiological parameters and image data of the injured and sick, integrate the current vital signs, and generate a preliminary health status report; based on the preliminary health status report, use a bidirectional encoder characterization algorithm to extract high-dimensional feature representation, perform in-depth processing and analysis, and use cross-domain transfer learning technology to identify key health indicators and change trends of the injured and sick, and generate a comprehensive health status index; based on the comprehensive health status index, use an extreme gradient boosting algorithm to build multiple decision tree models to determine the optimal emergency response level, use dynamic Bayesian network technology to continuously optimize the emergency response level matching, and generate an optimized emergency response level; based on the optimized emergency response level, quickly classify the injury, and generate treatment plan recommendations based on the results of the rapid injury classification.

[0175] 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.

[0176] 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.

[0177] Of course, the computing device may also include other components, such as input / output interfaces, display components, communication components, etc.

[0178] 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.

[0179] The communication component is configured to facilitate, among other things, wired or wireless communications between the computing device and other devices.

[0180] 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.

[0181] The present application also provides a computer storage medium storing a computer program, wherein the computer program can achieve the above-mentioned Figure 1 A method for quickly classifying injuries in an intelligent medical treatment box set according to the embodiment shown.

[0182] 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.

[0183] 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.

[0184] 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.

[0185] 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 rapid injury classification in an intelligent medical treatment box group, characterized in that: include: Receive physiological parameters and image data of the injured and sick, integrate current vital signs, and generate a preliminary health status report; Based on the preliminary health status report, a bidirectional encoder representation algorithm is used to extract high-dimensional feature representation, perform in-depth processing and analysis, and adopt cross-domain transfer learning technology to identify key health indicators and change trends of the injured and sick, and generate a comprehensive health status index; Based on the comprehensive health status index, using the extreme gradient boosting algorithm, constructing multiple decision tree models to determine the best emergency response level, using dynamic Bayesian network technology to continuously optimize the emergency response level matching and generate an optimized emergency response level; The injuries are quickly classified based on the optimized emergency response level, and treatment plan recommendations are generated based on the results of the rapid classification of the injuries.

2. The method according to claim 1, characterized in that Based on the preliminary health status report, the bidirectional encoder representation algorithm is used to extract high-dimensional feature representation, perform in-depth processing and analysis, and adopt cross-domain transfer learning technology to identify the key health indicators and change trends of the injured and sick, and generate a comprehensive health status index, including: Based on the preliminary health status report, performing standardized preprocessing on the physiological parameters and image data to generate a standardized data set; Based on the standardized data set, a bidirectional encoder representation algorithm is used for encoding, a multi-layer converter structure is used for deep processing and analysis, a high-dimensional feature representation is extracted to highlight important features, and an enhanced feature representation is generated; Based on the enhanced feature representation, cross-domain transfer learning technology is used to centrally learn large-scale medical data knowledge to transfer to current patient data, improve the generalization ability of the model, and generate adaptive feature representation; Based on the adaptive characteristic representation, the key health indicators and change trends of the injured and sick are identified to generate a comprehensive health status index.

3. The method according to claim 2, characterized in that Based on the standardized data set, the bidirectional encoder representation algorithm is used for encoding, a multi-layer converter structure is used for deep processing and analysis, a high-dimensional feature representation is extracted to highlight important features, and an enhanced feature representation is generated, including: Based on the standardized data set, the physiological parameters and image data are preliminarily encoded to generate an initial feature vector; Based on the initial feature vector, a bidirectional encoder representation algorithm is used to capture contextual information and long-distance dependencies in the data through a self-attention mechanism to generate a preliminary feature representation; Based on the preliminary feature representation, a multi-layer converter structure is used to perform in-depth processing and analysis to enhance feature expression capabilities, highlight important features, and generate high-dimensional feature representations; Based on the high-dimensional feature representation, a self-attention mechanism is used to automatically identify and highlight key health indicators to generate enhanced feature representation.

4. The method according to claim 2, characterized in that: Based on the enhanced feature representation, the cross-domain transfer learning technology is used to concentrate on learning large-scale medical data knowledge to migrate to the current patient data, improve the generalization ability of the model, and generate adaptive feature representation, including: Based on the enhanced feature representation, a large-scale medical data set is used to extract and learn common health indicator patterns and change trends to generate a pre-trained model; Based on the pre-trained model, the cross-domain transfer learning technology is used to transfer the knowledge in the pre-trained model to the current patient data, and the model parameters are adjusted to improve the fitness, so as to generate a knowledge transfer model; Based on the knowledge transfer model, the optimal parameter configuration is selected through a cross-validation method to improve the generalization ability of the model and generate an optimized transfer model; Based on the optimized migration model, applicability feature extraction processing is performed to generate adaptive feature representation.

5. The method according to claim 1, characterized in that Based on the comprehensive health status index, the extreme gradient boosting algorithm is used to construct multiple decision tree models to determine the best emergency response level, and the dynamic Bayesian network technology is used to continuously optimize the emergency response level matching degree to generate the optimized emergency response level, including: Based on the comprehensive health status index, feature selection and standardization processing are performed to generate a standardized feature vector; Based on the standardized feature vectors, an extreme gradient boosting algorithm is used to construct multiple decision tree models, and the correlation between the key health indicators and the changing trends of the injured and sick is learned by recursively segmenting the data set to generate a preliminary emergency response level; Based on the initial emergency response level, dynamic Bayesian network technology is used to continuously monitor, capture the time changes of the health status of the injured and sick, update the probability distribution in real time, and generate an optimized probability distribution; Based on the optimized probability distribution, the matching degree of emergency response levels is continuously optimized, the personalization and timeliness of first aid measures are improved, and an optimized emergency response level is generated.

6. The method according to claim 5, characterized in that Based on the standardized feature vector, the extreme gradient boosting algorithm is used to construct multiple decision tree models, and the correlation between the key health indicators and the change trend of the injured and sick is learned by recursively segmenting the data set to generate a preliminary emergency response level, including: Based on the standardized feature vector, initialization processing is performed, extreme gradient boosting algorithm parameters are set, and an initialization model is generated; Based on the initialization model, multiple decision tree models are constructed using the extreme gradient boosting algorithm, and the correlation between the key health indicators and the change trends of the injured and sick is learned by recursively segmenting the data set to generate initial prediction results; Based on the initial prediction results, a new decision tree model is constructed in combination with the model residuals, and the data set is continued to be recursively segmented to capture the part that is not fully explained by the previous decision tree, so as to generate an updated prediction result; Based on the updated prediction results, multiple decision tree prediction results are integrated through an ensemble learning method to generate a preliminary emergency response level.

7. The method according to claim 1, characterized in that The rapid classification of injuries based on the optimized emergency response level and the generation of treatment plan suggestions according to the rapid classification results of injuries include: Based on the optimized emergency response level, the injuries of the injured and sick are quickly classified, wherein the process of quickly classifying the injuries of the injured and sick includes taking the optimized emergency response level as an input condition, combining it with a pre-trained deep learning model to perform in-depth analysis and processing on the vital signs, physiological parameters and image data of the injured and sick, and obtaining a rapid classification result of the injuries of the injured and sick, including minor injuries, moderate injuries and severe injuries; Generate treatment plan recommendations based on the rapid injury classification results.

8. A rapid injury classification system in an intelligent medical treatment box set, characterized in that: include: The receiving module is used to receive the physiological parameters and image data of the injured and sick, integrate the current vital signs, and generate a preliminary health status report; An analysis module is used to extract high-dimensional feature representation based on the preliminary health status report using a bidirectional encoder representation algorithm, perform in-depth processing and analysis, and use cross-domain transfer learning technology to identify key health indicators and change trends of the injured and sick, and generate a comprehensive health status index; A construction module is used to construct multiple decision tree models based on the comprehensive health status index using an extreme gradient boosting algorithm to determine the best emergency response level, and to continuously optimize the emergency response level matching using a dynamic Bayesian network technology to generate an optimized emergency response level; A generation module is used to quickly classify injuries based on the optimized emergency response level, and generate treatment plan recommendations based on the results of the rapid injury classification.

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 rapid injury classification in an intelligent medical treatment box group 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 quickly classifying injuries in an intelligent medical treatment box group as described in any one of claims 1 to 7 is implemented.

Images