Medical institution full-process doctor-seeing path planning method and system based on artificial intelligence

Abstract

The invention belongs to the field of medical information systems, and provides a medical institution full-process doctor-seeing path planning method and system based on artificial intelligence, and the method comprises the steps: modeling each doctor-seeing link in a medical institution into a directed graph; the node features and the edge weights of the directed graph serve as input, the node features are updated through a graph neural network, and the weight of each node is dynamically generated; a queuing theory model is adopted for modeling according to the queuing condition of each node, and the average waiting time of each node is calculated; designing a multi-objective optimization function, and taking balance of the shortest path and the minimum waiting time as an optimization objective to generate an optimal doctor-seeing path scheme; the resource state of each node is dynamically monitored based on a real-time feedback mechanism of the Internet of Things equipment and a hospital information system, and when it is monitored that the resource state or waiting time of the node is changed, the graph neural network and the queuing theory model are recalled to update node features and priority weights.

Description

Technical Field

[0001] The present invention belongs to the field of medical information systems, and specifically relates to a method and system for planning the entire medical treatment path of a medical institution based on artificial intelligence. Background Art

[0002] In the current medical institution's medical treatment process, patients need to go through multiple links from admission to completion of treatment, such as registration, triage, examination, diagnosis, treatment, and medication collection. Due to the dynamic changes in resource distribution, availability of doctors or equipment, patient mobility, and waiting time in each link, the efficiency of the medical treatment process is often reduced, and patients have to wait too long, affecting the medical experience. Especially in large and medium-sized medical institutions, due to the large number of patients received per day, the load fluctuations of each medical treatment node are more obvious, and resource allocation is difficult to achieve real-time optimization. There is often a phenomenon of excessive concentration of resources in some links and long waiting times, resulting in low utilization of medical resources and poor patient experience.

[0003] Traditional medical path planning methods are mainly based on fixed processes, that is, patients are assigned to various departments and examination links according to their initial diagnosis of their condition, and lack a real-time adjustment mechanism, making it difficult to cope with dynamically changing patient flow and resource load. At the same time, traditional methods are slow to respond and lack flexibility when dealing with emergencies (such as an increase in emergency patients, equipment failure, temporary unavailability of doctors, etc.), resulting in an imbalance in resource allocation and a significant increase in the average waiting time of patients. In addition, the lack of effective data feedback means makes it difficult to monitor the status of each medical node in real time, so the potential for optimizing the path has not been fully tapped. Summary of the invention

[0004] In order to solve the problems in the prior art, the present invention provides a method for planning a full-process medical treatment path in a medical institution based on artificial intelligence, and the method comprises the following steps:

[0005] Modeling each medical consultation link in a medical institution as a directed graph, wherein the nodes of the directed graph include at least service rate, queue length and patient priority feature information, the directed edges between the nodes represent the transfer relationship of the medical consultation process, and the edge weight represents the transfer time and priority of the process;

[0006] The node features and edge weights of the directed graph are used as input, the node features are updated using a graph neural network, and the weight of each node is dynamically generated by combining the state information of neighboring nodes through an attention mechanism;

[0007] The node weight, service rate and queue length generated by the graph neural network are used as the input of the queuing theory model. The queuing theory model is used to model the queuing situation of each node and calculate the average waiting time of each node.

[0008] Combining the node weights output by the graph neural network and the node waiting time output by the queuing theory model, a multi-objective optimization function is designed to balance the shortest path and the lowest waiting time as the optimization goal, and the optimal medical path plan is generated;

[0009] Based on the real-time feedback mechanism of IoT devices and hospital information systems, the resource status of each node is dynamically monitored. When changes are detected in the resource status or waiting time of the node, the graph neural network and queuing theory model are re-called to update the node characteristics and priority weights.

[0010] On the other hand, the present invention also provides a medical institution full-process medical path planning system based on artificial intelligence, and the system includes the following modules:

[0011] A modeling module, used to model each medical consultation link in a medical institution as a directed graph, wherein the nodes of the directed graph include at least service rate, queue length and patient priority feature information, the directed edges between the nodes represent the transfer relationship of the medical consultation process, and the edge weight represents the transfer time and priority of the process;

[0012] A weight updating module, which is used to take the node features and edge weights of the directed graph as input, update the node features using a graph neural network, and dynamically generate the weight of each node by combining the state information of neighboring nodes through an attention mechanism;

[0013] The queuing model module is used to use the node weight, service rate and queue length generated by the graph neural network as the input of the queuing theory model, use the queuing theory model to model the queuing situation of each node, and calculate the average waiting time of each node;

[0014] The optimization module is used to combine the node weights output by the graph neural network and the node waiting time output by the queuing theory model, design a multi-objective optimization function, balance the shortest path and the lowest waiting time as the optimization goal, and generate the optimal medical path plan;

[0015] The update module is used to dynamically monitor the resource status of each node based on the real-time feedback mechanism of IoT devices and hospital information systems. When changes are detected in the resource status or waiting time of the node, the graph neural network and queuing theory model are re-called to update the node characteristics and priority weights.

[0016] The present invention combines graph neural network and queuing theory models to achieve real-time monitoring of the resource status of each consultation node in a medical institution and dynamic path optimization, which has the following beneficial effects:

[0017] The load and resource status of each node can be obtained in real time through IoT devices and hospital information systems. The graph neural network and queuing theory models can dynamically adjust the priority and resource allocation of service nodes to avoid excessive concentration or idleness of resources and improve the overall utilization of hospital resources.

[0018] Based on the queuing theory model, the average waiting time of each node is accurately calculated, and the shortest path and waiting time are balanced by combining the multi-objective optimization function to generate the optimal medical treatment path in real time, significantly reducing the waiting time of patients at each node and optimizing the efficiency of the medical treatment process.

[0019] When encountering emergencies (such as an increase in emergency patients, equipment failure, etc.), the system can quickly call graph neural networks and queuing theory models to update node weights and waiting times, and dynamically adjust the paths through real-time feedback mechanisms to ensure that each link responds quickly to changes and ensures the continuity and efficiency of the medical process.

[0020] The present invention greatly improves the process management efficiency within medical institutions, provides patients with a faster and more user-friendly medical experience, and significantly optimizes the allocation of medical resources. It is an innovative and practical medical process optimization solution. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0022] Figure 1 is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0023] Below, the invention is preferably described in conjunction with the accompanying drawings and specific implementation methods.

[0024] This embodiment solves the above problem through the following steps:

[0025] In one embodiment, reference Figure 1 The present invention provides a method for planning the entire medical treatment path in a medical institution based on artificial intelligence, including modeling, optimizing and dynamically adjusting each medical treatment link in the medical institution through a graph neural network and a queuing theory model to achieve efficient medical treatment path planning. The method comprehensively considers the patient's medical treatment needs, the status of medical resources and the real-time changing waiting situation, automatically plans and recommends the optimal medical treatment path, thereby reducing the patient's waiting time, improving the efficiency of medical treatment, and effectively allocating and utilizing the hospital's medical resources. Specifically, the method includes the following steps:

[0026] Each medical consultation link in a medical institution is modeled as a directed graph, wherein the nodes of the directed graph include at least characteristic information such as service rate, queue length, and patient priority, and the directed edges between the nodes represent the transfer relationship of the medical consultation process, and the edge weight represents the transfer time and priority of the process.

[0027] The nodes represent specific consultation links in a medical institution, and each node includes at least the following characteristic information:

[0028] Service rate: The service rate is the ability of a node to process patients, indicating the number of patients that the node can serve per unit time. The service rate is dynamically assigned by analyzing historical data and current resource conditions to reflect the current processing capacity of the node.

[0029] Queue length: The queue length is the number of patients waiting for service at the current node. The queue length is updated in real time according to the current service load of the node and is used to measure the waiting situation of patients at the node.

[0030] Patient priority: The patient priority is assigned based on factors such as the severity of the patient's condition, urgency, and appointment order, and is used to indicate the priority level of different patients at this node. Patient priority information is transmitted between nodes to ensure that patients with more urgent conditions receive priority services.

[0031] The nodes in the directed graph are connected by directed edges, which represent the transfer relationship of the medical treatment process. The directionality of the directed edges defines the flow order of the patient at each node in the medical treatment process, ensuring the rationality and coherence of the medical treatment path.

[0032] In addition, the weight of the directed edge represents the transfer time and priority between nodes, where:

[0033] Transfer time: The transfer time refers to the time required for a patient to transfer from one node to the next, including the actual movement time of the patient between different consultation links and the required preparation time. The transfer time is assigned based on historical data and real-time monitoring information, and can be dynamically adjusted to reflect changes in the actual process.

[0034] Priority: The priority is the transfer order weight between different nodes, which is used to give priority to specific transfer paths in path planning to optimize the overall treatment efficiency of patients. The priority is set based on comprehensive factors such as the service rate of each node, patient priority, and availability of medical resources to achieve reasonable resource allocation and patient flow optimization.

[0035] The node features and edge weights of the directed graph are used as input, the node features are updated using a graph neural network, and the weight of each node is dynamically generated by combining the state information of neighboring nodes through an attention mechanism.

[0036] In this step, the node features and edge weights of the directed graph are used as input, and the node features are updated using a graph neural network, which specifically includes the following steps:

[0037] The initial feature information of each node, including service rate, queue length, and patient priority, is converted into a feature vector. Each feature vector can be represented as a high-dimensional vector ,in Indicates the node number. To ensure data consistency, the feature values ​​of different units or dimensions are normalized, and all features are standardized to the same numerical range (such as [0,1]) to facilitate subsequent model processing and comparison.

[0038] The weight features of each directed edge, including transfer time and priority, are encoded into the edge weight matrix ,in and Indicates the node number of the connection. The edge weight matrix provides the model with time and priority information between nodes. The edge weight is also normalized to have a stable value range in network propagation.

[0039] Through the initialization layer of the graph neural network, the initial feature vector of each node is and the edge weight matrix As input, it is embedded into the graph neural network to form the input features for information propagation. The information of neighbor nodes is passed to each node, and the features from neighbors are accumulated in each layer of the graph neural network through the information aggregation function.

[0040] The graph attention network structure is used so that in each layer of the network, the node's neighbor information is calculated based on the attention mechanism. In the GAT structure, a multi-head attention mechanism is set up so that each node can use different weight allocations when aggregating the information of neighboring nodes, enhancing the flexibility of the model and the feature extraction ability.

[0041] For each pair of neighbor nodes and , calculate its attention coefficient , which represents the node

[0042] For Node importance.

[0043] Using a linear transformation matrix Perform linear transformation on node features to obtain transformed features , to ensure that the features are calculated in the same space. Next, the similarity score of the two node features is calculated and used as the input of the attention coefficient.

[0044] To ensure that the attention coefficients are comparable, The attention coefficients of all neighbor nodes are normalized to obtain the normalized attention weights , and its calculation formula is:

[0045]

[0046] in, Representation Node The normalization operation uses the softmax function to normalize the attention coefficients of all neighbors to the range of [0, 1] to ensure that the sum of all attention coefficients is 1.

[0047] Using the calculated attention coefficient Perform weighted summation on the neighbor features of each node to generate the node's feature update value. The specific aggregation formula is as follows:

[0048]

[0049] in, Indicates Layer Node The updated features of is an activation function (preferably, ReLU) used to enhance the nonlinear expression ability of the model.

[0050] Repeat the above feature update process multiple times, and gradually update the node features in each layer of the graph neural network. Through multi-layer attention aggregation, the features of each node can be integrated with information from more distant nodes, and global context information can be gradually obtained. Finally, by superimposing multi-layer attention calculations, a global node feature that integrates the status of neighboring nodes is generated, which can effectively represent the status information of each node in the medical process.

[0051] After the feature update of the multi-layer graph neural network, each node feature contains the state information of its neighbor nodes, and has integrated the results of multi-layer attention calculations. The final node feature is used as the node weight The output is used to guide node priority sorting and path selection in medical path planning.

[0052] The generated node weights can be used in the multi-objective optimization path planning model to dynamically determine the optimal path for patients, combined with other information such as patient waiting time and node service rate. Nodes with higher weights have higher priority in path planning, reflecting the urgency and importance of the patient at that node.

[0053] The node weight, service rate and queue length generated by the graph neural network are used as the input of the queuing theory model. The queuing theory model is used to model the queuing situation of each node and calculate the average waiting time of each node.

[0054] This paper uses the M / M / c queuing model to model the queuing conditions of multiple service nodes in medical institutions to calculate the average waiting time of each node and optimize the medical path. In order to more accurately reflect the actual medical needs, the model takes the node weight, service rate, queue length and other information generated by the graph neural network as input.

[0055] Specifically, the following parameters are used:

[0056] Reach :

[0057] in, Represents the unweighted basic arrival rate, which refers to the number of patients arriving at the node per unit time. After weighted processing, we get the weighted arrival rate .

[0058] A node with a higher weight guides more patients to flow to this node, reflecting its priority and importance in pathway planning.

[0059] Service Rate :

[0060] in, is the basic service rate of the node, which indicates the number of patients that a single service provider can handle per unit time.

[0061] Node Weight Directly adjust the service rate so that nodes with higher weights can obtain faster service speeds in the average waiting time calculation, ensuring that the waiting time of key nodes is shorter.

[0062] Number of Services

[0063] in, The basic service number of the node, which indicates the number of services that the node can serve concurrently (such as the number of doctors or devices).

[0064] By node weight The number of services is adjusted dynamically. Nodes with higher weights are assigned more services first to reduce the node load and waiting time. The number of services is rounded to ensure that the actual number of available services is an integer.

[0065] Utilization

[0066] Combining the above weighted arrival rate, service rate and number of services, the actual utilization of the node can be defined. Nodes with higher weights are dynamically allocated more resources to reduce utilization, thereby reducing the average waiting time.

[0067] System Idle Probability Indicates that at the node The probability that there are no patients in the queuing system. The calculation formula is as follows:

[0068]

[0069] After obtaining the idle probability, calculate the average waiting time

[0070]

[0071] Combining the node weights output by the graph neural network and the node waiting time output by the queuing theory model, a multi-objective optimization function is designed to balance the shortest path and the lowest waiting time as the optimization goal to generate the optimal medical path plan.

[0072] In the present invention, the key to multi-objective optimization is to balance two main objectives:

[0073] Shortest Path:

[0074] By taking the shortest path, we can reduce the distance and time patients travel within the hospital. This goal can improve overall process efficiency and avoid increasing waiting costs for patients due to long-distance travel.

[0075] The path length between nodes Represents the patient from the node Transfer to Node time or space distance.

[0076] Minimum waiting time:

[0077] By reducing the average waiting time of patients at each node, the waiting time of patients in each medical link is minimized, thereby optimizing the medical experience.

[0078] Node waiting time output by the queuing theory model (user exist The waiting time of each node is taken as the input of the optimization target to ensure that the waiting time of each node is controlled and optimized in path planning.

[0079] Define a multi-objective optimization function To balance the shortest path and the lowest waiting time. The specific optimization function is designed as follows:

[0080]

[0081] in:

[0082] A collection of nodes representing the selected path;

[0083] Representation Node To Node the length of the path (e.g., time or spatial distance);

[0084] Nodes calculated for the queuing theory model Average waiting time;

[0085] The node weight representing the output of the graph neural network indicates the priority or importance of the node;

[0086] and is a weight coefficient used to balance the relative importance between path length and waiting time. The value will give priority to the shortest path, while higher The value will give priority to the minimum waiting time.

[0087] The steps to generate the optimal treatment path plan include:

[0088] 1. Initialize the candidate path set:

[0089] Generate several candidate paths randomly or generate candidate paths through preliminary planning of graph neural network. Calculate the path length and node waiting time of each candidate path.

[0090] 2. Path evaluation:

[0091] Using the optimization function The candidate paths are evaluated and the comprehensive score of each path is calculated.

[0092] Based on the scores, the paths with better performance are selected to enter the next round of optimization.

[0093] 3. Iterative update:

[0094] Generate new candidate paths through operations such as crossover and mutation, and continuously iterate and optimize.

[0095] In each generation, the best performing path is retained, while appropriate diversity is introduced to prevent local optimality.

[0096] 4. Generate the optimal path plan:

[0097] When the iteration reaches the set termination condition or converges, the path plan with the best score is selected as the final optimal medical treatment path.

[0098] The generated optimal treatment path plan will be used to guide the patient's treatment process. The system recommends the treatment order based on the multi-objective optimization results, and displays the path information in real time on the patient's mobile terminal or self-service device, including the estimated waiting time and the travel time between nodes, to ensure that the patient can complete the entire process of treatment according to the optimal path.

[0099] Based on the real-time feedback mechanism of IoT devices and hospital information systems, the resource status of each node is dynamically monitored. When changes are detected in the resource status or waiting time of the node, the graph neural network and queuing theory model are re-called to update the node characteristics and priority weights.

[0100] IoT devices are installed at each medical node (such as registration, triage, examination, treatment, etc.) to collect key resource status information of the node in real time. The data collected by IoT devices include:

[0101] Equipment status: Check the usage and fault detection of equipment, treatment instruments, etc. Upload data in real time through equipment sensors to determine whether the equipment is in operation, idle or fault status.

[0102] Availability of doctors and medical staff: Monitor the location, status and current number of patients of medical staff in each node to ensure that the system can accurately grasp the real-time configuration of human resources.

[0103] Patient arrival and departure information: Use technologies such as RFID (radio frequency identification) or Bluetooth to track the patient's real-time location and record the patient's waiting and service time at each node.

[0104] The hospital information system (HIS) provides data such as daily scheduling, appointment status, patient medical records, and examination results for each node, and integrates data collected by IoT devices with HIS data:

[0105] Node appointment status: HIS data provides the number of patient appointments and estimated load for each node, assisting in determining changes in waiting time.

[0106] Patient condition level: The patient's medical record data provides a basis for updating the priority of each node to ensure that critically ill patients receive priority treatment.

[0107] Historical traffic data: Integrate historical arrival rate data to predict the future load of nodes and optimize in advance.

[0108] Based on the real-time data provided by IoT devices and HIS systems, the system dynamically monitors the following key parameters to determine whether the node resource status needs to be updated:

[0109] Node utilization: Real-time monitoring of the current utilization of the node, that is, the ratio of the number of patients to the number of services. When the utilization exceeds the set threshold, the system identifies it as a load peak.

[0110] Average waiting time: The average waiting time of a node is recalculated at regular intervals. If the waiting time increases significantly (for example, it exceeds 20% of the historical average), the system marks it as a potential anomaly.

[0111] Equipment and human resource status: If resource changes such as node equipment failure, doctor absence or addition are detected, the system automatically records the event and triggers the update process.

[0112] When the monitoring system detects a significant change in resource status or waiting time, the system automatically triggers the following steps, calling the graph neural network and queuing theory model to update the node characteristics and priority weights in real time:

[0113] (1) Graph neural network recalculates node weights

[0114] Once a dynamic update is triggered, the system first calls the graph neural network again to update the priority weights of the nodes based on the latest monitoring data:

[0115] Node feature update: Input new feature data such as arrival rate, service rate, service quantity, device status, etc. into the graph neural network. The graph neural network dynamically adjusts the weight of each node through the adjacency relationship so that the priority is more in line with the current resource status.

[0116] Weight adjustment: For nodes with long waiting time or high load, the graph neural network will increase their priority weight to ensure that resources are directed to these nodes first in path planning. For nodes with idle resources, their weights will be appropriately reduced to reduce unnecessary resource allocation.

[0117] (2) Queuing theory model updates average waiting time

[0118] After obtaining the latest node weights, the system further calls the queuing theory model to recalculate the average waiting time based on the new node arrival rate, service rate, and service quantity:

[0119] Weighted arrival rate adjustment: According to the node weights output by the graph neural network, the arrival rate of each node is dynamically adjusted to ensure that the arrival rate of high-priority nodes is increased accordingly.

[0120] Service rate adjustment: If the number of services (doctors, departments, instruments, etc.) is monitored to increase or decrease, or the status of doctors / equipment changes, the service rate of the node is dynamically updated and the average waiting time of the node is recalculated.

[0121] Waiting time recalculation: Based on the updated arrival rate and service rate, the queuing theory model recalculates the average waiting time of each node to ensure that the waiting time reflects the latest resource status.

[0122] After the graph neural network and queuing theory model re-output the node weights and waiting times, the system adjusts the path planning scheme in real time to optimize the patient flow path:

[0123] Path replanning: Re-call the multi-objective optimization function based on the updated node priority weights and average waiting time , while ensuring the shortest path, give priority to the path plan with the shortest waiting time.

[0124] Dynamic allocation of resource scheduling: After path replanning, the system dynamically allocates service resources, such as increasing the number of doctors or equipment at high-priority nodes, to ensure that patients can complete the medical treatment process with the shortest waiting time.

[0125] After completing the path replanning, the system will feed back the updated optimal path plan to the patient, guiding him to complete the medical treatment tasks of each node in sequence:

[0126] Through the hospital app or SMS notifications, new treatment routes and estimated waiting times at each node are pushed to patients in real time to ensure that patients are aware of the latest treatment arrangements.

[0127] The hospital's self-service terminals display information about changes in medical treatment routes in real time, helping patients to promptly respond to queue adjustments caused by changes in node load.

[0128] On the other hand, the present invention also provides a medical institution full-process medical path planning system based on artificial intelligence, including:

[0129] A modeling module, used to model each medical consultation link in a medical institution as a directed graph, wherein the nodes of the directed graph include at least service rate, queue length and patient priority feature information, the directed edges between the nodes represent the transfer relationship of the medical consultation process, and the edge weight represents the transfer time and priority of the process;

[0130] A weight updating module, which is used to take the node features and edge weights of the directed graph as input, update the node features using a graph neural network, and dynamically generate the weight of each node by combining the state information of neighboring nodes through an attention mechanism;

[0131] The queuing model module is used to use the node weight, service rate and queue length generated by the graph neural network as the input of the queuing theory model, use the queuing theory model to model the queuing situation of each node, and calculate the average waiting time of each node;

[0132] The optimization module is used to combine the node weights output by the graph neural network and the node waiting time output by the queuing theory model, design a multi-objective optimization function, balance the shortest path and the lowest waiting time as the optimization goal, and generate the optimal medical path plan;

[0133] The update module is used to dynamically monitor the resource status of each node based on the real-time feedback mechanism of IoT devices and hospital information systems. When changes are detected in the resource status or waiting time of the node, the graph neural network and queuing theory model are re-called to update the node characteristics and priority weights.

[0134] The prior art mentioned in the above background technology section and specific embodiment section of the present invention can be used as part of the present invention to understand the meaning of some technical features or parameters.

Claims

1. A full-process medical institution treatment path planning method based on artificial intelligence, characterized in that: The method comprises the following steps: Modeling each medical consultation link in a medical institution as a directed graph, wherein the nodes of the directed graph include at least service rate, queue length and patient priority feature information, the directed edges between the nodes represent the transfer relationship of the medical consultation process, and the edge weight represents the transfer time and priority of the process; The node features and edge weights of the directed graph are used as input, the node features are updated using a graph neural network, and the weight of each node is dynamically generated by combining the state information of neighboring nodes through an attention mechanism; The node weight, service rate and queue length generated by the graph neural network are used as the input of the queuing theory model. The queuing theory model is used to model the queuing situation of each node and calculate the average waiting time of each node. Combining the node weights output by the graph neural network and the node waiting time output by the queuing theory model, a multi-objective optimization function is designed to balance the shortest path and the lowest waiting time as the optimization goal, and the optimal medical path plan is generated; Based on the real-time feedback mechanism of IoT devices and hospital information systems, the resource status of each node is dynamically monitored. When changes are detected in the resource status or waiting time of the node, the graph neural network and queuing theory model are re-called to update the node characteristics and priority weights.

2. The method for planning the entire medical institution's medical treatment path based on artificial intelligence according to claim 1 is characterized in that: The node features and edge weights of the directed graph are used as input, the node features are updated using a graph neural network, and the weight of each node is dynamically generated by combining the state information of neighboring nodes through an attention mechanism, including: Convert the initial feature information of each node, including service rate, queue length, and patient priority, into a feature vector; Each feature vector is represented as a high-dimensional vector ,in Indicates the node number; The weight features of each directed edge, including transfer time and priority, are encoded into the edge weight matrix ,in and Indicates the node number of the connection; Through the initialization layer of the graph neural network, the initial feature vector of each node is and the edge weight matrix As input, it is embedded into the graph neural network to form input features for information propagation; Pass the information of neighbor nodes to each node, and accumulate features from neighbors in each layer of the graph neural network through the information aggregation function; Using a linear transformation matrix Perform linear transformation on node features to obtain transformed features ; For Node The attention coefficients of all neighbor nodes are normalized to obtain the normalized attention weights ; Using the calculated attention coefficient Perform weighted summation on the neighbor features of each node to generate the node’s feature update value; Repeat the above feature update process multiple times to gradually update the node features in each layer of the graph neural network; After the feature update of the multi-layer graph neural network, each node feature It contains the state information of its neighbor nodes, has integrated the results of multi-layer attention calculation, and uses the final node features as the node weight Output.

3. The method for planning the entire medical institution's medical treatment path based on artificial intelligence according to claim 2 is characterized in that: The node weight, service rate and queue length generated by the graph neural network are used as the input of the queuing theory model, and the queuing situation of each node is modeled using the queuing theory model, including: The M / M / c queuing model is used to model the queuing situation of multiple service nodes in medical institutions.

4. The method for planning the entire medical institution's medical treatment path based on artificial intelligence according to claim 3 is characterized in that: The calculation of the average waiting time of each node includes: Calculate the system idle probability.

5. The full-process medical institution treatment path planning system based on artificial intelligence is characterized by: The system includes the following modules: A modeling module, used to model each medical consultation link in a medical institution as a directed graph, wherein the nodes of the directed graph include at least service rate, queue length and patient priority feature information, the directed edges between the nodes represent the transfer relationship of the medical consultation process, and the edge weight represents the transfer time and priority of the process; A weight updating module, which is used to take the node features and edge weights of the directed graph as input, update the node features using a graph neural network, and dynamically generate the weight of each node by combining the state information of neighboring nodes through an attention mechanism; The queuing model module is used to use the node weights, service rates and queue lengths generated by the graph neural network as the input of the queuing theory model, and use the queuing theory model to model the queuing conditions of each node and calculate the average waiting time of each node; The optimization module is used to combine the node weights output by the graph neural network and the node waiting time output by the queuing theory model, design a multi-objective optimization function, balance the shortest path and the lowest waiting time as the optimization goal, and generate the optimal medical path plan; The update module is used to dynamically monitor the resource status of each node based on the real-time feedback mechanism of IoT devices and hospital information systems. When changes are detected in the resource status or waiting time of the node, the graph neural network and queuing theory model are re-called to update the node characteristics and priority weights.

6. The artificial intelligence-based full-process medical institution treatment path planning system according to claim 5 is characterized in that: The node features and edge weights of the directed graph are used as input, the node features are updated using a graph neural network, and the weight of each node is dynamically generated by combining the state information of neighboring nodes through an attention mechanism, including: Convert the initial feature information of each node, including service rate, queue length, and patient priority, into a feature vector; Each feature vector is represented as a high-dimensional vector ,in Indicates the node number; The weight features of each directed edge, including transfer time and priority, are encoded into the edge weight matrix ,in and Indicates the node number of the connection; Through the initialization layer of the graph neural network, the initial feature vector of each node is and the edge weight matrix As input, it is embedded into the graph neural network to form input features for information propagation; Pass the information of neighbor nodes to each node, and accumulate features from neighbors in each layer of the graph neural network through the information aggregation function; Using a linear transformation matrix Perform linear transformation on node features to obtain transformed features ; For Node The attention coefficients of all neighbor nodes are normalized to obtain the normalized attention weights ; Using the calculated attention coefficient Perform weighted summation on the neighbor features of each node to generate the node’s feature update value; Repeat the above feature update process multiple times to gradually update the node features in each layer of the graph neural network; After the feature update of the multi-layer graph neural network, each node feature It contains the state information of its neighbor nodes, has integrated the results of multi-layer attention calculation, and uses the final node features as the node weight Output.

7. The artificial intelligence-based full-process medical institution treatment path planning system according to claim 6 is characterized in that: The node weight, service rate and queue length generated by the graph neural network are used as the input of the queuing theory model, and the queuing situation of each node is modeled using the queuing theory model, including: The M / M / c queuing model is used to model the queuing situation of multiple service nodes in medical institutions.

8. The artificial intelligence-based full-process medical institution treatment path planning system according to claim 7 is characterized in that: The calculation of the average waiting time of each node includes: Calculate the system idle probability.

Images