Artificial intelligence auxiliary inquiry real-time data scheduling method and system

Abstract

The invention discloses an artificial intelligence auxiliary inquiry real-time data scheduling method and system. The method comprises the following steps: S1, forming a preliminary medical data set; s2, respectively constructing corresponding game tree models; s3, embedding the priority scheduling path into the game tree model; s4, establishing a cross-layer communication mechanism to realize state interaction and information transmission between the heterogeneous medical data hierarchical game tree models; s5, generating a final medical data scheduling plan; and S6, executing a final medical data scheduling plan, preferentially transmitting the medical data with high priority to a doctor end or a diagnosis model, enabling the medical data with low priority to enter a cache module, adjusting a subsequent transmission sequence of the medical data according to a global revenue function, and recording log information of medical data scheduling at the same time. According to the invention, the interference of low-priority data on key tasks is reduced, and the overall scheduling efficiency and the service quality of medical scenes are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of assisted medical consultation, and in particular to a method and system for real-time data scheduling of artificial intelligence assisted medical consultation. Background Art

[0002] In the existing technology, data scheduling in medical scenarios mainly relies on traditional rule settings and predefined processes. These methods are usually centered on fixed priorities and simple allocation strategies. They guide data transmission and processing by statically prioritizing medical data according to patient category, diagnostic complexity or data source. The rule-based static scheduling method has obvious shortcomings when processing multi-source, heterogeneous and real-time medical data. The correlation and priority requirements between different data types in complex medical scenarios are difficult to fully cover through static rules, resulting in delayed transmission of key data, affecting diagnosis and treatment efficiency and decision-making accuracy.

[0003] In order to solve some problems in data scheduling, some existing technologies have introduced dynamic scheduling methods based on distributed systems. Existing technologies use simple load balancing algorithms or resource allocation models to try to improve scheduling efficiency when medical resources are limited. However, existing technologies have weak response capabilities to dynamic changes in data and real-time needs. When faced with high-concurrency, multi-type medical data transmission tasks, resource allocation is prone to unevenness and key tasks are blocked. In addition, most existing distributed scheduling technologies focus on a single data type or a single task scenario and fail to effectively integrate the collaborative needs of multiple types of data, making it difficult to meet the diverse requirements of task complexity and urgency in actual medical scenarios.

[0004] Another attempt is to introduce artificial intelligence technology into medical data scheduling, predicting data load and optimizing scheduling paths through machine learning-based models. However, machine learning models are usually limited to single-time optimization of offline training models and lack the ability to dynamically adjust scheduling strategies in real-time environments. In addition, existing artificial intelligence scheduling technologies have low fusion processing capabilities for heterogeneous data and are unable to comprehensively analyze and optimize the comprehensive scheduling needs of structured data, unstructured data, and real-time data. Due to the lack of cross-task and cross-data type scheduling coordination mechanisms, in scenarios where high-priority tasks and low-priority tasks coexist, it is easy to cause waste of resources or delays in key tasks. Summary of the invention

[0005] One object of the present invention is to propose a real-time data scheduling method and system for artificial intelligence-assisted medical consultation, which reduces the interference of low-priority data on key tasks and improves the overall scheduling efficiency and service quality of medical scenarios.

[0006] According to an embodiment of the present invention, an artificial intelligence-assisted medical consultation real-time data scheduling method includes the following steps:

[0007] S1. Collect and classify different types of medical data, pre-process each type of medical data, and form a preliminary medical data set;

[0008] S2. Based on the heterogeneity of medical data types, the preliminary medical data set is divided into several independent decision levels, each decision level corresponds to a medical data type, and the corresponding game tree models are constructed respectively;

[0009] S3. Initialize the reinforcement learning strategy for the initial state of each game tree model, design a local optimization strategy based on the characteristic distribution of medical data, generate a priority scheduling path for medical data, and embed the priority scheduling path into the game tree model;

[0010] S4. Establish a cross-layer communication mechanism to enable state interaction and information transfer between heterogeneous medical data hierarchical game tree models;

[0011] S5. Based on the cross-layer communication mechanism, the local scheduling results of the game tree model at each layer are globally optimized, the real-time demand and resource allocation status of the medical scenario are mapped into a global benefit function, and the global scheduling strategy is iteratively optimized through reinforcement learning to generate the final medical data scheduling plan;

[0012] S6. Execute the final medical data scheduling plan, give priority to transmitting high-priority medical data to the doctor side or diagnostic model, and low-priority medical data enters the cache module. The subsequent transmission order of medical data is adjusted according to the global benefit function, and the log information of medical data scheduling is recorded at the same time.

[0013] Optionally, S1 includes the following steps:

[0014] S11. Collect medical data, including structured medical record data, unstructured medical imaging data, and real-time monitoring data, and mark the collected medical data as a medical data set:

[0015] D raw ={d1,d2,…,d n},1≤i≤n;

[0016] Among them, d i represents the i-th medical data, and n is the total number of medical data;

[0017] S12. Medical data set D raw The medical data in the data are classified and processed, and the medical data are divided into three subsets according to the data type: structured medical record data set D text , unstructured medical imaging data set D image and real-time monitoring data set D real_time :

[0018] D raw =D text ∪D image ∪D real_time ;

[0019] S13. Format the structured medical record data in the structured medical record data set, extract key fields and generate a standardized feature vector f i text ;

[0020] Perform image denoising, resolution standardization and ROI extraction on the unstructured medical image data in the unstructured medical image data set to generate the processed image tensor Where h, w, and c are the height, width, and number of channels of the image respectively;

[0021] Perform sampling rate alignment and time series interpolation processing on the real-time monitoring data in the real-time monitoring data set to generate a time series feature matrix Among them, t is the time step and k is the feature dimension;

[0022] S14. Combining the preprocessed medical data to form a preliminary medical data set D processed :

[0023]

[0024] where n text 、n image 、n real_time Respectively represent the number of structured medical record data, unstructured medical imaging data and real-time monitoring data, satisfying n text +n image +n real_time =n.

[0025] Optionally, S2 includes the following steps:

[0026] S21. Based on preliminary medical dataset D processed According to the heterogeneity of medical data types, the preliminary medical data set is divided into several independent decision levels, each decision level corresponds to a type of medical data;

[0027] Define the medical data type set K and the data type mapping function φ: D processed →K,φ(d i ) = k represents medical data d i The type is k, and the preliminary medical data set D is transformed according to the mapping function φ. processed Divided into decision-making levels L k , and subdivided into three decision-making levels:

[0028]

[0029] Among them, L text , L image , L real_time The decision-making levels corresponding to structured medical record data, unstructured medical imaging data, and real-time monitoring data respectively;

[0030] S22. Construct a game tree model for the medical data in each decision level:

[0031] T k =(N k ,E k );

[0032] Among them, N k is the game tree node set, E k is the edge set of the game tree;

[0033] For each medical data d i ∈L k Construct the corresponding game tree node n i ∈N k , node status S i Defined as:

[0034]

[0035] Among them, P i For medical data i Priority:

[0036]

[0037] U i Represents medical data i The urgency level is defined as:

[0038]

[0039] C i Represents medical data i Complexity:

[0040]

[0041] Among them, Size(d i ) is medical data d i The data size, For medical data i The processing complexity coefficient is Represents medical data i Timeliness:

[0042]

[0043] Among them, t now is the current time, t i For medical data i The generation time, κ is the time decay coefficient, λ1, λ2, λ3 are weight coefficients, which are used to balance the impact of urgency, complexity and timeliness on priority;

[0044] For medical data i Estimated transfer time:

[0045]

[0046] Among them, B network is the network bandwidth, η k is the network efficiency coefficient, which reflects the efficiency of medical data type k in network transmission;

[0047] R i For medical data i Processing resource requirements:

[0048]

[0049] Here, ρ is the amount of resources required per unit complexity.

[0050] Optionally, S3 includes the following steps:

[0051] S31. According to the specific needs of the medical scenario, the node state of the game tree model is initialized and the n of each node is calculated. i The comprehensive evaluation value V i :

[0052]

[0053] Among them, α, β, and γ are weight coefficients used to balance the impact of priority, expected transmission time, and processing resource requirements on the comprehensive evaluation value;

[0054] The comprehensive evaluation value V i As the node weight in the game tree model, initialize the game tree edge set E k Define the scheduling path between nodes, the edge weight W ij Calculated according to node weight:

[0055] W ij =V j -δC ij ;

[0056] Among them, C ij For slave node n i To node n jThe dispatch cost is:

[0057]

[0058] in, For slave node n i To node n j The actual transmission time of,δ,∈,is the adjustment parameter used to control the trade-off between scheduling benefit and cost;

[0059] S24. Input the states and path benefits of all nodes in the game tree model into the reinforcement learning model, and define the state space S of reinforcement learning = {S i}, action space A = {a ij}, a ij Represents slave node n i To node n j The scheduling action defines the strategy function π(a ij ∣S i ) indicates that in state S i Next select action a ij probability;

[0060] Define the value function Q(S i ,a ij ), satisfying the Bellman equation:

[0061]

[0062] Among them, R(S i ,a ij ) is the immediate reward, which is equal to the edge weight W ij , γ1 is the discount factor, P(S j ∣S i ,a ij ) is the state transition probability, which takes the value of 1 in a deterministic environment.

[0063] Optionally, S4 includes the following steps:

[0064] S41. Establish a cross-layer communication mechanism between hierarchical game tree models and define a cross-layer communication matrix M for the state interaction and information transmission of different types of medical data comm :

[0065] M comm =[m ij ];

[0066]

[0067] Among them, m ij Represents the slave node S i To Node S j The communication status, f(S i ,S j ) is the communication weight function, |P i -P j | indicates the difference in priority between nodes. The smaller the difference in priority, the better the communication status. is the timeliness weight of the timestamp difference between nodes, R j is the resource requirement of the target node, λ1, λ2, and λ3 are weight coefficients used to balance the impact of priority difference, timeliness, and resource requirements on the communication status;

[0068] S42. Based on the communication matrix M comm Establish a cross-layer scheduling path graph:

[0069] G cross =(V,E);

[0070] Where V = ∪ k N k represents the set of all game tree nodes, E = {e ij ∣m ij ≠0, i≠j} represents the scheduling path between nodes across layers;

[0071] The weight w of each path ij The communication weight function f(S i ,S j ) calculation, satisfying:

[0072]

[0073] S43. Define a task dependency matrix M used to represent the degree of dependency between different tasks task :

[0074] M task =[t ij ];

[0075]

[0076] Among them, t ij represents the dependency strength between task i and task j, μ1, μ2, and μ3 are weight coefficients, which are used to balance the impact of priority, resource demand, and transmission time on task dependency, respectively;

[0077] S44. Dynamically update the cross-layer communication mechanism according to the communication matrix M comm and the task dependency matrix M task Adjust the cross-layer path weights in real time to form an optimized cross-layer path matrix M path ;

[0078] S45. Using the optimized cross-layer path matrix Mpath Synchronize the global state of the hierarchical game tree model, update the priority, transmission time and resource requirements of each node, and redistribute the node weight V′ i .

[0079] Optionally, S5 includes the following steps:

[0080] S51. Based on cross-layer communication mechanism and path matrix M path Integrate the local scheduling results of each layer of the game tree model to construct the global profit function R global :

[0081]

[0082] Among them, P′ i , R′ i are the global post-synchronization priority, transmission time and resource requirements of node i, ω1, ω2, ω3 are the global benefit weight coefficients, which are used to balance the impact of priority, transmission time and resource consumption on the global benefit, and w ij is the cross-layer path weight from node i to node j, δ1 is the path weight penalty factor;

[0083] S52. The global profit function R global Mapping to reinforcement learning optimization objectives and defining the state space and action space A global , the policy function π for reinforcement learning global (a ij |S′ i ) means in state S′ i Next select action a ij probability;

[0084] S53. In the reinforcement learning process, the value function Q is updated through the global reward function global (S′ i ,a ij ) satisfies the Bellman equation:

[0085]

[0086] Among them, Q global (S′ j ,a jk ) is the next state S′ j The optimal action value function under , γ3 is the discount factor;

[0087] S54. Optimizing the policy function π through an iterative reinforcement learning process global (a ij |S′ i) Generate a global scheduling strategy to map the real-time demand and resource allocation status of the medical scenario into the final medical data scheduling plan:

[0088]

[0089] in, is the optimal scheduling action;

[0090] The path weight and resource consumption of each scheduling action are recorded to form the final medical data scheduling log.

[0091] An artificial intelligence-assisted consultation real-time data scheduling system includes the following modules:

[0092] The data collection and classification module is used to collect different types of medical data, including structured medical record data, unstructured medical imaging data and real-time monitoring data, and mark and classify the collected medical data into structured medical record data sets, unstructured medical imaging data sets and real-time monitoring data sets;

[0093] The data preprocessing and feature extraction module preprocesses the classified medical data, extracts key features and generates standardized data forms, including feature vectors of structured medical record data, processed tensors of unstructured medical imaging data and time series feature matrices of real-time monitoring data, ultimately forming a preliminary medical data set;

[0094] Hierarchical game tree modeling module, which divides the preliminary medical data set into several independent decision levels according to the heterogeneity of medical data types. Each decision level corresponds to a type of medical data. A hierarchical game tree model is constructed for each decision level. The node status includes the priority of medical data, the estimated transmission time and the processing resource requirements.

[0095] The cross-layer communication and state interaction module establishes a communication matrix and a task dependency matrix based on the cross-layer communication mechanism to realize state interaction and information transmission between hierarchical game tree models, dynamically adjusts the cross-layer path weights, generates an optimized cross-layer path matrix, and updates node status and priority in real time;

[0096] The global scheduling optimization module integrates the local scheduling results of the hierarchical game tree model, optimizes the global scheduling strategy through the reinforcement learning algorithm based on the global benefit function, generates the final medical data scheduling plan, and dynamically adjusts the scheduling path and task allocation;

[0097] The scheduling plan execution and log management module executes the final generated medical data scheduling plan, transmits high-priority data to the doctor or diagnosis model first, and low-priority data enters the cache module. It records the resource consumption, path information and scheduling time during the scheduling execution process, and generates a scheduling log.

[0098] The beneficial effects of the present invention are:

[0099] (1) The present invention adopts a multi-layer hierarchical game tree modeling method based on the heterogeneity of medical data types to perform hierarchical processing on heterogeneous medical data, and constructs exclusive game tree models for structured medical record data, unstructured medical imaging data, and real-time monitoring data, respectively. The state interaction and path optimization between hierarchical models are realized through a cross-layer communication mechanism. By introducing a cross-layer communication matrix and a task dependency matrix, the priority changes and resource demand coordination relationships between different data are dynamically evaluated, so that the scheduling strategy can accurately match the actual needs of medical data, thereby effectively avoiding the problems of data transmission delay and uneven resource allocation, and significantly improving the flexibility and efficiency of data scheduling under high concurrency of heterogeneous data.

[0100] (2) The present invention proposes a global benefit function, which comprehensively maps the priority, transmission time and resource requirements of medical data into scheduling optimization objectives. It dynamically adjusts the scheduling path and task allocation strategy by iteratively optimizing the global policy function through a reinforcement learning algorithm. It can respond to the dynamic changes of task priority and resource status in medical scenarios in real time, while maximizing the global benefit function and improving the robustness and adaptability of the scheduling strategy. Through iterative optimization of the global value function, the present invention demonstrates efficient resource utilization and key task guarantee capabilities in complex medical scenarios, ensuring the priority transmission and processing of key medical data.

[0101] (3) The present invention defines a comprehensive evaluation value of priority, transmission time and resource requirements for each node through the game tree node state initialization and path weight calculation mechanism, and uses the path weight to dynamically optimize the scheduling strategy between nodes. By dynamically adjusting the node weight and path weight matrix, the complexity of the task and resource requirements are fully considered during the scheduling process, so that the system can allocate limited medical resources more efficiently, while reducing the interference of low-priority data on key tasks, thereby improving the overall scheduling efficiency and service quality of medical scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0102] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0103] Figure 1 This is a flow chart of an artificial intelligence-assisted medical consultation real-time data scheduling method and system proposed by the present invention. DETAILED DESCRIPTION

[0104] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0105] refer to Figure 1 , an artificial intelligence-assisted consultation real-time data scheduling method, comprising the following steps:

[0106] S1. Collect and classify different types of medical data, pre-process each type of medical data, and form a preliminary medical data set;

[0107] S2. Based on the heterogeneity of medical data types, the preliminary medical data set is divided into several independent decision levels, each decision level corresponds to a medical data type, and the corresponding game tree models are constructed respectively;

[0108] S3. Initialize the reinforcement learning strategy for the initial state of each game tree model, design a local optimization strategy based on the characteristic distribution of medical data, generate a priority scheduling path for medical data, and embed the priority scheduling path into the game tree model;

[0109] S4. Establish a cross-layer communication mechanism to enable state interaction and information transfer between heterogeneous medical data hierarchical game tree models;

[0110] S5. Based on the cross-layer communication mechanism, the local scheduling results of the game tree model at each layer are globally optimized, the real-time demand and resource allocation status of the medical scenario are mapped into a global benefit function, and the global scheduling strategy is iteratively optimized through reinforcement learning to generate the final medical data scheduling plan;

[0111] S6. Execute the final medical data scheduling plan, give priority to transmitting high-priority medical data to the doctor side or diagnostic model, and low-priority medical data enters the cache module. The subsequent transmission order of medical data is adjusted according to the global benefit function, and the log information of medical data scheduling is recorded at the same time.

[0112] In this implementation, S1 includes the following steps:

[0113] S11. Collect medical data, including structured medical record data, unstructured medical imaging data, and real-time monitoring data, and mark the collected medical data as a medical data set:

[0114] D raw ={d1,d2,…,d n},1≤i≤n;

[0115] Among them, d i represents the i-th medical data, and n is the total number of medical data;

[0116] S12. Medical data set D raw The medical data in the data are classified and processed, and the medical data are divided into three subsets according to the data type: structured medical record data set D text, unstructured medical imaging data set D image and real-time monitoring data set D real_time :

[0117] D raw =D text ∪D image ∪D real_time ;

[0118] S13. Format the structured medical record data in the structured medical record data set, extract key fields and generate a standardized feature vector f i text ;

[0119] Perform image denoising, resolution standardization and ROI extraction on the unstructured medical image data in the unstructured medical image data set to generate the processed image tensor Where h, w, and c are the height, width, and number of channels of the image respectively;

[0120] Perform sampling rate alignment and time series interpolation processing on the real-time monitoring data in the real-time monitoring data set to generate a time series feature matrix Among them, t is the time step and k is the feature dimension;

[0121] S14. Combining the preprocessed medical data to form a preliminary medical data set D processed :

[0122]

[0123] Where n text 、n image 、n real_time Respectively represent the number of structured medical record data, unstructured medical imaging data and real-time monitoring data, satisfying n text +n image +n real_time =n.

[0124] In this implementation, S2 includes the following steps:

[0125] S21. Based on preliminary medical dataset D processed According to the heterogeneity of medical data types, the preliminary medical data set is divided into several independent decision levels, each decision level corresponds to a type of medical data;

[0126] Define the medical data type set K and the data type mapping function φ: D processed →K,φ(d i ) = k represents medical data d i The type is k, and the preliminary medical data set D is transformed according to the mapping function φ. processedDivided into decision-making levels L k , and subdivided into three decision-making levels:

[0127]

[0128] Among them, L text , L image , L real_time The decision-making levels corresponding to structured medical record data, unstructured medical imaging data, and real-time monitoring data respectively;

[0129] S22. Construct a game tree model for the medical data in each decision level:

[0130] T k =(N k ,E k );

[0131] Among them, N k is the game tree node set, E k is the edge set of the game tree;

[0132] For each medical data d i ∈L k Construct the corresponding game tree node n i ∈N k , node status S i Defined as:

[0133]

[0134] Among them, P i For medical data i Priority:

[0135]

[0136] U i Represents medical data i The urgency level is defined as:

[0137]

[0138] C i Represents medical data i Complexity:

[0139]

[0140] Among them, Size(d i ) is medical data d i The data size, For medical data i The processing complexity coefficient is Represents medical datai Timeliness:

[0141]

[0142] Among them, t now is the current time, t i For medical data i The generation time, κ is the time decay coefficient, λ1, λ2, λ3 are weight coefficients, which are used to balance the impact of urgency, complexity and timeliness on priority;

[0143] For medical data i Estimated transfer time:

[0144]

[0145] Among them, B network is the network bandwidth, η k is the network efficiency coefficient, which reflects the efficiency of medical data type k in network transmission;

[0146] R i For medical data i Processing resource requirements:

[0147]

[0148] Here, ρ is the amount of resources required per unit complexity.

[0149] In this implementation, S3 includes the following steps:

[0150] S31. According to the specific needs of the medical scenario, the node state of the game tree model is initialized and the n of each node is calculated. i The comprehensive evaluation value V i :

[0151]

[0152] Among them, α, β, and γ are weight coefficients used to balance the impact of priority, expected transmission time, and processing resource requirements on the comprehensive evaluation value;

[0153] The comprehensive evaluation value V i As the node weight in the game tree model, initialize the game tree edge set E k Define the scheduling path between nodes, the edge weight W ij Calculated according to node weight:

[0154] W ij =V j -δC ij ;

[0155] Among them, C ij For slave node n i To node n j The dispatch cost is:

[0156]

[0157] in, For slave node n i To node n j The actual transmission time of,δ,∈,is the adjustment parameter used to control the trade-off between scheduling benefit and cost;

[0158] S24. Input the states and path benefits of all nodes in the game tree model into the reinforcement learning model, and define the state space S of reinforcement learning = {S i}, action space A = {a ij}, a ij Represents slave node n i To node n j The scheduling action defines the strategy function π(a ij ∣S i ) indicates that in state S i Next select action a ij probability;

[0159] Define the value function Q(S i ,a ij ), satisfying the Bellman equation:

[0160]

[0161] Among them, R(S i ,a ij ) is the immediate reward, which is equal to the edge weight W ij , γ1 is the discount factor, P(S j ∣S i ,a ij ) is the state transition probability, which takes the value of 1 in a deterministic environment.

[0162] In this implementation, S4 includes the following steps:

[0163] S41. Establish a cross-layer communication mechanism between hierarchical game tree models and define a cross-layer communication matrix M for the state interaction and information transmission of different types of medical data comm :

[0164] M comm =[m ij ];

[0165]

[0166] Among them, m ij Represents the slave node S i To Node S j The communication status, f(S i ,S j ) is the communication weight function, |P i -P j | indicates the difference in priority between nodes. The smaller the difference in priority, the better the communication status. is the timeliness weight of the timestamp difference between nodes, R j is the resource requirement of the target node, λ1, λ2, and λ3 are weight coefficients used to balance the impact of priority difference, timeliness, and resource requirements on the communication status;

[0167] S42. Based on the communication matrix M comm Establish a cross-layer scheduling path graph:

[0168] G cross =(V,E);

[0169] Where V = ∪ k N k represents the set of all game tree nodes, E = {e ij ∣m ij ≠0, i≠j} represents the scheduling path between nodes across layers;

[0170] The weight w of each path ij The communication weight function f(S i ,S j ) calculation, satisfying:

[0171]

[0172] S43. Define a task dependency matrix M used to represent the degree of dependency between different tasks task :

[0173] M task =[t ij ];

[0174]

[0175] Among them, t ij represents the dependency strength between task i and task j, μ1, μ2, and μ3 are weight coefficients, which are used to balance the impact of priority, resource demand, and transmission time on task dependency, respectively;

[0176] S44. Dynamically update the cross-layer communication mechanism according to the communication matrix M comm and the task dependency matrix M task Adjust the cross-layer path weights in real time to form an optimized cross-layer path matrix Mpath ;

[0177] S45. Using the optimized cross-layer path matrix M path Synchronize the global state of the hierarchical game tree model, update the priority, transmission time and resource requirements of each node, and redistribute the node weight V′ i .

[0178] In this implementation, S5 includes the following steps:

[0179] S51. Based on cross-layer communication mechanism and path matrix M path Integrate the local scheduling results of each layer of the game tree model to construct the global profit function R global :

[0180]

[0181] Among them, P′ i , R′ i are the global post-synchronization priority, transmission time and resource requirements of node i, ω1, ω2, ω3 are the global benefit weight coefficients, which are used to balance the impact of priority, transmission time and resource consumption on the global benefit, and w ij is the cross-layer path weight from node i to node j, δ1 is the path weight penalty factor;

[0182] S52. The global profit function R global Mapping to reinforcement learning optimization objectives and defining the state space and action space A global , the policy function π for reinforcement learning global (a ij |S′ i ) means in state S′ i Next select action a ij probability;

[0183] S53. In the reinforcement learning process, the value function Q is updated through the global reward function global (S′ i ,a ij ) satisfies the Bellman equation:

[0184]

[0185] Among them, Q global (S′ j ,a jk ) is the next state S′ j The optimal action value function under , γ3 is the discount factor;

[0186] S54. Optimizing the policy function π through an iterative reinforcement learning processglobal (a ij |S′ i ) Generate a global scheduling strategy to map the real-time demand and resource allocation status of the medical scenario into the final medical data scheduling plan:

[0187]

[0188] in, is the optimal scheduling action;

[0189] The path weight and resource consumption of each scheduling action are recorded to form the final medical data scheduling log.

[0190] An artificial intelligence-assisted consultation real-time data scheduling system includes the following modules:

[0191] The data collection and classification module is used to collect different types of medical data, including structured medical record data, unstructured medical imaging data and real-time monitoring data, and mark and classify the collected medical data into structured medical record data sets, unstructured medical imaging data sets and real-time monitoring data sets;

[0192] The data preprocessing and feature extraction module preprocesses the classified medical data, extracts key features and generates standardized data forms, including feature vectors of structured medical record data, processed tensors of unstructured medical imaging data and time series feature matrices of real-time monitoring data, ultimately forming a preliminary medical data set;

[0193] Hierarchical game tree modeling module, which divides the preliminary medical data set into several independent decision levels according to the heterogeneity of medical data types. Each decision level corresponds to a type of medical data. A hierarchical game tree model is constructed for each decision level. The node status includes the priority of medical data, the estimated transmission time and the processing resource requirements.

[0194] The cross-layer communication and state interaction module establishes a communication matrix and a task dependency matrix based on the cross-layer communication mechanism to realize state interaction and information transmission between hierarchical game tree models, dynamically adjusts the cross-layer path weights, generates an optimized cross-layer path matrix, and updates node status and priority in real time;

[0195] The global scheduling optimization module integrates the local scheduling results of the hierarchical game tree model, optimizes the global scheduling strategy through the reinforcement learning algorithm based on the global benefit function, generates the final medical data scheduling plan, and dynamically adjusts the scheduling path and task allocation;

[0196] The scheduling plan execution and log management module executes the final generated medical data scheduling plan, transmits high-priority data to the doctor or diagnosis model first, and low-priority data enters the cache module. It records the resource consumption, path information and scheduling time during the scheduling execution process, and generates a scheduling log.

[0197] Embodiment 1:

[0198] This embodiment 1 selects the emergency department of a comprehensive tertiary hospital as the application scenario. The average number of patients admitted to the emergency department per day is about 300, of which 80% require rapid diagnosis support based on real-time medical data. The data types include patient medical history information (past medical history, drug allergy history), real-time monitoring data (electrocardiogram, blood pressure, blood oxygen saturation) and medical imaging data (CT, MRI scan results). The emergency scenario has extremely high requirements for the real-time and priority of medical data. The data scheduling process must not only quickly respond to the diagnostic needs of high-risk patients, but also take into account the resource allocation of other patients. Under traditional scheduling methods, data transmission is prone to delays due to unreasonable priority allocation or task congestion.

[0199] During the peak hours of emergency department (e.g., 19:00-23:00 in the evening), the hospital deployed the real-time data scheduling system for auxiliary consultation based on game tree reinforcement learning proposed in this invention. In one scenario, the emergency department received data requirements from multiple patients at the same time, including the following situations:

[0200] 1. A 67-year-old male patient was admitted to the hospital due to chest pain. Real-time monitoring showed abnormal electrocardiogram and suspected acute myocardial infarction. It was necessary to quickly retrieve past medical records and real-time electrocardiogram waveforms for diagnosis.

[0201] 2. A 32-year-old female patient suffered a right leg fracture due to a car accident. The CT scan data needs to be transmitted to the doctor as soon as possible to assess whether there is fracture displacement and vascular damage.

[0202] 3. A 10-year-old child is admitted to the hospital due to persistent high fever. Real-time monitoring of blood oxygen saturation and medical history information is required to rule out potential respiratory infections.

[0203] 4. Data requirements for other non-emergency patients include the retrieval and uploading of routine examination reports.

[0204] In the face of the high concurrency requirements of multi-source heterogeneous medical data, the method of the present invention initiates the following process:

[0205] The system collects the data requirements of each patient, including real-time electrocardiogram waveforms, CT imaging data, blood oxygen saturation and medical records, and classifies the data into structured medical record data, unstructured medical imaging data and real-time monitoring data. Each type of data enters the corresponding hierarchical game tree model.

[0206] In the hierarchical game tree, the 67-year-old male patient was given the highest priority due to chest pain and abnormal electrocardiogram. The scheduling of CT imaging data was given the second highest priority due to the urgent need of fracture patients. Real-time blood oxygen monitoring of children was assigned a medium priority, while the data of other non-urgent patients were temporarily cached. Through the calculation of the communication matrix and the task dependency matrix, the node weights and path weights of the hierarchical game tree model were dynamically adjusted.

[0207] Subsequently, the system generates the optimal scheduling plan through reinforcement learning based on the global benefit function. Key high-priority data (electrocardiograms and medical records of patients with chest pain) are transmitted to the doctor's side first, and CT data are quickly transmitted to the imaging department for automatic analysis. At the same time, medium-priority children's blood oxygen monitoring data are updated in real time according to the time sequence, so that doctors can obtain data in time and eliminate risks.

[0208] During the entire process, the system records the transmission delay, path consumption and resource allocation of each task in real time, and dynamically updates the scheduling strategy. All patient data scheduling tasks are completed within 15 minutes, and the response time for high-priority data is controlled within 3 minutes.

[0209] In actual use, the classification scheduling of heterogeneous data by the method of the present invention significantly improves the processing efficiency of key tasks:

[0210] 1. The time required to retrieve ECG data for patients with chest pain has been shortened from 4 minutes using traditional methods to 1 minute and 15 seconds, allowing doctors to diagnose the condition and arrange interventional treatment more quickly.

[0211] 2. The CT image analysis time for fracture patients is shortened to within 2 minutes, significantly improving the efficiency of surgical evaluation.

[0212] 3. Children's blood oxygen monitoring data is updated in real time, and the data transmission delay is less than 2 seconds, which effectively ensures the real-time diagnosis.

[0213] This embodiment 1 verifies the feasibility and effectiveness of the present invention through the application of the emergency department scenario, solves the delay problem caused by unreasonable priority allocation in the traditional scheduling method, significantly improves the real-time performance and resource utilization efficiency of heterogeneous data scheduling, and provides efficient and reliable technical support for hospital emergency.

[0214] The present invention adopts a multi-layer hierarchical game tree modeling method based on the heterogeneity of medical data types to perform hierarchical processing of heterogeneous medical data, and constructs exclusive game tree models for structured medical record data, unstructured medical imaging data and real-time monitoring data respectively, and realizes state interaction and path optimization between hierarchical models through a cross-layer communication mechanism. By introducing a cross-layer communication matrix and a task dependency matrix, the priority changes and resource demand coordination relationship between different data are dynamically evaluated, so that the scheduling strategy can accurately match the actual needs of medical data, thereby effectively avoiding the problems of data transmission delay and uneven resource allocation, and significantly improving the flexibility and efficiency of data scheduling under high concurrency of heterogeneous data.

[0215] The present invention proposes a global benefit function, which comprehensively maps the priority, transmission time and resource requirements of medical data into scheduling optimization objectives, and dynamically adjusts the scheduling path and task allocation strategy by iteratively optimizing the global policy function through a reinforcement learning algorithm. It can respond to the dynamic changes of task priority and resource status in medical scenarios in real time, while maximizing the global benefit function and improving the robustness and adaptability of the scheduling strategy. Through iterative optimization of the global value function, the present invention demonstrates efficient resource utilization capabilities and key task guarantee capabilities in complex medical scenarios, ensuring the priority transmission and processing of key medical data.

[0216] The present invention defines a comprehensive evaluation value of priority, transmission time and resource requirement for each node through game tree node state initialization and path weight calculation mechanism, and uses path weight to dynamically optimize the scheduling strategy between nodes. By dynamically adjusting node weights and path weight matrices, the complexity of tasks and resource requirements are fully considered during the scheduling process, so that the system can allocate limited medical resources more efficiently, while reducing the interference of low-priority data on key tasks, thereby improving the overall scheduling efficiency and service quality of medical scenarios.

[0217] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for real-time data scheduling of artificial intelligence-assisted medical consultation, characterized in that: The steps include: S1. Collect and classify different types of medical data, pre-process each type of medical data, and form a preliminary medical data set; S2. Based on the heterogeneity of medical data types, the preliminary medical data set is divided into several independent decision levels, each decision level corresponds to a medical data type, and the corresponding game tree models are constructed respectively; S3. Initialize the reinforcement learning strategy for the initial state of each game tree model, design a local optimization strategy based on the characteristic distribution of medical data, generate a priority scheduling path for medical data, and embed the priority scheduling path into the game tree model; S4. Establish a cross-layer communication mechanism to enable state interaction and information transfer between heterogeneous medical data hierarchical game tree models; S5. Based on the cross-layer communication mechanism, the local scheduling results of the game tree model at each layer are globally optimized, the real-time demand and resource allocation status of the medical scenario are mapped into a global benefit function, and the global scheduling strategy is iteratively optimized through reinforcement learning to generate the final medical data scheduling plan; S6. Execute the final medical data scheduling plan, give priority to transmitting high-priority medical data to the doctor side or diagnostic model, and low-priority medical data enters the cache module. The subsequent transmission order of medical data is adjusted according to the global benefit function, and the log information of medical data scheduling is recorded at the same time.

2. The method for real-time data scheduling of artificial intelligence-assisted medical consultation according to claim 1, characterized in that: The S1 comprises the following steps: S11. Collect medical data, including structured medical record data, unstructured medical imaging data, and real-time monitoring data, and mark the collected medical data as a medical data set: D raw ={d1,d2,…,d n },1≤i≤n; Among them, d i represents the i-th medical data, and n is the total number of medical data; S12. Medical data set D raw The medical data in the data are classified and processed, and the medical data are divided into three subsets according to the data type: structured medical record data set D text , unstructured medical imaging data set D image and real-time monitoring data set D real_time : D raw =D text ∪D image ∪D real_time ; S13. Format the structured medical record data in the structured medical record data set, extract key fields and generate a standardized feature vector f i text ; Perform image denoising, resolution standardization and ROI extraction on the unstructured medical image data in the unstructured medical image data set to generate the processed image tensor Where h, w, and c are the height, width, and number of channels of the image respectively; Perform sampling rate alignment and time series interpolation processing on the real-time monitoring data in the real-time monitoring data set to generate a time series feature matrix Among them, t is the time step and k is the feature dimension; S14. Combining the preprocessed medical data to form a preliminary medical data set D processed : Where n text 、n image 、n real_time Respectively represent the number of structured medical record data, unstructured medical imaging data and real-time monitoring data, satisfying n text +n image +n real_time =n.

3. The method for real-time data scheduling of artificial intelligence-assisted medical consultation according to claim 1 is characterized in that: The S2 comprises the following steps: S21. Based on preliminary medical dataset D processed According to the heterogeneity of medical data types, the preliminary medical data set is divided into several independent decision levels, each decision level corresponds to a type of medical data; Define the medical data type set K and the data type mapping function φ: D processed →K,φ(d i ) = k represents medical data d i The type is k, and the preliminary medical data set D is transformed according to the mapping function φ. processed Divided into decision-making levels L k , and subdivided into three decision-making levels: Among them, L text , L image , L real_time The decision-making levels corresponding to structured medical record data, unstructured medical imaging data, and real-time monitoring data respectively; S22. Construct a game tree model for the medical data in each decision level: T k =(N k ,E k ); Among them, N k is the game tree node set, E k is the edge set of the game tree; For each medical data d i ∈L k Construct the corresponding game tree node n i ∈N k , node status S i Defined as: Among them, P i For medical data i Priority: U i Represents medical data i The urgency level is defined as: C i Represents medical data i Complexity: Among them, Size(d i ) is medical data d i The data size, For medical data i The processing complexity coefficient is Represents medical data i Timeliness: Among them, t now is the current time, t i For medical data i The generation time, κ is the time decay coefficient, λ1, λ2, λ3 are weight coefficients, which are used to balance the impact of urgency, complexity and timeliness on priority; For medical data i Estimated transfer time: Among them, B network is the network bandwidth, η k is the network efficiency coefficient, which reflects the efficiency of medical data type k in network transmission; R i For medical data i Processing resource requirements: Here, ρ is the amount of resources required per unit complexity.

4. The method for real-time data scheduling of artificial intelligence-assisted medical consultation according to claim 1, characterized in that: The S3 comprises the following steps: S31. According to the specific needs of the medical scenario, the node state of the game tree model is initialized and the n of each node is calculated. i The comprehensive evaluation value V i : Among them, α, β, and γ are weight coefficients used to balance the impact of priority, expected transmission time, and processing resource requirements on the comprehensive evaluation value; The comprehensive evaluation value V i As the node weight in the game tree model, initialize the game tree edge set E k Define the scheduling path between nodes, the edge weight W ij Calculated according to node weight: W ij =V j -δC ij ; Among them, C ij For slave node n i To node n j The dispatch cost is: in, For slave node n i To node n j The actual transmission time of,δ,∈,is the adjustment parameter used to control the trade-off between scheduling benefit and cost; S24. Input the states and path benefits of all nodes in the game tree model into the reinforcement learning model, and define the state space S of reinforcement learning = {S i }, action space A = {a ij }, a ij Represents slave node n i To node n j The scheduling action defines the strategy function π(a ij ∣S i ) indicates that in state S i Next select action a ij The probability of Define the value function Q(S i ,a ij ), satisfying the Bellman equation: Among them, R(S i ,a ij ) is the immediate reward, which is equal to the edge weight W ij , γ1 is the discount factor, P(S j ∣S i ,a ij ) is the state transition probability, which takes the value of 1 in a deterministic environment.

5. The method and system for real-time data scheduling of artificial intelligence-assisted medical consultation according to claim 1, characterized in that: The S4 comprises the following steps: S41. Establish a cross-layer communication mechanism between hierarchical game tree models and define a cross-layer communication matrix M for the state interaction and information transmission of different types of medical data comm : M comm =[m ij ]; Among them, m ij Represents the slave node S i To Node S j The communication status, k≠l,f(S i ,S j ) is the communication weight function, |P i -P j | indicates the difference in priority between nodes. The smaller the difference in priority, the better the communication status. is the timeliness weight of the timestamp difference between nodes, R j is the resource requirement of the target node, λ1, λ2, and λ3 are weight coefficients used to balance the impact of priority difference, timeliness, and resource requirements on the communication status; S42. Based on the communication matrix M comm Establish a cross-layer scheduling path graph: G cross =(V,E); Where V = ∪ k N k represents the set of all game tree nodes, E = {e ij ∣m ij ≠0, i≠j} represents the scheduling path between nodes across layers; The weight w of each path ij The communication weight function f(S i ,S j ) calculation, satisfying: S43. Define a task dependency matrix M used to represent the degree of dependency between different tasks task : M task =[t ij ]; Among them, t ij represents the dependency strength between task i and task j, μ1, μ2, and μ3 are weight coefficients, which are used to balance the impact of priority, resource demand, and transmission time on task dependency, respectively; S44. Dynamically update the cross-layer communication mechanism according to the communication matrix M comm and the task dependency matrix M task Adjust the cross-layer path weights in real time to form an optimized cross-layer path matrix M path ; S45. Using the optimized cross-layer path matrix M path Synchronize the global state of the hierarchical game tree model, update the priority, transmission time and resource requirements of each node, and redistribute the node weight V′ i .

6. The method for real-time data scheduling of artificial intelligence-assisted medical consultation according to claim 1, characterized in that: The S5 comprises the following steps: S51. Based on cross-layer communication mechanism and path matrix M path Integrate the local scheduling results of each layer of the game tree model to construct the global profit function R global : Among them, P′ i , R′ i are the global post-synchronization priority, transmission time and resource requirements of node i, ω1, ω2, ω3 are the global benefit weight coefficients, which are used to balance the impact of priority, transmission time and resource consumption on the global benefit, and w ij is the cross-layer path weight from node i to node j, δ1 is the path weight penalty factor; S52. The global profit function R global Mapping to reinforcement learning optimization objectives and defining the state space and action space A global , the policy function π for reinforcement learning global (a ij |S′ i ) means in state S′ i Next select action a ij The probability of S53. In the reinforcement learning process, the value function Q is updated through the global reward function global (S′ i ,a ij ) satisfies the Bellman equation: Among them, Q global (S′ j ,a jk ) is the next state S′ j The optimal action value function under , γ3 is the discount factor; S54. Optimizing the policy function π through an iterative reinforcement learning process global (a ij |S′ i ) Generate a global scheduling strategy to map the real-time demand and resource allocation status of the medical scenario into the final medical data scheduling plan: in, is the optimal scheduling action; The path weight and resource consumption of each scheduling action are recorded to form the final medical data scheduling log.

7. An artificial intelligence-assisted medical consultation real-time data scheduling system, comprising the method according to any one of claims 1 to 6, characterized in that: Includes the following modules: The data collection and classification module is used to collect different types of medical data, including structured medical record data, unstructured medical imaging data and real-time monitoring data, and mark and classify the collected medical data into structured medical record data sets, unstructured medical imaging data sets and real-time monitoring data sets; The data preprocessing and feature extraction module preprocesses the classified medical data, extracts key features and generates standardized data forms, including feature vectors of structured medical record data, processed tensors of unstructured medical imaging data and time series feature matrices of real-time monitoring data, ultimately forming a preliminary medical data set; Hierarchical game tree modeling module, which divides the preliminary medical data set into several independent decision levels according to the heterogeneity of medical data types. Each decision level corresponds to a type of medical data. A hierarchical game tree model is constructed for each decision level. The node status includes the priority of medical data, the estimated transmission time and the processing resource requirements. The cross-layer communication and state interaction module establishes a communication matrix and a task dependency matrix based on the cross-layer communication mechanism to realize state interaction and information transmission between hierarchical game tree models, dynamically adjusts the cross-layer path weights, generates an optimized cross-layer path matrix, and updates node status and priority in real time; The global scheduling optimization module integrates the local scheduling results of the hierarchical game tree model, optimizes the global scheduling strategy through the reinforcement learning algorithm based on the global benefit function, generates the final medical data scheduling plan, and dynamically adjusts the scheduling path and task allocation; The scheduling plan execution and log management module executes the final generated medical data scheduling plan, transmits high-priority data to the doctor or diagnosis model first, and low-priority data enters the cache module. It records the resource consumption, path information and scheduling time during the scheduling execution process, and generates a scheduling log.

Images