Virtual reality-based ai emergency triage system

Abstract

This invention relates to the field of smart healthcare virtual reality technology, and discloses an AI-based rapid triage system for emergency care based on virtual reality. The system includes modules for virtual reality scene construction and data acquisition, multimodal data fusion processing, dynamic disease evolution modeling, triage priority calculation, resource adaptability adjustment, virtual reality interface interaction, and model parameter updating and self-optimization. The system collects patient physiological and behavioral data through a virtual reality environment, processes this data, and uses a dynamic disease evolution model to simulate disease development trends, generating triage priority suggestions. It optimizes decisions by integrating real-time hospital resource status and displays the triage results on the virtual reality interface. This system enables a shift from static assessment to dynamic prediction, continuously optimizing triage accuracy through a closed-loop learning mechanism and improving the efficiency of emergency resource allocation.

Description

Technical Field

[0001] This invention relates to the field of smart medical virtual reality technology, specifically to an AI-based rapid triage system for emergency care based on virtual reality. Background Technology

[0002] Current emergency triage primarily relies on nurses' on-site assessments, based on patient complaints, limited vital sign monitoring, and standardized triage scales. This model is highly dependent on the personal experience of healthcare professionals. During peak hours, with a large influx of patients, the limited number of staff makes it difficult to conduct continuous and in-depth observation of each patient. Assessment data is mostly single-point, static snapshots, failing to comprehensively reflect the continuous dynamic changes in a patient's condition. Patients' subjective descriptions may be biased or unclear, and some potentially serious conditions, especially those related to the neurological system or pain responses, may have their early symptoms easily overlooked during traditional rapid triage, leading to the risk of under-triage or over-triage.

[0003] Existing electronic triage systems or early AI-assisted triage tools are mostly based on rule engines or relatively simple machine learning models. These systems typically separate triage suggestions from the hospital's real-time resource status, with triage logic belonging to a different system than the dynamic allocation of resources such as beds, doctors, and equipment. After a triage decision is generated, resource scheduling still requires manual coordination, making it difficult to optimize overall efficiency. Furthermore, such systems generally lack effective continuous learning mechanisms; their model parameters are relatively fixed once deployed, failing to acquire knowledge from daily decision-making feedback from medical staff and actual treatment results. This makes it difficult to adapt to the specific clinical pathways and patient population characteristics of different hospitals, limiting the system's accuracy and adaptability. Summary of the Invention

[0004] The purpose of this invention is to provide an AI-based rapid triage system for emergency care based on virtual reality, in order to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides an AI-based rapid triage system for emergency care based on virtual reality, the system comprising: The virtual reality scene construction and data acquisition module is used to construct virtual reality emergency scenes and collect real-time physiological and behavioral data of emergency patients through virtual reality devices. The multimodal data fusion processing module is used to perform multimodal fusion processing on the collected raw data to generate a standardized patient status data stream; The dynamic disease evolution modeling module is used to build a dynamic disease evolution model based on patient status data streams to simulate the development trend of the disease. The triage priority calculation module is used to calculate the triage priority based on the output of the disease evolution model and generate preliminary triage suggestions. The resource adaptability adjustment module is used to adjust the triage suggestions based on the hospital's real-time resource status data. The virtual reality interface interaction module is used to display triage results and resource allocation plans through a virtual reality interface, and to receive feedback input from medical staff. The model parameter update and self-optimization module is used to update the parameters of the disease evolution model based on feedback data, thereby enabling the system to self-optimize.

[0006] Preferably, the virtual reality scene construction and data acquisition module constructs a virtual reality emergency scene including: Create a 3D environment model of the emergency room, including virtual mappings of functional areas such as the treatment area, waiting area, and resuscitation area; Multiple sensor simulation devices are deployed in a virtual environment to collect physiological parameters such as patient heart rate, blood oxygen saturation, and respiratory rate; Set up patient behavior capture nodes to record patients' limb movements, facial expression changes, and voice information; Establish an environmental parameter acquisition channel to obtain real-time data on temperature, humidity, and personnel density in the emergency room; The data from the 3D environment model, physiological parameters, patient behavior capture nodes, and environmental parameter acquisition channels are integrated into a spatiotemporally synchronized virtual reality scene data stream.

[0007] Preferably, the multimodal data fusion processing module includes: Receive virtual reality scene data streams and extract three sub-data streams: physiological data, behavioral data, and environmental data. Perform time series alignment on each sub-data stream to eliminate time offsets caused by different sampling frequencies; A feature-level fusion strategy is adopted to map the aligned multi-source data to a unified feature space; The feature sequence is segmented and standardized using a sliding window mechanism to eliminate dimensional differences; Using an attention-based weighted fusion method, the weights of each modality are dynamically adjusted according to data quality, and a standardized patient status data stream is output.

[0008] Preferably, the specific steps of the dynamic disease evolution modeling module in establishing a dynamic disease evolution model based on patient state data stream include: Acquire standardized patient status data streams and extract historical disease data sequences as training samples; Construct a dual-path model architecture that includes a spatiotemporal feature extraction network and a temporal prediction network; The spatiotemporal feature extraction network uses a graph convolutional structure to capture the disease transmission relationships between patients; Temporal prediction networks use deep recurrent neural networks to learn the temporal patterns of disease progression; The outputs of the two networks are spliced ​​together in the fusion layer, and the disease evolution prediction is generated through the fully connected layer. The prediction results are compared with real-time monitoring data, and the model parameters are dynamically adjusted.

[0009] Preferably, the operation steps of the disease evolution model further include: Set up a virtual timeline to simulate the disease progression in minutes, and calculate the probability of disease deterioration at each time step based on the current patient status. Multiple possible disease progression paths are generated based on the probability of deterioration, and risk assessment and resource demand prediction are performed for each path; The Monte Carlo simulation method is used to weight and fuse multiple paths to generate a comprehensive prediction of the disease's evolution trend.

[0010] Preferably, the specific steps for the triage priority calculation module to calculate the triage priority based on the output results of the disease evolution model include: Receive disease progression trend prediction data and extract key risk indicators; Construct a triage decision tree, where tree nodes contain characteristic judgment conditions for various emergency symptoms; Starting from the root node of the decision tree, the judgment conditions are matched and judged layer by layer based on the patient's real-time status data; Obtain the initial triage level at the leaf node, and adjust the level based on the speed of disease progression; The system calculates the degree of match between the adjusted triage level and the hospital's triage standards, and generates preliminary triage recommendations that include the triage level, expected waiting time, and suggested treatment measures.

[0011] Preferably, the specific steps of the resource adaptation adjustment module in adjusting the triage suggestions based on the hospital's real-time resource status data include: Establish a hospital resource status monitoring channel to obtain real-time information on the number of beds, staffing of medical personnel, and the usage status of medical equipment. The matching degree between the resource requirements in the initial triage recommendations and the real-time resource status is calculated, and in the case of resource shortage, the alternative solution generation mechanism is activated; The alternative plan generation mechanism calculates the optimal adjustment plan through a resource reallocation algorithm, taking into account the balance between the urgency of the patient's condition and the efficiency of resource utilization, and generates the final triage plan.

[0012] Preferably, the specific steps for the virtual reality interface interaction module to display triage results and resource allocation plans through a virtual reality interface include: The interface is designed with a hierarchical structure: the bottom layer displays the panoramic resource distribution of the emergency room, the middle layer displays the patient triage status, and the top layer displays detailed information. A color-coding system is used to distinguish patients at different triage levels: red represents critical, yellow represents urgent, and green represents normal. It provides interactive operation controls, allowing medical staff to view the simulated process of a specific patient's disease progression; The settings interface allows for manual modification of triage plans and recording of the reasons for modifications. Visual charts showcasing resource allocation plans, including bed allocation diagrams and staff scheduling.

[0013] Preferably, the specific steps for the virtual reality interface interaction module to receive feedback input from medical staff include: Record all actions taken by medical staff in relation to the triage plan, including confirmation, modification, and rejection. Extract key decision-making factors from operation records and construct a sample library of decision-making patterns; The decision-making model samples are compared and analyzed with the solutions automatically generated by the system to identify differences. These differences are then classified and stored according to their types to form a system optimization training dataset. The optimized training dataset is used periodically to perform incremental learning on the disease evolution model.

[0014] Preferably, the specific steps for the model parameter update and self-optimization module to achieve system self-optimization include: Set model performance monitoring metrics, including triage accuracy, response time, and resource utilization. When the monitored indicators fall below the threshold, the model retraining process is initiated. The retraining process involves cleaning and augmenting historical data, using the augmented data to optimize all parameters of the disease evolution model, verifying the performance of the optimized model in a virtual testing environment, and deploying the optimized model to the production environment through a progressive update strategy.

[0015] Compared with the prior art, the beneficial effects of the present invention are: Through the virtual reality scene construction and data acquisition module, high-dimensional behavioral data of patients in immersive environments can be obtained, such as limb motor coordination, eye-tracking trajectory, and the quality of interactive task completion. This data, along with real-time physiological data, is integrated into a continuous patient status data stream in the multimodal data fusion processing module. The dynamic disease evolution modeling module utilizes this data stream and employs a time series analysis model to simulate and predict the future development trend of the disease. This allows triage decisions to move beyond judging current static symptoms and proactively identify high-risk patients whose current vital signs appear stable but whose behavioral manifestations indicate an impending deterioration. This elevates the insight of triage from "current status assessment" to "risk warning," enhancing the ability to identify potentially critically ill patients.

[0016] The initial suggestions generated by the triage priority calculation module are further optimized by the resource adaptation adjustment module in conjunction with real-time hospital resource status data. The algorithm built into this module dynamically balances the urgency of the patient's condition with resource utilization efficiency, outputting a resource allocation scheme that considers both fairness and overall treatment efficiency. The virtual reality interface module visualizes this comprehensive decision-making process. Feedback from medical staff is continuously collected by the model parameter update and self-optimization module and used as training data to adjust and optimize the internal parameters of the patient condition evolution model and the triage algorithm. This mechanism enables continuous iteration of the system based on real-world feedback, allowing the triage model to continuously adapt to the actual operating modes of specific medical institutions and the characteristics of patient groups, gradually narrowing the gap between algorithmic decisions and clinical expert experience, and improving the system's practicality and reliability. Attached Figure Description

[0017] Figure 1 This is a schematic diagram illustrating the working principle of the AI-based rapid triage system for emergency care based on virtual reality as described in this invention. Figure 2 A flowchart for constructing a scene for the virtual reality scene construction and data acquisition module; Figure 3 A flowchart illustrating the data processing of the multimodal data fusion processing module; Figure 4 A distribution diagram of the risk and probability of deterioration in the disease evolution path of an AI-based emergency triage system; Figure 5 This is a comparison chart of the available and used hospital resources for the AI-based emergency triage system. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Please see Figure 1This invention provides an AI-based rapid triage system for emergency care based on virtual reality. The system includes: a virtual reality scene construction and data acquisition module, a multimodal data fusion and processing module, a dynamic disease evolution modeling module, a triage priority calculation module, a resource adaptation adjustment module, a virtual reality interface interaction module, and a model parameter update and self-optimization module. The virtual reality scene construction and data acquisition module is responsible for constructing a virtual reality emergency scene and collecting real-time physiological and behavioral data from patients. The multimodal data fusion and processing module fuses the raw data to generate a standardized patient status data stream. The dynamic disease evolution modeling module establishes a dynamic disease evolution model based on the patient status data stream to simulate the disease development trend. The triage priority calculation module calculates the triage priority based on the output results of the disease evolution model and generates preliminary triage suggestions. The resource adaptation adjustment module adjusts the triage suggestions based on real-time hospital resource status data. The virtual reality interface interaction module displays the triage results and resource allocation scheme and receives feedback input from medical staff. The model parameter update and self-optimization module updates the model parameters based on the feedback data to achieve system self-optimization.

[0020] Example 1: See Figure 2 In practical implementation, the core step in constructing a virtual reality emergency room scenario using the virtual reality scene construction and data acquisition module is creating a 3D environment model of the emergency room. This model precisely includes virtual mappings of the treatment area, waiting area, and resuscitation area, with the mapping relationships established through a spatial coordinate transformation matrix. Multiple sensor simulation devices are deployed within the virtual environment to collect data on the patient's heart rate, blood oxygen saturation, and respiratory rate. The data output frequency of these sensor simulation devices can be configured from 1 to 10 times per second. Patient behavior capture nodes are set up to record the patient's limb movements, facial expressions, and voice information. Limb movement data is acquired through a skeletal tracking algorithm, implemented using a depth camera and inertial measurement unit integrated into the virtual reality device. The depth camera captures a sequence of depth images of the patient's body contour, while the inertial measurement unit provides acceleration and angular velocity data for joint movements.

[0021] The algorithm uses a convolutional neural network to identify key human joints from depth images, including the head, shoulders, elbows, wrists, hips, knees, and ankles. It then fuses depth image data and inertial measurement unit (IMU) data using Kalman filtering to calculate the 3D coordinates and rotation angles of each joint in real time, forming a continuous skeletal motion trajectory. This trajectory data is mapped onto a patient avatar in the virtual environment, synchronously updating limb activity states for recording by behavior capture nodes. Facial expressions are described by facial feature point displacement vectors, and speech information is processed by a real-time speech-to-text engine. An environmental parameter acquisition channel is established to acquire real-time temperature, humidity, and personnel density data for the emergency room. Temperature data is in degrees Celsius, humidity data is in percentage, and personnel density is calculated as the ratio of real-time number of people to area. The 3D environment model data, physiological parameter data, patient behavior capture node data, and environmental parameter acquisition channel data are integrated into a spatiotemporally synchronized virtual reality scene data stream. Spatiotemporal synchronization is achieved through a unified timestamp mechanism with millisecond-level accuracy.

[0022] In some embodiments, the integration process of virtual reality scene data streams involves packet serialization, where each packet contains a unique sequence identifier and a timestamp accurate to the millisecond level. It can be understood that the 3D environment model data contains the vertex coordinates and texture information of all virtual objects, physiological parameter data is stored in key-value pairs, patient behavior capture node data contains skeletal joint rotation data and facial motion unit intensity values, and environmental parameter acquisition channel data consists of simple floating-point values. Optionally, the spatiotemporally synchronized virtual reality scene data stream can be verified for data integrity using the following formula: in: Represents the checksum. This represents the payload of the i-th data packet. Represents the timestamp of the i-th data packet. Represents the total number of data packets within a synchronization period, symbol This indicates a bitwise XOR operation.

[0023] In practical implementation, the operation of the virtual reality scene construction and data acquisition module relies on a high-performance graphics rendering engine, which is responsible for rendering the 3D environment model of the emergency room in real time. It can be understood that the data acquisition port of the virtual reality device and the data output port of the sensor simulation device are connected via a low-latency communication protocol, which ensures the real-time transmission of physiological parameter data. The data stream from the patient behavior capture node and the virtual reality scene data stream are synchronized frame-by-frame through middleware, which is responsible for buffering and aligning data from different sources. Optionally, data from the environmental parameter acquisition channel can be directly written to a shared memory area for direct reading by other components of the virtual reality scene construction and data acquisition module, reducing data copying overhead. In some embodiments, the virtual reality scene data stream is ultimately encapsulated into standard format data packets, which are sent to the multimodal data fusion processing module for subsequent processing via a high-speed network interface.

[0024] Example 2: See Figure 3 In its implementation, the multimodal data fusion processing module receives a virtual reality scene data stream, which originates from the output of the virtual reality scene construction and data acquisition module. The module extracts three sub-data streams: physiological data, behavioral data, and environmental data. Each sub-data stream undergoes time-series alignment, employing interpolation algorithms to compensate for time shifts caused by different sampling frequencies. Physiological data typically has a higher sampling frequency than behavioral data. A feature-level fusion strategy maps the aligned multi-source data to a unified feature space, the dimension of which is determined using principal component analysis. A sliding window mechanism is used to segment and standardize the feature sequences. The size of the sliding window is dynamically adjusted based on data stability, eliminating dimensional differences in parameters such as heart rate and blood oxygen saturation. Finally, an attention-based weighted fusion method dynamically adjusts the modal weights based on the data signal-to-noise ratio, outputting a standardized patient state data stream.

[0025] In some embodiments, the weight calculation of the attention mechanism relies on a learnable scoring function that evaluates the reliability of each modality's data at the current time step. Optionally, the weighted fusion process can be expressed as the following formula: in: The output vector of the standardized patient state data stream representing time point t. Represents the total number of modes. The eigenvector representing the k-th mode at time point t. Let represent the attention weights of the k-th modality at time point t, and satisfy . Attention weights The calculation is performed using a single-layer neural network, whose input is a feature vector. And its recent historical data.

[0026] In practical implementation, the dynamic disease evolution modeling module establishes a dynamic disease evolution model based on patient state data streams. This module acquires standardized patient state data streams and extracts historical disease data sequences as training samples, with the length of these sequences configured to be 60 minutes. A dual-pathway model architecture is constructed, comprising a spatiotemporal feature extraction network and a temporal prediction network. The spatiotemporal feature extraction network employs a graph convolutional structure, where nodes represent patients and edges represent spatial proximity relationships between patients, used to capture the disease propagation relationships among them. The temporal prediction network uses a deep recurrent neural network with 128 hidden layer units, used to learn the temporal patterns of disease development. The outputs of the two networks are concatenated in a fusion layer, resulting in a vector dimension equal to the sum of the dimensions of the two feature components. Disease evolution predictions are generated through fully connected layers, using the ReLU activation function. The prediction results are compared with real-time monitoring data, and model parameters are dynamically adjusted using gradient descent.

[0027] In some embodiments, the implementation of the dynamic disease evolution modeling module relies on a deep learning framework that manages the weight initialization of the graph convolutional structure and the deep recurrent neural network. Optionally, a batch normalization layer is typically applied after the feature concatenation operation to accelerate model convergence and improve stability. The process of dynamically adjusting model parameters is an online learning loop, where each new batch of data triggers a small parameter update. It can be understood that the output of the disease evolution prediction is a multi-dimensional vector, where each element represents the probability distribution of a certain physiological parameter at a specific future time point.

[0028] Example 3: In specific implementation, the disease evolution model operates using a virtual timeline, which advances the disease evolution simulation in minutes. At each time step, the probability of disease deterioration is calculated based on the current patient status; this probability is a value between 0 and 1. Multiple possible disease progression paths are generated based on the deterioration probability, each representing a potential physiological parameter change trajectory. Risk assessment and resource requirement prediction are performed on each path, with the risk assessment outputting a danger level score. Multiple paths are then weighted and fused using a Monte Carlo simulation method, with 1000 iterations to generate a comprehensive disease evolution trend prediction, which is a probability distribution function.

[0029] The triage priority calculation module calculates triage priorities based on the output of the disease evolution model. This module receives disease evolution trend prediction data and extracts key risk indicators, including the degree of abnormality in vital signs and the slope of the deterioration trend. A triage decision tree is constructed, with each node containing characteristic judgment conditions for various emergency conditions, based on clinical guidelines. Starting from the root node, the judgment conditions are matched layer by layer based on the patient's real-time status data, involving threshold comparisons and logical operations. A preliminary triage level is obtained at the leaf nodes and adjusted based on the disease evolution rate, which is the rate of change of risk indicators per unit time. The matching degree between the adjusted triage level and the hospital's triage standards is calculated using a cosine similarity algorithm, generating preliminary triage suggestions that include the triage level, expected waiting time, and suggested treatment measures. These preliminary triage suggestions exist in the form of structured data objects.

[0030] In some embodiments, the calculation of the probability of disease deterioration integrates historical statistical data and real-time physiological parameters, and its mathematical expression is a logistic regression function. Optionally, the weighted fusion process for predicting the overall disease evolution trend can be described by the following formula: in: This represents the overall predicted probability that the patient's condition will enter a critical state at time point c. This represents the total number of disease progression paths generated by the Monte Carlo simulation. The weight represents the j-th disease progression path, and the weight is determined by the initial deterioration probability of that path. This represents the risk assessment score of the j-th path at time point c. This represents a preset risk threshold. It is an indicator function, when the condition is met. The probability value is 1 if the condition is met, and 0 otherwise. The triage priority calculation module will be used as one of the key risk indicators.

[0031] It is understandable that the structure of the triage decision tree is predefined, with each decision condition corresponding to a Boolean expression. In some embodiments, the triage level adjustment operation introduces an adjustment coefficient, which is a linear function of the rate of disease progression. The initial triage suggestion output by the triage priority calculation module is encapsulated as a JSON object, containing fields such as triage level, expected waiting time, and suggested treatment measures. The estimated expected waiting time is based on the historical average processing time corresponding to the current triage level and the current load of the emergency department.

[0032] See Figure 4This figure presents the core analysis results of the dynamic disease evolution modeling module in a virtual reality-based AI-powered emergency rapid triage system. It generates five different disease progression paths and simultaneously displays the risk score and probability of deterioration for each path. The horizontal axis represents the disease progression path, and the vertical axis represents the risk score, with the corresponding probability of deterioration marked above each path. The risk score and probability of deterioration for each path are positively correlated, demonstrating the system's ability to quantitatively assess different development directions through the dynamic disease evolution model. This result will directly serve as a key input to the triage priority calculation module, helping the system identify high-risk patients and match them with appropriate resources. It is one of the core supporting data for achieving the transition from current status assessment to risk warning.

[0033] Example 4: In specific implementation, the resource adaptability adjustment module adjusts the triage suggestions based on real-time hospital resource status data. This module establishes a hospital resource status monitoring channel, which obtains real-time information on bed availability, staffing levels, and medical equipment usage status through the application programming interface provided by the hospital information system. The resource requirements in the initial triage suggestions are compared with the real-time resource status using a similarity metric based on Euclidean distance. For resource-constrained situations, an alternative solution generation mechanism is activated. This mechanism calculates the optimal adjustment plan using a resource reallocation algorithm that balances the urgency of the patient's condition with resource utilization efficiency, generating the final triage plan. The final triage plan is encapsulated in XML format.

[0034] The virtual reality (VR) interface module displays triage results and resource allocation plans through a VR interface. The module features a layered display: the bottom layer shows a panoramic view of the emergency room's resource distribution, the middle layer displays patient triage status, and the top layer displays detailed information. A color-coding system distinguishes patients at different triage levels: red for critical, yellow for urgent, and green for normal. Interactive controls allow medical staff to view a simulated progression of a patient's condition, presented as a 3D animation. A plan adjustment interface allows manual modification of the triage plan and recording of the reasons for modification. Reasons can be selected from predefined options via dropdown menus or voice input. Visual charts displaying the resource allocation plan include a bed allocation map and a staff schedule. The bed allocation map uses a topological representation of bed occupancy.

[0035] In some embodiments, the core of the resource reallocation algorithm is to solve a constrained optimization problem, the optimization objective of which is to maximize the weighted sum of overall resource utilization and patient urgency priority. When performing resource matching, the resource adaptability adjustment module refers to a static resource configuration table (see Table 1), which defines the standard resource requirements corresponding to different triage levels.

[0036] Table 1: Resource Requirements for Triage Level Standards Triage Level Standard treatment area Minimum required medical staff Standard equipment requirements Critical (Red) emergency area 1 doctor, 2 nurses Monitors, defibrillators, ventilators Emergency (yellow) Treatment area One doctor, one nurse Monitors, infusion pumps Normal (green) Waiting area / Simplified treatment area 1 nurse Basic diagnostic equipment It is understandable that the matching degree calculation compares the resource requirements in the initial triage suggestions with the standard requirements in the table above and the real-time available resources obtained from the hospital resource status monitoring channel. When resources are insufficient, the alternative solution generation mechanism will recalculate the resource allocation weights, and its objective function can be expressed as the following formula: in: Represents the overall utility value of resource allocation. Represents the total number of patients awaiting triage. The urgency score represents the triage level of the i-th patient. This represents the utilization efficiency score of the resource combination allocated to the i-th patient. This is a weighting coefficient between 0 and 1, used to adjust the relative importance of the urgency of the illness and the efficiency of resource utilization in the decision-making process. The resource suitability adjustment module determines the final triage plan by solving for the maximum value of the above utility function.

[0037] In practical implementation, the interactive operation controls of the virtual reality interface module include gesture recognition and voice commands. Gesture recognition is achieved through data from the gyroscope and accelerometer of the virtual reality controller. It is understood that the reasons for modifications recorded in the solution adjustment interface will be stored as structured data in the system log. Layers of the layered display interface can be zoomed and switched using pinch gestures. The visualized chart of the medical staff scheduling can dynamically highlight medical staff currently overloaded with tasks. The color saturation of the color coding system can be dynamically adjusted according to patient waiting time; the longer the waiting time, the higher the color saturation.

[0038] See Figure 5 This chart is a core data visualization result of the resource adaptability adjustment module, showing the comparison between the available and used resources of seven key resources in an emergency scenario. The chart presents resource status using grouped bar charts: dark bars represent available resources, and light bars represent used resources. The core value of this chart lies in providing real-time resource data support for the resource adaptability adjustment module. Based on this, the system can calculate resource matching degree. When the initial triage suggestion does not match the resource status, an alternative solution generation mechanism is activated to balance the urgency of the patient's condition with resource utilization efficiency, ultimately outputting the optimal triage plan. This is a crucial basis for achieving coordination between triage decisions and resource scheduling.

[0039] Example 5: In specific implementation, the virtual reality interface interaction module receives feedback input from medical staff. The module records all operations performed by medical staff on the triage plan, including confirmation, modification, and rejection. Each operation is accompanied by a precise timestamp and the operator's identity information. Key decision-making factors are extracted from the operation records, including the modified triage level, the adjusted resource allocation plan, and the reason for the modification. A decision-making pattern sample library is constructed and stored in the form of database tables. Each sample contains the system's original suggestion, the medical staff's final decision, and its contextual data. The decision-making pattern samples are compared and analyzed with the system's automatically generated plans. The comparison analysis calculates the differences between the two in triage level and resource allocation, identifies the differences, and stores them according to the type of difference (level upgrade, level downgrade, resource reallocation), forming a system optimization training dataset. This optimized training dataset is used periodically to incrementally learn the disease evolution model. The incremental learning process uses an online learning algorithm, updating the model weights with a batch of new samples each time.

[0040] The model parameter update and self-optimization module enables system self-optimization. This module sets model performance monitoring metrics, including triage accuracy, response time, and resource utilization. Triage accuracy is calculated based on the triage plan ultimately confirmed by medical staff. When a monitoring metric falls below a preset threshold, a model retraining process is initiated. This threshold is set by the system administrator based on historical operational data. The retraining process cleans and augments historical data. Data cleaning removes invalid and abnormal data records, while data augmentation expands sample diversity by adding noise and temporal transformations. The augmented data is then used to perform full-parameter optimization of the disease evolution model using a stochastic gradient descent algorithm with momentum. The performance of the optimized model is verified in a virtual testing environment that simulates the operation of a real emergency room. The optimized model is deployed to the production environment using a gradual update strategy. This strategy initially imports a small amount of traffic into the new model, and gradually expands the deployment after confirming stability.

[0041] In some embodiments, the classification and storage logic for discrepancies is based on a rule engine, which automatically adds labels according to the attributes of the discrepancies. It can be understood that the construction of the optimized training dataset is an ongoing process, and the old samples in the database are updated on a rolling basis according to timestamps. The thresholds for model performance monitoring metrics are typically set to a dynamic range, which is adjusted according to different time periods in the emergency room. Optionally, the update magnitude of model weights during incremental learning can be constrained using the following formula: in: Represents the model weights in this iteration. Update volume Represents the learning rate. This represents the number of samples in the current optimized training dataset batch. Represents the loss function Relative to weight gradient, The representative disease evolution model for input samples The predicted output, Represents the corresponding real label, It is the momentum coefficient. This is the weight update amount from the previous iteration. This formula ensures the stability of the incremental learning process.

[0042] In practice, the virtual testing environment includes a patient flow simulator that generates virtual patient flows based on historical data patterns. It's understood that implementing a progressive update strategy requires a traffic distributor that controls whether requests entering the production environment model are routed to the old or new model. Data cleaning rules include checking the reasonable range of physiological parameters and the logical consistency of behavioral data. Resource utilization is calculated as the ratio of actual resource usage to the system's recommended resource usage.

[0043] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0044] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An AI-based rapid triage system for emergency care based on virtual reality, characterized in that, Includes the following modules: The virtual reality scene construction and data acquisition module is used to construct virtual reality emergency scenes and collect real-time physiological and behavioral data of emergency patients through virtual reality devices. The multimodal data fusion processing module is used to perform multimodal fusion processing on the collected raw data to generate a standardized patient status data stream; The dynamic disease evolution modeling module is used to build a dynamic disease evolution model based on patient status data streams to simulate the development trend of the disease. The triage priority calculation module is used to calculate the triage priority based on the output of the disease evolution model and generate preliminary triage suggestions. The resource adaptability adjustment module is used to adjust the triage suggestions based on the hospital's real-time resource status data. The virtual reality interface interaction module is used to display triage results and resource allocation plans through a virtual reality interface, and to receive feedback input from medical staff. The model parameter update and self-optimization module is used to update the parameters of the disease evolution model based on feedback data, thereby enabling the system to self-optimize.

2. The implementation method of an AI-based rapid triage system for emergency care based on virtual reality according to claim 1, characterized in that, The virtual reality scene construction and data acquisition module constructs virtual reality emergency scenes including: Create a 3D environment model of the emergency room, including virtual mappings of functional areas such as the treatment area, waiting area, and resuscitation area; Multiple sensor simulation devices are deployed in a virtual environment to collect physiological parameters such as patient heart rate, blood oxygen saturation, and respiratory rate; Set up patient behavior capture nodes to record patients' limb movements, facial expression changes, and voice information; Establish an environmental parameter acquisition channel to obtain real-time data on temperature, humidity, and personnel density in the emergency room; The data from the 3D environment model, physiological parameters, patient behavior capture nodes, and environmental parameter acquisition channels are integrated into a spatiotemporally synchronized virtual reality scene data stream.

3. The implementation method of an AI-based rapid triage system for emergency care based on virtual reality according to claim 2, characterized in that, The multimodal data fusion processing module includes: Receive virtual reality scene data streams and extract three sub-data streams: physiological data, behavioral data, and environmental data. Perform time series alignment on each sub-data stream to eliminate time offsets caused by different sampling frequencies; A feature-level fusion strategy is adopted to map the aligned multi-source data to a unified feature space; The feature sequence is segmented and standardized using a sliding window mechanism to eliminate dimensional differences; Using an attention-based weighted fusion method, the weights of each modality are dynamically adjusted according to data quality, and a standardized patient status data stream is output.

4. The implementation method of an AI-based rapid triage system for emergency care based on virtual reality according to claim 3, characterized in that, The specific steps of the dynamic disease evolution modeling module in establishing a dynamic disease evolution model based on patient status data stream include: Acquire standardized patient status data streams and extract historical disease data sequences as training samples; Construct a dual-path model architecture that includes a spatiotemporal feature extraction network and a temporal prediction network; The spatiotemporal feature extraction network uses a graph convolutional structure to capture the disease transmission relationships between patients; Temporal prediction networks use deep recurrent neural networks to learn the temporal patterns of disease progression; The outputs of the two networks are spliced ​​together in the fusion layer, and the disease evolution prediction is generated through the fully connected layer. The prediction results are compared with real-time monitoring data, and the model parameters are dynamically adjusted.

5. The implementation method of an AI-based rapid triage system for emergency care based on virtual reality according to claim 4, characterized in that, The operation steps of the disease evolution model also include: Set up a virtual timeline to simulate the disease progression in minutes, and calculate the probability of disease deterioration at each time step based on the current patient status. Multiple possible disease progression paths are generated based on the probability of deterioration, and risk assessment and resource demand prediction are performed for each path; The Monte Carlo simulation method is used to weight and fuse multiple paths to generate a comprehensive prediction of the disease's evolution trend.

6. The implementation method of an AI-based rapid triage system for emergency care based on virtual reality according to claim 5, characterized in that, The specific steps of the triage priority calculation module in calculating triage priority based on the output results of the disease evolution model include: Receive disease progression trend prediction data and extract key risk indicators; Construct a triage decision tree, where tree nodes contain characteristic judgment conditions for various emergency symptoms; Starting from the root node of the decision tree, the judgment conditions are matched and judged layer by layer based on the patient's real-time status data; Obtain the initial triage level at the leaf node, and adjust the level based on the speed of disease progression; The system calculates the degree of match between the adjusted triage level and the hospital's triage standards, and generates preliminary triage recommendations that include the triage level, expected waiting time, and suggested treatment measures.

7. The implementation method of an AI-based rapid triage system for emergency care based on virtual reality according to claim 6, characterized in that, The specific steps of the resource adaptation adjustment module in adjusting the triage suggestions based on the hospital's real-time resource status data include: Establish a hospital resource status monitoring channel to obtain real-time information on the number of beds, staffing of medical personnel, and the usage status of medical equipment. The matching degree between the resource requirements in the initial triage recommendations and the real-time resource status is calculated, and in the case of resource shortage, the alternative solution generation mechanism is activated; The alternative plan generation mechanism calculates the optimal adjustment plan through a resource reallocation algorithm, taking into account the balance between the urgency of the patient's condition and the efficiency of resource utilization, and generates the final triage plan.

8. The implementation method of an AI-based rapid triage system for emergency care based on virtual reality according to claim 7, characterized in that, The specific steps for the virtual reality interface interaction module to display triage results and resource allocation plans through a virtual reality interface include: The interface is designed with a hierarchical structure: the bottom layer displays the panoramic resource distribution of the emergency room, the middle layer displays the patient triage status, and the top layer displays detailed information. A color-coding system is used to distinguish patients at different triage levels: red represents critical, yellow represents urgent, and green represents normal. It provides interactive operation controls, allowing medical staff to view the simulated process of a specific patient's disease progression; The settings interface allows for manual modification of triage plans and recording of the reasons for modifications. Visual charts showcasing resource allocation plans, including bed allocation diagrams and staff scheduling.

9. The implementation method of an AI-based rapid triage system for emergency care based on virtual reality according to claim 8, characterized in that, The specific steps for the virtual reality interface interaction module to receive feedback input from medical staff include: Record all actions taken by medical staff in relation to the triage plan, including confirmation, modification, and rejection. Extract key decision-making factors from operation records and construct a sample library of decision-making patterns; The decision-making model samples are compared and analyzed with the solutions automatically generated by the system to identify differences. These differences are then classified and stored according to their types to form a system optimization training dataset. The optimized training dataset is used periodically to perform incremental learning on the disease evolution model.

10. The method for implementing an AI-based rapid triage system for emergency care based on virtual reality according to claim 9, characterized in that, The specific steps for the model parameter update and self-optimization module to achieve system self-optimization include: Set model performance monitoring metrics, including triage accuracy, response time, and resource utilization. When the monitored indicators fall below the threshold, the model retraining process is initiated. The retraining process involves cleaning and augmenting historical data, using the augmented data to optimize all parameters of the disease evolution model, verifying the performance of the optimized model in a virtual testing environment, and deploying the optimized model to the production environment through a progressive update strategy.

Images