Artificial intelligence medical technology evaluation scheme making method and system

Abstract

The invention relates to the technical field of artificial intelligence medical treatment, and discloses an artificial intelligence medical technology evaluation scheme making method and system, and the method comprises the steps: collecting artificial intelligence medical technology input and output data, medical environment variables and preference information of multiple participants in real time; data is mapped to a dynamic game framework, the technology is simulated, doctors, patients and the like are game participants, and Nash equilibrium points are calculated through an adaptive utility function adjustment mechanism, a multi-scale game nested structure and a repeated game model; generating a proxy behavior branch prediction tree; integrating results and outputting a customized evaluation scheme containing a risk threshold value and the like; and updating parameters based on a deployment result to realize closed-loop optimization. According to the method, medical scene variation can be dynamically responded, and the flexibility, adaptability and continuous iteration capability of an evaluation scheme are improved.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence medical technology, and in particular to a method and system for formulating an evaluation scheme for artificial intelligence medical technology. Background Technology

[0002] With the deepening penetration of artificial intelligence (AI) technology in the medical field, the application scope of AI-based medical technologies is constantly expanding. Assessing the safety, effectiveness, and adaptability of these technologies has become a crucial step in ensuring medical quality and protecting patient rights. Currently, the industry has established a preliminary technical system for evaluating AI-based medical technologies, providing important support for promoting the standardized implementation of these technologies.

[0003] In existing technologies, AI-based medical technology evaluation schemes often adopt a fixed indicator and process model. These schemes typically pre-set core evaluation indicators such as diagnostic accuracy, response speed, and compliance, along with corresponding thresholds. They then construct standardized test datasets to simulate specific medical scenarios and conduct static tests on the technology.

[0004] However, in the process of realizing the technical solution of this application, the inventors of this application discovered that the above-mentioned technology has at least the following technical problems: the existing evaluation scheme lacks the ability to dynamically adapt to changes in patient group characteristics and fluctuations in equipment parameters in medical scenarios due to the lack of an interactive simulation mechanism involving multiple participants, resulting in insufficient flexibility of the evaluation scheme. Summary of the Invention

[0005] To overcome the above shortcomings, this invention provides a method and system for formulating an artificial intelligence medical technology evaluation scheme. It aims to improve the problem that existing evaluation schemes lack flexibility due to the absence of an interactive simulation mechanism involving multiple participants, which results in a lack of dynamic adaptability to variations such as changes in patient characteristics and fluctuations in equipment parameters in medical scenarios.

[0006] In a first aspect, the present invention provides the following technical solution: a method for formulating an artificial intelligence medical technology evaluation scheme, characterized by comprising the following steps: S1. The data acquisition module collects the input and output data of artificial intelligence medical technology, medical environmental variables, and preference information of multiple participants in real time. S2. In the game modeling module, the collected data is mapped to a dynamic game framework. Artificial intelligence medical technology, doctors, patients, medical institutions, and regulatory agencies are simulated as game participants. Each game participant defines a utility function. An adaptive utility function adjustment mechanism is introduced to dynamically adjust the utility function parameters according to real-time medical environmental variables. A multi-scale game nesting structure is adopted to simulate institutional-level games at the macro level and individual-level games at the micro level. The macro and micro levels are connected through a hierarchical transmission mechanism. Based on a repeated game model, multiple rounds of interactive simulation are conducted to calculate the Nash equilibrium point. S3. In the extended subsystem for predicting agent behavior, a branch prediction tree for agent behavior is generated using the historical trajectory of game simulation. S4. The solution generation module integrates the game modeling results and the branch prediction tree of the agent behavior to output a customized evaluation solution. S5. Through the feedback iteration module, the game parameters, utility function parameters, and branch prediction tree structure of agent behavior are updated based on the actual deployment results to achieve closed-loop optimization.

[0007] Preferably, in the game modeling, artificial intelligence medical technology, doctors, patients, medical institutions and regulatory agencies are simulated as game participants, and each game participant defines a utility function, which quantifies the payoff of each game participant under a given strategy combination.

[0008] Preferably, the game modeling incorporates an adaptive utility function adjustment mechanism, which dynamically adjusts the utility function parameters based on real-time medical environmental variables. This adaptive utility function adjustment mechanism allows the utility function to adapt to real-time medical environmental variables.

[0009] Preferably, the game modeling adopts a multi-scale game nesting structure, simulating institutional-level games at the macro level and individual-level games at the micro level, and connecting the macro and micro levels through a hierarchical transmission mechanism, which makes the equilibrium result at the macro level constrain the utility function at the micro level.

[0010] Preferably, the game modeling is based on a repeated game model, performing multiple rounds of interactive simulation to calculate the Nash equilibrium point. The Nash equilibrium point is a set of strategies that satisfy the condition that for all game participants and any alternative strategies, the utility under the current strategy is not lower than the utility under the alternative strategies.

[0011] Preferably, in the agent behavior prediction extension subsystem, a branch prediction tree for agent behavior is generated using the historical trajectory of game simulation. The branch prediction tree for agent behavior is represented by a tree structure, with each node corresponding to a game state. The branches on the tree represent the response variations of game participants in different scenarios. The expected utility of each path on the tree is calculated through dynamic programming.

[0012] Preferably, in the scheme generation module, the scheme generation process integrates the game modeling results and the branch prediction tree of the agent behavior to output a customized evaluation scheme, which includes a risk threshold, a testing protocol, a hierarchical evaluation framework, and an iteration path.

[0013] Preferably, in the feedback iteration, the feedback iteration module updates the game parameters, utility function parameters, and branch prediction tree structure of agent behavior based on the actual deployment results to achieve closed-loop optimization. This closed-loop optimization forms a loop with the data acquisition step and the game modeling step.

[0014] Preferably, the data acquisition process involves real-time acquisition of input and output data of artificial intelligence medical technology, medical environmental variables, and preference information of multiple participants through a data acquisition module. The medical environmental variables include patient group characteristics, equipment parameters, and information on emergencies. The preference information of multiple participants includes doctors' preferences for diagnostic efficiency, patients' preferences for safety, medical institutions' preferences for compliance, and risk threshold requirements of regulatory agencies.

[0015] Secondly, the present invention provides the following technical solution: an artificial intelligence medical technology evaluation scheme formulation system, comprising: The system includes a data acquisition module, a game modeling module, an agent behavior prediction extension subsystem, a solution generation module, and a feedback iteration module. The game modeling module includes an adaptive utility function adjustment mechanism and a multi-scale game nesting structure. The data acquisition module is used to collect input and output data of artificial intelligence medical technology, medical environmental variables, and preference information of multiple participants in real time; The game modeling module maps the collected data to a dynamic game framework, simulating AI medical technology, doctors, patients, medical institutions, and regulatory agencies as game participants. Each game participant defines a utility function, and the utility function parameters are dynamically adjusted according to real-time medical environmental variables through an adaptive utility function adjustment mechanism. A multi-scale game nesting structure is adopted to simulate institutional-level games at the macro level and individual-level games at the micro level. The macro and micro levels are connected through a hierarchical transmission mechanism. Based on a repeated game model, multiple rounds of interactive simulation are conducted to calculate the Nash equilibrium point. The agent behavior prediction extension subsystem is used to generate a branch prediction tree for agent behavior using historical trajectories from game simulations; The scheme generation module is used to integrate the game modeling results and the branch prediction tree of agent behavior to output a customized evaluation scheme; The feedback iteration module is used to update the branch prediction tree structure of game parameters, utility function parameters, and agent behavior based on the actual deployment results, so as to achieve closed-loop optimization.

[0016] The present invention has the following beneficial effects: 1. In this invention, a dynamic game framework is used to simulate multiple parties, such as artificial intelligence medical technology, doctors, and patients, as game participants and calculate the Nash equilibrium point. This can capture the uncertainty propagation of multi-agent interactions, realize dynamic response to changes in medical scenarios, and improve the flexibility of evaluation schemes.

[0017] 2. This invention introduces an adaptive utility function adjustment mechanism and a multi-scale game nesting structure, which adjusts the utility parameters according to real-time medical environmental variables, and connects macro-level institutional games and micro-level individual games through hierarchical transmission, thereby enhancing the adaptability of the solution to different medical levels and ensuring the comprehensiveness and consistency of the evaluation process.

[0018] 3. In this invention, the branch prediction tree and feedback iteration module of the agent behavior are used to predict potential conflict paths and update the game parameters based on the actual deployment results. This method forms a closed-loop optimization cycle, which improves the predictability and continuous iteration capability of the evaluation scheme. Attached Figure Description

[0019] Figure 1 This is a flowchart of a method for developing an artificial intelligence medical technology evaluation scheme proposed in this invention; Figure 2 A system architecture diagram is provided for the artificial intelligence medical technology evaluation scheme proposed in this invention. Detailed Implementation

[0020] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Example 1 Reference Figure 1 In the first embodiment of the present invention, the present invention provides a method for formulating an evaluation scheme for artificial intelligence medical technology, comprising the following steps: S1. The data acquisition module collects the input and output data of artificial intelligence medical technology, medical environmental variables, and preference information of multiple participants in real time. S2. In the game modeling module, the collected data is mapped to a dynamic game framework. Artificial intelligence medical technology, doctors, patients, medical institutions, and regulatory agencies are simulated as game participants. Each game participant defines a utility function. An adaptive utility function adjustment mechanism is introduced to dynamically adjust the utility function parameters according to real-time medical environmental variables. A multi-scale game nesting structure is adopted to simulate institutional-level games at the macro level and individual-level games at the micro level. The macro and micro levels are connected through a hierarchical transmission mechanism. Based on a repeated game model, multiple rounds of interactive simulation are conducted to calculate the Nash equilibrium point. S3. In the extended subsystem for predicting agent behavior, a branch prediction tree for agent behavior is generated using the historical trajectory of game simulation. S4. The solution generation module integrates the game modeling results and the branch prediction tree of the agent behavior to output a customized evaluation solution. S5. Through the feedback iteration module, the game parameters, utility function parameters, and branch prediction tree structure of agent behavior are updated based on the actual deployment results to achieve closed-loop optimization.

[0022] Specifically, the system architecture consists of a data acquisition module, a game modeling module, an agent behavior prediction extension subsystem, a solution generation module, and a feedback iteration module. The data acquisition module connects to the game modeling module to provide input data. The game modeling module includes an adaptive utility function adjustment mechanism and a multi-scale game nested structure to handle mapping and simulation processes. The agent behavior prediction extension subsystem connects the game modeling module and the solution generation module to generate a prediction tree. The solution generation module integrates the results and outputs a solution. The feedback iteration module connects all modules to update parameters. This method executes S1 to S5 sequentially to form a complete process. The data collected in S1 includes the input and output data of artificial intelligence medical technology and the medical environment. The data collection process includes variables and preference information from multiple participants. Medical environment variables include patient group characteristics, equipment parameters, and information on unforeseen events. Participant preference information includes physicians' preferences for diagnostic efficiency, patients' preferences for safety, healthcare institutions' preferences for compliance, and regulatory agencies' risk threshold requirements. The data collection process employs a real-time monitoring mechanism to extract data from electronic medical record systems or equipment logs. Data preprocessing includes normalization of continuous variables and vector encoding of discrete preference information. In S2, the data is mapped to a dynamic game framework, simulating AI-powered medical technology agents, physician agents, patient agents, healthcare institution agents, and regulatory agency agents. Each agent defines a utility function. (1); in, Describe the utility function of agent i. , , This represents the set of action strategies for all agents. Let represent the weight of the j-th objective function for agent i. Let j represent the objective function of agent i. Let i represent the target number, j represent the target index, and n represent the number of proxies. The adaptive utility function adjustment mechanism adjusts the weights based on environmental variables, and the formula is: ;in, This indicates the adjusted weights. Indicates adjusting the step size. This represents an incremental function based on environment variables, where e represents the environment variable.

[0023] The multi-scale game-theoretic nested structure simulates the game between regulatory agencies and medical institutions at the macro level, and the game between doctors and patients at the micro level. The hierarchical transmission mechanism uses the following formula: (2); in, This represents the adjusted utility of micro-agent i. Indicates micro-level basic utility. Denotes a subset of the micro-strategy set. This indicates a macro-level equilibrium strategy. Indicates nesting strength. The transfer function is used to calculate the Nash equilibrium through multiple rounds of simulation based on a repeated game model. The Nash equilibrium is a set of strategies. ,satisfy: .

[0024] In S3, a branch prediction tree is generated using historical trajectories, where each node in the tree structure corresponds to a game state. The branch represents the response variation, and the value function is calculated using dynamic programming, with the formula: ,in, Representing state The value function, Indicates immediate feedback. Let represent the probability of branch b. To represent the subsequent state, summation covers all branches. Representation strategy, This indicates the branch index, and the output prediction tree includes the equilibrium point and probability distribution.

[0025] In S4, the game results and prediction tree are integrated to generate a scheme including risk threshold, testing protocol, hierarchical evaluation framework and iterative path. In S5, the parameters are updated based on the deployment results, including game parameters, utility function weights and prediction tree structure, forming a loop with S1 and S2.

[0026] In game theory modeling, AI-powered medical technology, doctors, patients, medical institutions, and regulatory agencies are simulated as game participants. Each participant defines a utility function, which quantifies the payoff of each participant under a given strategy combination. An adaptive utility function adjustment mechanism is introduced into the game theory modeling to dynamically adjust the utility function parameters based on real-time medical environmental variables. This adaptive utility function adjustment mechanism allows the utility function to adapt to real-time medical environmental variables.

[0027] Specifically, the game modeling receives a structured dataset provided by the data collection step and maps it to a dynamic game framework. This framework defines a set of agents: an AI medical technology agent responsible for diagnosis output strategies, a doctor agent responsible for intervention decision strategies, a patient agent responsible for feedback acceptance strategies, a medical institution agent responsible for resource allocation strategies, and a regulatory agency agent responsible for compliance inspection strategies. The action strategy set of each agent is represented by a finite state machine. For example, the strategy of the AI ​​medical technology agent includes adjusting the output confidence threshold, and the strategy of the doctor agent includes accepting or modifying the diagnosis. The utility function of each agent is calculated by a weighted objective function (1), where... The j-th objective function for agent i is defined as follows: accuracy function is defined as diagnostic match rate, or safety function is defined as risk minimization. The adaptive utility function adjustment mechanism monitors environmental variables such as changes in patient population characteristics during each simulation period and updates the weights through incremental calculation. ;in, This indicates the adjusted weights. This indicates that the adjustment step size is set to a fixed value within the range of 0.01 to 0.1. The incremental function based on environment variables is defined as follows: ;in, This indicates that the sensitivity coefficient is set to a fixed value within the range of 0.05 to 0.2. This indicates the current environmental variables, such as noise level. The baseline environmental value is set from the initial data. This adjustment is triggered when the environmental variable exceeds a preset threshold, such as when the noise is greater than 5%, to ensure that the utility function reflects real-time changes, such as increasing the safety weight when noise increases. The multi-scale game nesting structure divides the game into macro-level, including institutional equilibrium strategies of interaction between regulatory agency agents and medical agency agents, and micro-level, including individual equilibrium strategies of interaction between artificial intelligence medical technology agents, doctor agents and patient agents. The hierarchical transmission mechanism transmits the macro equilibrium to the micro utility. The formula is (2), where, This indicates that the micro-level fundamental utility is calculated using the utility function described above. This indicates that the equilibrium strategy at the macro level is obtained through independent solutions. This indicates that the nesting strength is set to a fixed value within the range of 0.3 to 0.7. The transfer function is defined as a linear constraint, such as: Where k is a proportionality coefficient, hierarchical connections are achieved by first calculating the macro equilibrium and then constraining the micro calculations. Based on a repeated game model, T rounds of simulation are performed, where T is a fixed value within the range of 10 to 50. In each round, the strategy is updated and a Nash equilibrium is calculated, which is a set of strategies. satisfy For all agents i and any alternative strategies ,in, Let represent the optimal strategy for agent i. This represents the optimal policy for all agents except agent i. The solution uses a fixed-point iterative method to initialize the random policy and iterate until the convergence threshold, such as the utility difference being less than 0.001, is reached. This step outputs the equilibrium policy and utility value, which are then transmitted to the agent behavior prediction extension subsystem.

[0028] The game modeling employs a multi-scale nested game structure, simulating institutional-level games at the macro level and individual-level games at the micro level. A hierarchical transmission mechanism connects the macro and micro levels, ensuring that the equilibrium outcome at the macro level constrains the utility function at the micro level. The game modeling is based on a repeated game model, conducting multi-round interactive simulations to calculate the Nash equilibrium point. This Nash equilibrium point represents a set of strategies that, for all game participants and any alternative strategies, ensure that the utility under the current strategy is no less than the utility under the alternative strategies.

[0029] Specifically, this multi-scale game nested structure divides the dynamic game framework into two levels. First, at the macro level, the agent subset is defined as regulatory agency agent and medical institution agent, whose action strategy set includes compliance standard setting and resource allocation adjustment. The macro Nash equilibrium strategy is solved by independently calculating the macro utility function. This calculation uses a fixed-point iteration method to initialize the strategy and iterate until the utility difference is less than the 0.001 threshold. Then, at the micro level, the agent subset is defined as artificial intelligence medical technology agent, doctor agent and patient agent, whose action strategy set includes diagnostic output, intervention selection and feedback response. The micro utility function is obtained through basic calculation. The hierarchical transmission mechanism integrates macro equilibrium into micro utility adjustment, formula (2), where, Micro-agent The adjusted utility This indicates that the micro-level fundamental utility is obtained through a weighted objective function. This indicates that the set of micro-policies is limited to a subset of macro-policies. This indicates a macroeconomic equilibrium strategy. This indicates that the nesting strength is set to a fixed value within the range of 0.3 to 0.7, such as 0.5. The transfer function is defined as a linear interpolation calculation. ,in, With the scaling factor set to 1.0, a connection is achieved by first performing a macroscopic simulation and then applying constraints to the microscopic simulation, ensuring overall consistency such as macroscopic compliance constraints and microscopic diagnostic strategies. Based on a repeated game model, multiple rounds of interactive simulation are performed after the nested structure is completed. The number of rounds T is set to a fixed value within the range of 10 to 50, such as 20. In each round, all agents update their strategies according to the current utility, and the overall Nash equilibrium is calculated as a set of strategies. satisfy For agents Range 1 to 5 and any alternative strategies Select from the set of strategies, among which Indicates agent The optimal strategy This represents the optimal policy for the remaining agents. The solution process uses the gradient descent method to adjust the policy parameters until convergence. Random perturbations are included in the simulation to simulate uncertainties such as changes in environmental variables. Intermediate utility values ​​are output in each round for trajectory recording. This step transmits the equilibrium policy and trajectory to the agent behavior prediction extension subsystem for subsequent prediction tree generation.

[0030] In the extended agent behavior prediction subsystem, a branch prediction tree for agent behavior is generated using the historical trajectory of game simulation. This branch prediction tree is represented by a tree structure, with each node corresponding to a game state. The branches on the tree represent the response variations of game participants under different scenarios. The expected utility of each path on the tree is calculated through dynamic programming. In the solution generation module, the solution generation module integrates the game modeling results and the branch prediction tree for agent behavior to output a customized evaluation solution. This customized evaluation solution includes risk thresholds, testing protocols, hierarchical evaluation frameworks, and iteration paths.

[0031] Specifically, the agent behavior prediction extension subsystem receives historical trajectory data output from the game modeling steps. This trajectory includes strategy combinations, utility values, and state transition records for each round in the T-round simulation. It generates a branch prediction tree by first initializing the root node, corresponding to the initial game state, which includes all agent initial strategies. Then, it expands L layers of branches from the root node, where L is a fixed value between 3 and 5 (e.g., 4), and the number of branches per layer, B, is a fixed value between 2 and 4 (e.g., 3), representing variations in different scenarios such as high or low diagnostic accuracy or positive or negative patient feedback. Each node corresponds to a game state. Record the current strategy and utility. Branches represent response variations defined by conditional probability tables. For example, in the case of misdiagnosis by AI-powered medical technology agents, the doctor agent branches with probabilities of acceptance or intervention of 0.7 and 0.3 respectively. The tree structure is stored in adjacency list format for easy traversal. Dynamic programming is used to calculate the value function from bottom to top, with the formula as follows: ,in, Representing state The value function, The instant reward is defined as the utility gain of the current node. Indicates branch The probabilities are obtained from the conditional probability table. Indicates the subsequent state. This indicates that the optional strategy is selected from historical trajectories. Indicates the branch index range to The system calculates the expected utility and optimal branch path for each path recursively upwards from the leaf node to the root node. This prediction tree is then transmitted to the solution generation module. The solution generation module receives the Nsh equilibrium strategy, adjusted utility values, and the branch prediction tree. It first integrates the data to create a solution template, then calculates the risk threshold, setting it based on the minimum equilibrium utility value. Subtracting the standard deviation, the test protocol defines phased verification steps based on the optimal branch path, such as phase 2 simulation dataset verification and phase 2 clinical control testing, totaling 3 to 5 phases. The hierarchical evaluation framework corresponds to a multi-scale structure, including macro-level protocols such as institutional compliance checklists and micro-level paths such as individual test process tables. The iterative path extracts backup plans from non-optimal branches of the prediction tree, such as emergency intervention protocols corresponding to misdiagnosis branches. All content is formatted as an XML structure document and output to the feedback iteration module for deployment and execution.

[0032] In the feedback iteration, the feedback iteration module updates the game parameters, utility function parameters, and branch prediction tree structure of agent behavior based on the actual deployment results to achieve closed-loop optimization. This closed-loop optimization forms a loop with the data acquisition step and the game modeling step. In the data acquisition, the data acquisition module collects the input and output data of artificial intelligence medical technology, medical environmental variables, and the preference information of multiple participants in real time. The medical environmental variables include patient group characteristics, equipment parameters, and information on emergencies. The preference information of multiple participants includes doctors' preferences for diagnostic efficiency, patients' preferences for safety, medical institutions' preferences for compliance, and the risk threshold requirements of regulatory agencies.

[0033] Specifically, the feedback iteration module receives XML-formatted evaluation data of the clinical deployment results of the protocol, output from the protocol generation step. This data includes actual diagnostic accuracy deviations, intervention frequency records, and compliance / violation event logs, collected daily for a continuous period. Tianzhong Set as For fixed values ​​within the range of 30, such as 14, the update process is divided into three stages. First, the game parameters are updated. Then, the initial strategy set is adjusted by the deviation ratio. If the accuracy deviation exceeds... The threshold for the AI-driven medical technology agent strategy is increased by 0.1, and the utility function parameters are updated to feed the difference between actual and simulated utility into the weight adjustment formula. in This indicates an update to the weights. This indicates a step size of 0.01. This indicates a difference between actual and simulated utility. This represents the simulation utility; the adjustment covers all agent ranges 1 to 5 and target range 1 to 5. The third update branch prediction tree structure recalculates the conditional probability table based on actual path frequencies, adjusting the actual misdiagnosis intervention probability from 0.3 to 0.45 and reconstructing the L-layer tree structure. Closed-loop optimization feeds back the updated parameters to the data acquisition step as new baseline environmental values ​​and the game modeling step as initial weights to form a loop execution. This data acquisition module connects to the electronic medical record system and medical equipment via API interface, sampling input data such as CT images and output data such as diagnostic labels in real time. Medical environmental variables are collected, including patient group characteristics sampled from 1000 patients using age distribution histograms, and equipment parameters are read from sensors. CT scanner voltage and resolution, and information on sudden events such as network outages lasting more than 30 seconds are recorded in the event log. Preference information from multiple participants is collected through a questionnaire system. Doctors' preference ratings for diagnostic efficiency range from 1 to 10 with an average of 7.2, patients' preference ratings for safety average 9.1, and medical institutions' preference ratings for compliance average 8.5. The collected data is preprocessed by performing min-max normalization to the 0-1 range for continuous variables such as voltage values ​​and performing one-hot vector encoding on discrete preference ratings to generate 5-dimensional preference vectors. After preprocessing, the dataset is stored in JSON format and transmitted to the game modeling module to start a new round of S2 execution.

[0034] Reference Figure 2 In a second embodiment of the present invention, the present invention provides an artificial intelligence medical technology evaluation scheme formulation system, comprising: The system includes a data acquisition module, a game modeling module, an agent behavior prediction extension subsystem, a solution generation module, and a feedback iteration module. The game modeling module includes an adaptive utility function adjustment mechanism and a multi-scale game nesting structure. The data acquisition module is used to collect input and output data of artificial intelligence medical technology, medical environmental variables, and preference information of multiple participants in real time; The game modeling module maps the collected data to a dynamic game framework, simulating AI medical technology, doctors, patients, medical institutions, and regulatory agencies as game participants. Each game participant defines a utility function, and the utility function parameters are dynamically adjusted according to real-time medical environmental variables through an adaptive utility function adjustment mechanism. A multi-scale game nesting structure is adopted to simulate institutional-level games at the macro level and individual-level games at the micro level. The macro and micro levels are connected through a hierarchical transmission mechanism. Based on a repeated game model, multiple rounds of interactive simulation are conducted to calculate the Nash equilibrium point. The agent behavior prediction extension subsystem is used to generate a branch prediction tree for agent behavior using historical trajectories from game simulations; The scheme generation module is used to integrate the game modeling results and the branch prediction tree of agent behavior to output a customized evaluation scheme; The feedback iteration module is used to update the branch prediction tree structure of game parameters, utility function parameters, and agent behavior based on the actual deployment results, so as to achieve closed-loop optimization.

[0035] Specifically, the data acquisition module collects input and output data, patient group characteristics, and other medical environmental variables, as well as preference information from multiple participants, from the electronic medical record system and medical devices. This data is then transmitted to the game modeling module, which maps the data to a dynamic game framework to simulate multi-agent interactions. The module adjusts parameters in response to changes in environmental variables using an adaptive utility function adjustment mechanism and connects levels using a multi-scale game nesting structure. Based on a repeated game model, it calculates the equilibrium point and outputs it to the agent behavior prediction extension subsystem. This subsystem uses trajectory generation to create a branch prediction tree and transmits it to the scheme generation module. The scheme generation module integrates the results to generate an evaluation scheme including risk thresholds and testing protocols, which is then output to the feedback iteration module. The feedback iteration module updates parameters based on the deployment results and sends the updated scheme back to the data acquisition module and game modeling module, forming a closed-loop process. The modules are connected via API interfaces to ensure the sequential execution of data flow, including acquisition, mapping, simulation, prediction, generation, and iterative updates.

[0036] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for developing an evaluation scheme for artificial intelligence medical technology, characterized in that, Includes the following steps: S1. The data acquisition module collects the input and output data of artificial intelligence medical technology, medical environmental variables, and preference information of multiple participants in real time. S2. In the game modeling module, the collected data is mapped to a dynamic game framework. Artificial intelligence medical technology, doctors, patients, medical institutions, and regulatory agencies are simulated as game participants. Each game participant defines a utility function. An adaptive utility function adjustment mechanism is introduced to dynamically adjust the utility function parameters according to real-time medical environmental variables. A multi-scale game nesting structure is adopted to simulate institutional-level games at the macro level and individual-level games at the micro level. The macro and micro levels are connected through a hierarchical transmission mechanism. Based on a repeated game model, multiple rounds of interactive simulation are conducted to calculate the Nash equilibrium point. S3. In the extended subsystem for predicting agent behavior, a branch prediction tree for agent behavior is generated using the historical trajectory of game simulation. S4. The solution generation module integrates the game modeling results and the branch prediction tree of the agent behavior to output a customized evaluation solution. S5. Through the feedback iteration module, the game parameters, utility function parameters, and branch prediction tree structure of agent behavior are updated based on the actual deployment results to achieve closed-loop optimization.

2. The method for formulating an evaluation scheme for artificial intelligence medical technology according to claim 1, characterized in that, In the game modeling described above, artificial intelligence medical technology, doctors, patients, medical institutions, and regulatory agencies are simulated as game participants. Each game participant defines a utility function, which quantifies the payoff of each game participant under a given strategy combination.

3. The method for formulating an evaluation scheme for artificial intelligence medical technology according to claim 1, characterized in that, The game modeling introduces an adaptive utility function adjustment mechanism, which dynamically adjusts the utility function parameters based on real-time medical environmental variables. This adaptive utility function adjustment mechanism allows the utility function to adapt to real-time medical environmental variables.

4. The method for formulating an evaluation scheme for artificial intelligence medical technology according to claim 1, characterized in that, The game modeling adopts a multi-scale game nesting structure, simulating institutional-level games at the macro level and individual-level games at the micro level. The macro and micro levels are connected through a hierarchical transmission mechanism, which makes the equilibrium result at the macro level constrain the utility function at the micro level.

5. The method for formulating an evaluation scheme for artificial intelligence medical technology according to claim 1, characterized in that, The game modeling is based on a repeated game model, which performs multi-round interactive simulations and calculates the Nash equilibrium point. The Nash equilibrium point is a set of strategies that satisfy the condition that for all game participants and any alternative strategies, the utility of the current strategy is not lower than the utility of the alternative strategies.

6. The method for formulating an evaluation scheme for artificial intelligence medical technology according to claim 1, characterized in that, In the aforementioned agent behavior prediction extension subsystem, a branch prediction tree for agent behavior is generated using the historical trajectory of game simulation. This branch prediction tree for agent behavior is represented by a tree structure, with each node corresponding to a game state. The branches on the tree represent the response variations of game participants in different scenarios. The expected utility of each path on the tree is calculated through dynamic programming.

7. The method for formulating an evaluation scheme for artificial intelligence medical technology according to claim 1, characterized in that, The scheme generation module integrates the game modeling results and the branch prediction tree of the agent behavior to output a customized evaluation scheme, which includes risk threshold, testing protocol, hierarchical evaluation framework and iteration path.

8. The method for formulating an evaluation scheme for artificial intelligence medical technology according to claim 1, characterized in that, In the feedback iteration, the feedback iteration module updates the game parameters, utility function parameters, and branch prediction tree structure of agent behavior based on the actual deployment results to achieve closed-loop optimization. This closed-loop optimization forms a cycle with the data acquisition step and the game modeling step.

9. The method for formulating an evaluation scheme for artificial intelligence medical technology according to claim 1, characterized in that, The data acquisition process involves real-time collection of input and output data of artificial intelligence medical technology, medical environmental variables, and preference information of multiple participants through a data acquisition module. The medical environmental variables include patient group characteristics, equipment parameters, and information on emergencies. The preference information of multiple participants includes doctors' preferences for diagnostic efficiency, patients' preferences for safety, medical institutions' preferences for compliance, and risk threshold requirements of regulatory agencies.

10. A system for developing an evaluation scheme for artificial intelligence medical technology, characterized in that, The method for developing an artificial intelligence medical technology evaluation scheme according to any one of claims 1-9 includes: The system includes a data acquisition module, a game modeling module, an agent behavior prediction extension subsystem, a solution generation module, and a feedback iteration module. The game modeling module includes an adaptive utility function adjustment mechanism and a multi-scale game nesting structure. The data acquisition module is used to collect input and output data of artificial intelligence medical technology, medical environmental variables, and preference information of multiple participants in real time; The game modeling module maps the collected data to a dynamic game framework, simulating AI medical technology, doctors, patients, medical institutions, and regulatory agencies as game participants. Each game participant defines a utility function, and the utility function parameters are dynamically adjusted according to real-time medical environmental variables through an adaptive utility function adjustment mechanism. A multi-scale game nesting structure is adopted to simulate institutional-level games at the macro level and individual-level games at the micro level. The macro and micro levels are connected through a hierarchical transmission mechanism. Based on a repeated game model, multiple rounds of interactive simulation are conducted to calculate the Nash equilibrium point. The agent behavior prediction extension subsystem is used to generate a branch prediction tree for agent behavior using historical trajectories from game simulations; The scheme generation module is used to integrate the game modeling results and the branch prediction tree of agent behavior to output a customized evaluation scheme; The feedback iteration module is used to update the branch prediction tree structure of game parameters, utility function parameters, and agent behavior based on the actual deployment results, so as to achieve closed-loop optimization.

Images