A trauma deduction model construction method and system based on a multi-modal probability model

Abstract

The application provides a trauma deduction model construction method and system based on a multi-modal probability model, which comprises the following steps: constructing a current trauma deduction model; obtaining historical trauma injury data; determining a standard trauma injury change rule according to the historical trauma injury data; determining an evolution time by using an exponential distribution, generating vital signs by using a Poisson distribution, and generating transition vital signs by using a diffusion model based on the standard trauma injury change rule; respectively judging whether the evolution time, the vital signs and the transition vital signs conform to the standard trauma injury change rule; in the case that at least one of the evolution time, the vital signs and the transition vital signs does not conform to the standard trauma injury change rule, adjusting parameters of the current trauma deduction model until the evolution time, the vital signs and the transition vital signs all conform to the standard trauma injury change rule, and then outputting a corresponding current trauma deduction model as a target trauma deduction model.

Description

Technical Field

[0001] This invention relates to the field of trauma deduction technology, and in particular to a method and system for constructing a trauma deduction model based on a multimodal probability model. Background Technology

[0002] With the development of the times, the injury scenarios and injury conditions are becoming increasingly complex, and the occurrence and development of injuries and complications are also highly uncertain, which brings great challenges to rescue decisions. Therefore, the need to simulate the dynamic evolution of trauma and assist in rescue decisions has emerged. Researching dynamic evolution models of trauma can help rescuers better understand the dynamic changes of trauma and improve the level of practical rescue training.

[0003] However, research on the dynamic evolution of traumatic injuries faces three core challenges: discrete diagnostic and treatment data struggles to capture continuous pathological processes (e.g., seamless integration of pre-hospital emergency care, intraoperative monitoring, and ICU monitoring data); complex multi-system interaction mechanisms (e.g., the cascade reaction of hemorrhagic shock inducing coagulation dysfunction leading to multiple organ failure); and individual differences causing standardized predictions to fail (e.g., the rate of traumatic injury deterioration is significantly higher in elderly patients than in younger patients). This research background stems from the current passive nature of clinical decision-making—physicians rely heavily on static assessments (e.g., initial CT scan upon admission, ISS score), lacking the ability to predict the insidious deterioration of traumatic injuries (e.g., delayed intracranial hematoma, septic shock). Simultaneously, the injury environment is complex and variable, making continuous monitoring and recording of traumatic injuries difficult. Rescuers struggle to assess the evolution trend of traumatic injuries in emergency situations, potentially resulting in patients not receiving timely and effective treatment. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the present invention aims to provide a method and system for constructing a trauma deduction model based on a multimodal probabilistic model. By structurally integrating trauma pathology mechanisms and historical trauma injury data, it forms evolutionary rules that combine medical logical rigor with computational feasibility, providing core data-driven support for dynamic trauma injury evolution models. This model achieves dynamic trauma injury evolution based on rules: it generates key time nodes for changes in trauma injury status based on an exponential distribution, and simultaneously outputs continuous temporal fluctuation curves of vital signs based on a Poisson distribution and a diffusion model. This algorithm effectively connects the structured and rule-based modeling of trauma injury medical knowledge with the dynamic calculation of trauma injury evolution, overcoming the limitations of traditional static models in predicting complex battlefield timelines. Through explicit medical logic and quantified intervention effects, it provides methodological support for subsequent research on the timeliness assessment of emergency response plans and the optimal allocation of resources.

[0005] This invention provides a method for constructing a trauma inference model based on a multimodal probability model, the method comprising:

[0006] Step S1: Construct the current trauma simulation model;

[0007] Step S2: Obtain historical trauma data;

[0008] Step S3: Based on the historical trauma data, mine and determine the standard trauma change patterns covering the transition patterns of complication stages, the range of vital signs, and transitional characteristics;

[0009] Step S4: Based on the standard trauma condition change pattern, use the exponential distribution to determine the trauma condition evolution time, use the Poisson distribution to generate vital signs, and use the diffusion model to generate transitional vital signs of adjacent complication stages;

[0010] Step S5: Determine whether the traumatic injury evolution time, vital signs, and transitional vital signs meet the constraints of the standard traumatic injury change pattern.

[0011] Step S6: If at least one of the traumatic injury evolution time, vital signs, and transitional vital signs does not conform to the constraint range of the standard traumatic injury change law, adjust the parameters of the current traumatic injury simulation model, and repeat steps S4 to S6 until the traumatic injury evolution time, vital signs, and transitional vital signs all conform to the constraint range of the standard traumatic injury change law. Then, output the corresponding current traumatic injury simulation model as the target traumatic injury simulation model.

[0012] In one exemplary embodiment, the current trauma simulation model includes a current trauma evolution time submodule for determining the trauma evolution time, a current vital signs submodule for generating the vital signs, and a current transitional vital signs submodule for generating the transitional vital signs; the target trauma simulation model includes a target trauma evolution time submodule for determining the trauma evolution time, a target vital signs submodule for generating the vital signs, and a target transitional vital signs submodule for generating the transitional vital signs.

[0013] The parameters adjusted by the current trauma simulation model are the parameters of the sub-modules corresponding to the generation results that do not conform to the constraint range of the standard trauma injury change law;

[0014] The diffusion model is divided into a current diffusion model during training and a target diffusion model after training, based on the training phase, and is used to generate vital sign data for smooth transition between adjacent complication phases.

[0015] In one exemplary implementation, the training steps of the target trauma injury evolution time submodule specifically include:

[0016] The time range of any complication stage is determined based on the aforementioned standard trauma condition change pattern.

[0017] Based on the time range of any of the aforementioned complication stages, the corresponding complication stage index distribution parameters are calculated according to a preset clinical confidence probability.

[0018] Random numbers are drawn from a uniform distribution;

[0019] Candidate evolution times are calculated using a sampling formula for the evolution time of traumatic injuries;

[0020] Determine whether the candidate evolution time is within the time range of the corresponding complication stage;

[0021] If yes, the current traumatic injury evolution time submodule is output as the target traumatic injury evolution time submodule; if no, the random number is re-evaluated and calculated until the candidate evolution time falls within the corresponding complication stage change time range, then the corresponding current traumatic injury evolution time submodule is output as the target traumatic injury evolution time submodule.

[0022] In one exemplary embodiment, the time range for the change of the complication stage is determined by the standard trauma condition change pattern. And the candidate evolution time falls within the time range When the probability of occurrence is 95%, the training steps for the target trauma injury evolution time submodule specifically include:

[0023] Calculate the exponential distribution parameters of the complication stage. The calculation formula is as follows:

[0024] ;

[0025] From uniform distribution Random numbers drawn ;

[0026] right Scaling The calculation formula is as follows:

[0027] ;

[0028] exist arrive In between, at this time The evolution time of the candidate is calculated:

[0029] ;

[0030] Examine the candidate evolution time Is it within the time range? If the current traumatic injury evolution time submodule is within the specified range, it will be output as the target traumatic injury evolution time submodule; if it is outside the specified range, the random number will be re-generated and recalculated until the candidate evolution time is within the specified range. If the current traumatic injury evolution time submodule is used as the target traumatic injury evolution time submodule, then the current traumatic injury evolution time submodule will be output.

[0031] In one exemplary implementation, the training steps of the target vital signs submodule specifically include:

[0032] Based on the aforementioned standard trauma condition change patterns, determine the range of vital sign constraints for any stage of complication.

[0033] Based on the vital sign constraints and preset information parameters at the complication stage, calculate the mean and standard deviation of vital signs;

[0034] Candidate vital signs were calculated using the Poisson distribution sampling formula;

[0035] Determine whether the candidate vital signs are within the vital sign constraints of the complication stage;

[0036] If yes, the current vital signs submodule is output as the target vital signs submodule; if not, the standard normal random numbers in the Poisson distribution sampling formula are re-extracted and recalculated until the candidate vital signs are within the vital signs constraint range of the complication stage, then the corresponding current vital signs submodule is output as the target vital signs submodule.

[0037] In one exemplary embodiment, when the standard trauma condition change pattern determines that the vital signs at the stage of the complication are constrained within the range of [c, d], the mean vital signs... The calculation formula is:

[0038] ;

[0039] The standard deviation of vital signs The calculation formula is:

[0040] ;

[0041] in, The preset confidence parameter corresponds to the standard normal distribution quantile of the clinical confidence level;

[0042] The calculation formula for the Poisson distribution sampling formula is as follows:

[0043] ;

[0044] in, It is the standard normal random number.

[0045] In one exemplary implementation, the training process of the target diffusion model is as follows:

[0046] The current diffusion model is constructed using a multilayer perceptron.

[0047] Based on the standard trauma injury change pattern, the initial point of the previous stage and the target point of the next stage of vital signs are extracted for adjacent complication stages.

[0048] The current diffusion model generates an intermediate transition distribution from the initial point to the target point;

[0049] The intermediate transition distribution is compared with the actual transitional vital sign data in the historical trauma injury data to obtain the comparison results;

[0050] If the comparison result meets the preset conditions, the current diffusion model is output as the target diffusion model, and the current transitional vital signs submodule is output as the target transitional vital signs submodule. If the comparison result does not meet the preset conditions, the parameters of the current diffusion model are adjusted and the intermediate transitional distribution is regenerated for comparison until the comparison result meets the preset conditions. Then, the corresponding current diffusion model is output as the target diffusion model, and the corresponding current transitional vital signs submodule is output as the target transitional vital signs submodule.

[0051] In one exemplary implementation, the current diffusion model is a neural network model, and the parameters of the current diffusion model are adjusted, including the number of network layers and the learning rate.

[0052] In one exemplary implementation, the process of generating an intermediate transition distribution from the initial point to the target point using the current diffusion model specifically involves the current diffusion model transferring data from the initial distribution... Transform into target distribution The process is divided into two stages:

[0053] The forward diffusion process, which is a Markov chain, progressively diffuses data from the initial distribution. Convert to noise distribution Each step adds a preset amount of noise to gradually blur the features of the data. The calculation formula is as follows:

[0054] ;

[0055] in:

[0056] It is a time step Data;

[0057] It is the diffusion coefficient, which controls the amount of noise added at each step;

[0058] It is noise that follows a standard normal distribution;

[0059] The reverse diffusion process, which is the inverse of forward diffusion, aims to reduce noise distribution. Gradually restore to the initial distribution Each step involves learning to remove noise and gradually restore the characteristics of the data. The calculation formula is as follows:

[0060] ;

[0061] in:

[0062] It is noise predicted through a neural network;

[0063] The training objective of the current diffusion model is to learn the velocity field. This velocity field, used to guide data migration from an initial distribution to a target distribution, satisfies a specific differential equation:

[0064] ;

[0065] In discrete time, it can be represented as a difference equation:

[0066] ;

[0067] in, It is the time step.

[0068] In one exemplary implementation, the standard trauma condition change pattern is obtained by statistically analyzing the historical trauma condition data, including the duration distribution of each complication stage, the range of vital sign fluctuations, and the threshold of transition characteristics between adjacent stages, all of which conform to the medical logic of clinical trauma condition evolution.

[0069] This invention also provides a trauma inference model construction system based on a multimodal probability model, implemented using any of the aforementioned trauma inference model construction methods based on a multimodal probability model, the system comprising:

[0070] The model building module is used to build the current trauma simulation model;

[0071] The data acquisition module is used to acquire historical trauma injury data;

[0072] The change pattern determination module is used to mine and determine standard trauma change patterns based on the historical trauma injury data, covering the transition patterns of complication stages, the range of vital signs, and transitional characteristics.

[0073] The data generation module is used to determine the evolution time of traumatic injuries based on the standard traumatic injury change pattern, generate vital signs using the exponential distribution, generate transitional vital signs of adjacent complication stages using the Poisson distribution, and generate transitional vital signs of adjacent complication stages using the diffusion model.

[0074] The judgment module is used to determine whether the traumatic injury evolution time, the vital signs, and the transitional vital signs meet the constraints of the standard traumatic injury change pattern.

[0075] The repetitive training module is used to adjust the parameters of the current trauma simulation model when at least one of the trauma evolution time, vital signs, and transitional vital signs does not conform to the constraints of the standard trauma change law. The data generation module is then executed sequentially to the repetitive training module until the trauma evolution time, vital signs, and transitional vital signs all conform to the constraints of the standard trauma change law. Then, the corresponding current trauma simulation model is output as the target trauma simulation model.

[0076] Compared with the prior art, the present invention has the following beneficial effects:

[0077] This invention innovatively proposes a multidimensional modeling framework that integrates probabilistic models and stochastic processes. First, it utilizes the exponential distribution to characterize the temporal randomness of transitions between treatment stages. Its memoryless nature accurately reflects the clinical principle that the risk of a patient entering the next stage depends solely on their current state, providing a temporal probabilistic basis for resource allocation. Within specific treatment stages, the Poisson distribution is used to simulate discrete fluctuations in vital signs, such as heart rate and respiratory rate, dynamically reflecting the stochastic characteristics of physiological instability within each stage. Crucially, a diffusion model is used to construct a natural transition mechanism between stages, describing the continuous trajectory of vital signs using stochastic differential equations. This approach adheres to the deterministic trend of overall trauma recovery while incorporating stochastic factors such as individual differences, ensuring that no physiological indicator mutations that violate medical common sense occur at the critical points of treatment stage transitions.

[0078] The above structured modeling approach achieves a triple breakthrough at the methodological level: it grasps the stochastic time nodes of treatment phase transitions at the macro level, characterizes discrete life events within a phase at the micro level, and then achieves seamless connection between discrete and continuous states through a diffusion process, fully presenting the clinical time-series chain of "phase stay - intra-phase fluctuation - phase transition". The model combines the interpretability of parametric methods with the flexibility of nonparametric methods. The key parameters of the exponential and Poisson distributions directly correspond to the clinical risk rate and event incidence rate, while the diffusion model fits complex nonlinear physiological changes through a continuous stochastic process. When embedded with a Bayesian inference framework, the system can integrate new observational data in real time and dynamically update the probability distribution of traumatic injury status, providing quantitative support for the uncertainty of clinical decision-making over time. This architecture also supports personalized intervention strategy simulation, allowing for the simulation of traumatic injury development under different treatment pathways by adjusting parameters such as medication regimens or monitoring levels.

[0079] This multi-paradigm fusion modeling approach not only overcomes the inherent shortcomings of traditional methods in simulating continuous dynamic processes and handling the smoothness of transitions between stages, but also lays a theoretical foundation for building a high-fidelity trauma prediction system by strictly adhering to clinical treatment pathways and the essence of physiological changes. Its synergistic enhancement of dynamism, stochasticity, and interpretability significantly improves the precision and individualization of trauma treatment, and has important practical value for optimizing key aspects such as emergency triage and intensive care resource allocation. This marks a paradigm shift in trauma evolution modeling from discrete-stage analysis to continuous dynamic extrapolation. Attached Figure Description

[0080] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0081] Figure 1 This is a schematic diagram of the design of a pre-treatment trauma simulation model provided in an embodiment of the present invention;

[0082] Figure 2 This is a schematic diagram of the design of a post-treatment trauma simulation model provided in an embodiment of the present invention;

[0083] Figure 3 A technical framework diagram for pre-treatment trauma assessment provided by an embodiment of the present invention;

[0084] Figure 4 This is a technical framework diagram for post-treatment trauma assessment provided by an embodiment of the present invention. Detailed Implementation

[0085] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0086] The invention process will be introduced first.

[0087] Common methods for predicting traumatic injury progression include rule-based state transition models, Markov models, multi-state models, Bayesian networks, mixed-effects models, and dynamic system models. Research on these methods both domestically and internationally exhibits a differentiated development trend. Chinese scholars have made significant breakthroughs in the field of trauma. The state transition model developed by the Naval Medical University, by quantifying state transition probabilities, verified the crucial impact of the timeliness of tiered treatment on survival rates; however, the model's reliance on expert experience data needs to be overcome. In contrast, the Zhejiang University team used generalized Poisson regression to reveal the correlation mechanism between climate and fracture risk, discovering that the lag effect of perceived temperature is particularly significant in specific populations, providing a new perspective for predicting environment-related traumatic injury progression. Meanwhile, the trauma knowledge graph constructed by CSSC Neptune Corporation, integrating random field algorithms and temporal embedding technology, while achieving multi-source data integration, failed to solve the problem of smooth transitions between stages.

[0088] International research focuses more on technological integration and mechanistic innovation. The electric field-activated bandage developed by North Carolina State University enhances neutrophil migration efficiency, increasing the healing speed of chronic wounds by 30%, while a diabetes wound model based on aging parameters opens new avenues for personalized treatment. European researchers are advancing in both materials engineering and statistical modeling; Portugal has developed 3D-printed smart dressings that enable precise regulation of inflammatory factors. A team at KU Leuven in Belgium has developed an intracranial pressure crisis early warning system based on Long Short-Term Memory (LSTM) networks. By analyzing high-resolution physiological parameters, it predicts the risk of ICP > 22 mmHg within the next two hours, achieving a paradigm shift from static prediction to dynamic risk intervention.

[0089] Current mainstream methods still face common challenges: each has its own strengths and weaknesses in modeling complex physiological changes, injury progression, or complication risk assessment. Rule-based models are simple and intuitive, but cannot handle randomness; Markov models can model the probability of transitions between states and are suitable for multi-stage trauma progression, but their expressive power is limited by the assumption of no memory; mixed-effects models can capture individual differences but are computationally complex; Bayesian networks are suitable for small datasets but require a lot of prior knowledge for modeling. At the same time, although the above methods have made breakthroughs in specific fields, they still lack a systematic integration of the entire chain of "stochastic transitions between stages - discrete fluctuations within stages - continuous transitions between stages," making it difficult to balance clinical interpretability with the dynamic fitting accuracy of complex physiological processes.

[0090] To address the shortcomings of existing methods in simulating the continuous evolution of traumatic injuries, this study innovatively proposes a multidimensional modeling framework that integrates probabilistic models and stochastic processes. First, the exponential distribution is used to characterize the temporal randomness of transitions between treatment stages. Its memoryless nature accurately reflects the clinical principle that the risk of a patient entering the next stage depends solely on their current state, providing a temporal probabilistic basis for resource allocation. Within specific treatment stages, the Poisson distribution is used to simulate discrete fluctuations in vital signs, such as heart rate and respiratory rate, dynamically reflecting the stochastic characteristics of physiological instability within each stage. Crucially, a diffusion model is used to construct a natural transition mechanism between stages, describing the continuous trajectory of vital signs using stochastic differential equations. This approach adheres to the deterministic trend of overall traumatic injury recovery while incorporating stochastic factors such as individual differences, ensuring that no physiological indicator mutations that violate medical common sense occur at the critical points of treatment stage transitions.

[0091] The above structured modeling approach achieves a triple breakthrough at the methodological level: it grasps the stochastic time nodes of treatment phase transitions at the macro level, characterizes discrete life events within a phase at the micro level, and then achieves seamless connection between discrete and continuous states through a diffusion process, fully presenting the clinical time-series chain of "phase stay - intra-phase fluctuation - phase transition". The model combines the interpretability of parametric methods with the flexibility of nonparametric methods. The key parameters of the exponential and Poisson distributions directly correspond to the clinical risk rate and event incidence rate, while the diffusion model fits complex nonlinear physiological changes through a continuous stochastic process. When embedded with a Bayesian inference framework, the system can integrate new observational data in real time and dynamically update the probability distribution of traumatic injury status, providing quantitative support for the uncertainty of clinical decision-making over time. This architecture also supports personalized intervention strategy simulation, allowing for the simulation of traumatic injury development under different treatment pathways by adjusting parameters such as medication regimens or monitoring levels.

[0092] This multi-paradigm fusion modeling approach not only overcomes the inherent shortcomings of traditional methods in simulating continuous dynamic processes and handling the smoothness of transitions between stages, but also lays a theoretical foundation for building a high-fidelity trauma prediction system by strictly adhering to clinical treatment pathways and the essence of physiological changes. Its synergistic enhancement of dynamism, stochasticity, and interpretability significantly improves the precision and individualization of trauma treatment, and has important practical value for optimizing key aspects such as emergency triage and intensive care resource allocation. This marks a paradigm shift in trauma evolution modeling from discrete-stage analysis to continuous dynamic extrapolation.

[0093] First Embodiment

[0094] Please see Figures 1-4 This invention provides a method for constructing a trauma inference model based on a multimodal probability model, the method comprising:

[0095] Step S1: Construct the current trauma simulation model;

[0096] Step S2: Obtain historical trauma data;

[0097] Step S3: Based on historical trauma data, mine and determine the standard trauma change patterns that cover the transition patterns of complication stages, the range of vital signs, and transitional characteristics;

[0098] Step S4: Based on the standard trauma condition change pattern, use the exponential distribution to determine the trauma condition evolution time, use the Poisson distribution to generate vital signs, and use the diffusion model to generate transitional vital signs of adjacent complication stages;

[0099] Step S5: Determine whether the time of traumatic injury evolution, vital signs, and transitional vital signs meet the constraints of the standard traumatic injury change pattern.

[0100] Step S6: If the traumatic injury evolution time, vital signs, and transitional vital signs all conform to the constraints of the standard traumatic injury change law, output the current traumatic injury simulation model as the target traumatic injury simulation model; if at least one of the traumatic injury evolution time, vital signs, and transitional vital signs does not conform to the constraints of the standard traumatic injury change law, adjust the parameters of the current traumatic injury simulation model, and repeat steps S4 to S6 until the traumatic injury evolution time, vital signs, and transitional vital signs all conform to the constraints of the standard traumatic injury change law, then output the corresponding current traumatic injury simulation model as the target traumatic injury simulation model.

[0101] In an optional embodiment, the current trauma simulation model includes a current trauma evolution time submodule for determining the trauma evolution time, a current vital signs submodule for generating vital signs, and a current transitional vital signs submodule for generating transitional vital signs; the target trauma simulation model includes a target trauma evolution time submodule for determining the trauma evolution time, a target vital signs submodule for generating vital signs, and a target transitional vital signs submodule for generating transitional vital signs.

[0102] The parameters adjusted in the current trauma simulation model are those of the sub-modules corresponding to the generated results that do not conform to the standard trauma injury change pattern.

[0103] The diffusion model is divided into the current diffusion model during training and the target diffusion model after training, based on the training phase, and is used to generate vital sign data for smooth transition between adjacent complication phases.

[0104] In an optional embodiment, the training steps of the above-mentioned target trauma injury evolution time submodule specifically include:

[0105] Determine the time range of any complication stage based on the standard trauma injury progression pattern;

[0106] Based on the time range of any complication stage change, the corresponding complication stage index distribution parameters are calculated according to the preset clinical confidence probability.

[0107] Random numbers are drawn from a uniform distribution;

[0108] Candidate evolution times are calculated using a sampling formula for the evolution time of traumatic injuries;

[0109] Determine whether the candidate evolution time falls within the time range of the corresponding complication stage;

[0110] If yes, the current traumatic injury evolution time submodule will be output as the target traumatic injury evolution time submodule; if not, random numbers will be re-evaluated and calculated until the candidate evolution time falls within the corresponding complication stage change time range, and then the corresponding current traumatic injury evolution time submodule will be output as the target traumatic injury evolution time submodule.

[0111] In an optional embodiment, the time range for determining a complication stage based on the standard trauma condition change pattern is as follows: And the candidate evolution time falls within the time range When the probability of occurrence is 95%, the training steps for the target trauma injury evolution time submodule specifically include:

[0112] Calculate the complication stage index distribution parameters The calculation formula is as follows:

[0113] ;

[0114] From uniform distribution Random numbers drawn ;

[0115] right Scaling The calculation formula is as follows:

[0116] ;

[0117] exist arrive In between, at this time The candidate evolution time was calculated:

[0118] ;

[0119] Examining candidate evolution time Is it within the time range? If the current traumatic injury evolution time submodule is within the specified range, it will be output as the target traumatic injury evolution time submodule; if it is outside the specified range, random numbers will be re-generated and recalculated until the candidate evolution time is within the specified range. Within this context, the current traumatic injury evolution time submodule will be output as the target traumatic injury evolution time submodule.

[0120] In an optional embodiment, the training steps of the above-mentioned target vital signs submodule specifically include:

[0121] The range of vital signs constraints for any stage of complication is determined based on the standard pattern of changes in traumatic injury.

[0122] Based on the range of vital signs constraints and preset information parameters at the complication stage, the mean and standard deviation of vital signs were calculated.

[0123] Candidate vital signs were calculated using the Poisson distribution sampling formula;

[0124] Determine whether the candidate vital signs are within the vital sign constraints during the complication stage;

[0125] If yes, the current vital signs submodule is output as the target vital signs submodule; otherwise, the standard normal random numbers in the Poisson distribution sampling formula are re-extracted and recalculated until the candidate vital signs are within the vital signs constraint range of the complication stage, then the corresponding current vital signs submodule is output as the target vital signs submodule.

[0126] In an optional embodiment, when the standard trauma injury condition change pattern determines the vital signs within the complication stage with a constraint range of [c, d], the mean vital signs are... The calculation formula is:

[0127] ;

[0128] Standard deviation of vital signs The calculation formula is:

[0129] ;

[0130] in, These are preset confidence parameters, corresponding to the standard normal distribution quantiles of the clinical confidence level;

[0131] The formula for calculating the Poisson distribution sampling formula is as follows:

[0132] ;

[0133] in, It is a standard normally distributed random number.

[0134] In an optional embodiment, the training process of the above-described target diffusion model is as follows:

[0135] The current diffusion model is constructed using a multilayer perceptron.

[0136] Based on the standard trauma injury change pattern, the initial point of the previous stage and the target point of the next stage of vital signs are extracted for adjacent complication stages.

[0137] Generate an intermediate transition distribution from the initial point to the target point using the current diffusion model;

[0138] The intermediate transitional distribution is compared with the actual transitional vital sign data in historical trauma injury data to obtain the comparison results;

[0139] If the comparison results meet the preset conditions, the current diffusion model is output as the target diffusion model, and the current transitional vital signs submodule is output as the target transitional vital signs submodule. If the comparison results do not meet the preset conditions, the parameters of the current diffusion model are adjusted and an intermediate transitional distribution is regenerated for comparison until the comparison results meet the preset conditions. Then, the corresponding current diffusion model is output as the target diffusion model, and the corresponding current transitional vital signs submodule is output as the target transitional vital signs submodule.

[0140] In an optional embodiment, the current diffusion model is a neural network model, and the parameters of the current diffusion model are adjusted, including the number of network layers, learning rate, and other parameters.

[0141] In an optional embodiment, the process of generating an intermediate transition distribution from the initial point to the target point using the current diffusion model specifically involves the current diffusion model transferring data from the initial distribution... Transform into target distribution The process is divided into two stages:

[0142] The forward diffusion process, which follows a Markov chain, progressively diffuses data from its initial distribution. Convert to noise distribution Each step adds a preset amount of noise to gradually blur the features of the data. The calculation formula is as follows:

[0143] ;

[0144] in:

[0145] It is a time step Data;

[0146] It is the diffusion coefficient, which controls the amount of noise added at each step;

[0147] It is noise that follows a standard normal distribution;

[0148] The reverse diffusion process is the inverse of the forward diffusion process, and its goal is to diffuse from the noise distribution... Gradually restore to the initial distribution Each step involves learning to remove noise and gradually restore the characteristics of the data. The calculation formula is as follows:

[0149] ;

[0150] in:

[0151] It is noise predicted through a neural network;

[0152] The current training objective of the diffusion model is to learn the velocity field. This velocity field, used to guide data migration from an initial distribution to a target distribution, satisfies a specific differential equation:

[0153] ;

[0154] In discrete time, it can be represented as a difference equation:

[0155] ;

[0156] in, It is the time step.

[0157] The current training objective of the diffusion model is to learn the velocity field. This allows the vital signs trajectory obtained through ODE integration to be tracked. exist The time follows an initial normal distribution (mean) Standard deviation ),exist The time follows a target normal distribution (mean) Standard deviation ), and at any time in between Distribution It can also be accurately generated. The training process is optimized through neural networks to ensure... To capture the dynamic changes between two distributions.

[0158] In an optional embodiment, the above method may include a process of preprocessing historical trauma injury data, specifically including:

[0159] We collected clinical history data on trauma and injury from multiple tertiary hospitals' trauma departments and emergency centers, including but not limited to basic patient information (age, gender), type of trauma (blunt injury, sharp injury, etc.), injury information, type and stage of complications, time-series data of vital signs at each stage (sampling interval of 1 hour), and records of treatment measures.

[0160] The collected clinical historical trauma data were sequentially processed by denoising, missing value imputation, standardization, and data filtering to obtain historical trauma data.

[0161] Among them, noise reduction processing: remove data that is obviously inconsistent with clinical common sense (such as heart rate >200 beats / minute or <30 beats / minute).

[0162] Missing value imputation: Time-series interpolation is used to imput missing vital sign data to ensure data continuity;

[0163] Standardization processing: Normalize vital sign data (e.g., convert systolic blood pressure to [0,1] interval values) to facilitate model training;

[0164] Data filtering: retain case data that fully records the entire process of complication development (including the initial stage, intermediate stage, and final stage) to ultimately form an effective historical trauma injury dataset.

[0165] In an optional embodiment, the above-mentioned standard trauma condition change pattern is obtained by statistical analysis of historical trauma condition data, including the duration distribution of each complication stage, the range of vital sign fluctuations, and the threshold of transition characteristics between adjacent stages, all of which conform to the medical logic of clinical trauma condition evolution.

[0166] Specifically, based on preprocessed historical trauma data, combined with statistical analysis and clinical expert verification, standard patterns of trauma condition changes are identified, including:

[0167] Complication stage transition pattern: Using the Kaplan-Meier survival analysis method, the duration distribution of each complication (such as infection, shock) from one stage to the next was statistically analyzed to determine the time range of each stage (e.g., the time range from early infection to middle infection is 3-7 days).

[0168] Vital signs range: The fluctuation range of vital signs at each stage of complications was analyzed by box plot. After removing outliers, a 95% confidence interval was determined as the vital signs constraint range for that stage (e.g., the heart rate constraint range during the compensated shock period is 100-120 beats / minute).

[0169] Transition feature threshold: Temporal clustering algorithms (such as K-means) are used to mine the changes in vital signs during the transition between adjacent complication stages, and the feature threshold of the transition stage is determined (such as the blood pressure drop rate threshold of 5 mmHg / h from the compensated shock stage to the decompensated stage).

[0170] Clinical validation: Invite more than three chief physicians of trauma departments to review the patterns discovered, correct parameters that do not conform to medical logic (such as adjusting the range of vital signs at a certain complication stage to match actual clinical diagnosis and treatment experience), and finally form a standardized pattern of trauma injury changes.

[0171] It should be noted that, as Figures 1-4 As shown, the trauma simulation model of the present invention is applicable to the simulation of trauma conditions in two stages: before and after treatment. The difference is that the complication stage after treatment is the reverse stage of the complication stage before treatment.

[0172] In a real-world application scenario, the evolution of traumatic injuries before and after treatment is as follows:

[0173] Initial trauma data indicated drowning as the primary injury type, with asphyxiation as the complication. Since asphyxiation occurs immediately within 0-1 minute, with a random occurrence time of 0 minutes, basic life support techniques were initiated during respiratory arrest, including hands-only cardiopulmonary resuscitation (CPR).

[0174] Process 1: Initial "asphyxiation" state generates vital signs:

[0175] According to the simulation rules, the vital signs in the initial "asphyxiation" state are in the following range: "Heart rate: 60-100 beats / min, respiration: 16-20 breaths / min, systolic blood pressure: 90-140 mmHg, diastolic blood pressure: 60-90 mmHg, blood pressure: systolic / diastolic, consciousness: conscious."

[0176] Procedure 1.1: Determining vital signs based on Poisson distribution:

[0177] Heart rate range [60, 100], mean Standard deviation The random number Z = 0.324 is a standard normal random number. The heart rate is calculated using the sampling formula. beats / min; within the heart rate range, meeting the requirements;

[0178] The respiratory range is [16, 20], with a mean of Standard deviation The random number Z = 0.472 is a standard normal random number. The respiration rate is calculated using the sampling formula. Breathing rate; within the breathing range, meets the requirements;

[0179] The systolic blood pressure range is [90, 140], with a mean of [missing value]. Standard deviation Given a randomized standard normal number Z = -0.1, the systolic blood pressure is calculated using the sampling formula. mmHg; within the systolic blood pressure range, meets the requirements;

[0180] Diastolic blood pressure range [60, 90], mean Standard deviation Given a randomized standard normal number Z = -0.156, the diastolic blood pressure is calculated using the sampling formula. mmHg; within the diastolic blood pressure range, meeting the requirements;

[0181] Then 0 min, complications: asphyxia, heart rate: 83 beats / min, blood pressure: 114 / 74 mmHg, respiration: 18 breaths / min, consciousness: conscious.

[0182] Process 2, from the initial "asphyxiation" state to the "immediate mechanical injury period":

[0183] According to the extrapolation rules, within 0-1 minute, the complication will evolve into the "immediate mechanical injury period," and the vital signs will be in the following range: "Heart rate: 120-140 beats / min, respiration: 30-40 breaths / min, systolic blood pressure: 140-160 mmHg, diastolic blood pressure: 90-110 mmHg, blood pressure is expressed as systolic / diastolic, consciousness: mild impairment of consciousness."

[0184] Step 2.1, determining the time of traumatic injury evolution based on exponential distribution:

[0185] The time range is [0,1], that is , ;

[0186] according to Calculate the exponential distribution parameters ;

[0187] From uniform distribution Random numbers drawn , , ;

[0188] Scaling Calculate the time of change min;

[0189] If the condition is met within the time range [0,1], then the "instantaneous mechanical damage period" will begin at 0.3 min.

[0190] Procedure 2.2: Determining vital signs based on Poisson distribution:

[0191] Refer to procedure 1.1, mean heart rate Standard deviation The random number Z = -0.412, obtained from a standard normal distribution, is used to calculate a heart rate of 128 beats / min using the sampling formula; this is within the heart rate range and meets the requirements.

[0192] Reference procedure 1.1, mean respiratory rate Standard deviation The random number Z = 0.315, calculated using the sampling formula, indicates a respiratory rate of 36 breaths / min; this is within the respiratory range and meets the requirements.

[0193] Refer to procedure 1.1, mean systolic blood pressure Standard deviation The random number Z = 0.234 is obtained from a standard normal distribution. The systolic blood pressure is calculated to be 151 mmHg using the sampling formula. This is within the systolic blood pressure range and meets the requirements.

[0194] Refer to procedure 1.1, mean diastolic blood pressure Standard deviation The random number Z = -0.189 was obtained from a standard normal distribution. The diastolic blood pressure was calculated to be 99 mmHg using the sampling formula. This is within the diastolic blood pressure range and meets the requirements.

[0195] Then 0.3min, complications: asphyxiation - immediate mechanical injury period, heart rate: 128 beats / min, blood pressure: 151 / 99 mmHg, respiration: 36 breaths / min, consciousness: mild consciousness impairment.

[0196] Process 3: From the "immediate mechanical injury phase" to the "acute obstruction phase":

[0197] According to the extrapolation rules, within 0.5-3 minutes, the complication will evolve into the "acute obstruction period," and the vital signs will be in the following range: "Heart rate: 140-180 beats / min, Respiration: 40-10 breaths / min, Systolic blood pressure: 70-50 mmHg, Diastolic blood pressure: 40-20 mmHg, Blood pressure is expressed as systolic / diastolic, Consciousness: Severe impairment of consciousness."

[0198] Step 3.1: Determining the time of traumatic injury evolution based on exponential distribution:

[0199] The time range is [0.5, 3], that is... , ;

[0200] according to Calculate the exponential distribution parameters ;

[0201] From uniform distribution Random numbers drawn , , Scaling Calculate the time of change min;

[0202] If the condition is met within the time range [0.5, 3], then the "acute obstruction period" will begin at 0.3 + 1.46 = 1.76 min.

[0203] Process 3.2, generating transitional vital signs based on a diffusion model:

[0204] Within the time frame of 0.3–1.76 min, vital signs at time 1 min are generated based on a diffusion model. The time step is as follows: =1-0.3=0.7min.

[0205] Taking heart rate as an example, the average value of the initial "immediate mechanical injury period" The target is the mean value of the "acute obstruction period". , According to the difference equation Initial standard deviation The random number Z = -0.156 is a standard normal random number. The heart rate is calculated to be 144 beats / min using the sampling formula.

[0206] Taking respiration as an example, the average initial "immediate mechanical injury period" The target is the mean value of the "acute obstruction period". , According to the difference equation Initial standard deviation The random number Z = 0.234 is a standard normal random number. The respiratory rate is calculated to be 31 breaths / min using the sampling formula.

[0207] Taking systolic blood pressure as an example, the average initial "immediate mechanical damage period" The target is the mean value of the "acute obstruction period". , According to the difference equation Initial standard deviation The random number Z = 0.127 is a standard normal random number. The systolic blood pressure is calculated to be 107 mmHg using the sampling formula.

[0208] Taking diastolic blood pressure as an example, the average initial "immediate mechanical injury period" The target is the mean value of the "acute obstruction period". , According to the difference equation Initial standard deviation The random number Z = -0.412 is a standard normal random number. The diastolic blood pressure is calculated to be 62 mmHg using the sampling formula.

[0209] Then, within 1 minute, complications: asphyxiation - immediate mechanical injury period, heart rate: 144 beats / min, blood pressure: 107 / 62 mmHg, respiration: 31 breaths / min, consciousness: mild impairment of consciousness.

[0210] Step 3.3: Determining vital signs based on Poisson distribution:

[0211] Refer to procedure 1.1, mean heart rate Standard deviation The random number Z = 0.278 is obtained from a standard normal distribution. The heart rate calculated using the sampling formula is 163 beats / min; this is within the heart rate range and meets the requirements.

[0212] Reference procedure 1.1, mean respiratory rate Standard deviation The random number Z = -0.156, calculated using the sampling formula, indicates a respiratory rate of 24 breaths / min; this is within the respiratory range and meets the requirements.

[0213] Refer to procedure 1.1, mean systolic blood pressure Standard deviation The random number Z = 0.127 is obtained from a standard normal distribution. The systolic blood pressure is calculated to be 61 mmHg using the sampling formula. This is within the systolic blood pressure range and meets the requirements.

[0214] Refer to procedure 1.1, mean diastolic blood pressure Standard deviation The random number Z = 0.234, obtained from a standard normal distribution, was used to calculate the diastolic blood pressure to be 31 mmHg. This is within the diastolic blood pressure range and meets the requirements.

[0215] The time was 1.76 minutes. Complications: asphyxia-acute obstruction phase, heart rate: 163 beats / min, blood pressure: 61 / 31 mmHg, respiration: 24 breaths / min, consciousness: severe impairment of consciousness.

[0216] Process 4: From the "acute obstruction phase" to the "respiratory arrest phase":

[0217] According to the projection rules, within 1-2 minutes, the complication will evolve into "respiratory arrest," and the vital signs will be in the following range: "heart rate: 180-40 beats / min, respiration: 10-6 breaths / min, blood pressure: too low, consciousness: severe impairment of consciousness."

[0218] Step 4.1, determining the time of traumatic injury evolution based on exponential distribution:

[0219] The time range is [1,2], that is , ;

[0220] according to Calculate the exponential distribution parameters ;

[0221] From uniform distribution Random numbers drawn , , ;

[0222] Scaling Calculate the time of change min;

[0223] If the condition is met within the time range [1,2], then the "respiratory arrest period" will begin at 1.76+1.51=3.27min.

[0224] Step 4.2, generating transitional vital signs based on a diffusion model:

[0225] Within the range of 1.76–3.27 minutes, vital signs at 2 minutes and 3 minutes were generated based on the diffusion model. The time steps at these times were respectively… =2-1.76=0.24min、 =3-1.76=1.24min.

[0226] Refer to procedure 3.2, mean heart rate Standard deviation The random number Z = 0.156 is a standard normal random number. The heart rate is calculated to be 154 beats / min using the sampling formula.

[0227] Refer to procedure 3.2, mean respiratory rate Standard deviation The random number Z = -0.127 is a standard normal random number. The respiratory rate is calculated to be 21 breaths / min using the sampling formula.

[0228] Refer to procedure 3.2, mean systolic blood pressure Standard deviation The random number Z = 0.127 is a standard normal random number. The systolic blood pressure is calculated to be 52 mmHg using the sampling formula.

[0229] Refer to procedure 3.2, mean diastolic blood pressure Standard deviation The random number Z = 0.127 is a standard normal random number. The diastolic blood pressure is calculated to be 26 mmHg using the sampling formula.

[0230] Then, 2 minutes later, complications: asphyxia-acute obstruction, heart rate: 154 beats / min, blood pressure: 52 / 26 mmHg, respiration: 21 breaths / min, consciousness: severe impairment of consciousness.

[0231] Refer to procedure 3.2, mean heart rate Standard deviation The random number Z = -0.412 is a standard normal random number. The heart rate is calculated to be 116 beats / min using the sampling formula.

[0232] Refer to procedure 3.2, mean respiratory rate Standard deviation The random number Z = 0.315 is a standard normal random number. The breathing rate is calculated to be 15 breaths / min using the sampling formula.

[0233] Refer to procedure 3.2, mean systolic blood pressure Standard deviation The random number Z = -0.189 is a standard normal random number. The systolic blood pressure is calculated to be 10 mmHg using the sampling formula.

[0234] Refer to procedure 3.2, mean diastolic blood pressure Standard deviation The random number Z = 0.156 is a standard normal random number. The diastolic blood pressure is calculated to be 6 mmHg using the sampling formula.

[0235] Then, after 3 minutes, complications included asphyxiation-acute obstruction, heart rate of 116 beats / min, blood pressure of 10 / 6 mmHg, respiration of 15 breaths / min, and consciousness of severe impairment.

[0236] Step 4.3: Determining vital signs based on Poisson distribution:

[0237] Refer to procedure 1.1, mean heart rate Standard deviation The random number Z = -0.127, obtained from a standard normal distribution, is used to calculate a heart rate of 105 beats / min using the sampling formula; this is within the heart rate range and meets the requirements.

[0238] Reference procedure 1.1, mean respiratory rate Standard deviation The random number Z = -0.298, obtained from a standard normal distribution, is used to calculate a respiratory rate of 8 breaths / min using the sampling formula; this is within the respiratory range and meets the requirements.

[0239] Low blood pressure;

[0240] The time was 3.27 minutes. Complications: asphyxia-respiratory arrest, heart rate: 105 beats / min, blood pressure: too low, respiration: 8 breaths / min, consciousness: severe impairment of consciousness.

[0241] Step 5: Treatment begins during the "respiratory arrest period":

[0242] Process 5.1: Generating transitional vital signs during treatment:

[0243] Hands-only CPR was initiated at 3.27 minutes and lasted 1.25 minutes. The treatment ended at 3.27 + 1.25 = 4.52 minutes. Vital signs were generated at 4 minutes based on the fluctuations in the "respiratory arrest" state between 3.27 and 4.52 minutes.

[0244] Refer to process 4.3, mean heart rate Standard deviation The random number Z = -0.156 is a standard normal random number. The heart rate is calculated to be 111 beats / min using the sampling formula.

[0245] Refer to procedure 4.3, mean respiratory rate Standard deviation The random number Z = -0.234 is a standard normal random number. The breathing rate is calculated to be 8 breaths / min using the sampling formula.

[0246] Low blood pressure;

[0247] The patient was 4 minutes old. Complications included asphyxia-respiratory arrest, heart rate of 111 beats / min, blood pressure of hypotension, respiration of 8 breaths / min, and consciousness of severe impairment.

[0248] Procedure 5.2, determining vital signs based on Poisson distribution:

[0249] At the end of treatment, the vital signs still met the requirements for the "respiratory arrest period" in Process 4, referring to Process 4.3, with the mean heart rate as follows: Standard deviation The random number Z = -0.156, obtained from a standard normal distribution, is used to calculate a heart rate of 111 beats / min using the sampling formula; this is within the heart rate range and meets the requirements.

[0250] Mean respiratory rate Standard deviation The random number Z = -0.234, obtained from a standard normal distribution, is used to calculate a respiratory rate of 8 breaths / min using the sampling formula; this is within the respiratory range and meets the requirements.

[0251] Low blood pressure;

[0252] The time was 4.52 minutes. Complications: asphyxia-respiratory arrest, heart rate: 111 beats / min, blood pressure: too low, respiration: 8 breaths / min, consciousness: severe impairment of consciousness.

[0253] Step 5.3, Judging the Trend of Traumatic Injury Changes Based on a Large Model:

[0254] Although the injured person had entered respiratory arrest before treatment, exhibiting extremely weak breathing (8 breaths / min), undetectable blood pressure, and severe altered consciousness, placing them in a highly critical state, basic life support, including hands-only CPR, was immediately initiated at this critical moment and continued for 1.25 minutes. This is a core method used in emergency care to maintain brain and cardiac perfusion and save lives. Despite being a primary intervention, its timely initiation, closely following the onset of respiratory arrest, maximized the window of opportunity for resuscitation and prevented the condition from progressing to cardiac arrest and an irreversible state. Combined with the fact that the heart rate, although decreased, was still present (105 beats / min), it indicates that circulatory function had not been completely lost, providing a certain foundation for resuscitation. With timely CPR intervention, the injured person has the potential to regain spontaneous breathing and circulation, and their condition is expected to stabilize with further medical support. Therefore, the post-treatment trend of the traumatic injury is judged to be improving.

[0255] Process 6: Improvement from "respiratory arrest" to "acute obstruction":

[0256] Based on the results of the 5.3 model, the complications will improve within 1-3 minutes and will evolve into the "acute obstructive phase". The vital signs will be in the following range: "Heart rate: 140-180 beats / min, Respiration: 40-10 breaths / min, Systolic blood pressure: 70-50 mmHg, Diastolic blood pressure: 40-20 mmHg, Blood pressure is expressed as systolic / diastolic, Consciousness: severe impairment of consciousness".

[0257] Procedure 6.1, determining the time of traumatic injury evolution based on exponential distribution:

[0258] The time range is [1,3], that is , ;

[0259] according to Calculate the exponential distribution parameters ;

[0260] From uniform distribution Random numbers drawn , , Scaling Calculate the time of change min;

[0261] If the condition is met within the time range [1,3], then the "acute obstruction period" will begin at 4.52+1.51=6.03min.

[0262] Step 6.2, generating transitional vital signs based on a diffusion model:

[0263] Within the range of 4.52-6.03 minutes, vital signs at 5 minutes and 6 minutes were generated based on the diffusion model. The time steps at these times were respectively... =5-4.52=0.48min、 =6-4.52=1.48min.

[0264] Refer to process 4.3, mean heart rate Standard deviation The random number Z = -0.156 is a standard normal random number. The heart rate is calculated to be 120 beats / min using the sampling formula.

[0265] Refer to procedure 4.3, mean respiratory rate Standard deviation The random number Z = -0.127 is a standard normal random number. The respiratory rate is calculated to be 14 breaths / min using the sampling formula.

[0266] Refer to process 4.3, mean systolic blood pressure Standard deviation The random number Z = -0.234 is a standard normal random number. The systolic blood pressure is calculated to be 21 beats / min using the sampling formula.

[0267] Refer to procedure 4.3, mean diastolic blood pressure Standard deviation The random number Z = -0.156 is a standard normal random number. The diastolic blood pressure is calculated to be 10 beats / min using the sampling formula.

[0268] If the patient remains unconscious for 5 minutes, complications include asphyxia-respiratory arrest, heart rate of 120 beats / min, blood pressure of hypotension, respiration of 14 breaths / min, and consciousness of severe impairment.

[0269] Refer to process 4.3, mean heart rate Standard deviation The random number Z = 0.189 is a standard normal random number. The heart rate is calculated to be 165 beats / min using the sampling formula.

[0270] Refer to procedure 4.3, mean respiratory rate Standard deviation The random number Z = -0.234 is a standard normal random number. The respiratory rate is calculated to be 24 breaths / min using the sampling formula.

[0271] Refer to process 4.3, mean systolic blood pressure Standard deviation The random number Z = -0.127 is a standard normal random number. The systolic blood pressure is calculated to be 58 beats / min using the sampling formula.

[0272] Refer to procedure 4.3, mean diastolic blood pressure Standard deviation The random number Z = 0.412 is a standard normal random number. The diastolic blood pressure is calculated to be 32 beats / min using the sampling formula.

[0273] The patient was 6 minutes old. Complications included asphyxia-respiratory arrest, heart rate of 165 bpm, blood pressure of 58 / 32 mmHg, respiration of 24 breaths / min, and consciousness of severe impairment.

[0274] Procedure 6.3: Determining vital signs based on Poisson distribution:

[0275] Refer to procedure 1.1, mean heart rate Standard deviation The random number Z = 0.278 is obtained from a standard normal distribution. The heart rate calculated using the sampling formula is 163 beats / min; this is within the heart rate range and meets the requirements.

[0276] Reference procedure 1.1, mean respiratory rate Standard deviation The random number Z = -0.156, calculated using the sampling formula, indicates a respiratory rate of 24 breaths / min; this is within the respiratory range and meets the requirements.

[0277] Refer to process 1.1, mean systolic blood pressure Standard deviation The random number Z = 0.127 is obtained from a standard normal distribution. The systolic blood pressure is calculated to be 61 mmHg using the sampling formula. This is within the systolic blood pressure range and meets the requirements.

[0278] Refer to procedure 1.1, mean diastolic blood pressure Standard deviation The random number Z = 0.234, obtained from a standard normal distribution, was used to calculate the diastolic blood pressure to be 31 mmHg. This is within the diastolic blood pressure range and meets the requirements.

[0279] The patient's condition was assessed at 6.03 min, with complications including asphyxia-acute obstruction, heart rate of 163 bpm, blood pressure of 61 / 31 mmHg, respiration of 24 breaths / min, and consciousness of severe impairment.

[0280] Process 7: Improvement from the "acute obstruction phase" to the "immediate mechanical injury phase":

[0281] Complications will improve within 0.5-3 minutes and will evolve into the "immediate mechanical injury period". Vital signs will be in the following range: "Heart rate: 120-140 beats / min, Respiration: 30-40 breaths / min, Systolic blood pressure: 140-160 mmHg, Diastolic blood pressure: 90-110 mmHg, Blood pressure is expressed as systolic / diastolic, Consciousness: mild impairment of consciousness".

[0282] Procedure 7.1, determining the time of traumatic injury evolution based on exponential distribution:

[0283] The time range is [0.5, 3], that is... , ;

[0284] according to Calculate the exponential distribution parameters ;

[0285] From uniform distribution Random numbers drawn , , Scaling Calculate the time of change min;

[0286] If the condition is met within the time range [0.5, 3], then the "immediate mechanical damage period" will begin at 6.03 + 1.46 = 7.49 min.

[0287] Process 7.2, generating transitional vital signs based on a diffusion model:

[0288] Within the time frame of 6.03-7.49 minutes, vital signs at time 7 minutes were generated based on the diffusion model. The time step at this point was... =7-6.03=0.97min.

[0289] Refer to process 4.3, mean heart rate Standard deviation The random number Z = 0.189 is a standard normal random number. The heart rate is calculated to be 146 beats / min using the sampling formula.

[0290] Refer to procedure 4.3, mean respiratory rate Standard deviation The random number Z = -0.234 is a standard normal random number. The respiratory rate is calculated to be 30 breaths / min using the sampling formula.

[0291] Refer to process 4.3, mean systolic blood pressure Standard deviation The random number Z = -0.127 is a standard normal random number. The systolic blood pressure is calculated to be 119 beats / min using the sampling formula.

[0292] Refer to procedure 4.3, mean diastolic blood pressure Standard deviation The random number Z = 0.412 is a standard normal random number. The diastolic blood pressure is calculated to be 79 beats / min using the sampling formula.

[0293] Then, 7 minutes later, complications: asphyxia-acute obstruction period, heart rate: 146 beats / min, blood pressure: 119 / 79 mmHg, respiration: 30 breaths / min, consciousness: severe impairment of consciousness (because the immediate mechanical injury period has not been fully entered).

[0294] Procedure 7.3, determining vital signs based on Poisson distribution:

[0295] Refer to procedure 1.1, mean heart rate Standard deviation The random number Z = 0.412 is obtained from a standard normal distribution. The heart rate calculated using the sampling formula is 132 beats / min; this is within the heart rate range and meets the requirements.

[0296] Reference procedure 1.1, mean respiratory rate Standard deviation The random number Z = 0.1, a standard normal distribution, was used to calculate the respiratory rate as 35 breaths / min using the sampling formula; this is within the respiratory range and meets the requirements.

[0297] Refer to procedure 1.1, mean systolic blood pressure Standard deviation The random number Z = 0.234 is obtained from a standard normal distribution. The systolic blood pressure is calculated to be 151 mmHg using the sampling formula. This is within the systolic blood pressure range and meets the requirements.

[0298] Refer to procedure 1.1, mean diastolic blood pressure Standard deviation The random number Z = 0.189 was obtained from a standard normal distribution. The diastolic blood pressure was calculated to be 101 mmHg using the sampling formula. This is within the diastolic blood pressure range and meets the requirements.

[0299] The time was 7.49 minutes. Complications: asphyxia - immediate mechanical injury period, heart rate: 132 beats / min, blood pressure: 151 / 101 mmHg, respiration: 35 breaths / min, consciousness: mild impairment of consciousness.

[0300] Process 8: From the "immediate mechanical injury period" to the initial state of "asphyxiation" improvement:

[0301] The complication will improve within 0-1 minute and will evolve into the initial state of "asphyxia". Vital signs will be in the following range: "Heart rate: 60-100 beats / min, Respiration: 16-20 breaths / min, Systolic blood pressure: 90-140 mmHg, Diastolic blood pressure: 60-90 mmHg, Blood pressure: Systolic / Diastolic, Consciousness: Alert".

[0302] Procedure 8.1, determining the time of traumatic injury evolution based on exponential distribution:

[0303] The time range is [0,1], that is , ;

[0304] according to Calculate the exponential distribution parameters ;

[0305] From uniform distribution Random numbers drawn , , Scaling Calculate the time of change min;

[0306] If the condition is met within the time range [0,1], then "asphyxiation" will occur at 7.49 + 70.1 = 7.59 min.

[0307] Procedure 7.2, determining vital signs based on Poisson distribution:

[0308] Refer to procedure 1.1, mean heart rate Standard deviation The random number Z = 0.412 is obtained from a standard normal distribution. The heart rate calculated using the sampling formula is 84 beats / min; this is within the heart rate range and meets the requirements.

[0309] Reference procedure 1.1, mean respiratory rate Standard deviation The random number Z = 0.1, a standard normal distribution, was used to calculate the respiratory rate as 18 breaths / min using the sampling formula; this is within the respiratory range and meets the requirements.

[0310] Refer to procedure 1.1, mean systolic blood pressure Standard deviation The random number Z = 0.234 is obtained from a standard normal distribution. The systolic blood pressure is calculated to be 118 mmHg using the sampling formula. This is within the systolic blood pressure range and meets the requirements.

[0311] Refer to procedure 1.1, mean diastolic blood pressure Standard deviation The random number Z = 0.189 was obtained from a standard normal distribution. The diastolic blood pressure was calculated to be 76 mmHg using the sampling formula. This is within the diastolic blood pressure range and meets the requirements.

[0312] The patient's condition was assessed at 7.59 minutes. Complications included asphyxia, heart rate of 84 beats / min, blood pressure of 118 / 76 mmHg, respiration of 18 breaths / min, and consciousness of conscious individuals.

[0313] And so on.

[0314] In summary, the deduction results within 8 minutes can be summarized as follows:

[0315] 0 min, Complications: Asphyxia, Heart rate: 83 bpm, Blood pressure: 114 / 74 mmHg, Respiration: 18 breaths / min, Consciousness: Alert.

[0316] 0.3 min, complications: asphyxia - immediate mechanical injury period, heart rate: 128 beats / min, blood pressure: 151 / 99 mmHg, respiration: 36 breaths / min, consciousness: mild impairment of consciousness.

[0317] 1 min, complications: asphyxia - immediate mechanical injury period, heart rate: 144 bpm, blood pressure: 107 / 62 mmHg, respiration: 31 breaths / min, consciousness: mild impairment of consciousness.

[0318] 1.76 min, complications: asphyxia - acute obstruction, heart rate: 163 bpm, blood pressure: 61 / 31 mmHg, respiration: 24 breaths / min, consciousness: severe impairment of consciousness.

[0319] 2 minutes, complications: asphyxia - acute obstruction, heart rate: 154 beats / min, blood pressure: 52 / 26 mmHg, respiration: 21 breaths / min, consciousness: severe impairment of consciousness.

[0320] 3 minutes, complications: asphyxia - acute obstruction, heart rate: 116 beats / min, blood pressure: 10 / 6 mmHg, respiration: 15 breaths / min, consciousness: severe impairment of consciousness.

[0321] 3.27 min, complications: asphyxia-respiratory arrest, heart rate: 105 beats / min, blood pressure: too low, respiration: 8 breaths / min, consciousness: severe impairment of consciousness.

[0322] 4 min, complications: asphyxia-respiratory arrest, heart rate: 111 beats / min, blood pressure: too low, respiration: 8 breaths / min, consciousness: severe impairment of consciousness.

[0323] 4.52 min, complications: asphyxia-respiratory arrest, heart rate: 111 beats / min, blood pressure: too low, respiration: 8 breaths / min, consciousness: severe impairment of consciousness.

[0324] 5 minutes, complications: asphyxia-respiratory arrest, heart rate: 120 beats / min, blood pressure: too low, respiration: 14 breaths / min, consciousness: severe impairment of consciousness.

[0325] 6 minutes, complications: asphyxia-respiratory arrest, heart rate: 165 bpm, blood pressure: 58 / 32 mmHg, respiration: 24 breaths / min, consciousness: severe impairment of consciousness.

[0326] 6.03 min, complications: asphyxia - acute obstruction, heart rate: 163 bpm, blood pressure: 61 / 31 mmHg, respiration: 24 breaths / min, consciousness: severe impairment of consciousness.

[0327] 7 min, complications: asphyxia - acute obstruction period, heart rate: 146 beats / min, blood pressure: 119 / 79 mmHg, respiration: 30 breaths / min, consciousness: severe impairment of consciousness (due to not fully entering the immediate mechanical injury period).

[0328] 7.49 min, Complications: Asphyxia - Immediate mechanical injury period, Heart rate: 132 bpm, Blood pressure: 151 / 101 mmHg, Respiration: 35 breaths / min, Consciousness: Mild impairment of consciousness.

[0329] 7.59 min, Complications: Asphyxia, Heart rate: 84 bpm, Blood pressure: 118 / 76 mmHg, Respiration: 18 breaths / min, Consciousness: Alert.

[0330] Second Embodiment

[0331] Corresponding to the first embodiment, this embodiment of the invention also provides a trauma inference model construction system based on a multimodal probability model, implemented using the trauma inference model construction method based on a multimodal probability model provided in the first embodiment. The system includes:

[0332] The model building module is used to build the current trauma simulation model;

[0333] The data acquisition module is used to acquire historical trauma injury data;

[0334] The module for determining the pattern of change is used to mine and determine the standard pattern of change of traumatic injury based on historical traumatic injury data, which includes the pattern of transition of complication stages, the range of vital signs, and transitional characteristics.

[0335] The data generation module is used to determine the evolution time of traumatic injuries based on the standard traumatic injury change pattern, generate vital signs using the exponential distribution, generate transitional vital signs using the Poisson distribution, and generate transitional vital signs using the diffusion model.

[0336] The judgment module is used to determine whether the time of traumatic injury evolution, vital signs, and transitional vital signs of adjacent complication stages meet the constraints of the standard traumatic injury change pattern.

[0337] The repetitive training module is used to adjust the parameters of the current trauma simulation model when at least one of the trauma evolution time, vital signs, and transitional vital signs does not conform to the constraints of the standard trauma change law. The data generation module is then executed sequentially to the repetitive training module until the trauma evolution time, vital signs, and transitional vital signs all conform to the constraints of the standard trauma change law. Then, the corresponding current trauma simulation model is output as the target trauma simulation model.

[0338] It should be noted that the system provided in this embodiment is only illustrated by the division of the above-mentioned functional modules when implementing its functions. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the system can be divided into different functional modules to complete all or part of the functions described above. In addition, the system provided in this embodiment and the method provided in the first embodiment belong to the same concept, and the specific implementation process can be found in the first embodiment, which will not be repeated here.

Claims

1. A method for constructing a trauma inference model based on a multimodal probability model, characterized in that, The method includes: Step S1: Construct the current trauma simulation model; Step S2: Obtain historical trauma data; Step S3: Based on the historical trauma data, mine and determine the standard trauma change patterns covering the transition patterns of complication stages, the range of vital signs, and transitional characteristics; Step S4: Based on the standard trauma condition change pattern, use the exponential distribution to determine the trauma condition evolution time, use the Poisson distribution to generate vital signs, and use the diffusion model to generate transitional vital signs of adjacent complication stages; Step S5: Determine whether the traumatic injury evolution time, vital signs, and transitional vital signs meet the constraints of the standard traumatic injury change pattern. Step S6: If at least one of the traumatic injury evolution time, vital signs, and transitional vital signs does not conform to the constraint range of the standard traumatic injury change law, adjust the parameters of the current traumatic injury deduction model, and repeat steps S4 to S6 until the traumatic injury evolution time, vital signs, and transitional vital signs all conform to the constraint range of the standard traumatic injury change law. Then, output the corresponding current traumatic injury deduction model as the target traumatic injury deduction model. The current trauma simulation model includes a current trauma evolution time submodule for determining the trauma evolution time, a current vital signs submodule for generating the vital signs, and a current transitional vital signs submodule for generating the transitional vital signs; the target trauma simulation model includes a target trauma evolution time submodule for determining the trauma evolution time, a target vital signs submodule for generating the vital signs, and a target transitional vital signs submodule for generating the transitional vital signs. The parameters adjusted by the current trauma simulation model are the parameters of the sub-modules corresponding to the generation results that do not conform to the constraint range of the standard trauma injury change law; The diffusion model is divided into a current diffusion model during training and a target diffusion model after training, based on the training stage, and is used to generate vital sign data for smooth transition between adjacent complication stages. The training steps for the target trauma injury evolution time submodule specifically include: The time range of any complication stage is determined based on the aforementioned standard trauma condition change pattern. Based on the time range of any of the aforementioned complication stages, the corresponding complication stage index distribution parameters are calculated according to a preset clinical confidence probability. Random numbers are drawn from a uniform distribution; Candidate evolution times are calculated using a sampling formula for the evolution time of traumatic injuries; Determine whether the candidate evolution time is within the time range of the corresponding complication stage; If yes, the current traumatic injury evolution time submodule is output as the target traumatic injury evolution time submodule; if no, the random number is re-evaluated and calculated until the candidate evolution time is within the corresponding complication stage change time range, then the corresponding current traumatic injury evolution time submodule is output as the target traumatic injury evolution time submodule. The training steps for the target vital signs submodule specifically include: Based on the aforementioned standard trauma condition change patterns, determine the range of vital sign constraints for any stage of complication. Based on the vital sign constraints and preset information parameters at the complication stage, calculate the mean and standard deviation of vital signs; Candidate vital signs were calculated using the Poisson distribution sampling formula; Determine whether the candidate vital signs are within the vital sign constraints of the complication stage; If yes, the current vital signs submodule is output as the target vital signs submodule; if no, the standard normal random numbers in the Poisson distribution sampling formula are re-extracted and recalculated until the candidate vital signs are within the vital signs constraint range of the complication stage, then the corresponding current vital signs submodule is output as the target vital signs submodule. The training process of the target diffusion model is as follows: The current diffusion model is constructed using a multilayer perceptron. Based on the standard trauma injury change pattern, the initial point of the previous stage and the target point of the next stage of vital signs are extracted for adjacent complication stages. The current diffusion model generates an intermediate transition distribution from the initial point to the target point; The intermediate transition distribution is compared with the actual transitional vital sign data in the historical trauma injury data to obtain the comparison results; If the comparison result meets the preset conditions, the current diffusion model is output as the target diffusion model, and the current transitional vital signs submodule is output as the target transitional vital signs submodule. If the comparison result does not meet the preset conditions, the parameters of the current diffusion model are adjusted and the intermediate transitional distribution is regenerated for comparison until the comparison result meets the preset conditions. Then, the corresponding current diffusion model is output as the target diffusion model, and the corresponding current transitional vital signs submodule is output as the target transitional vital signs submodule.

2. The method for constructing a trauma inference model based on a multimodal probability model according to claim 1, characterized in that, The time range for the complication stage is determined by the standard trauma condition change pattern. And the candidate evolution time falls within the time range When the probability of occurrence is 95%, the training steps for the target trauma injury evolution time submodule specifically include: Calculate the exponential distribution parameters of the complication stage. The calculation formula is as follows: ； From uniform distribution Random numbers drawn ; right Scaling The calculation formula is as follows: ； exist arrive In between, at this time The evolution time of the candidate is calculated: ； Examine the candidate evolution time Is it within the time range? If the current traumatic injury evolution time submodule is within the specified range, it will be output as the target traumatic injury evolution time submodule; if it is outside the specified range, the random number will be re-generated and recalculated until the candidate evolution time is within the specified range. If the current traumatic injury evolution time submodule is used as the target traumatic injury evolution time submodule, then the current traumatic injury evolution time submodule will be output.

3. The method for constructing a trauma inference model based on a multimodal probability model according to claim 1, characterized in that, Given that the standard trauma condition change pattern determines the range of vital signs within the complication stage as [c, d], the mean vital signs... The calculation formula is: ； The standard deviation of vital signs The calculation formula is: ； in, The preset confidence parameter corresponds to the standard normal distribution quantile of the clinical confidence level; The calculation formula for the Poisson distribution sampling formula is as follows: ； in, It is the standard normal random number.

4. The method for constructing a trauma inference model based on a multimodal probability model according to claim 1, characterized in that, The current diffusion model is a neural network model, and the parameters of the current diffusion model include the number of network layers and the learning rate.

5. The method for constructing a trauma inference model based on a multimodal probability model according to claim 4, characterized in that, The process of generating an intermediate transitional distribution from the initial point to the target point using the current diffusion model specifically involves the current diffusion model transferring data from the initial distribution... Transform into target distribution The process is divided into two stages: The forward diffusion process, which is a Markov chain, progressively diffuses data from the initial distribution. Convert to noise distribution Each step adds a preset amount of noise to gradually blur the features of the data. The calculation formula is as follows: ； in: It is a time step Data; It is the diffusion coefficient, which controls the amount of noise added at each step; It is noise that follows a standard normal distribution; The reverse diffusion process, which is the inverse of forward diffusion, aims to reduce noise distribution. Gradually restore to the initial distribution Each step involves learning to remove noise and gradually restore the characteristics of the data. The calculation formula is as follows: ； in: It is noise predicted through a neural network; The training objective of the current diffusion model is to learn the velocity field. This velocity field, used to guide data migration from an initial distribution to a target distribution, satisfies a specific differential equation: ； In discrete time, it can be represented as a difference equation: ； in, It is the time step.

6. A trauma deduction model construction system based on a multimodal probability model, implemented using the trauma deduction model construction method based on a multimodal probability model as described in any one of claims 1-5, characterized in that, The system includes: The model building module is used to build the current trauma simulation model; The data acquisition module is used to acquire historical trauma injury data; The change pattern determination module is used to mine and determine standard trauma change patterns based on the historical trauma injury data, covering the transition patterns of complication stages, the range of vital signs, and transitional characteristics. The data generation module is used to determine the evolution time of traumatic injuries based on the standard traumatic injury change pattern, generate vital signs using the exponential distribution, generate transitional vital signs of adjacent complication stages using the Poisson distribution, and generate transitional vital signs of adjacent complication stages using the diffusion model. The judgment module is used to determine whether the traumatic injury evolution time, the vital signs, and the transitional vital signs meet the constraints of the standard traumatic injury change pattern. The repetitive training module is used to adjust the parameters of the current trauma simulation model when at least one of the trauma evolution time, vital signs, and transitional vital signs does not conform to the constraints of the standard trauma change law. The data generation module is then executed sequentially to the repetitive training module until the trauma evolution time, vital signs, and transitional vital signs all conform to the constraints of the standard trauma change law. Then, the corresponding current trauma simulation model is output as the target trauma simulation model.

Images