Physical examination personnel intelligent hospital guide method based on graph neural network

Abstract

The invention discloses a physical examination personnel intelligent hospital guide method based on a graph neural network, and the method comprises the following steps: obtaining real-time operation data and physical examination personnel individual data, and constructing a dynamic heterogeneous hospital guide graph structure; receiving a physical examination process event flow, updating the dynamic heterogeneous hospital guide map structure, and generating a hospital guide map state representation; constructing an improved HGT network, executing state sensing calculation, and generating a personnel node state embedding expression and a physical examination queue node state embedding expression; constructing a candidate executable hospital guide state unit set; constructing a physical reachable propagation channel and a load propagation channel, executing propagation reasoning, and generating a load propagation reasoning result; based on the candidate executable hospital guide state unit set, generating an optimal hospital guide state in combination with a load propagation reasoning result; and mapping the optimal hospital guide state into a hospital guide behavior event and updating the dynamic heterogeneous hospital guide map structure. According to the invention, dynamic collaborative decision-making and closed-loop optimization of physical examination hospital guidance are realized.

Description

Technical Field

[0001] This invention relates to the field of medical examination guidance and decision support technology, and in particular to an intelligent medical examination guidance method based on graph neural networks. Background Technology

[0002] As health checkup centers operate on a large scale and the complexity of checkup items continues to increase, the efficiency of personnel movement between multiple examination departments has gradually become a crucial factor affecting the health checkup experience and operational efficiency. Existing health checkup guidance systems mostly rely on fixed process configurations or path recommendation methods based on simple rules to pre-arrange the examination sequence for personnel, which is difficult to adapt to the operating environment where there are changes in the number of people, fluctuations in equipment status, and individual differences at the health checkup site.

[0003] Some studies have attempted to optimize the physical examination process by introducing queuing theory models or traditional operations research methods. However, these methods usually rely on simplification assumptions and are difficult to model in a unified manner the various types of entities, multiple constraints, and real-time changing states that exist in the physical examination process. Furthermore, they lack stable global reasoning capabilities when facing high-dimensional and dynamic data, which can easily lead to local optimization or even new congestion problems.

[0004] In addition, existing triage technologies generally fail to effectively handle the dependencies between physical examination items, the coupling effects between physiological constraints and spatial accessibility, and triage decisions are mostly based on a single indicator, lacking a systematic portrayal of the overall operational status of the physical examination center. This results in repeated waiting for examinees, frequent back-and-forth travel, uneven distribution of departmental load, and difficulty in improving overall operational efficiency.

[0005] Therefore, how to provide an intelligent triage method for physical examination personnel based on graph neural networks is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose an intelligent triage method for medical examination personnel based on graph neural networks. By constructing a dynamic heterogeneous triage graph structure, it unifies the modeling of personnel nodes, medical examination queue nodes, and their relationships. Based on an improved HGT network, it performs two-stage inference computation on the triage graph state representation, achieving state perception, state competition, and closed-loop updates during the medical examination triage process. This invention, through the synergistic effect of candidate executable triage state unit sets, load propagation inference results, and triage behavior events, forms a continuous triage computation process driven by the event flow of the medical examination process. This process can adapt to the dynamic changes in the operating state of the medical examination center, improve the overall consistency and stability of triage decisions, reduce redundant waiting and queue load imbalance problems, and possesses the advantages of strong continuity in the triage process, complete state representation, and clear decision-making closed loop.

[0007] According to an embodiment of the present invention, an intelligent triage method for physical examination personnel based on graph neural networks includes the following steps:

[0008] Acquire real-time operational data of the physical examination center and individual data of examinees, and construct a dynamic heterogeneous triage graph structure, including personnel nodes, physical examination queue nodes, queue space relationship edges, project dependency relationship edges, and personnel reachability relationship edges;

[0009] Receive the event stream of the physical examination process, update the dynamic heterogeneous triage diagram structure, and generate the triage diagram status representation;

[0010] An improved HGT network is constructed, including a first-stage HGT network and a second-stage HGT network. The state representation of the patient guidance map is input, and state-aware computation is performed based on the first-stage HGT network to generate the state embedding representation of personnel nodes and the state embedding representation of physical examination queue nodes.

[0011] Based on the personnel node state embedding representation and the physical examination queue node state embedding representation obtained by filtering through personnel reachability relationship edges, a set of candidate executable triage state units is constructed.

[0012] In the improved HGT network, relation decomposition is performed on queue space relation edges, physical reachability propagation channels and load propagation channels are constructed, and reachability propagation calculation, load propagation calculation and queue load propagation inference are performed on the candidate executable diagnostic state unit set to generate load propagation inference results.

[0013] The set of candidate executable triage state units is input into the second-stage HGT network as triage state nodes, message passing and competition calculation are performed, and the optimal triage state is determined by combining the load propagation inference results.

[0014] The optimal triage state is mapped to triage behavior events and the dynamic heterogeneous triage graph structure is updated, thus entering the next round of triage calculation process driven by the event flow of the physical examination process.

[0015] Optionally, the generation of the dynamic heterogeneous triage map structure includes:

[0016] Based on the physical layout information of the physical examination center and the configuration data of the physical examination business, the set of node types and the set of edge types of the dynamic heterogeneous triage diagram structure are determined in advance. The set of node types is fixed to include personnel nodes and physical examination queue nodes, and the set of edge types is fixed to include queue space relationship edges, project dependency relationship edges and personnel reachability relationship edges.

[0017] For personnel nodes, individual characteristics are constructed based on individual data of physical examination personnel, and these characteristics are loaded into the dynamic heterogeneous triage graph structure as the initial node features of personnel nodes.

[0018] For the physical examination queue node, the queue operation characteristics are constructed based on the real-time operation data of each examination department in the physical examination center, and are loaded into the dynamic heterogeneous triage diagram structure as the initial node characteristics of the physical examination queue node.

[0019] For queue spatial relationship edges, based on the physical layout information between the examination departments corresponding to each physical examination queue in the physical examination center, spatial distance features are constructed and configured to queue spatial relationship edges;

[0020] For project dependency edges, dependency features are constructed based on physical examination business rules and medical process constraints, and configured to project dependency edges;

[0021] For personnel reachability edges, based on the current location information of the examinee and the queue operation characteristics corresponding to the physical examination queue node, reachability features are constructed and configured to the personnel reachability edges.

[0022] Optionally, the generation of the triage map status representation includes:

[0023] During the physical examination guidance process, the event flow of the physical examination process is continuously received. Each event carries an event type identifier, an event occurrence time identifier, and a corresponding personnel node identifier and physical examination queue node identifier.

[0024] When a personnel registration event is triggered in the event flow of the physical examination process, the individual characteristics of the personnel in the dynamic heterogeneous triage diagram structure are updated based on the personnel node identifier corresponding to the personnel registration event.

[0025] When a personnel location update event is triggered, the reachability features associated with the personnel node in the dynamic heterogeneous triage map structure are updated based on the personnel node identifier corresponding to the personnel location update event and in combination with the queue operation characteristics.

[0026] When a queue number change event is triggered, the corresponding queue operation characteristics in the dynamic heterogeneous triage diagram structure are updated based on the physical examination queue node identifier corresponding to the queue number change event.

[0027] When an inspection completion event is triggered, the individual characteristics, queue operation characteristics, and reachability characteristics of the personnel are updated synchronously based on the personnel node identifier and physical examination queue node identifier corresponding to the inspection completion event.

[0028] After any event in the physical examination process is triggered and the corresponding node features and edge features are updated, the system performs structured aggregation and unified state encapsulation based on the dynamic heterogeneous triage graph structure to generate a triage graph state representation.

[0029] Optionally, the generation of the personnel node state embedding representation and the physical examination queue node state embedding representation includes:

[0030] During the physical examination guidance process, the dynamic heterogeneous guidance map structure corresponding to the state of the guidance map is used as the input of the improved HGT network;

[0031] In the improved HGT network, the first-stage HGT network is invoked to perform node state perception calculation on the dynamic heterogeneous triage map structure. Based on the node type perception attention calculation mechanism, independent node type attention weights are established for personnel nodes and physical examination queue nodes respectively.

[0032] In the first-stage HGT network, message passing computation is performed on different relation types in the dynamic heterogeneous triage graph structure based on the relation type-aware message passing mechanism.

[0033] After completing the relation type-aware message passing calculation, weighted aggregation is performed on the message results from different relation types. Based on the aggregation results, the node representations of personnel nodes and physical examination queue nodes are updated, and the state embedding representations of personnel nodes and physical examination queue nodes are generated.

[0034] The embedded representations of personnel node states and physical examination queue node states are used as the outputs of the first-stage HGT network.

[0035] Optionally, the generation of the candidate executable triage state unit set includes:

[0036] After completing the state awareness computation of the first stage HGT network, the state embedding representation of personnel nodes is used as the main input for constructing the triage state. Combined with the reachability feature, a filtering operation is performed on the state embedding representation of physical examination queue nodes.

[0037] Based on the embedded representations of personnel node states, the embedded representations of physical examination queue node states, and the corresponding reachability features, a joint mapping calculation is performed to generate the expected completion state representation.

[0038] While constructing the expected completion status representation, the project adaptation relationship between personnel nodes and physical examination queue nodes is encoded to generate a compatible relationship representation.

[0039] For the same personnel node, its corresponding target physical examination queue identifier, expected completion status representation and compatibility relationship representation are combined to generate a candidate executable triage state unit. The construction process is repeated to generate a set of candidate executable triage state units.

[0040] Perform the candidate executable triage state unit construction operation on all personnel nodes respectively, and gather the candidate executable triage state unit sets corresponding to each personnel node to form a candidate executable triage state unit set covering all personnel nodes in the dynamic heterogeneous triage graph structure.

[0041] Optionally, the generation of the load propagation inference result includes:

[0042] Based on the spatial distance characteristics and queue operation characteristics simultaneously carried by the queue spatial relationship edges, a relationship decomposition process is performed to split the original queue spatial relationship edges into physical reachability propagation channels and load propagation channels.

[0043] Based on the physical reachability propagation channel, for each candidate executable triage state unit in the candidate executable triage state unit set, combined with the spatial distance characteristics of the corresponding physical examination queue node and the reachability characteristics of the personnel reachability relationship edge, reachability propagation calculation is performed to obtain the propagation result characterizing the triage state under the spatial reachability constraint;

[0044] Based on the load propagation channel, for each candidate executable triage state unit, combined with the corresponding queue operation characteristics, load propagation calculation is performed to obtain the propagation result characterizing the triage state under the queue operation load constraint.

[0045] Independent attention calculation mechanisms are configured for the physical reachability propagation channel and the load propagation channel, respectively;

[0046] Based on the reachability propagation calculation results of the physical reachability propagation channel and the load propagation calculation results of the load propagation channel, the queue load propagation inference is executed, and the propagation results of each candidate executable diagnostic state unit are comprehensively calculated to generate the load propagation inference results.

[0047] Optionally, the generation of the optimal triage status includes:

[0048] Map each candidate executable triage state unit in the candidate executable triage state unit set to a triage state node, and construct a triage state graph structure that only contains triage state nodes;

[0049] The triage state graph structure is used as the input structure of the second-stage HGT network, and the load propagation inference result is used as the exogenous modulation information input to the second-stage HGT network. Message passing calculation between triage state nodes is performed, and state update calculation is performed on the triage state nodes.

[0050] After message passing computation and state update computation, competitive computation is performed on the triage state nodes based on the second-stage HGT network. The state representations of each triage state node in the candidate executable triage state unit set are compared and evaluated to generate competitive results.

[0051] Based on the competition results of the triage state nodes, a unique optimal triage state is determined from the set of candidate executable triage state units.

[0052] Optionally, the generation of the triage behavior events and the updating of the dynamic heterogeneous triage map structure include:

[0053] Bind the target physical examination queue identifier in the optimal triage state to the corresponding personnel node identifier, and generate a triage behavior event based on the current triage calculation time, including the personnel node identifier, the target physical examination queue identifier, and the event time identifier;

[0054] Based on triage behavior events, a state write-back operation is performed on the reachability relationship edges of personnel in the dynamic heterogeneous triage graph structure, and the corresponding reachability features are updated according to the personnel node identifier and the target physical examination queue identifier.

[0055] Based on the same triage behavior event, perform a status write-back operation on the physical examination queue node and update the corresponding queue operation characteristics according to the target physical examination queue identifier;

[0056] The process of guiding patients to the next round of physical examination is driven by an event flow based on the updated dynamic heterogeneous triage graph structure.

[0057] The beneficial effects of this invention are:

[0058] First, this invention constructs a dynamic heterogeneous triage map structure and uses the triage map state as the unified input state within the triage calculation cycle. Driven by the event flow of the physical examination process, it continuously depicts the personnel nodes, physical examination queue nodes and their relationship states, which can fully reflect the overall operating status of the physical examination center at any time. This avoids the local distortion problem caused by triage based on static rules or single indicators in the prior art, and improves the adaptability of triage decisions to changes in the real-time operating environment.

[0059] Secondly, this invention introduces an improved HGT network and adopts a two-stage inference mechanism. In the first stage, the state perception model of the dynamic heterogeneous triage map structure is completed. In the second stage, triage state competition calculation is performed around the set of candidate executable triage state units, and the optimal triage state is determined by combining the load propagation inference results. This makes triage decision no longer rely on simple queuing prediction, but on a comprehensive judgment based on the joint inference results of spatial reachability constraints and queue operation load constraints, thereby effectively reducing the probability of repeated waiting by examinees and unbalanced load in the examination queue.

[0060] Furthermore, this invention maps the optimal triage state to triage behavior events and writes them back to a dynamic heterogeneous triage graph structure, forming a closed-loop triage calculation process of decision-making, state update, and re-decision. This allows the triage results to continuously influence the construction space and reasoning basis of subsequent triage state units, significantly improving the continuity and overall consistency of the physical examination triage process. Compared with existing technologies that rely on one-time recommendations or single-round predictions, this invention has higher stability and practical value. Attached Figure Description

[0061] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0062] Figure 1 This is an overall flowchart of an intelligent triage method for physical examination personnel based on graph neural networks proposed in this invention;

[0063] Figure 2 This is a schematic diagram of the structure of the improved HGT network in this invention;

[0064] Figure 3 This is a schematic diagram of the relation decomposition and dual-channel propagation reasoning mechanism in this invention. Detailed Implementation

[0065] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0066] refer to Figures 1-3 A method for intelligent triage of medical examinees based on graph neural networks includes the following steps:

[0067] Real-time operational data of the physical examination center and individual data of examinees are acquired to construct a dynamic heterogeneous triage map structure. The dynamic heterogeneous triage map structure includes personnel nodes, physical examination queue nodes, queue spatial relationship edges, item dependency relationship edges, and personnel reachability relationship edges. Personnel nodes are configured with individual characteristics, physical examination queue nodes are configured with queue operation characteristics, queue spatial relationship edges are configured with spatial distance characteristics, item dependency relationship edges are configured with dependency relationship characteristics, and personnel reachability relationship edges are configured with reachability characteristics. Individual characteristics consist of a package item list, physiological status, and priority. Queue operation characteristics consist of real-time queue length, estimated average examination time, department capacity, and examination equipment status. Reachability characteristics consist of estimated waiting time and path reachability.

[0068] Receive the physical examination process event stream, which consists of personnel registration event, personnel location update event, queue number change event, and examination completion event. Based on the physical examination process event stream, update the individual characteristics of personnel, queue operation characteristics, and reachability characteristics of personnel reachable relationship edges in the dynamic heterogeneous triage map structure to generate a triage map state representation.

[0069] The state representation of the triage map is input into the improved HGT network, which consists of a first-stage HGT network and a second-stage HGT network. Based on the node type-aware attention calculation mechanism and the relationship type-aware message passing mechanism of the first-stage HGT network, state-aware calculation is performed on the dynamic heterogeneous triage map structure to generate personnel node state embedding representation and physical examination queue node state embedding representation.

[0070] Based on the personnel node state embedding representation and the physical examination queue node state embedding representation obtained by filtering through personnel reachability relationship edges, a set of candidate executable triage state units is constructed. Each candidate executable triage state unit in the set consists of a target physical examination queue identifier, an expected completion state representation, and a compatibility relationship representation. The expected completion state representation is used to characterize the estimated waiting time and expected completion information after the fusion of the physical examination queue node state embedding representation and reachability features. The compatibility relationship representation is used to characterize the adaptation information between the package item list, physiological state, and the dependency relationship features of the item dependency relationship edges.

[0071] In the improved HGT network, the queue spatial relation edges are decomposed to construct physical reachability propagation channels and load propagation channels. The physical reachability propagation channel performs reachability propagation calculation on the candidate executable triage state unit set based on spatial distance features and path reachability. The load propagation channel performs load propagation calculation on the candidate executable triage state unit set based on queue operation features. Based on the independent attention calculation mechanism of the physical reachability propagation channel and the load propagation channel, queue load propagation inference is performed on the candidate executable triage state unit set to generate load propagation inference results.

[0072] The set of candidate executable triage state units is input as triage state nodes into the second-stage HGT network. Message passing and competition calculation are performed on the triage state nodes based on the second-stage HGT network. The optimal triage state is determined by combining the load propagation inference results. The optimal triage state is used to indicate the target physical examination queue identifier and expected completion status of the physical examination personnel.

[0073] The optimal triage state is mapped to triage behavior events and the dynamic heterogeneous triage graph structure is updated. The triage behavior events consist of personnel node identifiers, target physical examination queue identifiers, and event time identifiers. Based on the triage behavior events, the reachability characteristics of personnel reachable relationship edges are updated and the queue operation characteristics of physical examination queue nodes are updated. Based on the updated dynamic heterogeneous triage graph structure, the next round of physical examination process event flow-driven triage calculation process is entered.

[0074] In this embodiment, the generation of the dynamic heterogeneous diagnostic map structure includes:

[0075] Based on the physical layout information and physical examination business configuration data of the physical examination center, the set of node types and the set of edge types of the dynamic heterogeneous triage map structure are predetermined. The set of node types is fixed to include personnel nodes and physical examination queue nodes, and the set of edge types is fixed to include queue space relationship edges, project dependency relationship edges and personnel reachability relationship edges. Within a single physical examination triage cycle, the set of node types and the set of edge types remain unchanged, and only the corresponding node features and edge features change with the physical examination process.

[0076] For personnel nodes, individual characteristics are constructed based on the individual data of the examinees. These individual characteristics are composed of the list of package items corresponding to the physical examination package, physiological status, and priority. The individual characteristics are then loaded into the dynamic heterogeneous triage graph structure as the initial node characteristics of the personnel nodes. During the physical examination, when a personnel registration event or examination completion event is triggered in the event flow of the physical examination process, the individual characteristics of the personnel nodes are updated in an event-triggered manner.

[0077] For the physical examination queue node, queue operation characteristics are constructed based on the real-time operation data of each examination department in the physical examination center. The queue operation characteristics are composed of real-time queue length, estimated average examination time, department capacity and examination equipment status. The queue operation characteristics are loaded into the dynamic heterogeneous triage graph structure as the initial node characteristics of the physical examination queue node. During the physical examination, when the event of queue number change or examination completion is triggered in the physical examination process event stream, the queue operation characteristics corresponding to the physical examination queue node are updated in an event-triggered manner.

[0078] For queue spatial relationship edges, based on the physical layout information between the examination departments corresponding to each physical examination queue in the physical examination center, a spatial distance feature is constructed, and the spatial distance feature is configured to queue spatial relationship edges. The spatial distance feature remains unchanged within a single physical examination guidance cycle.

[0079] For project dependency edges, dependency features are constructed based on physical examination business rules and medical process constraints. These dependency features are used to characterize the sequential constraints between physical examination items and are configured to project dependency edges. The dependency features remain unchanged within a single physical examination guidance cycle.

[0080] For personnel reachability edges, based on the current location information of the examinee and the queue operation characteristics corresponding to the examination queue node, a reachability feature is constructed. The reachability feature consists of the estimated waiting time and path reachability. The reachability feature is configured to the personnel reachability edge. During the examination process, when a personnel location update event or a queue number change event is triggered in the examination process event flow, the reachability feature of the personnel reachability edge is dynamically updated to realize the dynamic opening and closing of the personnel reachability edge at different stages of the examination.

[0081] In this embodiment, the generation of the triage map status representation includes:

[0082] During the physical examination guidance process, the physical examination process event stream is continuously received. The physical examination process event stream is processed as a time-ordered event sequence. Each event in the physical examination process event stream carries an event type identifier, an event occurrence time identifier, and a personnel node identifier and a physical examination queue node identifier corresponding to the event, which are used to determine the object of the event.

[0083] When a personnel registration event is triggered in the event flow of the physical examination process, the individual characteristics of the corresponding personnel node in the dynamic heterogeneous triage diagram structure are updated based on the personnel node identifier corresponding to the personnel registration event. The list of package items, physiological status and priority corresponding to the personnel after registration are written into the individual characteristics of the personnel node.

[0084] When a personnel location update event is triggered in the event flow of the physical examination process, based on the personnel node identifier corresponding to the personnel location update event and combined with the queue operation characteristics of the physical examination queue node, the reachability characteristics of the personnel reachability relationship edges associated with the personnel node in the dynamic heterogeneous triage diagram structure are updated. The update only applies to the personnel reachability relationship edges that are associated with the personnel node.

[0085] When a queue number change event is triggered in the physical examination process event stream, the queue operation characteristics of the corresponding physical examination queue node in the dynamic heterogeneous triage diagram structure are updated based on the physical examination queue node identifier corresponding to the queue number change event. The real-time queue length and the estimated average examination time in the queue operation characteristics change with the queue number change event.

[0086] When a check-up completion event is triggered in the physical examination process event flow, based on the personnel node identifier and physical examination queue node identifier corresponding to the check-up completion event, the individual characteristics of the personnel node and the queue operation characteristics of the physical examination queue node are updated synchronously, and the reachability characteristics of the personnel reachability relationship edge associated with the personnel node are also updated.

[0087] After any physical examination process event is triggered and the corresponding node and edge features are updated, the individual characteristics of personnel nodes, the queue operation characteristics of physical examination queue nodes, and the reachability characteristics of personnel reachable relationship edges of the dynamic heterogeneous triage graph structure at that moment are structurally aggregated and uniformly encapsulated to generate a triage graph state representation. The triage graph state representation is used to characterize the complete operating state of the dynamic heterogeneous triage graph structure at the current moment and serves as the sole input state source for subsequent triage inference in the improved HGT network. During this period, in the process of performing triage inference calculation based on the triage graph state representation, no update operation driven by the physical examination process event flow is performed on the dynamic heterogeneous triage graph structure.

[0088] In this embodiment, the generation of the personnel node state embedding representation and the physical examination queue node state embedding representation includes:

[0089] During the physical examination guidance process, the dynamic heterogeneous guidance graph structure in the corresponding state is used as the input of the improved HGT network. The guidance graph state represents the joint state formed at the current moment by the individual characteristics of the personnel nodes, the queue operation characteristics of the physical examination queue nodes, and the reachability characteristics of the personnel reachable relationship edges in the corresponding dynamic heterogeneous guidance graph structure. The joint state serves as the initial feature configuration for the improved HGT network to perform graph computation.

[0090] In the improved HGT network, the first-stage HGT network is invoked to perform node state-aware calculation on the dynamic heterogeneous triage map structure. The first-stage HGT network is based on the node type-aware attention calculation mechanism, and independent node type attention weights are established for personnel nodes and physical examination queue nodes respectively, which are used to distinguish the degree of influence of different node types in state-aware calculation.

[0091] In the first-stage HGT network, based on the relationship type-aware message passing mechanism, message passing calculations are performed on different relationship types in the dynamic heterogeneous triage graph structure. Among them, personnel reachability relationship edges participate in message weight calculation based on reachability features, project dependency relationship edges participate in message weight calculation based on dependency relationship features, queue space relationship edges participate in message weight calculation based on spatial distance features, and the message weight calculation paths corresponding to different relationship types are independent of each other.

[0092] After completing the relation type-aware message passing calculation, weighted aggregation is performed on the message results from different relation types. Based on the aggregation results, the node representations of personnel nodes and physical examination queue nodes are updated, and the state embedding representations of personnel nodes and physical examination queue nodes that correspond one-to-one with the personnel nodes in the dynamic heterogeneous triage diagram structure are generated.

[0093] The personnel node state embedding representation and the physical examination queue node state embedding representation are used as the output results of the first-stage HGT network. The personnel node state embedding representation and the physical examination queue node state embedding representation are only used to characterize the state perception result of the dynamic heterogeneous triage map structure under the current triage map state representation. They do not participate in the triage decision calculation and serve as the input basis for the subsequent construction of candidate executable triage state units and the triage state competition calculation of the second-stage HGT network.

[0094] In this embodiment, the generation of the candidate executable triage state unit set includes:

[0095] After completing the state-aware computation of the first stage HGT network, the state embedding representation of personnel nodes is used as the main input for constructing the triage state. Combined with the reachability features of the personnel reachability relationship edges, the state embedding representation of the physical examination queue nodes corresponding to the physical examination queue nodes that have a reachability relationship with the personnel node in the dynamic heterogeneous triage graph structure is filtered to limit the range of physical examination queue node state embedding representations used for subsequent triage state construction.

[0096] Based on the embedded representation of the personnel node state, the embedded representation of the physical examination queue node state, and the reachability features of the corresponding personnel reachable relationship edges, a joint mapping calculation is performed to generate the expected completion state representation corresponding to the personnel node and the physical examination queue node. The expected completion state representation is used to characterize the estimated waiting time and expected completion information of the physical examination queue under the current triage map state representation.

[0097] While constructing the expected completion status representation, based on the package item list and physiological status corresponding to the personnel node, and combined with the dependency relationship features of the item dependency relationship edge, the item adaptation relationship between the personnel node and the physical examination queue node is encoded to generate a compatible relationship representation corresponding to the personnel node and the physical examination queue node. The compatible relationship representation is used as a structured representation to participate in the subsequent triage reasoning process.

[0098] For the same personnel node, the target physical examination queue identifier, expected completion status representation and compatibility relationship representation corresponding to it are combined to generate a candidate executable triage state unit. The process of constructing the state embedding representation of all physical examination queue nodes in the candidate set of personnel nodes and physical examination queue nodes is repeated to generate a set of candidate executable triage state units corresponding to the personnel node.

[0099] The candidate executable triage state unit construction operation is performed on all personnel nodes respectively, and the candidate executable triage state unit sets corresponding to each personnel node are collected to form a candidate executable triage state unit set covering all personnel nodes in the dynamic heterogeneous triage graph structure. The candidate executable triage state units in the candidate executable triage state unit set serve as input objects for subsequent load propagation inference and second-stage HGT network triage state competition calculation.

[0100] In this embodiment, the generation of the load propagation inference result includes:

[0101] Given that the set of candidate executable triage state units has been constructed, for the queue space relationship edges in the dynamic heterogeneous triage graph structure, based on the spatial distance features and queue operation features simultaneously carried by the queue space relationship edges, the relationship decomposition process is performed on the queue space relationship edges, splitting the original queue space relationship edges into physical reachability propagation channels and load propagation channels. Among them, the physical reachability propagation channels only retain the relationship attributes related to spatial distance features and path reachability, and the load propagation channels only retain the relationship attributes related to the operation status of the physical examination queue nodes.

[0102] Based on the physical reachability propagation channel, for each candidate executable triage state unit in the candidate executable triage state unit set, the reachability propagation calculation is performed by combining the spatial distance characteristics of the corresponding physical examination queue node and the reachability characteristics of the personnel reachability relationship edge, so as to obtain the propagation result used to characterize the triage state under the spatial reachability constraint.

[0103] The reachability propagation calculation is used to propagate and infer the reachability constraints of candidate executable triage state unit sets at the spatial and path levels in a dynamic heterogeneous triage graph structure. This calculation is performed based on the physical reachability propagation channel formed after relation decomposition. The physical reachability propagation channel is composed of relation attributes related to spatial distance features and path reachability in queue spatial relation edges. When performing reachability propagation calculation, the candidate executable triage state unit is used as the propagation subject. Combining the spatial distance features of the corresponding physical examination queue node and the reachability features of the personnel reachable relation edges, the spatial reachability of the triage state in the dynamic heterogeneous triage graph structure is propagated and updated layer by layer. This propagation process does not introduce queue operation features and only completes the state constraint propagation based on spatial and path related information. It is used to characterize the spatial reachability propagation result of the triage state in the current triage calculation cycle, thereby providing a stable spatial constraint input for subsequent triage state inference.

[0104] Based on the load propagation channel, for each candidate executable triage state unit in the candidate executable triage state unit set, combined with the queue operation characteristics of the corresponding physical examination queue node, load propagation calculation is performed to obtain the propagation result that characterizes the triage state under the queue operation load constraint.

[0105] The load propagation calculation is used to propagate and infer the influence relationship of the candidate executable triage state unit set at the queue operation load level in a dynamic heterogeneous triage graph structure. This calculation is performed based on the load propagation channel formed after relation decomposition. The load propagation channel is composed of relation attributes related to the operation status of the physical examination queue node in the queue space relation edge. When performing load propagation calculation, the candidate executable triage state unit is used as the propagation subject. Combined with the queue operation characteristics of the corresponding physical examination queue node, the load influence relationship of the triage state in the dynamic heterogeneous triage graph structure is propagated and updated. This propagation process does not introduce spatial distance features and path reachability features. It only performs load-level propagation reasoning on the triage state around the operation status of the physical examination queue node. It is used to characterize the queue operation load propagation result of the triage state in the current triage calculation cycle and serves as an important input basis for subsequent triage state competition calculation.

[0106] Independent attention calculation mechanisms are configured for the physical reachability propagation channel and the load propagation channel, respectively. The attention calculation of the physical reachability propagation channel is based solely on the spatial distance feature and the path reachability feature to perform weight allocation, while the attention calculation of the load propagation channel is based solely on the queue operation feature and the state representation of the candidate executable triage state unit to perform weight allocation. The two types of attention calculation processes are independent of each other in the parameter space.

[0107] Based on the reachability propagation calculation results of the physical reachability propagation channel and the load propagation calculation results of the load propagation channel, queue load propagation inference is performed to comprehensively calculate the propagation results of each candidate executable triage state unit in the candidate executable triage state unit set, and generate load propagation inference results. The load propagation inference results are used to characterize the propagation inference state of each candidate executable triage state unit under the combined action of spatial reachability constraints and queue load constraints.

[0108] The queue load propagation inference includes: for each candidate executable triage state unit in the candidate executable triage state unit set, aligning the reachability propagation calculation result and load propagation calculation result corresponding to the candidate executable triage state unit one by one; performing weight calculation on the aligned reachability propagation calculation result and load propagation calculation result based on the independent attention calculation mechanism configured in the physical reachability propagation channel and the load propagation channel, respectively; performing weighted aggregation on the reachability propagation calculation result and load propagation calculation result based on the corresponding weights; performing channel fusion calculation on the weighted aggregation result to generate the propagation inference state representation of the candidate executable triage state unit; and aggregating the propagation inference state representations corresponding to each candidate executable triage state unit in the candidate executable triage state unit set to form the load propagation inference result, which serves as the input basis for the subsequent second-stage HGT network triage state competition calculation.

[0109] In this embodiment, the generation of the optimal triage state includes:

[0110] Given that the candidate executable triage state unit set has been constructed and the load propagation inference result has been obtained, each candidate executable triage state unit in the candidate executable triage state unit set is mapped to a triage state node, and a triage state graph structure containing only triage state nodes is constructed. The triage state graph structure does not contain personnel nodes and physical examination queue nodes.

[0111] The triage state graph structure is used as the input structure of the second-stage HGT network, and the load propagation inference result is used as the exogenous modulation information corresponding one-to-one with the triage state nodes. In the second-stage HGT network, based on the message passing mechanism between the triage state nodes, message passing calculation is performed between the triage state nodes, and state update calculation is performed on the triage state nodes based on the message passing calculation result.

[0112] After completing the message passing calculation and state update calculation between the triage state nodes, the competition calculation is performed on the triage state nodes based on the second-stage HGT network. The state representations of each triage state node in the candidate executable triage state unit set are compared and evaluated to generate the competition result corresponding to each triage state node.

[0113] Based on the competition results of the triage status nodes, a unique optimal triage status is determined from the set of candidate executable triage status units. The optimal triage status consists of the target physical examination queue identifier and the expected completion status representation, and serves as the triage output result within the current triage calculation cycle.

[0114] In this embodiment, the generation of the triage behavior events and the updating of the dynamic heterogeneous triage map structure include:

[0115] Given that the optimal triage state has been determined, the target physical examination queue identifier in the optimal triage state is bound to the corresponding personnel node identifier, and a triage behavior event is generated in combination with the current triage calculation time, including the personnel node identifier, the target physical examination queue identifier, and the event time identifier.

[0116] The generation of the triage behavior event specifically involves: performing a consistency check on the personnel node identifier and the target physical examination queue identifier to confirm that there is a valid correlation between the two in the dynamic heterogeneous triage graph structure; under the premise that the consistency check passes, generating an event time identifier based on the system time corresponding to the current triage calculation cycle; and structurally encapsulating the personnel node identifier, the target physical examination queue identifier, and the event time identifier according to the preset event field order to form a triage behavior event. The triage behavior event serves as the sole event input for performing a state write-back operation on the dynamic heterogeneous triage graph structure, triggering the update of the reachability features of the personnel reachable relationship edges and the update of the queue operation features of the physical examination queue nodes, and participating in the triage calculation process driven by the event flow of the next round of physical examination process.

[0117] Based on the triage behavior event, a state write-back operation is performed on the reachability relationship edges of personnel in the dynamic heterogeneous triage graph structure. The reachability features of the corresponding personnel reachability relationship edges are updated according to the personnel node identifier and the target physical examination queue identifier. The reachability features are used to reflect the path reachability status change of the personnel node after completing the triage behavior.

[0118] Based on the same triage behavior event, a status write-back operation is performed on the physical examination queue node in the dynamic heterogeneous triage diagram structure. The queue operation characteristics of the corresponding physical examination queue node are updated according to the target physical examination queue identifier. The queue operation characteristics are used to reflect the change in the operation status of the physical examination queue node after receiving the triage behavior.

[0119] After updating the reachability features of the reachable edges of the personnel and the queue operation features of the physical examination queue nodes, the next round of physical examination process event flow-driven triage calculation process is entered based on the updated dynamic heterogeneous triage graph structure. The triage behavior events, as historical decision results, participate in the continuous updating of the dynamic heterogeneous triage graph structure by the subsequent physical examination process event flow.

[0120] Example 1: To verify the feasibility of this invention in practice, it was applied to the intelligent triage process for medical examinees during peak hours at a health checkup center. In this scenario, the health checkup center receives a large number of examinees daily, with uneven arrival times, significant differences in health checkup packages, and clear dependencies between checkup items. Furthermore, the geographically dispersed locations of checkup departments and the dynamic changes in equipment status and department capacity easily lead to problems such as localized queue congestion, unreasonable personnel travel paths, and excessively long average waiting times. Traditional triage methods often rely on fixed rules or static queuing strategies, making it difficult to coordinate and schedule personnel flow and queue load in the context of continuous evolution during the health checkup process. This results in limited overall health checkup efficiency and unstable user experience.

[0121] In this embodiment, the system continuously acquires real-time operational data of the health checkup center and individual data of the examinees, including the real-time queue length, estimated average examination time, department capacity, and examination equipment status of each queue, as well as the list of package items, physiological status, and priority information corresponding to each examinee. Based on this, a dynamic heterogeneous triage graph structure is constructed, abstracting examinees as personnel nodes and the queues corresponding to examination items as examination queue nodes. Simultaneously, queue spatial relationship edges, item dependency relationship edges, and personnel reachability relationship edges are constructed to characterize spatial distance constraints, item order constraints, and the set of queues reachable by the examinee at the current moment. As the health checkup process progresses, the system continuously receives a stream of events, including events such as personnel registration, changes in personnel location, changes in the number of people in the queue, and completion of the examination. Upon event triggering, only the reachability characteristics of the associated personnel node features, examination queue node features, and personnel reachability relationship edges are updated, and a triage graph status representation is generated at that moment to characterize the overall operational status of the health checkup center.

[0122] The triage map state representation is input into the improved HGT network. In the first stage, the HGT network performs state-aware computation only on the dynamic heterogeneous triage map structure. Through node-type-aware attention and relationship-type-aware message passing, it learns the state embedding representations of personnel nodes and medical examination queue nodes. This stage does not participate in any triage decisions; it is only used for unified modeling of the current operating state. After obtaining the node state embedding representations, the system, based on the personnel node state embedding representation and combined with the personnel reachable relationship edges, filters the medical examination queue node state embedding representations to construct a set of candidate executable triage state units. Each candidate executable triage state unit corresponds to a medical examination queue selection that a person can choose in the current state.

[0123] Subsequently, the system decomposes the spatial relationships of the queues into physical reachability propagation channels and load propagation channels. The physical reachability propagation channel, based on spatial distance features and path reachability, performs reachability propagation calculations on the set of candidate executable triage state units to characterize the rationality of personnel reaching different medical examination queues under spatial constraints. The load propagation channel, based on queue operation characteristics, performs load propagation calculations on the set of candidate executable triage state units to reflect the capacity of different medical examination queues to handle new personnel under the current load state. The system further performs queue load propagation inference on the two types of propagation results, generating load propagation inference results to comprehensively reflect the combined impact of spatial reachability and queue load on triage decisions.

[0124] Based on this, the set of candidate executable triage state units is input as triage state nodes into the second-stage HGT network. The second-stage HGT network only performs message passing and competition computation on the triage state nodes, and uses the load propagation inference results as exogenous modulation information to participate in the competition process, ultimately determining a unique optimal triage state. This optimal triage state is mapped as a triage behavior event and written back to the dynamic heterogeneous triage graph structure to update the reachability features of personnel reachable edges and the queue operation features of physical examination queue nodes, thus entering the next round of physical examination process event flow-driven triage computation process, forming a complete closed-loop triage mechanism.

[0125] During the experimental verification process, physical examination data within a continuous operating cycle were selected to compare the performance of the three strategies—the method of this invention, the guidance method based on static rules, and the guidance method based on single waiting time prediction—under the same operating conditions. Several core operating indicators were statistically analyzed, and the results are summarized in the table below.

[0126] Table 1. Performance Comparison of Different Patient Guidance Methods in the Operation Scenario of a Physical Examination Center

[0127] Indicator Name Static rule-based triage method Waiting time prediction and triage methods Method of the present invention Average total waiting time per person (minutes) 68.4 61.7 56.9 Maximum queue waiting time (minutes) 142.3 128.6 118.2 Queue load standard deviation 21.5 18.9 15.6 Number of round trips per person 3.2 2.7 2.3 Peak period completion rate (%) 82.4 86.1 89.3

[0128] As shown in Table 1, the static rule-based triage method suffers from significant differences in queue load due to the lack of real-time state modeling, with both the average total waiting time and the maximum queue waiting time remaining at relatively high levels. The triage method based on waiting time prediction alleviates local congestion to some extent, reducing both the average total waiting time and the maximum queue waiting time. However, because it does not coordinate spatial accessibility with queue load modeling, the standard deviation of queue load remains large, and the reduction in the number of round-trip paths for personnel is limited.

[0129] The method of this invention demonstrates more stable improvements across the aforementioned indicators. The average total waiting time per person is reduced by approximately 16.8% compared to the static rule-based method and by approximately 7.8% compared to the waiting time prediction method. Both the maximum queue waiting time and the standard deviation of queue load show a decreasing trend, indicating a more balanced distribution of queue load. Simultaneously, the number of round trips for individuals is significantly reduced, suggesting that the triage results are more spatially reasonable, and the completion rate during peak hours is steadily improved.

[0130] The aforementioned performance improvement stems from the fact that this invention uses a dynamic heterogeneous triage graph structure to uniformly model personnel, queues, and their relationships. It also utilizes a two-stage HGT network to functionally decouple state perception from triage decision-making. Furthermore, it introduces relationship decomposition and a dual-channel propagation reasoning mechanism to separately model spatial reachability constraints and queue load constraints, integrating them during the decision-making stage. This avoids the problem of a single indicator dominating triage results. Through a closed-loop update mechanism formed by writing back triage behavior events, the system can continuously adapt to the dynamic changes in the physical examination process, achieving a balanced and interpretable improvement in overall efficiency and stability.

[0131] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for intelligent triage of medical examinees based on graph neural networks, characterized in that, Includes the following steps: Acquire real-time operational data of the physical examination center and individual data of examinees, and construct a dynamic heterogeneous triage graph structure, including personnel nodes, physical examination queue nodes, queue space relationship edges, project dependency relationship edges, and personnel reachability relationship edges; Receive the event stream of the physical examination process, update the dynamic heterogeneous triage diagram structure, and generate the triage diagram status representation; An improved HGT network is constructed, including a first-stage HGT network and a second-stage HGT network. The state representation of the patient guidance map is input, and state-aware computation is performed based on the first-stage HGT network to generate the state embedding representation of personnel nodes and the state embedding representation of physical examination queue nodes. Based on the personnel node state embedding representation and the physical examination queue node state embedding representation obtained by filtering through personnel reachability relationship edges, a set of candidate executable triage state units is constructed. In the improved HGT network, relation decomposition is performed on queue space relation edges, physical reachability propagation channels and load propagation channels are constructed, and reachability propagation calculation, load propagation calculation and queue load propagation inference are performed on the candidate executable diagnostic state unit set to generate load propagation inference results. The set of candidate executable triage state units is input into the second-stage HGT network as triage state nodes, message passing and competition calculation are performed, and the optimal triage state is determined by combining the load propagation inference results. The optimal triage state is mapped to triage behavior events and the dynamic heterogeneous triage graph structure is updated, thus entering the next round of triage calculation process driven by the event flow of the physical examination process.

2. The intelligent triage method for physical examination personnel based on graph neural networks according to claim 1, characterized in that, The generation of the dynamic heterogeneous triage map structure includes: Based on the physical layout information of the physical examination center and the configuration data of the physical examination business, the set of node types and the set of edge types of the dynamic heterogeneous triage diagram structure are determined in advance. The set of node types is fixed to include personnel nodes and physical examination queue nodes, and the set of edge types is fixed to include queue space relationship edges, project dependency relationship edges and personnel reachability relationship edges. For personnel nodes, individual characteristics are constructed based on individual data of physical examination personnel, and these characteristics are loaded into the dynamic heterogeneous triage graph structure as the initial node features of personnel nodes. For the physical examination queue node, the queue operation characteristics are constructed based on the real-time operation data of each examination department in the physical examination center, and are loaded into the dynamic heterogeneous triage diagram structure as the initial node characteristics of the physical examination queue node. For queue spatial relationship edges, based on the physical layout information between the examination departments corresponding to each physical examination queue in the physical examination center, spatial distance features are constructed and configured to queue spatial relationship edges; For project dependency edges, dependency features are constructed based on physical examination business rules and medical process constraints, and configured to project dependency edges; For personnel reachability edges, based on the current location information of the examinee and the queue operation characteristics corresponding to the physical examination queue node, reachability features are constructed and configured to the personnel reachability edges.

3. The intelligent triage method for physical examination personnel based on graph neural networks according to claim 1, characterized in that, The generation of the triage map status representation includes: During the physical examination guidance process, the event flow of the physical examination process is continuously received. Each event carries an event type identifier, an event occurrence time identifier, and a corresponding personnel node identifier and physical examination queue node identifier. When a personnel registration event is triggered in the event flow of the physical examination process, the individual characteristics of the personnel in the dynamic heterogeneous triage diagram structure are updated based on the personnel node identifier corresponding to the personnel registration event. When a personnel location update event is triggered, the reachability features associated with the personnel node in the dynamic heterogeneous triage map structure are updated based on the personnel node identifier corresponding to the personnel location update event and in combination with the queue operation characteristics. When a queue number change event is triggered, the corresponding queue operation characteristics in the dynamic heterogeneous triage diagram structure are updated based on the physical examination queue node identifier corresponding to the queue number change event. When an inspection completion event is triggered, the individual characteristics, queue operation characteristics, and reachability characteristics of the personnel are updated synchronously based on the personnel node identifier and physical examination queue node identifier corresponding to the inspection completion event. After any event in the physical examination process is triggered and the corresponding node features and edge features are updated, the system performs structured aggregation and unified state encapsulation based on the dynamic heterogeneous triage graph structure to generate a triage graph state representation.

4. The intelligent triage method for physical examination personnel based on graph neural networks according to claim 1, characterized in that, The generation of the personnel node state embedding representation and the physical examination queue node state embedding representation includes: During the physical examination guidance process, the dynamic heterogeneous guidance map structure corresponding to the state of the guidance map is used as the input of the improved HGT network; In the improved HGT network, the first-stage HGT network is invoked to perform node state perception calculation on the dynamic heterogeneous triage map structure. Based on the node type perception attention calculation mechanism, independent node type attention weights are established for personnel nodes and physical examination queue nodes respectively. In the first-stage HGT network, message passing computation is performed on different relation types in the dynamic heterogeneous triage graph structure based on the relation type-aware message passing mechanism. After completing the relation type-aware message passing calculation, weighted aggregation is performed on the message results from different relation types. Based on the aggregation results, the node representations of personnel nodes and physical examination queue nodes are updated, and the state embedding representations of personnel nodes and physical examination queue nodes are generated. The embedded representations of personnel node states and physical examination queue node states are used as the outputs of the first-stage HGT network.

5. The intelligent triage method for physical examination personnel based on graph neural networks according to claim 1, characterized in that, The generation of the candidate executable triage state unit set includes: After completing the state awareness computation of the first stage HGT network, the state embedding representation of personnel nodes is used as the main input for constructing the triage state. Combined with the reachability feature, a filtering operation is performed on the state embedding representation of physical examination queue nodes. Based on the embedded representations of personnel node states, the embedded representations of physical examination queue node states, and the corresponding reachability features, a joint mapping calculation is performed to generate the expected completion state representation. While constructing the expected completion status representation, the project adaptation relationship between personnel nodes and physical examination queue nodes is encoded to generate a compatible relationship representation. For the same personnel node, its corresponding target physical examination queue identifier, expected completion status representation and compatibility relationship representation are combined to generate a candidate executable triage state unit. The construction process is repeated to generate a set of candidate executable triage state units. Perform the candidate executable triage state unit construction operation on all personnel nodes respectively, and gather the candidate executable triage state unit sets corresponding to each personnel node to form a candidate executable triage state unit set covering all personnel nodes in the dynamic heterogeneous triage graph structure.

6. The intelligent triage method for physical examination personnel based on graph neural networks according to claim 1, characterized in that, The generation of the load propagation inference result includes: Based on the spatial distance characteristics and queue operation characteristics simultaneously carried by the queue spatial relationship edges, a relationship decomposition process is performed to split the original queue spatial relationship edges into physical reachability propagation channels and load propagation channels. Based on the physical reachability propagation channel, for each candidate executable triage state unit in the candidate executable triage state unit set, combined with the spatial distance characteristics of the corresponding physical examination queue node and the reachability characteristics of the personnel reachability relationship edge, reachability propagation calculation is performed to obtain the propagation result characterizing the triage state under the spatial reachability constraint; Based on the load propagation channel, for each candidate executable triage state unit, combined with the corresponding queue operation characteristics, load propagation calculation is performed to obtain the propagation result characterizing the triage state under the queue operation load constraint. Independent attention calculation mechanisms are configured for the physical reachability propagation channel and the load propagation channel, respectively; Based on the reachability propagation calculation results of the physical reachability propagation channel and the load propagation calculation results of the load propagation channel, the queue load propagation inference is executed, and the propagation results of each candidate executable diagnostic state unit are comprehensively calculated to generate the load propagation inference results.

7. The intelligent triage method for physical examination personnel based on graph neural networks according to claim 1, characterized in that, The generation of the optimal triage state includes: Map each candidate executable triage state unit in the candidate executable triage state unit set to a triage state node, and construct a triage state graph structure that only contains triage state nodes; The triage state graph structure is used as the input structure of the second-stage HGT network, and the load propagation inference result is used as the exogenous modulation information input to the second-stage HGT network. Message passing calculation between triage state nodes is performed, and state update calculation is performed on the triage state nodes. After message passing computation and state update computation, competitive computation is performed on the triage state nodes based on the second-stage HGT network. The state representations of each triage state node in the candidate executable triage state unit set are compared and evaluated to generate competitive results. Based on the competition results of the triage state nodes, a unique optimal triage state is determined from the set of candidate executable triage state units.

8. The intelligent triage method for physical examination personnel based on graph neural networks according to claim 1, characterized in that, The generation of patient guidance behavior events and the updating of the dynamic heterogeneous patient guidance map structure include: Bind the target physical examination queue identifier in the optimal triage state to the corresponding personnel node identifier, and generate a triage behavior event based on the current triage calculation time, including the personnel node identifier, the target physical examination queue identifier, and the event time identifier; Based on triage behavior events, a state write-back operation is performed on the reachability relationship edges of personnel in the dynamic heterogeneous triage graph structure, and the corresponding reachability features are updated according to the personnel node identifier and the target physical examination queue identifier. Based on the same triage behavior event, perform a status write-back operation on the physical examination queue node and update the corresponding queue operation characteristics according to the target physical examination queue identifier; The process of guiding patients to the next round of physical examination is driven by an event flow based on the updated dynamic heterogeneous triage graph structure.

Images