Search

CN-122000004-A - Intelligent diagnosis guiding method for physical examination personnel based on graph neural network

CN122000004ACN 122000004 ACN122000004 ACN 122000004ACN-122000004-A

Abstract

The invention discloses an intelligent guided diagnosis method for physical examination staff based on a graphic neural network, which comprises the following steps of obtaining real-time operation data and individual data of the physical examination staff, constructing a dynamic heterogeneous guided diagnosis graphic structure, receiving an event stream of a physical examination process, updating the dynamic heterogeneous guided diagnosis graphic structure, generating a guided diagnosis graphic state representation, constructing an improved HGT network, executing state sensing calculation, generating a staff node state embedded representation and a physical examination queue node state embedded representation, constructing a candidate executable guided diagnosis state unit set, constructing a physical reachable propagation channel and a load propagation channel, executing propagation reasoning, generating a load propagation reasoning result, generating an optimal guided diagnosis state based on the candidate executable guided diagnosis state unit set and combining the load propagation reasoning result, mapping the optimal guided diagnosis state into a guided diagnosis behavior event, and updating the dynamic heterogeneous guided diagnosis graphic structure. The invention realizes dynamic collaborative decision and closed-loop optimization of physical examination diagnosis.

Inventors

  • LIU YULIANG

Assignees

  • 陕西森昂科技有限公司

Dates

Publication Date
20260508
Application Date
20260130

Claims (8)

  1. 1. The intelligent diagnosis guiding method for the physical examination staff based on the graph neural network is characterized by comprising the following steps of: acquiring real-time operation data of a physical examination center and individual data of physical examination personnel, and constructing a dynamic heterogeneous diagnosis guiding graph structure, wherein the dynamic heterogeneous diagnosis guiding graph structure comprises personnel nodes, physical examination queue nodes, queue space relationship sides, project dependency relationship sides and personnel reachable relationship sides; Receiving a physical examination process event stream, updating a dynamic heterogeneous diagnosis guiding chart structure, and generating a diagnosis guiding chart state representation; Constructing an improved HGT network, comprising a first-stage HGT network and a second-stage HGT network, inputting a diagnosis-guiding diagram state representation, and executing state sensing calculation based on the first-stage HGT network to generate a personnel node state embedded representation and a physical examination queue node state embedded representation; Constructing a candidate executable diagnosis guiding state unit set based on the personnel node state embedded representation and the physical examination queue node state embedded representation obtained through personnel reachable relation edge screening; In the improved HGT network, performing relationship decomposition on the queue space relationship edges, constructing a physical reachable propagation channel and a load propagation channel, performing reachable propagation calculation, load propagation calculation and queue load propagation reasoning on the candidate executable diagnosis guiding state unit set, and generating a load propagation reasoning result; Inputting the candidate executable guided diagnosis state unit set as a guided diagnosis state node into a second-stage HGT network, performing message transfer and competition calculation, and determining an optimal guided diagnosis state by combining a load propagation reasoning result; mapping the optimal diagnosis guiding state into a diagnosis guiding behavior event and updating a dynamic heterogeneous diagnosis guiding graph structure, and entering a diagnosis guiding calculation process driven by the event flow in the next round of examination process.
  2. 2. The intelligent guided diagnosis method for physical examination staff based on the graph neural network according to claim 1, wherein the generating of the dynamic heterogeneous guided diagnosis graph structure comprises the following steps: based on physical layout information and physical examination business configuration data of a physical examination center, a node type set and an edge type set of a dynamic heterogeneous diagnosis guide structure are predetermined, wherein the node type set fixedly comprises personnel nodes and physical examination queue nodes, and the edge type set fixedly comprises queue space relation edges, project dependency relation edges and personnel reachable relation edges; Aiming at personnel nodes, constructing personnel individual features based on physical examination personnel individual data, and loading a dynamic heterogeneous diagnosis guide graph structure as initial node features of the personnel nodes; aiming at physical examination queue nodes, constructing queue operation characteristics based on real-time operation data of each examination department of a physical examination center, and loading the queue operation characteristics serving as initial node characteristics of the physical examination queue nodes into a dynamic heterogeneous diagnosis guiding graph structure; aiming at the queue space relation edge, constructing a space distance characteristic based on physical layout information among corresponding examination departments of each physical examination queue of the physical examination center, and configuring the space distance characteristic to the queue space relation edge; Aiming at the project dependency edges, building dependency characteristics based on physical examination business rules and medical procedure constraints, and configuring the dependency characteristics to the project dependency edges; aiming at the personnel accessibility relation edge, based on the current position information of the physical examination personnel and the queue operation characteristics corresponding to the physical examination queue nodes, accessibility characteristics are constructed and configured to the personnel accessibility relation edge.
  3. 3. The method for intelligently guiding a medical examination person based on a graphic neural network according to claim 1, wherein the generating of the status representation of the guiding map comprises: in the physical examination and diagnosis guiding process, continuously receiving an event stream in the physical examination process, wherein each event carries an event type identifier, an event occurrence time identifier, a personnel node identifier corresponding to the event and a physical examination queue node identifier; When a personnel registration event is triggered in the physical examination process event stream, updating individual characteristics of personnel in the dynamic heterogeneous diagnosis-guiding graph structure based on personnel node identifiers corresponding to the personnel registration event; When triggering a personnel position updating event, updating the reachability characteristic associated with the personnel node in the dynamic heterogeneous diagnosis guiding graph structure based on the personnel node identifier corresponding to the personnel position updating event and combining the queue operation characteristic; When a queue number change event is triggered, updating the corresponding queue operation characteristics in the dynamic heterogeneous diagnosis guide map structure based on the physical examination queue node identification corresponding to the queue number change event; when an inspection completion event is triggered, based on a personnel node identifier and a physical examination queue node identifier corresponding to the inspection completion event, synchronously updating personnel individual characteristics, queue operation characteristics and reachability characteristics; after event flow event triggering and corresponding node feature and edge feature updating of any integral detection process are completed, structured collection and unified state packaging are carried out based on the dynamic heterogeneous diagnosis guiding graph structure, and diagnosis guiding graph state representation is generated.
  4. 4. The intelligent guided diagnosis method for physical examination staff based on the graph neural network according to claim 1, wherein the generation of the staff node state embedded representation and the physical examination queue node state embedded representation comprises the following steps: In the physical examination diagnosis guiding process, taking a dynamic heterogeneous diagnosis guiding chart structure in a corresponding state of the diagnosis guiding chart state representation as the input of the improved HGT network; In the improved HGT network, a first-stage HGT network is called to execute node state sensing calculation on the dynamic heterogeneous diagnosis-guiding graph structure, and independent node type attention weights are respectively established for personnel nodes and physical examination queue nodes based on a node type sensing attention calculation mechanism; In the first stage HGT network, based on a relation type perception message transfer mechanism, respectively executing message transfer calculation on different relation types in the dynamic heterogeneous diagnosis-guiding graph structure; After the relation type perception message transfer calculation is completed, carrying out weighted aggregation on message results from different relation types, updating node representations of personnel nodes and physical examination queue nodes based on the aggregation results, and generating personnel node state embedded representations and physical examination queue node state embedded representations; And embedding the personnel node state embedded representation and the physical examination queue node state embedded representation as output results of the first-stage HGT network.
  5. 5. The method for intelligently guiding a physical examination person based on a graph neural network according to claim 1, wherein the generating of the candidate executable guiding state unit set comprises: After the state sensing calculation of the HGT network in the first stage is completed, the personnel node state embedded representation is used as a main item input for constructing the diagnosis guiding state, and the screening operation is executed on the physical examination queue node state embedded representation by combining the reachability characteristic; Based on the personnel node state embedded representation, the physical examination queue node state embedded representation and the corresponding reachability characteristics, performing joint mapping calculation to generate an expected completion state representation; While constructing the expected completion state representation, performing encoding processing on the project adaptation relationship between the personnel node and the physical examination queue node to generate a compatibility relationship representation; Combining the corresponding target physical examination queue identification, the expected completion state representation and the compatibility relation representation aiming at the same personnel node to generate a candidate executable guided diagnosis state unit, and repeating the construction process to generate a candidate executable guided diagnosis state unit set; And respectively executing candidate executable guided diagnosis state unit construction operation on all personnel nodes, and collecting candidate executable guided diagnosis state unit sets corresponding to all personnel nodes to form a candidate executable guided diagnosis state unit set covering all personnel nodes in the dynamic heterogeneous guided diagnosis map structure.
  6. 6. The intelligent diagnosis guiding method for physical examination staff based on the graph neural network according to claim 1, wherein the generating of the load propagation reasoning result comprises the following steps: Based on the spatial distance characteristic and the queue operation characteristic carried by the queue spatial relationship side, executing relationship decomposition processing, and splitting the original queue spatial relationship side into a physical reachable propagation channel and a load propagation channel; Based on the physical reachable propagation channel, performing reachable propagation calculation on each candidate executable guided diagnosis state unit in the candidate executable guided diagnosis state unit set by combining the space distance characteristic of the corresponding physical examination queue node and the reachability characteristic of the personnel reachable relation edge to obtain a propagation result of the characterization guided diagnosis state under the space-level reachable constraint; Based on the load propagation channel, performing load propagation calculation on each candidate executable diagnosis guiding state unit in combination with the corresponding queue operation characteristic to obtain a propagation result of the characterization diagnosis guiding state under the queue operation load constraint; for a physical reachable propagation channel and a load propagation channel, respectively configuring independent attention computing mechanisms; Based on the reachable propagation calculation result of the physical reachable propagation channel and the load propagation calculation result of the load propagation channel, queue load propagation reasoning is executed, comprehensive calculation is carried out on the propagation result of each candidate executable diagnosis guiding state unit, and a load propagation reasoning result is generated.
  7. 7. The intelligent guided diagnosis method for physical examination staff based on the graphic neural network according to claim 1, wherein the generating of the optimal guided diagnosis state comprises the following steps: Mapping each candidate executable guided diagnosis state unit in the candidate executable guided diagnosis state unit set into a guided diagnosis state node, and constructing a guided diagnosis state diagram structure only comprising the guided diagnosis state node; Taking the diagnosis guiding state diagram structure as an input structure of the second-stage HGT network, taking a load propagation reasoning result as exogenous modulation information to be input into the second-stage HGT network, executing message transfer calculation among diagnosis guiding state nodes, and executing state update calculation on the diagnosis guiding state nodes; after the message passing calculation and the state updating calculation, performing competition calculation on the guided diagnosis state nodes based on the second-stage HGT network, and performing comparison evaluation on the state representation of each guided diagnosis state node in the candidate executable guided diagnosis state unit set to generate a competition result; based on the competing results of the triage status nodes, a unique optimal triage status is determined from the candidate executable triage status unit set.
  8. 8. The intelligent guided diagnosis method for physical examination staff based on the graphic neural network according to claim 1, wherein the generation of the guided diagnosis behavior event and the updating of the dynamic heterogeneous guided diagnosis graphic structure comprise the following steps: Binding a target physical examination queue identifier in an optimal diagnosis guiding state with a corresponding personnel node identifier, and generating a diagnosis guiding behavior event by combining the current diagnosis guiding calculation time, wherein the diagnosis guiding behavior event comprises the personnel node identifier, the target physical examination queue identifier and an event time identifier; based on the diagnosis guiding behavior event, executing state write-back operation on the personnel reachable relation edge in the dynamic heterogeneous diagnosis guiding graph structure, and updating the corresponding reachability characteristics according to the personnel node identification and the target physical examination queue identification; Based on the same diagnosis guiding behavior event, executing state write-back operation on the physical examination queue node, and updating corresponding queue operation characteristics according to the target physical examination queue identifier; And entering a diagnosis guiding calculation process driven by the event flow in the next round of examination process based on the updated dynamic heterogeneous diagnosis guiding graph structure.

Description

Intelligent diagnosis guiding method for physical examination personnel based on graph neural network Technical Field The invention relates to the technical field of physical examination guiding and diagnosing and decision support, in particular to an intelligent guiding and diagnosing method for physical examination staff based on a graph neural network. Background Along with the continuous improvement of the large-scale operation of the physical examination center and the complexity of physical examination projects, the circulation efficiency of physical examination staff among multiple examination departments gradually becomes an important factor influencing the physical examination experience and the operation efficiency. The existing physical examination guiding and diagnosing system is mostly dependent on fixed flow configuration or path recommending mode based on simple rules, and the examination sequence of physical examination personnel is preset and arranged, so that the system is difficult to adapt to the concurrent operation environment of personnel quantity change, equipment state fluctuation and individual difference in the physical examination site. Part of researches try to introduce queuing theory models or traditional operation research methods to optimize physical examination flows, but the methods generally rely on simplifying assumptions, are difficult to uniformly model multiple types of entities, multiple constraint relations and real-time change states existing in physical examination processes, and lack stable global reasoning capability when facing high-dimensional and dynamic data, so that local optimization and even new congestion problems are easily caused. In addition, the existing diagnosis guiding technology cannot effectively treat the dependency relationship among physical examination projects, the coupling influence among physiological state constraint and space accessibility conditions, diagnosis guiding decisions are based on single indexes, systematic description of the overall operation state of a physical examination center is lacked, repeated waiting of physical examination personnel, frequent route round trip, uneven distribution of department loads are caused, and overall operation efficiency is difficult to improve. Therefore, how to provide an intelligent diagnosis guiding method for physical examination staff based on a graph neural network is a problem to be solved by the person skilled in the art. Disclosure of Invention The invention aims to provide an intelligent diagnosis guiding method for physical examination staff based on a graphic neural network, which is characterized in that a dynamic heterogeneous diagnosis guiding graph structure is constructed, unified modeling is carried out on staff nodes, physical examination queue nodes and relations thereof, and two-stage reasoning calculation is carried out on the state representation of the diagnosis guiding graph based on an improved HGT network, so that state perception, state competition and closed loop updating in the physical examination diagnosis guiding process are realized. According to the invention, through the synergistic effect of the candidate executable guided diagnosis state unit set, the load propagation reasoning result and the guided diagnosis behavior event, a continuous guided diagnosis calculation process is formed under the driving of the physical examination process event stream, so that the dynamic change of the running state of the physical examination center can be adapted, the overall consistency and stability of guided diagnosis decision can be improved, the problems of repeated waiting and queue load unbalance are reduced, and the method has the advantages of strong continuity of the guided diagnosis process, complete state expression and clear decision closed loop. According to the embodiment of the invention, the intelligent diagnosis guiding method for the physical examination personnel based on the graph neural network comprises the following steps: acquiring real-time operation data of a physical examination center and individual data of physical examination personnel, and constructing a dynamic heterogeneous diagnosis guiding graph structure, wherein the dynamic heterogeneous diagnosis guiding graph structure comprises personnel nodes, physical examination queue nodes, queue space relationship sides, project dependency relationship sides and personnel reachable relationship sides; Receiving a physical examination process event stream, updating a dynamic heterogeneous diagnosis guiding chart structure, and generating a diagnosis guiding chart state representation; Constructing an improved HGT network, comprising a first-stage HGT network and a second-stage HGT network, inputting a diagnosis-guiding diagram state representation, and executing state sensing calculation based on the first-stage HGT network to generate a personnel node state embedded representation and a physic