Emergency rescue real-time visualization and intelligent operation management platform based on virtual reality

Abstract

The invention provides an emergency rescue real-time visualization and intelligent operation management platform based on virtual reality, and relates to the technical field of augmented reality. The technical problem that accurate attribution and prospective prediction cannot be performed on systematic risks due to information dimension splitting and static analysis methods in the prior art is solved. The fundamental transformation from discrete and single-dimensional state monitoring to unified and multi-dimensional systematic risk attribution is realized; secondly, by introducing quantitative analysis on a dynamic trend, the early recognition and prediction capability of a team and task interaction failure mode in a complex high-voltage scene is improved; finally, decision information with diagnosis and prediction functions is provided for managers, so that the accuracy of remote command decision and the effectiveness of intervention measures are enhanced.

Description

Technical Field

[0001] This invention relates to the field of extended reality technology, specifically to a real-time visualization and intelligent operation management platform for emergency rescue based on virtual reality. Background Technology

[0002] In high-risk, time-sensitive, and complex human-machine collaborative fields such as emergency response and air traffic control, leveraging data-driven intelligent operation management platforms to improve decision-making efficiency and system resilience has become an important direction for technological development. In particular, with the advancement of extended reality (XR) and digital twin technologies, how to transform massive amounts of multimodal field data into visual decision-making data that provides intuitive guidance for operators is currently at the forefront of technological research.

[0003] However, existing technologies still face several challenges in achieving this goal. Specifically: Information fragmentation and attribution difficulties: Existing technical solutions typically treat the analysis of external tasks (such as patient conditions) and the evaluation of internal teams (such as collaboration status) as two separate technical paths. This information fragmentation makes it difficult for managers to accurately determine whether poor system performance is due to excessive task difficulty or a failure of team collaboration processes, thus hindering the implementation of targeted interventions.

[0004] Static limitations of risk assessment: Most risk assessment models rely on snapshots of the system's state at specific points in time, lacking quantitative analysis of the "trend" of system state evolution. This limits the system to passive, reactive alerts, preventing it from predicting and warning of potential state transition risks based on dynamic trends, thus missing the optimal opportunity for proactive intervention.

[0005] Limitations of the Single Subject in the Analytical Framework: Even though some existing technologies attempt to integrate multi-dimensional indicators, such as the patent application document with publication number CN118942641A, which uses joint analysis of rehabilitation movement quality and rehabilitation effect evaluation index to warn of risks, its analytical framework is still essentially limited to multiple attributes of a single subject (i.e., the patient themselves). This type of technology fails to establish a dynamic correlation model between the two different subject dimensions of "external objective challenges" and "internal team responsiveness," and therefore cannot reveal the systemic failure modes caused by the mismatch between the two. Summary of the Invention

[0006] The purpose of this invention is to provide a real-time visualization and intelligent operation management platform for emergency rescue based on virtual reality, so as to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: A virtual reality-based real-time visualization and intelligent operation management platform for emergency resuscitation includes: S1: Acquire clinical entity time-series data representing the patient's condition, generated by natural language processing of voice data at the emergency scene within a preset time window; S2: Obtain interactive network topology data representing the collaborative state of multiple entities, generated by performing graph network analysis on log data representing the interactive behavior between multiple participating entities in an emergency rescue event within the preset time window; S3: Based on the time-series data of the clinical entities, a first indicator is calculated, which is used to quantify the predicted risk of adverse events occurring in the patient in the future; S4: Based on the interactive network topology data, a second indicator is calculated, which is used to quantify the current cognitive load of the multiple participating entities; S5: Using a preset diagnostic state mapping model, the first indicator and the second indicator are jointly processed to generate a diagnostic state signal that drives the extended reality user interface to perform adaptive state presentation.

[0008] Compared with existing technologies, the beneficial effects of this invention are as follows: A "Predictive Clinical Risk Index (PCRI)" is generated through hierarchical dynamic calculation of clinical entity time-series data to quantify the risk of patient disease evolution; in parallel, a "Team Cognitive Load Index (TCLI)" is generated through graph network analysis of multi-agent interactive behaviors to quantify team collaboration effectiveness and cognitive load. These two indices constitute an orthogonal basis for a unique two-dimensional "diagnostic state space" created in this invention. The fundamental breakthrough of this invention lies in the subsequent introduction of a "diagnostic state mapping model" based on dynamic potential field theory. This model can not only diagnose the macroscopic state of the system based on the current coordinates of the PCRI and TCLI, including but not limited to "efficient response" or "process failure"; but also, by introducing the analysis of the historical trajectory of state points, i.e., calculating a "state velocity vector," it achieves a forward-looking prediction of the future evolution direction and probability of the system state. This shift from "static position diagnosis" to "dynamic trend prediction" directly solves the problem of delayed response in existing technologies. By integrating PCRI, which represents task difficulty, with TCLI, which represents team capability, we can quantitatively distinguish and accurately attribute "task-inherent difficulty failures" and "team process-driven failures," providing a solid technical foundation for achieving truly intelligent and forward-looking operation management. Attached Figure Description

[0009] Figure 1 This is a schematic diagram of the overall application modules of the present invention; Figure 2 A schematic diagram of the execution logic of steps S1 to S4 of the present invention; Figure 3 This is a schematic diagram illustrating the execution logic of step S5 of the present invention. Detailed Implementation

[0010] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0011] It is understood that the terms “first,” “second,” etc., used in this application may be used herein to describe various elements, but unless otherwise stated, these elements are not limited by these terms. These terms are used only to distinguish one element from another. Example

[0012] Please see Figures 1 to 3 This invention provides a technical solution: a real-time visualization and intelligent operation management platform for emergency rescue based on virtual reality, comprising the following steps: S1: Acquire clinical entity time-series data representing the patient's condition, generated by natural language processing of voice data at the emergency scene within a preset time window; S2: Obtain interactive network topology data representing the collaborative state of multiple parties by performing graph network analysis on log data representing the interactive behavior between multiple participating parties in an emergency rescue event within a preset time window; S3: Based on clinical entity time series data, a primary indicator is calculated. This primary indicator is used to quantify the predictive risk of patients experiencing adverse events in the future. S4: Based on the interactive network topology data, a second indicator is calculated, which is used to quantify the current cognitive load of multiple participating entities; S5: Through a preset diagnostic state mapping model, the first and second indicators are jointly processed to generate a diagnostic state signal for driving the extended reality (XR) user interface to perform adaptive state presentation.

[0013] In emergency medical care scenarios, acquiring and understanding the massive, heterogeneous, and dynamically changing data on-site is crucial for effective command and management. Existing technologies face bottlenecks in processing two key types of data: first, for verbal communications from on-site personnel, it is difficult to accurately assess the severity of the patient's condition and its dynamic evolution from continuous voice streams in real time; second, for complex interactions among multiple participants, there is a lack of an objective and reproducible method to quantify the team's collaborative effectiveness and cognitive stress.

[0014] It should be noted that: Figure 1The "Real-time Multimodal Data Acquisition" section corresponds to step S1; this block describes the starting point of the entire process, namely, acquiring raw data from multiple modalities in real time. The core task of S1 is to continuously collect information from various data modalities such as patient vital signs monitors, team communication records, and environmental sensors as the basis for subsequent analysis.

[0015] The "Dual-Index Model Diagnosis and Prediction" corresponds to the logical sum of steps S2, S3, S4, and S5; this block describes the process of transforming raw data into two key quantitative indicators. A functional summary of the three consecutive steps: S2 (Calculation of Predictive Clinical Risk Index (PCRI)): Generates the first core index. S3 (Calculation of Team Cognitive Load Index (TCLI)): Generates the second core index. S4 (Index Normalization and Fusion): Standardizes these two indices to prepare for the next step of joint analysis. This also corresponds to the part of S5 that uses a potential field model to generate the "diagnostic state identifier" and "confidence score." It points to the calculation of the "state transition probability" based on "historical trajectory" and "state velocity vector" in S5. The essence of S5 is its advanced model capable of processing time-series data and capturing the dynamic evolution trend of the system.

[0016] The adaptive XR interface rendering step is the downstream application directly driven by the output of S5; this block describes the final application and effect presentation of the entire method. The final output of step S5 is the "interface configuration instruction." This instruction is sent to the XR system, which performs specific rendering operations based on this instruction, that is, "adaptively" adjusts the interface. Therefore, this block represents the final execution stage of the system that follows S5 and is directly controlled by the decision results of S5.

[0017] Further explanation: The clinical entity time-series data includes at least four types of clinical entities identified from the speech data: symptoms, signs, drugs, and medical procedures, as well as the timestamp of each clinical entity being mentioned. Clinical entities are dynamically clustered into one or more clinical syndrome clusters based on a pre-defined clinical knowledge graph; the calculation of the first indicator is based on the activation status of one or more clinical syndrome clusters.

[0018] Further explanation: The activation state of a clinical syndrome cluster is determined by calculating a cluster threat accumulation value that changes dynamically over time; the calculation of the cluster threat accumulation value takes into account at least the basic threat value of each clinical entity within the cluster, as well as a time decay function used to characterize the decay of the basic threat value over time.

[0019] Further explanation: Interactive network topology data includes at least: multi-agent interaction entropy calculated based on log data, which characterizes the uncertainty of information flow among multiple participating entities, and task process conformity, which characterizes the degree of difference between the actual interaction process and the standard operating procedure.

[0020] Further explanation: The calculation of task process compliance includes: identifying the current primary clinical task based on clinical entity time-series data; selecting a target SOP corresponding to the primary clinical task from a library containing multiple standard operating procedures (SOPs); and comparing the real-time interactive process with the target SOP to determine task process compliance.

[0021] Further explanation: The calculation of multi-agent interaction entropy includes: identifying multiple predefined interaction pattern kernels in the interaction network topology data, and performing weighted summation based on the occurrence frequency of multiple interaction pattern kernels and their respective preset entropy contribution weights.

[0022] The following is a detailed description of the implementation of the above content: In this embodiment of the present invention, all configurable parameters, including the various thresholds, coefficients, weights, and rule bases described below, are predefined and stored in a structured external spreadsheet file. Before the method is executed, a data loading module reads this file and loads its contents into the working memory. All subsequent calculation steps query the required parameters from this memory data structure, thus achieving technical decoupling between the core algorithm and the application strategy.

[0023] In step S1, the real-time audio stream is converted into text using the speech recognition (ASR) module. Subsequently, the clinical named entity recognition (NER) model based on the BERT architecture processes the text to extract entities such as "dyspnea" (symptom), "cold and clammy skin" (sign), "adrenaline" (drug), and "tracheal intubation" (medical procedure), and records the time of their occurrence to form a structured time-series data stream.

[0024] 1) The first indicator is represented as a predictive clinical risk index. The calculation parameters required for the predictive clinical risk index are defined as follows: 1.1) Entity-Based Threat Value (ETV): This is a dimensionless numerical value used to characterize the potential severity or urgency of a single clinical entity when mentioned once. The essence of ETV is to transform the qualitative concept of "severity" in clinical medicine into a quantitative value usable in the engineering field. Its determination process integrates large-scale objective data analysis with the knowledge of senior experts. This invention employs a hybrid modeling method combining data-driven approaches and expert verification to determine this parameter. This invention aims to establish an ETV calibration system capable of learning from large-scale real-world data and being ultimately calibrated by expert consensus. In a large-scale emergency medical record dataset, the frequency of a clinical entity (symptoms, signs) in cases of poor prognosis is positively correlated with its inherent risk level. A statistical learning model is used to quantify this correlation, and the quantification result is used as an initial estimate of the ETV. Specifically, data from the emergency module of the publicly available, large-scale, anonymized intensive care database MIMIC-IV is used. Each record is labeled as a "high-risk event" (death or transfer to ICU) or a "non-high-risk event" based on its final outcome within 24 hours of admission. Using the clinical named entity recognition (NER) model described in the previous embodiments, the text portion of all records is processed to extract all mentions of clinical entities, forming a "medical record-entity" matrix. A logistic regression model is employed. Using the "medical record-entity" matrix as input features and "whether a high-risk event occurred" as the target variable, the logistic regression model is trained. After training, the positive regression coefficients corresponding to each clinical entity are extracted from the model. These coefficients represent the positive correlation strength between the entity and the high-risk event. Subsequently, all these coefficients are normalized to their maximum value, linearly mapping them to the [0,1] interval. This normalized value is the initial calibration value of the "Entity Baseline Threat Value (ETV)". A structured expert consensus process based on the Delphi method was adopted. An expert panel of at least 10 senior physicians with over 10 years of experience in pre-hospital emergency care or emergency departments was recruited to conduct multiple rounds of anonymous questionnaires. In the first round, initial entity baseline threat values ​​(ETVs) for all entities were distributed to the experts, who independently rated the reasonableness of the values ​​and provided suggestions for correction. In subsequent rounds, the statistical results (mean, median, and point of divergence) and the anonymous reasons for corrections from the previous round were fed back to the expert panel for their reference and reassessment until the expert panel's opinions converged (the change in values ​​between two consecutive rounds was less than 5%). The value obtained from the final round of expert consensus was used as the final ETV and stored in the configuration file.

[0025] 1.2) Clinical Syndrome Cluster (CSC): This represents a data structure used to organize multiple clinically related entities together, representing a specific pathological state or syndrome, including "Acute Coronary Syndrome Cluster" or "Traumatic Shock Cluster". The construction method for the Clinical Syndrome Cluster (CSC) and Standard Operating Procedure (SOP) library is as follows: The CSC and SOP are structured knowledge bases. The construction process transforms informal medical and procedural knowledge into machine-readable formal models. Using the principles of knowledge engineering, through collaboration between domain experts and knowledge engineers, the diagnostic logic and operational procedures contained in authoritative guidelines (including but not limited to the American Heart Association (AHA) Cardiopulmonary Resuscitation Guidelines and Advanced Trauma Life Support (ATLS) Guidelines) are formalized into "condition-conclusion" rule sets and directed graph structures, respectively. This embodiment takes the construction of an "Acute Coronary Syndrome (ACS)" cluster and a target SOP for "Cardiac Resuscitation (CPR)" as an example: For the CSC construction steps: A knowledge engineer and a cardiovascular emergency specialist jointly study the AHA diagnostic guidelines for ACS. ACS symptoms include "chest pain," "chest tightness," "difficulty breathing," and "profuse sweating." Engineers translated this logic into a rule stored in a configuration file: "Define a cluster named CSC-ACS. The activation condition for this cluster is: within the past 180 seconds, the sum of (the attenuated ETV of 'chest pain' plus the attenuated ETV of 'chest tightness') is greater than 0.9, or the sum of (the attenuated ETV of 'difficulty breathing' plus the attenuated ETV of 'profuse sweating') is greater than 0.8." Simultaneously, the cluster activation threshold CAT is set to 1.0.

[0026] For the target SOP construction steps: Engineers obtain the standard CPR flowchart from the AHA guidelines. Experts and engineers jointly define the nodes of the graph (containing "Assess Response", "Call for Help", "Check Breathing", "Chest Compressions", "Open Airway") and edges (representing the order of actions or conditional branches). Engineers convert this flowchart into a directed graph data structure stored in a configuration file. Adjacency lists are used to represent this: "Assess Response: [Call for Help, Check Breathing]; Check Breathing: [Chest Compressions];". This structure is the SOP-CPR graph.

[0027] 1.3) Cluster activation threshold, denoted as CAT: This is a dimensionless value. When the accumulated threat value within a Clinical Syndrome Cluster (CSC) exceeds this cluster activation threshold, the CSC is determined to be "activated". In this embodiment, this parameter is set to 0.7. The technical consideration for this value is to achieve a technical balance between ensuring sufficient sensitivity to combinations of key signs and avoiding false alarms caused by isolated, nonspecific signs. This embodiment uses an optimization method based on Receiver Operating Characteristic (ROC) analysis to determine the preferred value of the cluster activation threshold CAT. The specific process of this method is as follows: Using the aforementioned emergency room medical record dataset labeled with "high-risk events" for ETV calibration, for each medical record in the dataset, the algorithm of this invention is used to calculate the peak "cluster threat accumulation value" of each "Clinical Syndrome Cluster (CSC)". The "cluster activation" problem is regarded as a diagnostic classification task, that is, when the accumulation value exceeds CAT, the medical record is predicted to be a "high-risk event". Then, the value of CAT was iteratively changed within an interval from 0.01 to 1.00 in steps of 0.01, and the predicted true positive rate and false positive rate were calculated at each value point. These data points were plotted as an ROC curve. Finally, the point on the curve that maximized the difference between the true positive rate and the false positive rate was identified. The threshold corresponding to this point is considered the optimal value of the cluster activation threshold CAT, which achieves the best balance between diagnostic sensitivity and specificity.

[0028] 1.4) Threat Half-Life Parameter, denoted as THP: This is a time parameter, measured in seconds, used to define the rate at which the threat influence of an entity or cluster naturally decays over time. The half-life refers to the time required for its threat value to decrease to half of its initial value. In this embodiment, this parameter is set to 180 seconds. The technical consideration for this value is that three minutes is a critical assessment and decision-making cycle in pre-hospital emergency care. If no new relevant information is added or confirmed within three minutes, the urgency of the initial information should be appropriately reduced. The preferred value of the Threat Half-Life Parameter THP is determined through statistical analysis of large-scale clinical data, aiming to make it conform to the true timescale of disease evolution in critical events. The specific process of this determination method is as follows: All cases marked as "high-risk events" in the aforementioned emergency medical record dataset are selected. For each case, the first mention timestamps of all relevant clinical entities belonging to the ultimately activated "Clinical Syndrome Cluster (CSC)" are identified. Then, the time difference between the minimum time of appearance of the first relevant sign and the maximum time of appearance of the last relevant sign among these timestamps is calculated; this difference is defined as the "disease development window" for that case. After performing this calculation on multiple high-risk cases, a statistical distribution of the "disease development window" is obtained. To ensure that the algorithm can capture most of the relevant information within a sufficiently long time window, the 75th percentile of this distribution is selected as the benchmark. In the research and development dataset of this invention, the value of this 75th percentile is 360 seconds. Considering that half-life is a core parameter of exponential decay, this benchmark value is divided by 2, resulting in an optimal value of 180 seconds for the threat half-life parameter THP. This value ensures that early signs still maintain a significant impact during the critical illness development cycle.

[0029] 2) The second indicator is represented as the team cognitive load index, and its calculation includes the following parameter definitions: 2.1) Interaction Pattern Kernel, denoted as IPK: It represents a predefined subgraph structure representing a specific collaboration pattern. This includes a "command-confirmation" kernel representing efficient communication and a "command-question-clarification" kernel representing communication barriers. The Interaction Pattern Kernel IPK is a set of structured data patterns, serving as the "dictionary" for understanding the micro-behaviors of team communication in this invention. Each IPK defines a minimal interaction pattern with specific semantics. Their structure is predefined and belongs to the core rule base of this invention. Real-time team interactions are abstracted as a dynamically growing directed graph. IPKs are defined as a series of smaller, fixed directed graph templates. By matching these IPK templates in real-time within the dynamic interaction graph, unstructured communication flows can be parsed into a sequence of communication events with clear semantics. In specific embodiments of this invention, multiple IPKs are predefined and their structures are stored in a configuration file. These definitions are loaded for pattern matching. The following are detailed definitions of three representative IPKs: IPK-1: "Command-Confirmation" Mode; this mode represents an efficient, closed-loop command transmission. This mode consists of two nodes with different roles (node ​​A is the "team leader," and node B is "team member 1") and two edges in opposite directions. Specifically: there exists a directed edge of type "command" pointing from node A to node B; simultaneously, there exists a directed edge of type "confirmation" or "received" pointing from node B back to node A.

[0030] IPK-2: "Instruction-Conflict" Pattern; This pattern indicates obstruction in instruction delivery, misunderstandings, resource conflicts, or differing opinions, signaling increased cognitive load within the team. This pattern consists of two nodes (A and B) and two edges. Specifically: there exists a directed edge from node A to node B of type "Instruction"; simultaneously, there exists a directed edge from node B back to node A of type "Question," "Negation," or "Unexecutable."

[0031] IPK-3: "Information Request-Provide" Pattern; this pattern represents proactive information exchange among team members to establish shared contextual awareness. This pattern consists of two nodes (A and B) and two edges. Specifically, there is a directed edge of type "Inquiry" from node A to node B; simultaneously, there is a directed edge of type "Information Provide" from node B back to node A. By clearly defining these basic patterns, the computational basis of "Multi-Agent Interaction Entropy" is ensured to be clear, reproducible, and have clear technical implications. When calibrating the entropy contribution weights (ECW), communication patterns are encoded and counted in the interaction log based on these specific graph structure definitions.

[0032] 2.2) Entropy Contribution Weight, denoted as ECW: This represents a dimensionless coefficient associated with each IPK, used to characterize the degree to which the interaction pattern contributes to the overall cognitive load or communication disorder of the team. The determination of the entropy contribution weight ECW is a core algorithm parameter that determines the contribution of different communication patterns to the "Team Cognitive Load Index (TCLI)". In team collaboration, the "instruction-conflict-clarification" communication pattern consumes more cognitive resources than the "instruction-confirmation" pattern and better reflects the degree of team disorder. In this embodiment, to ensure that the calculation of "multi-subject interaction entropy" accurately captures these differences, a weight is assigned to each "interaction pattern kernel IPK". The core principle of this method is to measure the mental, physical, and time stress experienced by humans when performing tasks using a standardized and validated subjective assessment scale (NASA-TLX scale). By correlating the subjective cognitive load scores of subjects under different communication patterns with the frequency of occurrence of those communication patterns in a highly simulated experimental environment, the entropy contribution weight ECW for each pattern can be determined; the specific determination method is as follows: The objective impact weights of different interaction mode kernel (IPK) values ​​on team cognitive load were determined. At least 20 qualified first aid personnel were recruited and randomly divided into 10 pairs. A medical simulation center equipped with high-fidelity simulated patients, real medical equipment, and a complete audio-visual recording system was used. Multiple standardized first aid scenario scripts with increasing difficulty, inducing different types of IPKs (including missing information, conflicting instructions, and process deviations), were designed. Each pair completed all the pre-set simulation scenarios sequentially. All voice and operation logs throughout the process were fully recorded by the system. Immediately after each scenario, each participant was required to independently complete an electronic NASA-TLX Cognitive Load Assessment Questionnaire, which rated the completed task from six dimensions (mental needs, physical needs, time needs, performance, effort, and frustration). Two trained researchers independently coded the "interaction mode kernel" of all scenarios' audio-visual and log records, identifying and marking the time and frequency of each predefined IPK. Inter-coder reliability was calculated to ensure the reliability of the coding. A multiple linear regression model was constructed. The dependent variable of this model is the total NASA-TLX score submitted by the participants, and the independent variable is the frequency of occurrence of each IPK in the scenario. After training the regression model, standardized regression coefficients corresponding to the frequency of each IPK are extracted. The absolute value of this coefficient represents the independent contribution strength of that IPK to cognitive load. All these coefficients are normalized to their maximum value and mapped to the [0,1] interval, and the result is the final "Entropy Contribution Weight ECW".

[0033] 2.3) The Standard Operating Procedures (SOP) library is a database that stores standard procedures for various clinical tasks, including "Cardiopulmonary Resuscitation" and "Severe Trauma Management." Each SOP is represented as an idealized interactive network diagram. These SOP diagrams are drawn by experienced emergency care instructors based on the latest international emergency care guidelines and are stored in multiple "SOP" worksheets within the configuration file.

[0034] Furthermore, 3.1) Traditional risk assessment methods, which simply count the frequency of high-risk words, have two fundamental flaws: first, they ignore the inherent correlation between clinical signs and cannot identify specific critical illnesses pointed to by multiple non-specific signs; second, they are a static assessment and cannot reflect the dynamic evolution of the disease, i.e., the "timeliness" of information. This invention constructs the following hierarchical dynamic calculation mechanism.

[0035] The first layer is entity threat value generation: receiving the clinical entity time-series data stream output by S1. For each entity appearing in the clinical entity time-series data stream, querying the "entity dictionary" in the configuration file to obtain its corresponding "entity basic threat value (ETV)".

[0036] The second layer is Clinical Syndrome Cluster Activation: For each predefined "Clinical Syndrome Cluster (CSC)," its "Cluster Threat Cumulative Value" is calculated in real time. This calculation process is as follows: At the current time point, for each entity within the cluster, if it has been recently mentioned, its decayed threat value is calculated based on the "Threat Half-Life Parameter (THP)." Then, all decayed threat values ​​are summed to obtain the "Cluster Threat Cumulative Value" for that Clinical Syndrome Cluster CSC. This cluster threat cumulative value is compared with the "Cluster Activation Threat (CAT)." If the cluster threat cumulative value is greater than the cluster activation threshold (CAT), the Clinical Syndrome Cluster CSC is marked as "Activated."

[0037] The third layer is the final index normalization output: The "cluster threat cumulative value" of all active Clinical Syndrome Clusters (CSCs) is obtained. The maximum value of these active cumulative values ​​is taken to obtain a raw risk score. This raw risk score is mapped using a sigmoid function to ensure that the output "Predictive Clinical Risk Index (PCRI)," representing the first indicator, is smoothly normalized to the [0,1] interval. The following context-adaptive calculation mechanism for the Team Cognitive Load Index (TCLI) is performed: 3.2) Traditional methods for assessing team cognitive load involve calculating global communication frequency or entropy. Their limitations lie in not considering the specific tasks the team is currently performing, and failing to identify the key communication patterns that have the greatest impact on cognitive load. This invention introduces the concepts of context adaptation and pattern recognition. Its core principle is that the assessment criteria should dynamically change with the task, and the assessment focus should be on micro-interaction patterns that significantly impact collaboration efficiency, specifically including: Receive the identifier of the "Clinical Syndrome Cluster (CSC)" with the highest cumulative threat value, currently in an "active" state, from the output of the PCRI calculation module. Use this CSC identifier as the index for the current "Primary Clinical Task". Based on this index, select and load the corresponding target SOP diagram from the "Standard Operating Procedure Library" as the benchmark for subsequent process compliance calculations.

[0038] 3.21) Calculate Task Flow Compliance: Obtain the real-time interactive network topology graph generated by S2. Execute a graph edit distance algorithm to calculate the structural difference between the real-time interactive network topology graph and the target SOP graph loaded in the previous step. The definition and determination method of the graph edit distance cost function are as follows: The graph edit distance cost function is a set of core algorithm parameters that provide a quantitative evaluation standard for calculating "task flow compliance". Its essence is to transform various operational deviations in the clinical process into calculable numerical costs. In the emergency process, the impact of three deviation behaviors—"omission of critical steps", "execution of incorrect steps", and "increased redundant communication"—on patient safety and team efficiency is completely different. A general, cost-equal graph edit distance algorithm cannot distinguish this difference. Therefore, it is necessary to design and calibrate a non-equivalent cost function that reflects clinical importance. This invention regards the calibration of the cost function as an expert weighting process based on the severity of risk. Its core idea is that the cost of any editing operation (addition, deletion, modification) should be proportional to the degree of negative impact that the corresponding actual operational behavior may cause. To implement the graph editing distance algorithm, this invention predefines the following cost function. All values ​​of this function are determined by the aforementioned expert group through multiple rounds of evaluation and consensus (Delphi method) and stored in a configuration file. The node operation costs are set as follows: Node insertion cost: For nodes that appear in the real-time interactive graph but not in the standard SOP graph, the insertion cost is set to 0.2. The technical consideration for this value is that adding extra operations or communication usually represents redundancy or uncertainty, and should be moderately penalized, but its severity is lower than missing a critical step.

[0039] Node deletion cost: For nodes that exist in the standard SOP graph but not in the real-time interaction graph, their deletion cost is set to 1.0. The technical consideration behind this value is that omitting a step in the standard process is a serious error and must be assigned the highest cost weight. Edge operation costs are set as follows: Edge insertion cost: set to 0.1. This represents the interaction order of non-standard processes, with the lowest severity.

[0040] Edge deletion cost: set to 0.8. This indicates that the interaction required by the standard process failed to occur, representing a more serious process interruption.

[0041] The cost of setting a replacement operation is as follows: Node replacement cost: The cost of node replacement is not a fixed value, but is dynamically calculated based on the attributes of the two nodes being replaced. In this embodiment, each node is assigned a "criticality" attribute (including "chest compressions" as "high criticality" and "delivery equipment" as "medium criticality"). The cost of node replacement is set as the absolute value of the difference in the "criticality" levels of the two nodes multiplied by a base replacement cost of 0.5. This design ensures that replacing a high-criticality operation with a low-criticality operation will result in a higher cost than replacing two operations of the same level.

[0042] Edge replacement cost: The cost of replacing an edge depends on the difference in the "interaction type" (including "instruction" and "feedback") between the two edges.

[0043] By setting the aforementioned cost function, the graph edit distance algorithm can differentiate the penalties for process deviations of different natures during calculation, thereby enabling the final "task process compliance" component to more accurately reflect the true quality of the team's task execution. When calculating the total cost of transforming from a real-time graph to a SOP graph, the calculation logic is as follows: first, identify all insertion, deletion, and replacement operations of nodes and edges that need to be executed; second, for each identified operation, query its corresponding cost value according to the aforementioned rules; finally, sum the costs of all operations to obtain the final, unnormalized graph edit distance. Normalizing this difference yields the process compliance component in the [0,1] interval.

[0044] 3.22) For multi-agent interaction entropy calculation: In the real-time interactive network topology graph, the occurrence count of all predefined "interaction mode kernels (IPKs)" is identified and counted using a subgraph matching algorithm. For each identified IPK, its occurrence count is multiplied by the "entropy contribution weight (ECW)" defined for it in the configuration file. The calculation results of all IPKs are summed to obtain a total raw entropy score, which is then normalized to the [0,1] interval to obtain the interaction entropy component. The fusion weights of the process compliance component and the interaction entropy component are obtained from the configuration file. The two components are weighted and summed to obtain the final "team cognitive load index (TCLI)" representing the second indicator, whose value range is also limited to the [0,1] interval.

[0045] It should be noted that the experimental calibration method for the Team Cognitive Load Index (TCLI) fusion weights is as follows: The TCLI fusion weights balance the contributions of two different dimensions, "process deviation" and "communication confusion," to the overall cognitive load. Their determination process is based on a quantitative analysis of their objective impact on the team's final performance. The goal is to find a set of weights such that the Team Cognitive Load Index (TCLI) calculated from these weights can become the most effective predictor of the team's final objective performance.

[0046] In the aforementioned simulation experiment used to calibrate the entropy contribution weight ECW, an additional data collection dimension was added. Specifically, at least two senior first aid instructors were invited as observers to independently and objectively assess each group's performance in each scenario using a pre-defined "Key Operation Checklist." This score is a value between [0, 100], with higher scores indicating better performance. After the experiment, three core data points were compiled for each simulated scenario instance: the calculated "Task Flow Compliance" component (denoted as Ccomp), the "Multi-Agent Interaction Entropy" component (denoted as Cent), and the average value of the "Team Performance Score TPS." A weight parameter, denoted as Wcomp, was initialized to represent the weight of "Task Flow Compliance," ranging from 0.01 to 1.00 with a step size of 0.01. The other weight, Went, for "Multi-Agent Interaction Entropy," was set to the difference between Wcomp and 1, ensuring that the sum of the two weights is always 1. A loop calculation process was then initiated. In each iteration, using the current set of Wcomp and Went values, for each scenario instance in the dataset, the following calculations are performed: multiply the instance's Ccomp value by Wcomp to obtain the first product; multiply the Cent value by Went to obtain the second product; sum these two products to obtain the trial value of the Team Cognitive Load Index (TCLI) for that scenario. After completing the TLI trial value calculations for all scenario instances, the Pearson correlation coefficient between the TLI trial value sequence and the TPS sequence in the entire dataset is calculated. After traversing all possible Wcomp values, the Wcomp value that maximizes the absolute value of the Pearson correlation coefficient is selected as the final preferred weight. In this embodiment, through the above experiments and optimization process, the preferred weights determined are: a weight of 0.6 for the "Task Process Compliance" component (Wcomp) and a weight of 0.4 for the "Multi-Agent Interaction Entropy" component (Went). This result indicates that, in this technical scenario, whether team behavior deviates from the standard process is a slightly stronger indicator for predicting final task performance than the degree of confusion in communication itself.

[0047] The overall computational flow of this embodiment follows a parallel, ultimately converging data processing architecture. Data acquisition and processing in steps S1 and S2 are performed in parallel. In the first parallel branch, the management platform continuously processes the audio stream and, through a hierarchical dynamic calculation mechanism, outputs a "Predictive Clinical Risk Index (PCRI)" that continuously varies within the [0,1] interval in real time. Simultaneously, the identified major "clinical syndrome clusters" are sent to the second branch as a contextual signal.

[0048] In the second parallel branch, the system synchronously processes various log data and constructs an interactive network. Specifically, it receives context signals from the first branch to determine the current task and selects the correct target SOP benchmark accordingly. Through a context-adaptive calculation mechanism, it integrates two components: task flow compliance and multi-agent interaction entropy, and outputs a "Team Cognitive Load Index (TCLI)" that also varies within the [0,1] interval in real time. These two independent, deeply processed, and dimensionally consistent indices (PCRI and TCLI) are simultaneously input into the diagnostic state mapping model in step S5, serving as the final basis for achieving the core of this invention: systemic risk attribution diagnosis.

[0049] To decouple the core algorithm of this invention from specific application strategies and ensure the configurability and ease of debugging of the technical solution, in the specific implementation path of this invention, all configurable operating parameters, such as the cluster activation threshold CAT, threat half-life parameter THP, and various weights used for fusion calculation mentioned above, are predefined and stored in a structured external data carrier. In a preferred embodiment, this data carrier is a spreadsheet file named Emergency Model Configuration, whose internal structure is organized through multiple worksheets with clearly defined functions. The content of this spreadsheet file is specifically described as follows: It contains a worksheet named "GlobalConfig" for storing global operating configuration parameters. For example, it contains a worksheet named "EntityDictionary," where each row represents a clinical entity, and the columns define the standard terminology, entity type, and "Entity Base Threat Value (ETV)" obtained by the aforementioned method. There is also a worksheet named "ClusterRules" for defining all "Clinical Syndrome Clusters (CSCs)," where each row represents an activation rule, detailing the entity combination logic required to constitute the cluster and the corresponding "Cluster Activation Threat Threat CAT." Similarly, a worksheet named "InteractionPatterns" defines the structure of all "Interaction Pattern Kernels (IPKs)" and their corresponding "Entropy Contribution Weights (ECWs)". Finally, the file also contains several worksheets prefixed with "SOP", such as "SOP-CPR", which fully define the graph structure of the corresponding standard operating procedures in the form of adjacency lists or association matrices. The core output of this technical solution is two normalized key indicators: the Predictive Clinical Risk Index (PCRI) and the Team Cognitive Load Index (TCLI). Both indicators have a value range of [0,1].

[0050] The first indicator is characterized as the Predictive Clinical Risk Index (PCRI): used to quantify in real time the severity of a patient's condition and the risk of deterioration based on on-site voice data. When the output value of the Predictive Clinical Risk Index (PCRI) is closer to 1, it indicates that the patient's current clinical condition is more critical or the degree of deterioration is more severe, and the probability of adverse events (including but not limited to cardiac arrest, need for advanced airway intervention, etc.) is higher. This state is triggered by the system detecting that one or more high-threat clinical syndrome clusters (CSCs) are continuously activated within a short period of time. When the output value of the Predictive Clinical Risk Index (PCRI) is closer to 0, it indicates that the current clinical status of the patient is relatively stable and it is less likely to capture clear evidence pointing to a serious pathological state from the voice information. This state corresponds to the absence of meaningful clinical entities mentioned on site, or the mention of entities having a low base threat value (ETV), or the mention of entities being temporally discrete and failing to form a clinically meaningful cluster, or the relevant mentions exceeding the effective time window defined by the threat half-life parameter (THP).

[0051] The second indicator is represented by the Team Cognitive Load Index (TCLI): used to quantify the overall mental burden, collaboration efficiency, and process standardization of the emergency response team in real time during the execution of tasks. When the output value of the Team Cognitive Load Index (TCLI) approaches 1, it indicates that the team's collaboration is more likely to be in or near a state of disorder, and cognitive resources are being over-consumed. This state is the result of two or more of the following situations: First, there is a large deviation between the team's actual operating procedures and the standard operating procedures (SOPs) corresponding to the current clinical task; Second, communication among team members is filled with a large number of high-entropy contribution weight ECW interaction pattern kernels (IPKs), such as conflicting instructions, repeated questioning, and lack of information confirmation, reflecting poor information exchange.

[0052] When the output value of the Team Cognitive Load Index (TCLI) approaches 0, it indicates that the team's collaboration is more efficient, smooth, and standardized, and the overall cognitive load is at a lower ideal level. This state corresponds to the team's operational behavior being highly consistent with the Standard Operating Procedure (SOP), and the communication mode being mainly composed of the interaction mode kernel IPK (including "instruction-confirmation") with low entropy contribution weight ECW, reflecting a high degree of tacit understanding and shared situational awareness among team members.

[0053] Analysis of the Impact of Entity Base Threat Value (ETV) on Predictive Clinical Risk Index (PCRI): A clear positive correlation exists between Entity Base Threat Value (ETV) and the Predictive Clinical Risk Index (PCRI). According to a detailed embodiment of the present invention, the calculation basis of PCRI is the "cluster threat cumulative value" of the Clinical Syndrome Cluster (CSC). This cumulative value is calculated by summing the time-decayed ETVs of all recently mentioned entities within the cluster. Therefore, when other conditions (mention frequency, time) remain constant, if the ETV setting value of one or more entities is higher, the threat value contributed by each mention will be greater, directly leading to an increase in the "cluster threat cumulative value." When this cumulative value exceeds the cluster activation threshold (CAT), its increase will, through a subsequent normalization function, ultimately lead to a monotonically increasing PCRI. This positive correlation design accurately maps real-world clinical logic. ETV itself is a quantitative representation of the intrinsic severity of a clinical entity, determined through large-scale data analysis and expert consensus. An entity with a higher ETV (including but not limited to "no pulse") should have a greater impact on the final risk assessment than an entity with a lower ETV (including but not limited to "dizziness").

[0054] An analysis of the impact of the threat half-life parameter THP on the predictive clinical risk index PCRI revealed a positive correlation between the magnitude of the threat half-life parameter THP and the duration and peak height of the predictive clinical risk index PCRI. THP defines the rate at which the ETV decays over time. A larger THP value (300 seconds vs. 180 seconds) indicates a slower decay of the threat value. When calculating the "cluster threat accumulation value," a system with slower decay results in entities mentioned earlier maintaining a higher threat value for a longer period. This leads to two consequences: first, the threat value is more likely to accumulate above the cluster activation threshold; second, it takes longer for the accumulated value to fall below the threshold after entity mentions cease. Therefore, increasing the THP value makes the PCRI more durable in response to clinical events and more likely to reach a higher peak value in the case of continuous mentions. This can be adapted to different types of clinical events. For events with rapid disease progression (anaphylactic shock), a shorter THP can be configured; while for events with relatively slow progression (acute exacerbation of heart failure), a longer THP can be configured. This parameterized time decay mechanism enables the risk assessment model of this invention to adapt to different pathophysiological processes, which is superior to statistical methods with fixed time windows.

[0055] Analysis of the Impact of Entropy Contribution Weight ECW on Team Cognitive Load Index TCLI: A clear positive correlation exists between the entropy contribution weight ECW and the team cognitive load index TCLI. The calculation of TCLI includes a "multi-agent interaction entropy" component. The calculation logic for this component involves multiplying the frequency of occurrence of various interaction pattern kernels (IPKs) identified in the interaction network with their respective preset ECWs, and then summing the results. ECW plays a direct weighting coefficient role in this calculation. When other conditions remain unchanged, increasing the ECW value of an IPK will lead to a greater contribution to the "multi-agent interaction entropy" component each time that IPK occurs, thus directly causing a monotonically increasing TCLI. This design accurately reflects the objective fact that different communication behaviors consume different amounts of team cognitive resources. ECW is the "cognitive cost" of different communication patterns, calibrated through human factors engineering experiments. Quantifying this cost and introducing it as a weight in the calculation makes TCLI no longer simply a measure of the "quantity" of communication, but rather a measure of the "quality" of communication. This gives the cognitive load assessment of this invention a deeper organizational behavioral connotation compared to traditional network centrality or density analysis.

[0056] Analysis of the Impact of Standard Operating Procedure (SOP) Deviation on Team Cognitive Load Index (TCLI): A clear positive correlation exists between the structural deviation between the real-time interactive process and the selected SOP and the Team Cognitive Load Index (TCLI). The calculation of TCLI includes a "task flow conformity" component. This component is determined by calculating the graph edit distance between the real-time interactive network diagram and the baseline SOP diagram. Graph edit distance is an indicator that quantifies the structural differences between two graphs; the larger the distance, the greater the difference, i.e., the higher the deviation. This deviation, after normalization, is directly used as a weighting factor in the final calculation of TCLI. Therefore, when the difference between the team's actual operating process and the standard process increases, the graph edit distance increases accordingly, leading to an increase in the "task flow conformity" component value, ultimately causing a monotonically increasing TCLI. This design is based on the fact that deviation from standard operating procedures is one of the main causes of errors and inefficiency. When a team deviates from the established process, it indicates unexpected situations, insufficient knowledge, or internal coordination problems, all of which increase the team's cognitive load. By introducing graph edit distance to quantify this deviation and directly linking it to cognitive load, we can achieve an objective and automated assessment of the standardization of team process execution.

[0057] This embodiment will elaborate on the core technical features and further limitations of step S5. Its fundamental purpose is to solve the technical problem of how to transform two independent, continuously changing quantitative indicators into a qualitative diagnostic conclusion with clear semantics that can accurately reflect the overall operating state of the system and directly guide intervention actions. This requires understanding the essence of the current state, predicting its future evolution trend, and generating the most effective response strategy within a dynamically evolving two-dimensional state space.

[0058] Further explanation: The diagnostic state mapping model is configured to map a two-dimensional coordinate pair consisting of a first indicator and a second indicator to one of at least four predefined diagnostic states: efficient response state, controllable challenge state, process failure state, and system overload state.

[0059] In a preferred embodiment, the diagnostic state mapping model is a two-dimensional lookup table stored in a spreadsheet file or a decision boundary model trained using a support vector machine (SVM). The model constructs a two-dimensional state space with the Predictive Clinical Risk Index (PCRI) (first indicator) as the X-axis and the Team Cognitive Load Index (TCLI) (second indicator) as the Y-axis. When a set of (PCRI, TCLI) values ​​is received, the diagnostic state mapping model determines which region defined by the decision boundary it falls into, and thus outputs the corresponding diagnostic state.

[0060] Further explanation: The diagnostic state mapping model is further configured as follows: in the state space composed of two-dimensional coordinate pairs, a potential field center is set for each predefined diagnostic state; the mapping process includes calculating the potential energy from the current two-dimensional coordinate pair to each potential field center, and mapping the current state to the diagnostic state with the lowest potential energy to generate a diagnostic state identifier.

[0061] Further explanation: The diagnostic status signal includes at least: a diagnostic status identifier, and a confidence score for quantifying the diagnostic confidence; wherein, the confidence score is calculated based on the relative potential energy relationship between the current two-dimensional coordinate pair and the center of the potential field.

[0062] Further explanation: The diagnostic state mapping model also includes a state transition probability module, which is configured to calculate the state transition probability to each other predefined diagnostic state based on the current diagnostic state and the historical trajectory of the two-dimensional coordinate pair in the state space; the diagnostic state signal further includes the state transition probability.

[0063] Further explanation: The diagnostic state mapping model also includes an interface configuration instruction generation step, which is configured to: match and generate an interface configuration instruction for adaptively adjusting the extended reality (XR) user interface based on the diagnostic state identifier and state transition probability contained in the diagnostic state signal; wherein, when the state transition probability to a worse state exceeds a preset state transition alarm threshold, the interface configuration instruction is configured to trigger the presentation of a predictive alarm interface element.

[0064] The following is a detailed implementation description of the above content: In this embodiment, all configurable parameters are predefined and stored in a structured external spreadsheet file called Diagnosis and Intervention Configuration, which is read into memory by a data loading module during system initialization.

[0065] 3.1) The center coordinates of the diagnostic state potential field are designated as PFC: This is a two-dimensional coordinate value used to define the most stable "gravity center" for each predefined diagnostic state in the two-dimensional state space composed of the Predictive Clinical Risk Index (PCRI) (X-axis) and the Team Cognitive Load Index (TCLI) (Y-axis). The center point of the "Efficient Response State" is located in the low-risk, low-load region. The core idea of ​​this embodiment is that, among a large number of historical data points that have been labeled by experts, the distribution of data points belonging to the same state category in the state space is not completely random, but will naturally cluster towards one or more center points. By unbiasedly finding the center of these clusters, i.e., the "centroid," it is used as the center coordinates (PFC) of the diagnostic state potential field for that state. For the four predefined diagnostic states of "Efficient Response," "Controllable Challenge," "Process Failure," and "System Overload," a unique and most representative coordinate point is determined for each in the two-dimensional space composed of the Predictive Clinical Risk Index (PCRI) and the Team Cognitive Load Index (TCLI). This embodiment uses the K-Means unsupervised clustering algorithm. The training dataset is described in detail below: It is a dataset used to determine the center coordinates (PFC) of the diagnostic state potential field. It contains a proprietary corpus of complete pre-hospital emergency care cases. For each case in the corpus, the following data preparation steps were performed: The complete audio and operation logs of each case are input into the processing pipeline described in steps S1 to S4, generating a continuous sequence of PCRI and TCLI numerical values ​​aligned with timestamps, forming (PCRI, TCLI) data points. An expert review panel of at least three members is organized, including a senior emergency physician, a senior emergency care specialist, and a human factors engineering expert. The expert panel retrospectively analyzes the complete process video and data for each case and independently annotates them according to a pre-defined, standardized "system status assessment criterion." Finally, through majority voting or consensus, a unique diagnostic status label is assigned to the stable stage or final outcome of each case. This invention further introduces a quantitative quality control mechanism for inter-rater reliability (IRR) in the annotation process. The specific implementation process is as follows: Before formal annotation begins, 5% of the samples are randomly selected from the corpus as a reliability test set. The three members of the expert panel perform completely independent, unannounced preliminary annotations on this test set. Subsequently, the Fleiss-Kappa statistical method was used to calculate the consistency coefficient of the annotation results among the three experts. Fleiss-Kappa is a well-established metric for evaluating the degree of consistency among two or more evaluators when classifying nominal variables. In this embodiment, a clear quality threshold was set: the calculated Kappa value must be greater than 0.75, which is considered a level of consistency from "good" to "excellent." Only after this standard is met is the expert group authorized to continue with subsequent large-scale formal annotation. If the Kappa value is lower than 0.75 in the initial test, the expert group must reconvene to clarify and refine the ambiguous clauses in the "System State Evaluation Criteria" and conduct a second round of reliability testing until it passes. For the few cases of disagreement that arise during the formal annotation process, an arbitration and consensus process involving a fourth, more senior domain authority expert is initiated to finalize the labels. The annotated dataset is randomly divided into an 8:2 ratio into a PFC determination set (80%) and a subsequent model validation set (20%). For the data in the PFC determination set, for each diagnostic state label (e.g., processing all case data points labeled "system overload"), independently perform the following K-means calculation procedure, setting K to 1 since the goal is to find a center for each state: Obtain the (PCRI, TCLI) coordinates of all data points belonging to the current diagnostic state label. Calculate the arithmetic mean of the PCRI values ​​of all these data points to obtain the X-coordinate of the PFC. Calculate the arithmetic mean of the TCLI values ​​of all these data points to obtain the Y-coordinate of the PFC. Combine these two arithmetic means into a two-dimensional coordinate system, which is then determined as the diagnostic state potential field center coordinates (PFC) for that diagnostic state.Repeat this process for all four diagnostic states to obtain four unique PFC coordinates.

[0066] 3.2) State-space distortion coefficient, denoted as SDC: It represents the distortion coefficient of a dimensionless two-dimensional parameter vector containing two components, the X-axis (PCRI) and the Y-axis (TCLI), denoted as SDCx and SDCy respectively. It is used to asymmetrically weight distances along different axes when calculating potential energy, reflecting the different importance of different indicators in risk assessment. When constructing the potential field model, if standard Euclidean distance is used, it implicitly assumes that unit distances on the PCRI and TCLI axes have equal importance. However, this is not a valid assumption in real emergency risk assessment. The deterioration of the patient's vital signs (increased PCRI) and friction in teamwork (increased TCLI) contribute differently to the overall system risk. This invention introduces the state-space distortion coefficient SDC, the core idea of ​​which is that attributes in space vary with direction. By assigning different weight coefficients to the PCRI and TCLI axes respectively, a "distorted" state space that better reflects the true risk contribution is constructed, thereby improving diagnostic accuracy. The specific determination steps are as follows: An optimal SDC parameter vector (SDCx, SDCy) is determined such that the diagnostic model achieves the highest classification accuracy when dealing with a set of expert-recognized fuzzy cases at state boundaries. From the validation set (20%) of the aforementioned dataset, an expert panel further selects one hundred "boundary cases." These cases are characterized by their (PCRI, TCLI) trajectories lingering for extended periods within the theoretical boundary regions of two or more diagnostic states, representing samples most prone to misclassification by the model. The complete diagnostic model is then deployed, with SDC set as an adjustable parameter.

[0067] The baseline value of SDC is set to (1.0, 1.0), representing an isotropic space without distortion. Using the distortion coefficient SDCy of the TCLI axis fixed at 1.0 as the baseline, the distortion coefficient SDCx of the PCRI axis is used as the sole variable, and scanning is performed within a preset interval. In this embodiment, this interval is set from 1.0 to 2.0, with a scan step size of 0.1. For each SDCx value (1.0, 1.1, 1.2...) within the scan interval, the following sub-processes are executed: The current SDC value (1.1, 1.0) is configured into the diagnostic model; complete data from all one hundred "boundary cases" are input into the model; the model's diagnostic output for each case is recorded and compared with the expert-annotated "gold standard"; the overall diagnostic accuracy under this SDC configuration is calculated. The previous step is repeated until all SDCx values ​​have been scanned. A two-dimensional curve is plotted between each SDCx value and its corresponding diagnostic accuracy. The curve is observed to identify the point where the accuracy reaches its peak. In the calibration experiment of this embodiment, it was observed that the diagnostic accuracy reached its peak when SDCx was 1.5. Therefore, the state-space distortion coefficient SDC ultimately adopted in this invention was determined to be (1.5, 1.0).

[0068] 3.3) State transition alarm threshold, denoted as STAT: This represents a dimensionless value between 0 and 1, used to determine whether a potential state deterioration trend needs to be warned through the XR interface. The predictive alarm function of this invention relies on a threshold to determine when to report a potential "deterioration trend" to the user. This is a binary classification decision problem, the core challenge of which lies in balancing two types of errors: missed detection (i.e., failure to warn of a real deterioration trend, which may lead to catastrophic consequences) and false alarm (i.e., misjudging random fluctuations as deterioration trends, leading to "alarm fatigue" and causing users to lose trust in the system). This embodiment uses a systematic and quantifiable method to determine this decision threshold. This invention uses Receiver-Operating-Characteristic (ROC) analysis to determine this state transition alarm threshold. ROC analysis is the gold standard in signal detection theory and diagnostic testing evaluation. By systematically evaluating all possible decision thresholds and plotting the relationship curve between the "true positive rate" (sensitivity) and the "false positive rate" (1-specificity), it provides an objective basis for selecting the optimal threshold. On the ROC curve, find an optimal equilibrium point; the probability threshold corresponding to this point is the state transition alarm threshold (STAT). The specific determination steps are as follows: The validation set (20%) of the aforementioned dataset was used. This validation set was then re-annotated by an expert panel to specifically identify and mark all "time points when the state began to deteriorate." This included a case where a "controllable challenge" eventually evolved into a "system overload," where experts would mark the timestamps when the system state began to show an irreversible deterioration trend. These markers constituted "positive samples." A candidate threshold variable was set, with a scan range of 0.01 to 0.99 and a step size of 0.01. For each candidate threshold, the following evaluation procedure was performed: the candidate threshold was set as the model's temporary STAT. All case data from the validation set were input into the model. Two key values ​​were calculated: the first was the total number of "true positives," i.e., the number of times the time point when the model issued an alarm (the calculated transition probability was greater than the current candidate threshold) matched the time points of "positive samples" marked by experts (within a preset time window). The second was the total number of "false positives," i.e., the number of times the model issued an alarm, but there were no nearby positive samples marked by experts. Based on the above statistics, the current "true positive rate" (total number of true positives divided by the total number of positive samples labeled by experts) and "false positive rate" (total number of false positives divided by the total number of all non-positive time points) are calculated. The previous step is repeated until all candidate thresholds have been evaluated. The coordinates of (false positive rate, true positive rate) corresponding to all candidate thresholds are plotted on a two-dimensional graph to form an ROC curve. To select the optimal point from the curve, this invention uses the "Yorden index" as an evaluation metric. The Yoden index is calculated as follows: for each point on the curve, its true positive rate is subtracted from its false positive rate. The point that maximizes the Yoden index is found. In this embodiment, the Yoden index is maximized when the probability threshold is 0.65. Therefore, the state transition alarm threshold STAT is ultimately determined to be 0.65. In this embodiment, the optimal state transition alarm threshold STAT determined by the Yoden index is 0.65, but in actual deployment, the value of this parameter is adjusted within a preferred range to adapt to specific monitoring strategies. This preferred numerical range is determined to be between 0.60 and 0.75. The determination of this range is based on the analysis of the ROC curve generated in this embodiment. All threshold points within this range exhibit a high true positive rate and a low false positive rate, indicating stable system performance. Specifically, lowering the state transition alarm threshold STAT will increase the alarm sensitivity, making it suitable for high-risk scenarios where no risk can be overlooked; conversely, raising the state transition alarm threshold STAT (approaching 0.75) will decrease the sensitivity, making it suitable for scenarios where users need to focus highly on core tasks.

[0069] 3.4) The system refresh rate is denoted as SRR: This parameter defines the time interval at which the entire diagnostic and decision-making process described in step S5 of this invention is repeatedly executed. It directly determines the system's response speed and time resolution to changes in the on-site state. In this embodiment, the preferred value of the system refresh rate is set to 5 seconds. Its value ranges from 3 seconds to 10 seconds. The technical consideration for setting this value is to achieve a technical balance between ensuring clinically meaningful timeliness and avoiding unnecessary consumption of computing resources. On the one hand, the typical time scale of critical state changes in the emergency process (including significant deterioration of vital signs and severe interruption of team communication) is on the order of tens of seconds to several minutes, and a refresh rate of less than 10 seconds is sufficient to capture these meaningful trends; on the other hand, too frequent refreshes will not only cause instability in the state velocity vector calculation due to random noise in the input data, but also cause unnecessary continuous load on the system's processor and network bandwidth. Therefore, the 5 seconds set in this embodiment is considered to be a best practice value that takes into account both clinical needs and engineering feasibility.

[0070] 3.5) The state history queue is denoted as SHQ: It is a fixed-length data structure used to store the coordinates of the most recent N diagnostic states. Its function is to provide stable and reliable historical data input for calculating the state velocity vector. In this embodiment, the state history queue is implemented as a first-in, first-out queue, and its queue depth (i.e., maximum storage capacity) is set to 2. This means that the queue always stores the state coordinate pairs of the current moment and the immediately preceding moment. Its data update operation process is as follows: When the system refresh rate SRR arrives each time, after a new state coordinate pair is calculated, the coordinate pair is pushed to the tail of the queue; at the same time, if the number of elements in the queue exceeds the set depth of 2, the oldest element is removed from the head of the queue. When calculating the state velocity vector, all two elements are extracted from the queue, and the oldest element is subtracted from the newest element. The technical consideration of setting the depth to 2 is to minimize data storage overhead and computational complexity while providing the most direct historical information required to calculate instantaneous velocity.

[0071] To further explain, the core idea of ​​this invention is state transition prediction, specifically including: receiving the predictive clinical risk index (PCRI) value and the team cognitive load index (TCLI) value at the current moment from steps S3 and S4, forming a current state coordinate pair. For each predefined diagnostic state ("efficient response", "system overload", etc.), the following operations are performed: obtaining the diagnostic state potential field center coordinates (PFC) of that diagnostic state. Then, calculating the distorted Euclidean distance between the current state coordinate pair and the PFC. The calculation process is as follows: obtaining the difference between the current PCRI and the X coordinate of the PFC, multiplying this difference by the X component of the state space distortion coefficient (SDC), and then squaring it; similarly, obtaining the difference between the current TCLI and the PFC by the Y coordinate, multiplying this difference by the Y component of the SDC, and then squaring it; finally, adding these two squared values. This result is the "potential energy" from the current point to the PFC. Comparing the "potential energy" values ​​from the current point to all four PFCs, selecting the diagnostic state with the smallest potential energy value, and using its identifier as the "diagnostic state identifier" at the current moment. It should be further explained that since the Predictive Clinical Risk Index (PCRI) and Team Cognitive Load Index (TCLI) output from the preceding steps S3 and S4 are both normalized to the range of zero to one, the theoretical range of the two coordinate components of the diagnostic state potential field center coordinates (PFC) determined by the above clustering method should also be between 0 and 4. In historical data statistics and validation, the preferred value range for these center coordinates was determined to be between 0.1 and 0.9. The potential energy values ​​from the current point to all four PFCs are summed to obtain a total potential energy. The determined minimum potential energy value is obtained. This minimum potential energy value is subtracted from the total potential energy, and then the result is divided by the total potential energy. This calculation result is the "diagnostic confidence score," and its value range is normalized to the [0,1] interval. The closer a point is to its corresponding potential field center, and the farther it is from other centers, the higher its confidence score.

[0072] Obtain the current state coordinate pair and the previous state coordinate pair from the state history queue. Calculate the difference between the two time points on the PCRI (X-axis) and TCLI (Y-axis) axes to form a two-dimensional velocity vector, which represents the direction and speed of movement of the state point in the state space. For each target diagnostic state other than the current home state, perform the following operations: Calculate the direction vector from the current coordinate point to the target state PFC. Then, calculate the dot product of the "state velocity vector" and the "direction vector". The result of this dot product represents the consistency between the movement trend of the current state point and the trend towards the target state. Finally, normalize all these dot product results using the Softmax function to generate a set of probability values, each representing the probability that the system state will transition to the corresponding target state in the next time step. Specifically, perform a multi-valued normalized probability transformation process on all these dot product results to generate a set of probability values ​​that sum to 1. The specific calculation steps of this process are as follows: Obtain multiple dot product values ​​representing the trend of the current state point towards each target diagnostic state; independently calculate the natural exponent value for each dot product value to obtain a set of intermediate exponent results; sum all the intermediate exponent results to obtain a sum normalization factor; divide each intermediate exponent result by this sum normalization factor to obtain its corresponding final transition probability value, which is between 0 and 1. This set of final transition probability values, with a constant sum of 1, quantifies the relative probability of the system evolving towards each potential state in the next time step.

[0073] The "Diagnostic Status Identifier," "Diagnostic Confidence Score," and "State Transition Probability" sets are encapsulated into a structured "Diagnostic Status Signal" data object. This signal is then passed to a rule engine module. This module queries the "Enhanced Intervention Rule Base" in the diagnostic and intervention configuration file. This rule base contains a series of "IF-THEN" rules. A rule is defined as: "IF 'Diagnostic Status Identifier' equals 'Process Failure State' AND the probability in its 'State Transition Probability' pointing to 'System Overload State' is greater than the state transition alarm threshold STAT; THEN generate the corresponding interface configuration instruction." The rule engine matches applicable rules and generates the final "Interface Configuration Instruction."

[0074] In this embodiment, step S5, as a core processing unit, is executed cyclically every 5 seconds with a refresh rate. It continuously receives real-time PCRI and TCLI data streams from S3 and S4, and outputs a diagnostic status signal containing rich information through the aforementioned dynamic potential field diagnosis and prediction model, and further generates specific instructions to drive the XR interface.

[0075] Examples include: when the input is consistently low PCRI=0.1 and low TCLI=0.2, the state point remains stably near the potential center of the "efficient response state". The model will consistently output a high-confidence "efficient response" diagnosis, and all state transition probabilities will be close to zero.

[0076] If the initial state is "Controllable Challenge" (PCRI=0.6, TCLI=0.3), the PCRI rapidly increases to 0.9 within 10 seconds, while the TCLI remains unchanged. The state point moves rapidly to the right in space. The diagnostic state mapping model switches the diagnostic state to "System Overload," and the state velocity vector points to the right. Since the direction of movement does not point to other PFCs, the state transition probability module calculates that the probability of transitioning to other states is lower.

[0077] If the system is in a "Process Failure" state (PCRI=0.4, TCLI=0.8), team communication is chaotic, but the patient's vital signs are still manageable. Subsequently, the PCRI begins to slowly but steadily climb. The state point starts moving towards the "System Overload" PFC in the upper right corner. At this point, although the immediate diagnosis is still "Process Failure," the state transition probability module detects this clear trend and calculates that the probability of transitioning to "System Overload" will continue to increase. Once this probability exceeds the state transition alarm threshold STAT, a predictive alarm is triggered on the XR interface, providing an early warning that the current team problem is about to cause the patient's condition to spiral out of control.

[0078] Furthermore, in order to decouple the core algorithm of this invention from the specific application strategy, and to ensure the configurability and ease of debugging of the technical solution, in the specific implementation path of this invention, all configurable running parameters are predefined and stored in a local spreadsheet file.

[0079] The following is a detailed implementation description of the above content: One of the most critical outputs of this embodiment is the "probability of transitioning to a worse state", which has a value range of [0,1]. This probability value is not a simple repetition of the current state, but a forward-looking quantification of the dynamic evolution trend of the system in the next time step.

[0080] As the probability of transitioning to a worse state approaches zero, it indicates that the diagnostic state mapping model of this invention judges the current system's operating trajectory to be more stable, or is evolving towards a better state (including recovery from "process failure" to "controllable challenge"). In this state, the system not only has controllable risks, but also no inherent momentum for deterioration.

[0081] As the probability of transitioning to a worse state approaches 1, the characterization diagnostic state mapping model indicates that the current system's trajectory exhibits a stronger momentum to evolve towards a worse state (including sliding from "controllable challenge" to "system overload"). This high probability value not only reflects the current state's location but, more importantly, captures that its "velocity vector" points to a high-risk region, indicating that the system's state is about to deteriorate.

[0082] The core inputs influencing the "state transition probability" are a two-dimensional coordinate pair consisting of the predictive clinical risk index PCRI and the team cognitive load index TCLI, along with their historical trajectories.

[0083] When other parameters remain constant, an increase in PCRI or TCLI, if causing the current two-dimensional coordinate pair to move closer to the "potential center" of a worse state in the state space, will decrease the "potential energy" of that coordinate pair towards that potential center, thus leading to a positively correlated increase in the probability of transitioning to that worse state. In this embodiment, the potential field model design places the potential center of high-risk states such as "system overload" in the high-value region of the state space. Therefore, an increase in PCRI (representing patient risk) or TCLI (representing team load) logically causes the system to move towards a high-risk region.

[0084] It should be further explained that the "state velocity vector" is a vector defined in a two-dimensional "diagnostic state space." The two orthogonal coordinate axes of this space are formed by the Predictive Clinical Risk Index (PCRI) and the Team Cognitive Load Index (TCLI), respectively. At any time t, the current state is described by a unique state position vector P(t), with coordinates (PCRI(t), TCLI(t)). The essence of the state velocity vector V(t) is the rate of change of the state position vector P(t) with respect to time. It quantitatively describes the direction and speed of the system state's movement in the diagnostic state space. In discrete-time systems, the state velocity vector V(t) is approximated by performing a difference operation on the state position vectors of two consecutive time steps. The specific calculation logic is as follows: Obtain the state position vector P(t) at the current time step t, with coordinates (PCRI(t), TCLI(t)). Obtain the state position vector P(t-Δt) at the previous time step t-Δt, with coordinates (PCRI(t-Δt), TCLI(t-Δt)).

[0085] The calculation process of the state velocity vector V(t) is to subtract the previous state position vector from the current state position vector.

[0086] The x-component (PCRI direction) is calculated as: PCRI(t) minus PCRI(t-Δt).

[0087] The y-component (TCLI direction) is calculated as: TCLI(t) minus TCLI(t-Δt).

[0088] The final state-velocity vector is V(t) = (PCRI(t) - PCRI(t-Δt), TCLI(t) - TCLI(t-Δt)).

[0089] The magnitude (modulus) and direction of the state velocity vector together reveal the dynamic trend of the system state: The magnitude of the vector ||V(t)|| represents the rate of change of the system state. A large magnitude indicates that PCRI and / or TCLI have changed drastically in a short period of time, suggesting that the system is in an unstable and rapidly evolving phase. Conversely, a magnitude close to zero indicates that the system state is stable.

[0090] Pointing to high-risk regions: If the vector points to the region where the potential center of high-risk states such as "system overload" or "process failure" is located (the upper right quadrant where both PCRI and TCLI values ​​are high), it indicates that the system is evolving towards a worse state. The state transition probability module in the model captures this trend and increases the probability of transitioning to a worse state.

[0091] Pointing to low-risk regions: If the vector's direction points to the region where the potential center of a low-risk state, such as "efficient response," is located (the lower left quadrant), it indicates that the system is on a recovery or improvement trajectory. The model will then reduce the probability of transitioning to a worse state.

[0092] Zero vector: When the vector is zero, it indicates that the system state is "stationary" at the current position, with no trend of deterioration or improvement. Even if the current position is in a high-load region, the model will judge that the risk of deterioration is low due to its "stationary" characteristic.

[0093] The direction of the "state velocity vector" is crucial in determining predictability. When this vector's direction aligns with the "direction vector pointing to the center of a worse state" (i.e., their dot product is positive), the "state transition probability" is enhanced; conversely, when the velocity vectors are in opposite directions (the dot product is negative), the transition probability is suppressed. This design is the core difference between this system and traditional static thresholding systems. A system based solely on the current position cannot distinguish between two drastically different scenarios: "high load but improving conditions" and "medium load but rapidly deteriorating conditions." By introducing the state velocity vector, representing "momentum" and "trend," the model gains predictive capabilities. A strong velocity vector pointing to a high-risk region is a clear signal that the system is about to become unstable and is given high weight.

[0094] Furthermore, to quantitatively verify the superiority of the mapping model of this invention compared to existing technologies, a computational experimental environment was constructed in this embodiment. This environment generated a series of PCRI and TCLI time-series data representing typical emergency rescue scenarios using numerical methods, and input these data into the mapping model of this invention and an existing technology model serving as a control group. The control group adopted a common "static dual-threshold alarm model," whose alarm logic is: when PCRI exceeds 0.8 or TCLI exceeds 0.85, an alarm is immediately triggered. The table below shows the output comparison of the two models under three typical scenarios. Each scenario includes multiple consecutive time steps to observe dynamic changes, as detailed in Table 1 below: Table 1: Performance Comparison of the Invention Model and Existing Technology Models in Typical Scenarios

[0095] In Table 1, the early warning lead time is denoted as WLT; WLT = (the time step in which the system state actually reaches the "alarm" condition) - (the time step in which the model first issues an effective early warning). The larger the WLT value, the stronger the model's predictive ability. This parameter is used to quantify how many units of time the model can issue a warning compared to the actual occurrence point of the event, and is a key indicator for measuring the core value of predictive systems.

[0096] In contrast to scenario 1 (stable high load): at times T2 and T3, both PCRI and TCLI consistently exceeded the alarm thresholds of existing technologies, causing continuous alarm triggering. However, since the values ​​remained unchanged, the "state velocity vector" calculated by the model of this invention was close to zero, thus the "deterioration transition probability" remained at a low level of 0.15, far below the alarm threshold of 0.65. This data demonstrates that this invention can effectively suppress false positive alarms caused by the system being in a stable high load state. Compared to existing technologies, this invention reduces user "alarm fatigue," allowing them to focus more on handling practical problems rather than being distracted by invalid information.

[0097] Comparison Scenario 2 (Rapid Deterioration): This set of data decisively demonstrates the predictive advantage of this invention compared to existing technologies, an advantage that can be precisely quantified through the "Warning Lead Time (WLT)". The detailed demonstration process is as follows: Define the event occurrence point Tevent: In this experiment, an "event" is defined as the alarm condition in which the system state first reaches the existing technical model. According to the table data, at time T3, the current PCRI is 0.88 (>0.8), and the current TCLI is 0.89 (>0.85). Therefore, Tevent is determined to be T3. "0.88 (>0.8)" means that 0.88 is greater than 0.8, and the same applies below. The "probability of deterioration transition" of the model in this invention reaches 0.71 at time T2, exceeding the alarm threshold of 0.65, therefore its alarm time point Twarn is T2. Based on this, the calculation is performed: obtain the event occurrence point Tevent=T3 and the alarm time point Twarn=T2 of this invention; then, calculate the difference in their time steps. This difference is 1 time step. Therefore, WLT=1 time step. The prior art model first triggers an alarm at time T3, its alarm time point is T3. Based on this, the calculation is performed: obtain the event occurrence point Tevent=T3 and the alarm time point T3 of the prior art; then, calculate the difference in their time steps. This difference is 0 time steps. Therefore, the warning lead time is 0 time steps. The comparison results show that the warning lead time of this invention is 1 time step, while that of the prior art is 0. This indicates that the prior art is a purely "reactive" system, only reacting when the situation reaches a critical point. This invention is a "predictive" system. In this embodiment, if each time step is set to 5 seconds, a WLT of 1 time step means that the present invention can predict the impending deterioration of the system state 5 seconds earlier than the prior art, which is the key to realizing the transformation from "passive response" to "active prevention".

[0098] Comparison Scenario 3 (High-Risk Recovery): At time T2, the system is at a high risk level, with PCRI and TCLI still above the threshold, and existing technologies continue to issue warnings. The model of this invention detects a slight recovery trend at this time (the velocity vector begins to point towards the low-risk zone), and the probability has dropped to 0.25. At time T3, the recovery trend is more pronounced, and existing technologies continue to issue warnings because TCLI is still above the threshold, while the probability of deterioration transition in this invention has further decreased to 0.12, confirming the benign trend of the system. This data demonstrates that this invention can more quickly eliminate unnecessary warnings during system recovery from the high-risk zone, providing users with more accurate and timely status feedback.

[0099] To transform the "state transition probability to a worse state" (hereinafter referred to as Pworsen) output by the model into specific, operable instructions that drive the XR interface, this invention defines the following three-level application range based on data analysis and expert experience: Within the safety monitoring range (Pworsen ∈ [0, 0.30)): the system operates stably with no visible signs of deterioration. No active intervention is required. The XR interface maintains its baseline state, recording data only in the background to ensure a highly concise user interface.

[0100] Trend focus range, Pworsen ∈ [0.30, 0.65): The system shows a slight, noteworthy deterioration trend, but has not yet reached the level of urgency requiring immediate intervention. Initiate a primary interface prompt. Generate interface configuration instructions to gradually transition the color of the core status indicator in the XR interface from the usual green to yellow. This operation aims to convey the implicit message of "remain vigilant" to the user in a non-intrusive manner.

[0101] Predicted alarm range, Pworsen ∈ [0.65, 1.0]: The system's deterioration trend is clear and strong, and it will enter a worse operating state in the short term. Initiate a secondary predictive alarm. Generate interface configuration instructions to trigger one or more highly significant predictive alarm interface elements. This includes: switching the core status indicator's color to red and giving it a slight flashing or pulsating effect; rendering a semi-transparent arrow or highlighted area pointing to the root cause of the deterioration (including specific physiological abnormalities) at the user's field of vision; and triggering the haptic feedback module associated with the XR device to provide the user with clear tactile cues.

[0102] The computational logic involved in this application can be constructed using algorithms such as regression analysis in machine learning, establishing a mathematical model by analyzing the inherent trends and interrelationships of the collected parameters. This process can be implemented using specialized computational tools (such as Python's Scikit-learn library or the R language environment). Throughout all calculations, to eliminate the influence of different physical dimensions and ensure that data is compared and analyzed on the same scale, the input parameters in each formula are dimensionless. The dimensionless techniques used include, but are not limited to, max-min normalization or Z-score standardization.

[0103] The algorithm of this invention is implemented as a Python script. Before executing the core logic, the program first executes a data loading module (e.g., using the widely used pandas library in Python) configured to read the aforementioned spreadsheet file and load its contents into the program's working memory (e.g., a DataFrame data structure). Subsequent algorithm steps will directly query and retrieve the required configuration parameters from this in-memory data structure.

[0104] It should be emphasized that the foregoing embodiments are merely illustrative of preferred implementations of the present invention and are not intended to limit the scope of protection of the present invention. This application also provides a computer-readable storage medium having computer program instructions stored thereon.

Claims

1. A real-time visualization and intelligent operation management platform for emergency resuscitation based on virtual reality, characterized in that: Specifically, it includes: S1: Acquire clinical entity time-series data representing the patient's condition, generated by natural language processing of voice data at the emergency scene within a preset time window; S2: Obtain interactive network topology data representing the collaborative state of multiple entities, generated by performing graph network analysis on log data representing the interactive behavior between multiple participating entities in an emergency rescue event within the preset time window; S3: Based on the time-series data of the clinical entities, a first indicator is calculated, which is used to quantify the predicted risk of adverse events occurring in the patient in the future; S4: Based on the interactive network topology data, a second indicator is calculated, which is used to quantify the current cognitive load of the multiple participating entities; S5: By using a preset diagnostic state mapping model, the first indicator and the second indicator are jointly processed to generate a diagnostic state signal for driving the extended reality user interface to perform adaptive state presentation.

2. The real-time visualization and intelligent operation management platform for emergency rescue based on virtual reality as described in claim 1, characterized in that: The clinical entity time-series data includes at least four types of clinical entities identified from the voice data: symptoms, signs, drugs, and medical procedures, as well as the timestamp of each clinical entity being mentioned. The clinical entities are dynamically clustered into one or more clinical syndrome clusters based on a preset clinical knowledge graph; wherein, the calculation of the first indicator is based on the activation state of the one or more clinical syndrome clusters. The activation state of the clinical syndrome cluster is determined by calculating a cluster threat accumulation value that changes dynamically over time. The calculation of the cluster threat accumulation value takes into account at least the basic threat value of each clinical entity within the clinical syndrome cluster, and a time decay function used to characterize the decay of the basic threat value over time.

3. The real-time visualization and intelligent operation management platform for emergency rescue based on virtual reality as described in claim 2, characterized in that: The interactive network topology data includes at least: the multi-subject interaction entropy calculated based on the log data, which characterizes the uncertainty of information flow among the multiple participating subjects, and the task process conformity degree, which characterizes the difference between the actual interaction process and the standard operating procedure. The calculation of the task flow compliance includes: identifying the current primary clinical task based on the clinical entity time-series data; selecting a target SOP corresponding to the primary clinical task from a library containing multiple standard operating procedures (SOPs); and comparing the real-time interactive process with the target SOP to determine the task flow compliance.

4. The real-time visualization and intelligent operation management platform for emergency rescue based on virtual reality as described in claim 3, characterized in that: The calculation of the multi-subject interaction entropy includes: identifying multiple predefined interaction pattern kernels in the interaction network topology data, and performing weighted summation based on the occurrence frequency of the multiple interaction pattern kernels and their respective preset entropy contribution weights.

5. The real-time visualization and intelligent operation management platform for emergency rescue based on virtual reality as described in claim 4, characterized in that: The first indicator is characterized as a predictive clinical risk index, which is used to quantify in real time the severity of the patient's condition and the risk of deterioration based on on-site voice data. The higher the output value of the predictive clinical risk index, the more critical the patient's current clinical condition is, or the more severe the deterioration, and the higher the probability of adverse events in the future. The smaller the output value of the predictive clinical risk index, the more stable the current clinical status of the patient is, and the less likely it is to capture clear evidence pointing to a serious pathological state from the voice information. The second indicator is represented by the team cognitive load index, which is used to quantify the overall mental burden, collaboration efficiency and process standardization of the emergency rescue team in real time during the execution of the mission. The higher the output value of the team cognitive load index, the more likely the team's collaboration is to be in or on the verge of disorder, and the more cognitive resources are over-consumed. The smaller the output value of the team cognitive load index, the more efficient, smooth and standardized the team's collaboration is, and the lower the overall cognitive load is at an ideal level.

6. The real-time visualization and intelligent operation management platform for emergency rescue based on virtual reality as described in claim 5, characterized in that: The diagnostic state mapping model is configured to map a two-dimensional coordinate pair consisting of the first indicator and the second indicator to one of at least four predefined diagnostic states: efficient response state, controllable challenge state, process failure state, and system overload state. The diagnostic state mapping model is further configured to: in the state space formed by the two-dimensional coordinate pairs, set a potential field center for each predefined diagnostic state; the mapping process includes: calculating the potential energy from the current two-dimensional coordinate pair to each potential field center, and mapping the current state to the diagnostic state with the lowest potential energy to generate a diagnostic state identifier.

7. The real-time visualization and intelligent operation management platform for emergency rescue based on virtual reality as described in claim 6, characterized in that: The diagnostic status signal includes at least: the diagnostic status identifier, and a confidence score for quantifying diagnostic confidence; wherein the confidence score is calculated based on the relative potential energy relationship between the current two-dimensional coordinate pair and the center of the potential field; The diagnostic state mapping model further includes a state transition probability module, which is configured to calculate the state transition probability to each other predefined diagnostic state based on the current diagnostic state and the historical trajectory of the two-dimensional coordinate pair in the state space; the diagnostic state signal further includes the state transition probability.

8. The real-time visualization and intelligent operation management platform for emergency rescue based on virtual reality as described in claim 7, characterized in that: The step of calculating the state transition probability to each other predefined diagnostic state specifically includes: calculating a state velocity vector, based on the position of the two-dimensional coordinate pair at the current time step and the previous time step, to characterize the rate and direction of change of the two-dimensional coordinate pair in the state space; and determining the state transition probability based on the projection relationship between the state velocity vector and a direction vector pointing from the current diagnostic state to the target diagnostic state.

9. The real-time visualization and intelligent operation management platform for emergency rescue based on virtual reality as described in claim 8, characterized in that: The diagnostic state mapping model further includes an interface configuration instruction generation step, configured to: based on the diagnostic state identifier and the state transition probability contained in the diagnostic state signal, match and generate an interface configuration instruction for adaptively adjusting the extended reality user interface from a preset rule base; wherein, when the state transition probability to a worse state exceeds a preset state transition alarm threshold, the interface configuration instruction is configured to trigger the presentation of a predictive alarm interface element.

10. The real-time visualization and intelligent operation management platform for emergency rescue based on virtual reality as described in claim 9, characterized in that: The smaller the "probability of transitioning to a worse state", the more stable the current trajectory is judged by the characterization diagnostic state mapping model, or the more it is evolving towards a better state; the larger the "probability of transitioning to a worse state", the stronger the momentum of the current trajectory is judged by the characterization diagnostic state mapping model to evolve towards a worse state.

Images