Large model agent system for neurosurgery operation

Abstract

The invention provides a neurosurgery operation large model agent system. The system comprises an operation role definition module, an operation behavior generation module and an operation operation interaction module. The operation role definition module is used for defining agent roles in the agent system as a surgeon, an operation assistant, an operating room nurse, a circulating nurse and an anesthetist; the operation behavior generation module is used for calling a large language model to generate an operation behavior of the agent role in the operation scene; the operation interaction module is used for controlling the agent role to execute the operation behavior according to real operation case data input in real time, and the operation behavior comprises interaction of the agent role in the operation scene and execution of the corresponding operation. According to the application, the defect that in the prior art, research on an operation simulation system still does not realize a vivid operation environment and does not perfect an interaction mechanism is overcome.

Description

Technical Field

[0001] The present invention relates to the field of medical technology, and in particular to a neurosurgery large model proxy system. Background Art

[0002] In today's medical field, intelligent agent technology is rapidly gaining popularity, providing unprecedented automation and intelligent support for medical systems. However, despite the outstanding performance of intelligent agents in many medical applications, the lack of surgical simulation agents is still a significant gap. For example, in the field of neurosurgery, a comprehensive surgical simulation agent can accurately simulate the entire medical operation process, provide accurate and effective decision-making for real surgical scenarios, and provide excellent scalability, efficiently train medical students and junior doctors, improve surgical process management, and provide, significantly improve doctors' surgical skills and emergency response capabilities, while promoting further research on medical large language models and intelligent agents. However, current research on surgical simulation systems has not yet achieved a realistic surgical environment and a perfect interaction mechanism. Summary of the invention

[0003] The present invention provides a neurosurgery large model proxy system to overcome the defects in the prior art that the research on the surgery simulation system still fails to realize a realistic surgery environment and a perfect interaction mechanism.

[0004] The present invention provides a neurosurgery large model agent system, which includes the following modules: Surgical role definition module, surgical behavior generation module, surgical operation interaction module; The surgical role definition module is used to define the proxy roles in the proxy system as the chief surgeon, surgical assistant, operating room nurse, circulating nurse and anesthesiologist; The surgical behavior generation module is used to call the large language model to generate the surgical behavior of the agent character in the surgical scene; The surgical operation interaction module is used to control the agent character to perform the surgical behavior according to the real surgical case data input in real time, wherein the surgical behavior includes the agent character interacting in the surgical scene and performing the corresponding surgical operation.

[0005] Furthermore, the surgical role definition module is also used to define an agent assistant in the agent system, and the agent assistant is used to send surgical instructions to the agent role in real time in the surgical scene to guide the surgical operations performed by each of the agent roles.

[0006] Furthermore, the agent assistant is also used to compare the surgical operation performed by the agent role with the actual standard operation in the real surgical case data in real time. When a deviation occurs in the comparison, the surgical operation causing the deviation is located and the deviation is corrected.

[0007] Furthermore, the agent system executes the surgical process in the surgical scene through a long-short-term memory mechanism, wherein the long-short-term memory mechanism includes a short-term memory mechanism, and the processing process of the short-term memory mechanism includes at least one of the following: Record the latest interactions between agent roles in real time for issuing surgical instructions; The agent assistant makes decisions on subsequent surgical steps based on the current surgical steps of each agent role.

[0008] Furthermore, the decision-making on subsequent surgical steps includes: Obtain historical surgical case data; A correlation search is performed on the surgical status after each surgical step is performed based on the historical surgical case data to obtain historical status data similar to the surgical status, and a decision is made on subsequent surgical steps based on the historical status data.

[0009] Furthermore, the long-short term memory mechanism also includes a long-term memory mechanism, and the processing process of the long-term memory mechanism includes: After the agent role successfully completes the current operation or fails to perform the current operation, the current operation process is evaluated and recorded as historical operation case data.

[0010] Furthermore, the historical surgical case data includes successful surgical case data and failed surgical case data. The successful surgical case data is collected based on historical case data after the agent system successfully completed the surgery in the historical surgical scenario, and the failed surgical case data is collected based on case data after the agent system failed to perform the surgery in the historical surgical scenario.

[0011] Furthermore, the surgical scene is generated by simulation based on a pre-constructed surgical environment sandbox model, and the surgical environment sandbox model is used to simulate multiple different surgical scenes.

[0012] Furthermore, the surgical environment sandbox model is constructed based on a plurality of real surgical case data, and the construction process of the surgical environment sandbox model includes: Manually analyze each real surgical case data to obtain a corresponding surgical scene library, where each surgical scene in the surgical scene library includes surgical type, surgical status, and surgical environment; The surgical status refers to various surgical behaviors and surgical interaction operations performed in the surgical environment. The surgical status is used to feedback surgical events that are likely to occur during the surgery, and to complete the surgery according to the surgical event with the highest probability to form the surgical environment.

[0013] The surgical scenario consisting of surgical type, surgical status, and surgical environment is used as a surgical environment sandbox model.

[0014] The present invention also provides a neurosurgery large model proxy method, the method comprising the following steps: Defining proxy roles, wherein the proxy roles include the surgeon, surgical assistant, operating room nurse, circulating nurse, and anesthesiologist; Calling a large language model to generate surgical behaviors of the agent character in the surgical scene; According to the real surgical case data input in real time, the agent character is controlled to perform the surgical behavior, wherein the surgical behavior includes the agent character interacting in the surgical scene and performing corresponding surgical operations.

[0015] The neurosurgery operation model agent system provided by the present invention includes a surgical role definition module,

[0016] The agent roles in the agent system are defined as the surgeon, surgical assistant, operating room nurse, circulating nurse, and anesthesiologist, which are used to simulate various surgical personnel in a real surgical environment. Then, the large language model is called to generate the surgical behavior of the agent roles in the surgical scene, and the agent roles are controlled to interact in the surgical scene and perform corresponding surgical operations under the real surgical case data input in real time. In this way, the process of surgical simulation agent is realized by creating a realistic surgical environment and improving the interaction mechanism. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced one by one below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0018] Figure 1 It is a structural schematic diagram of the neurosurgery large model proxy system provided by the present invention.

[0019] Figure 2 It is a schematic diagram of the principle of the agent system provided by the present invention adapting to the surgical sandbox model.

[0020] Figure 3It is a schematic diagram of generating a surgical plan provided by the present invention.

[0021] Figure 4 It is a flow chart of the neurosurgery large model proxy method provided by the present invention. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0023] like Figure 1 As shown, Figure 1 The schematic diagram of the structure of the neurosurgery large model proxy system provided by the present invention is as follows. The neurosurgery large model proxy system of the present application specifically includes: a surgical role definition module, a surgical behavior generation module, and a surgical operation interaction module, which are described in detail below.

[0024] The surgical role definition module is used to define the proxy roles in the proxy system as the surgeon, surgical assistant, operating room nurse, circulating nurse, and anesthesiologist. The proxy roles here are virtual roles set up in the virtual environment of the proxy system through a computer or server. According to the real scene of neurosurgery, during the operation, the surgeon, surgical assistant, operating room nurse, circulating nurse, and anesthesiologist are generally involved in the operation. Therefore, these proxy roles also need to be defined in the virtual environment to simulate the completion of the operation.

[0025] Among them, the settings of each agent role are as follows: The surgeon in charge is responsible for performing key parts of neurosurgery or making some medical decisions. In real-life scenarios, this would be a neurosurgery technologist who specializes in pituitary tumors, meningiomas, and other complex neurosurgery procedures.

[0026] The role of a surgical assistant is to observe the surgeon's surgery, assist with some surgical procedures, and handle preoperative and postoperative preparations. In real life, this would be a neurosurgery intern who has relatively little clinical experience and needs to learn by personally participating in surgery.

[0027] The operating room nurse is responsible for handing over the surgical patient to the ward nurse (if any), keeping surgical records, and assisting the anesthesiologist in completing anesthesia. In real scenarios, this corresponds to a nurse with rich neurosurgery knowledge and clinical surgical participation experience.

[0028] The duty of the circulating nurse is to provide the surgical instruments required for the operation in a timely manner and to sterilize the instruments. In real scenarios, this corresponds to neurosurgery interns who have relatively little clinical experience and need to learn by personally participating in the operation.

[0029] An anesthesiologist is responsible for analyzing surgical plans, adjusting anesthetic drugs in a timely manner according to real-time conditions, and completing preoperative anesthesia. In real scenarios, this corresponds to an anesthesiologist who has extensive clinical experience in complex neurosurgery, is good at general anesthesia management, is proficient in various anesthesia techniques and emergency treatment measures, and can quickly make correct judgments and responses.

[0030] Of course, in some embodiments, other proxy roles can be defined according to actual needs, such as ward nurses, whose responsibilities are to monitor and record various physiological and biochemical data of surgical patients, transfer surgical patients to the operating room, and conduct patient handover with operating room nurses, etc. In real scenarios, the corresponding roles are nurses with rich clinical nursing experience.

[0031] When defining roles in the agent system, it is necessary to define the responsibilities, skills, and typical operations of each agent role, and enable the agents to interact with each other through pre-defined communication protocols to ensure coordinated operations and information sharing. In addition, a wealth of neurosurgery expertise and medical literature are collected to provide these agent roles with learning so that they can adapt to the corresponding surgical scenarios.

[0032] The surgical behavior generation module is used to call the large language model to generate the surgical behavior of the agent character in the surgical scene.

[0033] Here, when the agent system simulates the surgical scene and performs the surgical process, the corresponding operation prompt words can be manually input to the agent character in the agent system. The agent system is deployed with a large language model, which generates the corresponding surgical behavior based on the input operation prompt words. This enables the agent character to interact in the surgical environment.

[0034] The surgical operation interaction module is used to control the agent role to perform surgical actions according to the real surgical case data input in real time. In an embodiment of the present invention, the agent system simulates the surgical scene to perform the surgical process based on the input real surgical case data. The real surgical case data can be a surgical report, which records all surgical processes in a surgical scene, specifically recording the type of surgery, surgical participants, various surgical states in the surgical scene, and the surgical operations performed by each surgical participant. Through these real surgical case data, the agent system can represent the corresponding surgical participants through the defined agent role, and then simulate the surgical scene to perform the surgical process.

[0035] Therefore, after inputting the real surgical case data, the surgical operation interaction module in the agent system can parse these real surgical case data, split them into various surgical steps and surgical states, and then substitute various agent roles to perform corresponding surgical behaviors under the input prompt words. Among them, surgical behaviors include the agent roles interacting in the surgical scene and performing corresponding surgical operations, that is, the agent roles interact with each other in action, language, and corresponding surgical operations.

[0036] Through the three modules of the agent system, namely, the surgical role definition module, the surgical behavior generation module, and the surgical operation interaction module, the real surgical scene can be simulated and restored according to the input real surgical case data, and the corresponding surgical operation can be completed through the interaction and behavior between the agent roles. In this way, the process of surgical simulation agent is realized by creating a realistic surgical environment and improving the interaction mechanism.

[0037] In some embodiments, considering that there are too many proxy roles in some surgical scenes, and the surgical process and operations are complicated, it is necessary to coordinate and plan the entire surgical process, provide comprehensive technical support and guidance for the surgical team composed of proxy roles, and ensure the accuracy and efficiency of each surgical step during the operation. Therefore, the surgical role definition module is also used to define the proxy assistant in the proxy system. The proxy assistant plays the role of an intelligent assistant, has rich surgical experience and knowledge, and has the ability to continuously learn. Its responsibilities are to analyze the surgical plan, monitor the surgical behavior of each proxy role, be responsible for the overall management of the entire operation, and coordinate and plan the entire surgical process. The proxy assistant is used to send surgical instructions to the proxy role in real time in the surgical scene to guide the surgical operations performed by each proxy role. In the event of an unexpected situation or potential risk, the best emergency response is provided in a timely manner based on the rich experience in emergency handling. Thus, the surgical operation pressure of the proxy role can be reduced.

[0038] The proxy assistant does not exist in the real scenario. Specifically, it can be a database with rich neurosurgery knowledge and huge clinical and surgical experience, and it deploys various detection algorithms, recognition algorithms and decision-making models required for medical treatment.

[0039] Specifically, the agent assistant is also used to compare the surgical operations performed by the agent character with the actual standard operations in the input real surgical case data in real time. When deviations occur in the comparison, the surgical operations that cause the deviations are located and the deviations are corrected.

[0040] During the operation, the agent assistant monitors the surgical operations performed by each agent role in real time, and then compares the surgical operations performed by the agent role with the actual standard operations in the input real surgical case data to identify whether the corresponding surgical operations have deviations or even errors. When deviations are found after comparison, the deviations can be located in time, and the deviations can be corrected or corrected. In addition, the agent assistant can also provide an early warning function. During the operation, the agent assistant will also continuously monitor the various parameters and indicators of surgical equipment, instruments, and surgical patients. Once an abnormality is found, such as cerebrospinal fluid leakage in the patient, the agent assistant will immediately issue an early warning, suspend the operation process, and notify the corresponding responsible agent role to handle it.

[0041] Most importantly, the agent assistant has the ability to continuously learn and evolve. Every operation will become training data for the agent assistant to learn. By integrating a large amount of surgical case data and a constantly updated medical knowledge base, the agent assistant can continuously optimize its own algorithms and decision-making models during each operation, enabling it to perform better in future operations.

[0042] In some embodiments, in order to simulate various surgical scenarios and provide corresponding surgical decision support in different surgical scenarios, the present invention provides a method for a surgical sandbox model, in which a surgical scenario is simulated and generated by a pre-built surgical environment sandbox model in an agent system. The surgical environment sandbox model is constructed based on a plurality of different real surgical case data and is used to simulate a plurality of different surgical scenarios.

[0043] The following is an introduction to the construction process of the surgical environment sandbox model. Here we first propose a parallel world theory, because during the operation, by performing different surgical operations or making different medical decisions, different surgical states will be generated, which will give rise to different ordinary worlds, form different surgical scenes, and ultimately form different surgical results. Specifically, the essence of the parallel world is the surgical state. Therefore, when constructing the surgical environment sandbox model, it is necessary to parse the corresponding surgical state from each real surgical case data.

[0044] Here, manual analysis is performed on each real surgical case data to obtain the corresponding surgical scene library. The real surgical case data can be a surgical report that records the entire surgical process, and manual analysis can be performed by authoritative experts in the medical field to analyze and interpret each surgical report to obtain the corresponding surgical scene library, and each surgical scene in the surgical scene library includes the type of surgery, surgical status, and surgical environment. Among them, the surgical type can be various neurosurgery operations, and the surgical status is the surgical status obtained by performing different surgical steps before or during the operation, that is, a state formed by the agent role after performing various surgical behaviors and surgical interactions. The surgical status is used to feedback the surgical events that are likely to occur during the operation. These surgical events require the agent assistant to make corresponding decisions. Here, the operation is completed according to the surgical event with the highest probability to form a surgical environment.

[0045] Because the surgery type, surgery status, and surgery environment can all be obtained by manually analyzing real surgery case data, each real surgery case data is analyzed to obtain a surgery scene. Therefore, the surgery scene composed of the surgery type, surgery status, and surgery environment is used as a surgery environment sandbox model.

[0046] See also Figure 2 , Figure 2 It is a schematic diagram of the principle of the agent system provided by the present invention adapting to the surgical sandbox model. Figure 2 The specific example shows the process of simulating one of the surgical scenarios through the surgical sandbox model. In the preoperative preparation stage, the agent assistant understands the surgical plan based on the real surgical case data (including document records and picture records) and formulates the corresponding surgical plan. The surgical step performed at this time is to perform task 1, that is, patient handover, and the surgical status is to enter the operating room. The next surgical step to be performed is task 2, which is still the ward nurse and the operating room nurse to perform the patient handover, and the surgical status is to enter the operating room. After entering the operating room, the surgical step is to perform task 3, that is, patient anesthesia, and the operating room nurse interacts with the anesthesiologist to complete the anesthesia. At this time, the surgical status is to complete the anesthesia. After the anesthesia is completed, the surgical step is to perform task 4, that is, surgical instrument preparation, and the surgical assistant interacts with the anesthesiologist to correct the patient's posture, and the surgical status is to correct the posture.

[0047] At the beginning of the operation, the surgical steps are to perform the operation, and the operation status is in progress. Under the instructions issued by the agent assistant, the surgeon and the surgical assistant jointly complete the operation. At the same time, the agent assistant receives feedback on the probability of surgical events during the operation through real-time monitoring, and completes the operation based on the surgical event with the highest probability, and then makes a decision, such as whether the brain tumor is completely removed, and further determines whether there is cerebrospinal fluid leakage. According to the theory of parallel worlds, different parallel worlds will make different decisions, thus forming different surgical scenes.

[0048] At the end of the operation, the corresponding operation step is to write an operation report, and the operation status is to transfer to the ward. Thus, the introduction of the operation process in the corresponding operation scene is completed.

[0049] The embodiments of the present invention can construct a large number of surgical sandbox models by collecting a large amount of real surgical case data. The surgical sandbox models can be used to simulate various surgical scenarios, thereby training the decision-making ability and medical response ability of agent roles and agent assistants in different surgical scenarios in the agent system.

[0050] In order to further optimize the simulation and decision-making during the operation, the embodiment of the present invention introduces a long-term and short-term memory mechanism in the operation scene to assist in completing the operation. The long-term and short-term memory mechanism includes a short-term memory mechanism and a long-term memory mechanism. Among them, the short-term memory acts on information retrieval during a single operation, focusing on key information such as the agent role, the interaction and behavior of the agent assistant. The specific processing process includes: The latest interactions between agent roles are recorded in real time for issuing surgical instructions. Figure 2 As shown, there may be multiple interactions between agent roles. Therefore, through short-term memory, the latest interaction content between agent roles is recorded in real time for issuing surgical instructions. This short-term memory mechanism can be applied to intelligent assistants to ensure the timeliness and effectiveness of instruction issuance.

[0051] In addition, the short-term memory mechanism can also enable the agent assistant to make decisions on subsequent surgical steps based on the current surgical steps of each agent role. During the operation, the agent assistant monitors the surgical operations and surgical steps of each agent role in real time, and then conducts a comprehensive analysis of the overall situation, integrates the operating steps of each agent role, provides the next step of instructions or completes the decision on the subsequent surgical steps, thereby ensuring the comprehensiveness and rationality of the decision.

[0052] Among them, the specific process of making decisions on subsequent surgical steps is: first, obtain historical surgical case data, which includes successful surgical case data and failed surgical case data. The successful surgical case data is collected based on the historical case data after the agent system successfully completed the operation in the historical surgical scenario, and the failed surgical case data is collected based on the case data after the agent system failed to perform the operation in the historical surgical scenario.

[0053] Then, the surgical status after each surgical step is retrieved based on the historical surgical case data, and the historical status data similar to the surgical status is obtained, and the next surgical step is decided based on the historical status data. Here, similar historical status data can be retrieved from the historical surgical case data according to the current surgical step for reference, to learn from the predecessors' practices, and to make decisions on the subsequent surgical steps based on the historical status data.

[0054] The decision-making process here is realized by referring to historical surgical case data, which is a learning process for the agent role and the agent assistant. Using historical surgical case data to help process each surgical step can provide accuracy in decision-making. Through successful surgical case data, the agent role and the agent assistant can learn from the experience of successful surgery and improve their own capabilities. And through failed surgical case data, the agent role and the agent assistant can recognize and summarize the lessons of surgical failure and make improvements.

[0055] The long-term memory mechanism is used in multiple surgical procedures. During each surgical procedure, the agent role and the agent assistant can pay attention to the historical surgical experience related to themselves according to their own responsibilities, learn and summarize, and continuously improve their own abilities during each surgical procedure. The historical surgical experience needs to be obtained using the long-term memory mechanism.

[0056] Specifically, the processing process of the long-term memory mechanism includes: after the agent role successfully completes the current operation or fails to perform the current operation, the agent assistant is also used to evaluate the current operation process and record the current operation process as historical operation case data. Here, after each operation process, regardless of whether the operation is successful or not, the agent assistant scores the operation report corresponding to the current operation to evaluate the learning effect. The current operation is recorded as historical operation case data. If the operation is successful, it is recorded as successful operation case data, otherwise it is recorded as failed operation case data.

[0057] In this way, in the subsequent surgical process, the agent role and the agent assistant can learn based on the historical surgical case data, that is, continuously learn successful experiences or summarize failed experiences from successful surgical case data and failed surgical case data. Figure 3 As shown in the figure, during a surgical procedure in a certain surgical scenario, the agent assistant filters out the surgical plan from the input real surgical case data based on the long-term memory mechanism, that is, based on the medical experience learned from the historical surgical case data, and makes better decisions in each surgical step based on the surgical plan, thereby improving the medical level of the entire agent team.

[0058] The present invention also provides a neurosurgery large model agent method, which is applied to an agent system for performing neurosurgery or other systems that simulate neurosurgery scenes. Figure 4 Schematic diagram of the process of the neurosurgery large model proxy method provided by the present invention, such as Figure 4 As shown, the method includes the following steps 401 to 403. The following is a specific description using an agent system as an example.

[0059] Step 401: define proxy roles, where the proxy roles include the surgeon, surgical assistant, operating room nurse, circulating nurse, and anesthesiologist.

[0060] Here, various proxy roles can be freely defined in the proxy system according to the actual needs of neurosurgery scenarios, including the surgeon, surgical assistant, operating room nurse, circulating nurse, and anesthesiologist. Other proxy roles can also be set in other scenarios, such as ward nurses and proxy assistants, which are not limited here. By defining these proxy roles, the proxy system can simulate real surgical scenarios.

[0061] Step 402: Call the large language model to generate the surgical behavior of the agent character in the surgical scene.

[0062] Next, the agent system receives the input prompt words in real time, where the prompt words are used to instruct the agent character to perform the corresponding surgical behavior in the surgical scene. Here, the large language model deployed in the agent system can be called, and the input prompt words can be input into the large language model for generation to obtain the corresponding surgical behavior, so that the agent character can perform the corresponding surgical behavior and perform relevant interactions in the surgical scene.

[0063] Step 403: Control the agent character to perform surgical behavior according to the real surgical case data input in real time, wherein the surgical behavior includes the agent character interacting in the surgical scene and performing corresponding surgical operations.

[0064] Ultimately, the agent character can be controlled to perform surgical behaviors based on real-time input of real surgical case data, where the surgical behaviors include the agent character interacting in the surgical scene and performing corresponding surgical operations.

[0065] The agent system simulates the surgical scene and performs the surgical process based on the input real surgical case data. The real surgical case data can be a surgical report, which records all the surgical processes in a real surgical scene, specifically recording the type of surgery, surgical participants, various surgical states in the surgical scene, and the surgical operations performed by each surgical participant. Through these real surgical case data, the agent system can represent the corresponding surgical participants through the defined agent roles, and then simulate the surgical scene to perform the surgical process.

[0066] Therefore, after inputting real surgical case data, the surgical operation interaction module in the agent system can parse the data, and then substitute various agent roles accordingly to perform corresponding surgical behaviors under the input prompt words. The surgical behaviors include the agent roles interacting in the surgical scene and performing corresponding surgical operations, that is, the agent roles perform action interactions, language interactions and corresponding surgical operations.

[0067] It should be noted that the beneficial effects of the neurosurgery large model proxy method here and the neurosurgery large model proxy system mentioned above can correspond to each other, so the beneficial effects of the neurosurgery large model proxy method will not be repeated here.

[0068] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A neurosurgery large model agent system, characterized in that: The agent system includes: a surgical role definition module, a surgical behavior generation module, and a surgical operation interaction module; The surgical role definition module is used to define the proxy roles in the proxy system as the chief surgeon, surgical assistant, operating room nurse, circulating nurse and anesthesiologist; The surgical behavior generation module is used to call the large language model to generate the surgical behavior of the agent character in the surgical scene; The surgical operation interaction module is used to control the agent character to perform the surgical behavior according to the real surgical case data input in real time, wherein the surgical behavior includes the agent character interacting in the surgical scene and performing the corresponding surgical operation.

2. The neurosurgery large model agent system according to claim 1, characterized in that: The surgical role definition module is also used to define an agent assistant in the agent system, and the agent assistant is used to send surgical instructions to the agent role in real time in the surgical scene to guide the surgical operations performed by each agent role.

3. The neurosurgery large model agent system according to claim 2, characterized in that: The agent assistant is also used to compare the surgical operation performed by the agent role with the actual standard operation in the real surgical case data in real time. When a deviation occurs in the comparison, the surgical operation causing the deviation is located and the deviation is corrected.

4. The neurosurgery large model agent system according to claim 2, characterized in that: The agent system executes the surgical process in the surgical scene through a long-short-term memory mechanism, wherein the long-short-term memory mechanism includes a short-term memory mechanism, and the processing process of the short-term memory mechanism includes at least one of the following: Record the latest interactions between agent roles in real time for issuing surgical instructions; The agent assistant makes decisions on subsequent surgical steps based on the current surgical steps of each agent role.

5. The neurosurgery large model agent system according to claim 4, characterized in that: The decision on subsequent surgical steps includes: Obtain historical surgical case data; A correlation search is performed on the surgical status after each surgical step is performed based on the historical surgical case data to obtain historical status data similar to the surgical status, and a decision is made on subsequent surgical steps based on the historical status data.

6. The neurosurgery large model agent system according to claim 4, characterized in that: The long-term and short-term memory mechanism also includes a long-term memory mechanism, and the processing process of the long-term memory mechanism includes: After the agent role successfully completes the current operation or fails to perform the current operation, the current operation process is evaluated and recorded as historical operation case data.

7. The neurosurgery large model agent system according to claim 5, characterized in that: The historical surgical case data includes successful surgical case data and failed surgical case data. The successful surgical case data is collected based on historical case data after the agent system successfully completed the surgery in the historical surgical scenario, and the failed surgical case data is collected based on case data after the agent system failed to perform the surgery in the historical surgical scenario.

8. The neurosurgery large model agent system according to claim 1, characterized in that: The surgical scene is generated by simulation based on a pre-built surgical environment sandbox model, and the surgical environment sandbox model is used to simulate multiple different surgical scenes.

9. The neurosurgery large model agent system according to claim 8, characterized in that: The surgical environment sandbox model is constructed based on multiple real surgical case data. The construction process of the surgical environment sandbox model includes: Manually analyze each real surgical case data to obtain a corresponding surgical scene library, where each surgical scene in the surgical scene library includes surgical type, surgical status, and surgical environment; The surgical status refers to various surgical behaviors and surgical interaction operations performed in the surgical environment. The surgical status is used to feedback surgical events that are likely to occur during the surgery, and to complete the surgery according to the surgical event with the highest probability to form the surgical environment. The surgical scenario consisting of surgical type, surgical status, and surgical environment is used as a surgical environment sandbox model.

10. A neurosurgery large model proxy method, characterized in that: The method comprises: Defining proxy roles, wherein the proxy roles include the surgeon, surgical assistant, operating room nurse, circulating nurse, and anesthesiologist; Calling a large language model to generate surgical behaviors of the agent character in the surgical scene; According to the real surgical case data input in real time, the agent character is controlled to perform the surgical behavior, wherein the surgical behavior includes the agent character interacting in the surgical scene and performing corresponding surgical operations.

Images