Methods, systems, electronic devices, and storage media for determining treatment plans
By employing a method of collaborative work among multiple intelligent agents, and utilizing intelligent agents with different structures to generate tumor treatment plans, this approach solves the problem in existing technologies where tumor treatment plans struggle to balance tumor killing with the protection of organs at risk, thus achieving more efficient and accurate treatment plan formulation.
Patent Information
- Application Number
- CN202510641966.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-16
- Publication Date
- 2026-11-17
AI Technical Summary
Existing technologies struggle to effectively balance tumor killing and organ protection when developing cancer treatment plans, resulting in insufficient precision and low efficiency in planning, especially in clinical applications with significant individual differences, requiring extensive manual adjustments.
A multi-agent collaborative approach is adopted. By acquiring patient case information, agents with different structures (such as agents based on constraint networks, covariance matrix adaptive evolutionary algorithms, graph neural networks, and large language models) are used for comprehensive analysis and information sharing to generate treatment plans.
It improves the accuracy and efficiency of treatment plan generation, reduces information collection and processing time, avoids misdiagnosis or missed diagnosis by a single intelligent agent, and enhances the overall quality and efficiency of treatment plans.
Smart Images

Figure CN123120252A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of medical technology, and in particular to a method, system, electronic device, and storage medium for determining a treatment plan. Background Technology
[0002] Currently, one of the key technologies in cancer treatment is balancing tumor killing with the protection of organs at risk. During treatment, to achieve the desired therapeutic effect, a sufficiently high radiation dose needs to be delivered to the tumor target area to achieve local control and elimination of the tumor. However, tumors are often surrounded by various vital normal tissues and organs, such as the lungs, heart, and esophagus in lung cancer treatment. These organs at risk have low radiation tolerance, and excessively high radiation doses can lead to severe radiation damage, causing complications and potentially endangering life.
[0003] In clinical treatment, radiation physicists often need to repeatedly weigh the relationship between the probability of tumor control and the probability of complications in normal tissues when developing treatment plans based on their experience. This takes a lot of time for manual adjustments and optimizations.
[0004] To improve the efficiency of treatment planning, a common approach is to use deep learning models to generate treatment plans. These models generate plans based on learned case data, offering improved efficiency compared to manual generation. However, in real-world clinical settings, individual patient differences exist, resulting in insufficient accuracy in treatment plans generated using this method. Before clinical application, radiation oncology physicists still need to spend considerable time adjusting and optimizing these plans, thus maintaining a relatively low overall efficiency in treatment planning. Summary of the Invention
[0005] The purpose of this application is to provide a method, system, electronic device, and storage medium for determining a treatment plan, with the aim of improving the efficiency of treatment plan development.
[0006] In the first aspect, a method for determining a treatment plan is provided, comprising: acquiring patient medical information; and, based on the patient medical information, coordinating the work of multiple intelligent agents to generate a treatment plan for the patient; wherein the multiple intelligent agents have different structures.
[0007] The treatment plan determination method in this application involves acquiring patient medical information and collaborating with multiple intelligent agents to generate a treatment plan based on that information. These agents, with their different configurations, are adept at handling different types of medical information. Through collaborative work, they can comprehensively analyze the medical information from multiple dimensions, avoiding misdiagnosis or missed diagnosis due to the functional limitations of a single agent, thereby improving the accuracy of the treatment plan. Furthermore, the agents can share and interact with each other, reducing the time spent repeatedly collecting and processing information and improving the overall efficiency of generating the treatment plan.
[0008] In some embodiments, the various agents differ in at least one of the following: the model structure adopted by the agent, the reasoning ability of the agent, the reasoning speed of the agent, the hardware requirements of the agent, the functions of the agent, and the complexity of the information that the agent supports processing.
[0009] In some embodiments, the multiple intelligent agents include at least one of the following: a first intelligent agent, a second intelligent agent, a third intelligent agent, and a fourth intelligent agent; wherein, the first intelligent agent is an intelligent agent constructed based on a constraint network; the second intelligent agent is an intelligent agent constructed based on a covariance matrix adaptive evolution algorithm; the third intelligent agent is an intelligent agent constructed based on a graph neural network; and the fourth intelligent agent is an intelligent agent constructed based on a large language model.
[0010] In some embodiments, based on a patient's case information, the work of multiple agents is coordinated to generate a treatment plan for the patient, including: identifying the complexity of the case information, whereby the complexity characterizes the level of conflict between the patient's required dose and protection of organs at risk; identifying a target agent from among the multiple agents whose complexity matches that of the case information; and generating a treatment plan using the case information through the target agent.
[0011] In some embodiments, the complexity, from low to high, includes a first complexity, a second complexity, and a third complexity; the multiple intelligent agents include: a first intelligent agent, a second intelligent agent, and a third intelligent agent; the first intelligent agent is used to process case information with a complexity of the first complexity; the second intelligent agent is used to process case information with a complexity of the second complexity; and the third intelligent agent is used to process case information with a complexity of the third complexity; wherein the first intelligent agent, the second intelligent agent, and the third intelligent agent satisfy at least one of the following relationships: the reasoning speed of the first intelligent agent is greater than the reasoning speed of the second intelligent agent, and the reasoning speed of the second intelligent agent is greater than the reasoning speed of the third intelligent agent; the reasoning ability of the first intelligent agent is weaker than the reasoning ability of the second intelligent agent, and the reasoning ability of the second intelligent agent is weaker than the reasoning ability of the third intelligent agent.
[0012] In some embodiments, based on the patient's medical records, multiple agents work in coordination to generate a treatment plan for the patient, including: generating corresponding candidate treatment plans by using the medical records through the multiple agents respectively; and determining a treatment plan from the candidate treatment plans generated by the multiple agents respectively.
[0013] In some embodiments, the method further includes: evaluating the treatment plan to determine the predicted treatment effect of the treatment plan; if the predicted treatment effect of the treatment plan does not meet the expected effect, performing multiple optimization processes on the candidate treatment plans in the candidate treatment plan set until a preset stopping condition is met; wherein the preset stopping condition includes at least one of the following: the predicted treatment effect of the optimized treatment plan meets the expected effect; the duration of the optimization process is greater than a preset time threshold.
[0014] In some embodiments, during each optimization process, candidate treatment plans from the first treatment plan set are input into the target agent, and the target agent optimizes the candidate treatment plans to obtain optimized treatment plans. The first treatment plan set is the set of candidate treatment plans remaining in the candidate treatment plan set excluding the treatment plan corresponding to the target agent.
[0015] In some embodiments, the assessment includes: dose assessment and / or motion assessment of the multileaf grating blades, wherein the dose assessment is used to determine whether the dose required for tumor radiotherapy is greater than a preset target area threshold and whether the dose to organs at risk is less than a preset organ at risk threshold, and the motion assessment of the multileaf grating blades is used to determine whether the matching degree between the irradiation field formed by the multileaf grating blades and the tumor target area is greater than a preset matching degree threshold.
[0016] Secondly, a treatment plan determination system is also provided, comprising: a master intelligent agent and multiple intelligent agents for determining the treatment plan; the master intelligent agent is configured to: acquire patient medical information; and, based on the patient medical information, coordinate the work of multiple intelligent agents to generate a treatment plan for the patient.
[0017] In some embodiments, the various agents differ in at least one of the following: the model structure adopted by the agent, the reasoning ability of the agent, the reasoning speed of the agent, the hardware requirements of the agent, the functions of the agent, and the complexity of the information that the agent supports processing.
[0018] In some embodiments, the multiple intelligent agents include at least one of the following: a first intelligent agent, a second intelligent agent, a third intelligent agent, and a fourth intelligent agent; wherein, the first intelligent agent is an intelligent agent constructed based on a constraint network; the second intelligent agent is an intelligent agent constructed based on a covariance matrix adaptive evolution algorithm; the third intelligent agent is an intelligent agent constructed based on a graph neural network; and the fourth intelligent agent is an intelligent agent constructed based on a large language model.
[0019] In some embodiments, multiple agents are deployed on one electronic device, or multiple agents are deployed on different electronic devices.
[0020] Thirdly, an electronic device is also provided, comprising: a processor and a memory configured to store processor-executable instructions; wherein the processor is configured to execute the instructions to implement any of the alternative methods for determining a treatment plan described in the first aspect above.
[0021] Fourthly, this application provides a computer-readable storage medium storing instructions that, when executed by a device, enable the device to perform any of the optional treatment plan determination methods described in the first aspect.
[0022] Fifthly, this application provides a computer program product including computer instructions that, when executed on a processor of a device, enable the device to perform a method for determining a treatment plan as described in any of the optional methods in the first aspect above. Attached Figure Description
[0023] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0024] Figure 1 A schematic diagram of a treatment plan determination system provided in this application embodiment;
[0025] Figure 2 A schematic diagram of a treatment plan determination system provided in an embodiment of this application;
[0026] Figure 3 A schematic diagram illustrating the flow of information in a treatment plan determination system provided in this application embodiment;
[0027] Figure 4 A schematic diagram of the architecture of a treatment plan determination system provided in an embodiment of this application;
[0028] Figure 5 A flowchart illustrating a method for determining a treatment plan provided in an embodiment of this application;
[0029] Figure 6 A flowchart illustrating another method for determining a treatment plan provided in an embodiment of this application;
[0030] Figure 7 A flowchart illustrating another method for determining a treatment plan provided in an embodiment of this application;
[0031] Figure 8A flowchart illustrating another method for determining a treatment plan provided in an embodiment of this application;
[0032] Figure 9 A schematic diagram of a treatment plan determination device provided in an embodiment of this application;
[0033] Figure 10 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation
[0034] In the embodiments of this application, the terms "first," "second," "third," "fourth," "fifth," and "sixth" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined with "first," "second," "third," "fourth," "fifth," and "sixth" may explicitly or implicitly include one or more of that feature.
[0035] In embodiments of this application, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.
[0036] "A and / or B" includes the following three combinations: A only, B only, and a combination of A and B.
[0037] In the embodiments of this application, "parallel," "perpendicular," and "equal" include the described situation and situations similar to the described situation, the range of which is within an acceptable deviation range, said acceptable deviation range being determined by those skilled in the art taking into account the measurement under discussion and the error associated with the measurement of a particular quantity (i.e., the limitations of the measurement system). For example, "parallel" includes absolute parallelism and approximate parallelism, wherein the acceptable deviation range for approximate parallelism may be, for example, a deviation within 5°; "perpendicular" includes absolute perpendicularity and approximate perpendicularity, wherein the acceptable deviation range for approximate perpendicularity may also be, for example, a deviation within 5°. "Equal" includes absolute equality and approximate equality, wherein the acceptable deviation range for approximate equality may be, for example, a difference between the two equals being less than or equal to 5% of either one.
[0038] Currently, one of the key technologies in cancer treatment is balancing tumor killing with the protection of organs at risk. During treatment, to achieve the desired therapeutic effect, a sufficiently high radiation dose needs to be delivered to the tumor target area to achieve local control and elimination of the tumor. However, tumors are often surrounded by various vital normal tissues and organs, such as the lungs, heart, and esophagus in lung cancer treatment. These organs at risk have low radiation tolerance, and excessively high radiation doses can lead to severe radiation damage, causing complications and potentially endangering life.
[0039] In clinical treatment, radiation physicists often need to repeatedly weigh the relationship between the probability of tumor control and the probability of complications in normal tissues when developing treatment plans based on their experience. This takes a lot of time for manual adjustments and optimizations.
[0040] To improve the efficiency of treatment planning, a common approach is to use deep learning models to generate treatment plans. These models generate plans based on learned case data, offering improved efficiency compared to manual generation. However, in real-world clinical settings, individual patient differences exist, resulting in insufficient accuracy in treatment plans generated using this method. Before clinical application, radiation oncology physicists still need to spend considerable time adjusting and optimizing these plans, thus maintaining a relatively low overall efficiency in treatment planning.
[0041] Based on this, this application proposes a method, system, electronic device, and storage medium for determining a treatment plan, including: acquiring patient medical information; and, based on the patient's medical information, coordinating the work of multiple intelligent agents to generate a treatment plan for the patient; wherein the multiple intelligent agents have different structures. By acquiring the patient's medical information and coordinating multiple intelligent agents to generate a treatment plan based on the medical information, the intelligent agents with different structures are adept at handling different types of medical information. Through collaborative work, these intelligent agents can comprehensively analyze the medical information from multiple dimensions, avoiding misdiagnosis or missed diagnosis caused by the functional limitations of a single intelligent agent, thereby improving the accuracy of the treatment plan. Furthermore, the intelligent agents can share and interact with each other, reducing the time spent on repeatedly collecting and processing information and improving the overall efficiency of generating the treatment plan.
[0042] The following description, in conjunction with the accompanying drawings, introduces the method, system, electronic equipment, and storage medium for determining the treatment plan as described in this application.
[0043] In some embodiments, this application proposes a treatment plan determination system, including: a master agent 101 and multiple agents 102 for determining the treatment plan.
[0044] As one possible implementation, the master intelligent agent 101 is used to acquire the patient's medical records and, based on the patient's medical records, coordinates the work of multiple intelligent agents 102 to generate a treatment plan for the patient.
[0045] The master control agent 101 serves as the central coordinating unit of the treatment plan determination system, responsible for integrating patient case information and scheduling other agents to work collaboratively. Multiple agents act as execution units, responsible for generating treatment plans based on the case information integrated by the master control agent.
[0046] In some embodiments, the various agents 102 differ in at least one of the following: the model structure adopted by the agent, the reasoning ability of the agent, the reasoning speed of the agent, the hardware requirements of the agent, the functions of the agent, and the complexity of the information that the agent supports processing.
[0047] It should be understood that different agents may be designed for different tasks and problems. For example, an agent for image recognition might use a convolutional neural network (CNN) structure because CNNs excel at processing local features and spatial structure of image data. An agent for natural language processing might use a Transformer structure, which is advantageous in capturing long-range dependencies and semantic understanding in text. Therefore, depending on the characteristics of the specific task, the agent will choose different model structures to achieve optimal performance.
[0048] The reasoning ability of an intelligent agent depends on its design purpose. For example, in a smaller hospital, tumor images in patient medical records are generally simple in structure, so the agent only needs to make simple decisions based on simple medical information, and the requirement for reasoning ability is relatively low. However, in a larger hospital, the agent needs to process complex medical information and perform complex logical reasoning and judgment, so a stronger reasoning ability is required.
[0049] The difference in reasoning speed between agents is mainly related to the complexity of the case information they process. Case information containing a large amount of detailed tumor images, examination reports, and other data requires the agent to process and analyze this massive amount of information, extracting key features and correlations, which demands more computing resources and time. Simple cases, on the other hand, may only contain basic symptoms and a few examination results, making them easier for the agent to process and resulting in faster reasoning speeds.
[0050] Different intelligent agents have different hardware requirements due to their varying model structures and computational complexity. Complex deep learning models, such as large language models or complex image generation models, require powerful computing resources, such as high-performance GPUs, to support their operation, while some simple rule-based intelligent agents may only require a regular CPU to meet their operational needs.
[0051] The functions of an intelligent agent are determined based on its design goals. Driven by different application scenarios and needs, the responsibilities and capabilities of intelligent agents vary. For example, some intelligent agents are specifically designed to determine treatment plans; they deeply analyze patient case information, combining medical knowledge bases and clinical experience to develop personalized treatment plans for patients. Other agents optimize treatment plans, evaluating existing plans and continuously adjusting and improving aspects such as dosage distribution to make them more scientific and rational. Still other agents handle the scheduling of multiple agents, requiring real-time monitoring of each agent's operational status, processing capacity, and task progress. Based on the urgency, complexity, and priority of tasks, they rationally allocate tasks, coordinate collaboration between different agents, and ensure the efficient and stable operation of the entire system.
[0052] The complexity of information that an intelligent agent can process is related to its reasoning ability. The stronger the agent's reasoning ability, the higher the complexity of the information it can process; conversely, the lower the agent's reasoning ability, the lower the complexity of the information it can process.
[0053] Different intelligent agents may have different knowledge reserves. Intelligent agents with rich professional knowledge graphs and a large amount of training data can better understand and analyze information when dealing with complex information in their professional fields through knowledge retrieval and matching. Conversely, intelligent agents with insufficient knowledge reserves are prone to misjudgment or inability to handle complex information that is beyond their cognitive scope.
[0054] In other embodiments, see again Figure 1 The multiple intelligent agents 102 include at least one of the following: a first intelligent agent 1021, a second intelligent agent 1022, a third intelligent agent 1023, and a fourth intelligent agent 1024.
[0055] Among them, the first intelligent agent 1021 is an intelligent agent built based on a constraint network, the second intelligent agent 1022 is an intelligent agent built based on an adaptive evolutionary algorithm of covariance matrix, the third intelligent agent 1023 is an intelligent agent built based on a graph neural network, and the fourth intelligent agent 1024 is an intelligent agent built based on a large language model.
[0056] The first agent 1021 is built upon constraint networks. A constraint network is a structure used to represent and process constraints, solving problems by defining constraints between variables. When building an agent, using constraint networks allows it to follow specific rules and restrictions, thus better addressing tasks with multiple constraints. For example, in determining a treatment plan, the agent can use constraint networks to ensure that the dosage meets preset conditions, such as the target dose exceeding a preset threshold.
[0057] The second agent, 1022, is constructed based on the covariance matrix adaptive evolutionary algorithm. This algorithm is an evolutionary algorithm that guides the search direction by adjusting the covariance matrix of the search space to find the optimal solution. For example, when optimizing treatment plans, an agent built based on this algorithm exhibits strong optimization capabilities, continuously evolving and adapting within a complex search space to find more effective treatment plans.
[0058] The third agent, 1023, is built upon a graph neural network. A graph neural network is a type of neural network specifically designed for processing graph-structured data; it can model and learn the nodes and edges in a graph. In the process of determining a treatment plan, case information often exists in the form of images, and an agent built based on a graph neural network can effectively process this case information.
[0059] The third agent 1023 can also be constructed by combining the covariance matrix adaptive evolution algorithm and the graph neural network. This allows the third agent 1023 to utilize the optimization capability of the covariance matrix adaptive evolution algorithm and the graph neural network to efficiently process and analyze data with complex relationships (such as graph data), thereby improving the performance and decision-making ability of the third agent 1023.
[0060] The fourth intelligent agent, 1024, is built upon a large language model. A large language model is a deep learning-based language processing model that learns from large amounts of text data to understand and generate natural language. In addition to image data, case information also includes text information. An intelligent agent built upon a large language model can understand and reason based on the input text information and generate corresponding responses or processing results.
[0061] The differences between different intelligent agents can be seen in Table 1:
[0062] Table 1
[0063]
[0064]
[0065] In some embodiments, multiple agents are deployed on one electronic device, or multiple agents are deployed on different electronic devices.
[0066] As one possible implementation, deploying multiple agents on a single electronic device means installing and running multiple agents of different types within the same device, allowing them to collaborate within the same hardware resources. With multiple agents residing within the same device, data interaction and communication between them are relatively convenient. However, this also places high demands on the device's hardware performance, as the simultaneous operation of multiple agents can consume significant computing resources and storage space.
[0067] As another possible implementation, multiple intelligent agents are deployed on different electronic devices. These agents, distributed across different devices, communicate and collaborate via networks (such as the Internet, local area networks, etc.). The advantage of this deployment method is that it can select the most suitable device for deployment based on the characteristics and needs of each intelligent agent, fully leveraging the performance advantages of different devices, and avoiding the performance impact of a single device being overloaded.
[0068] For example, such as Figure 2 As shown, the system for determining the treatment plan includes: a first intelligent agent, a second intelligent agent, a third intelligent agent, a fourth intelligent agent, and a master intelligent agent.
[0069] When generating a treatment plan, the master agent acquires case information, then evaluates the complexity of the case information to obtain the first, second, and third complexities from low to high, and sends the case information to the agent corresponding to the complexity.
[0070] In the clinical phase, case information that can be used to generate treatment plans according to standardized procedures is relatively low in complexity, so it is input into the first intelligent agent. Case information requiring the generation of personalized treatment plans is typically more complex, requiring an agent with stronger reasoning capabilities to generate the treatment plan; therefore, it is input into the second intelligent agent. Case information requiring research and analysis is usually for complex and rare cases, requiring an agent with even stronger reasoning capabilities; therefore, it is input into the third intelligent agent.
[0071] Understandably, the intelligent agent determines treatment plans through complex model algorithms deployed on it. These models integrate advanced technologies such as deep learning, reinforcement learning, and knowledge graphs, enabling in-depth analysis and feature extraction of patient case information to generate treatment plans. However, medical knowledge is constantly being updated, new treatment plans are constantly emerging, and patient characteristics are constantly changing. To ensure that the intelligent agent can always provide accurate and effective treatment plans, the model needs continuous updates and iterations. By continuously collecting new case data and employing strategies such as online learning, the model's parameters and structure can be optimized and adjusted, allowing the model to adapt to the evolving medical field and improve its decision-making and generalization abilities.
[0072] As one possible implementation, the fourth agent can learn online from pre-annotated case information and corresponding treatment plans through experience or historical data. During the learning process, the fourth agent will transform the extracted key knowledge and experience into optimization directions for model parameters, and then pass the updated parameters to the first, second, and third agents through a federated learning framework. After receiving the updated parameters, these agents will adjust the deployed model, ultimately achieving iterative upgrades to the model.
[0073] like Figure 3 As shown, in the treatment plan generation process, the information flow in the treatment plan determination system is as follows: First, the case information is preprocessed, including image preprocessing and Natural Language Processing (NLP) parsing of clinical protocols (case information, dosage standards, etc.). Feature extraction is then performed on the preprocessed case information to obtain 3D hash codes, anatomical topology, and semantic features. Then, each agent generates a treatment plan based on the required features. Finally, the Monte Carlo verification method is used to verify the treatment plan, including dosage verification and kinematic verification. Dosage verification is used to determine whether the generated treatment plan meets the preset requirements, and kinematic verification is used to determine whether collisions will occur between the multi-leaf grating blades and whether the velocity curve of each blade is smooth.
[0074] like Figure 4 As shown, each healthcare institution, in addition to deploying a treatment plan determination system, can also deploy a federated learning center. The federated learning center aggregates models in a pattern database and then distributes the models to the treatment plan determination system. The federated learning center can learn from local data, training the model to obtain updated parameters. It can also obtain model parameters from other healthcare institutions' federated learning centers and compare them with its local model to obtain updated parameters. These updated parameters are then sent to the treatment plan determination system via the federated learning framework, allowing the system to update its plans based on these parameters.
[0075] like Figure 5 As shown in the embodiments of this application, the method for determining a treatment plan includes the following steps:
[0076] S501. Obtain the patient's medical records.
[0077] Among them, case information is the patient's multimodal medical data, including information such as imaging, pathology, and medical history. For example, case information may include computed tomography (CT) images and / or magnetic resonance imaging (MRI) of tumors, treatment records, etc.
[0078] As one possible approach, patient medical records can be stored in a medical institution's medical record system. The controlling agent can interface with the medical record system and, using a dedicated application programming interface (API) or data access protocol, read the patient's medical records from the system's database according to authorized and prescribed permissions.
[0079] As another possible implementation, the intelligent agent is deployed on an electronic device that can provide an interface for interaction with the user, allowing medical staff to directly input case information.
[0080] As another possible approach, various other information systems exist in the healthcare environment, such as Laboratory Information Systems (LIS) and Picture Archiving and Communication Systems (PACS). Intelligent agents can integrate with these systems and automatically retrieve data from the appropriate systems when patient medical records are needed.
[0081] S502. Based on the patient's medical records, coordinate the work of multiple intelligent agents to generate a treatment plan for the patient.
[0082] Among them, the structures of various intelligent agents are different.
[0083] One possible approach is to assign treatment plan generation tasks to each agent based on case information and the characteristics of different agents. Specifically, each agent has a different architecture, and consequently, each agent possesses unique characteristics and strengths. Some agents may excel in processing medical images, while others may be more capable in natural language processing, quickly understanding and analyzing textual descriptions in medical records. Assigning appropriate tasks to different agents based on their characteristics to generate treatment plans can fully leverage the strengths of each agent, improving the quality and efficiency of treatment plan generation.
[0084] In one possible implementation, a data-sharing mechanism can be established between the agents to share treatment plans. The agents can then optimize the shared treatment plans to ultimately determine the optimal treatment plan.
[0085] In one possible implementation, multiple agents can be controlled to perform collaborative reasoning according to a predetermined process. For example, in the first stage, the first agent generates a treatment plan using the case information; in the second stage, the second agent generates a treatment plan using the case information; and in the third stage, the third agent combines the results of the first two stages to perform a comprehensive analysis and obtain the final treatment plan.
[0086] Therefore, by acquiring patient medical information, multiple intelligent agents can collaborate to generate treatment plans based on this information. These agents, with their different configurations, are adept at handling different types of medical information. Through collaborative work, they can comprehensively analyze the medical information from multiple dimensions, avoiding misdiagnosis or missed diagnosis caused by the functional limitations of a single agent, thus improving the accuracy of the treatment plan. Furthermore, the agents can share and interact with each other, reducing the time spent repeatedly collecting and processing information and improving the overall efficiency of generating treatment plans.
[0087] In some embodiments, the various agents differ in at least one of the following: the model structure adopted by the agent, the reasoning ability of the agent, the reasoning speed of the agent, the hardware requirements of the agent, the functions of the agent, and the complexity of the information that the agent supports processing.
[0088] In some embodiments, the multiple intelligent agents include at least one of the following: a first intelligent agent, a second intelligent agent, a third intelligent agent, and a fourth intelligent agent; wherein, the first intelligent agent is an intelligent agent constructed based on a constraint network; the second intelligent agent is an intelligent agent constructed based on a covariance matrix adaptive evolution algorithm; the third intelligent agent is an intelligent agent constructed based on a graph neural network; and the fourth intelligent agent is an intelligent agent constructed based on a large language model.
[0089] In some embodiments, such as Figure 6 As shown, different agents differ in construction and function, each excelling at handling case information of varying complexity. Therefore, S502 above may include the following steps:
[0090] S601, The complexity of identifying case information.
[0091] Complexity is used to characterize the level of conflict between the dose required by the patient and the protection of organs at risk.
[0092] One possible approach is to first determine the complexity based on the extracted tumor's size, shape, location, and distance and relationship to surrounding organs at risk. For example, a large, irregularly shaped tumor located close to vital organs, such as the brainstem or eyes, may mean that administering a sufficient tumor dose would more easily affect these organs, thus increasing the complexity of the case information.
[0093] As another possible approach, the complexity can be determined by considering the patient's past medical history. For example, if a patient has previously received radiotherapy or chemotherapy, their tolerance to further radiotherapy may be reduced, or they may have other chronic diseases such as heart disease or lung disease. This increases the number of factors that need to be considered during treatment, increases the conflict between dosage and organ protection, and also increases the complexity of the case information.
[0094] S602. Identify the target agent from among multiple agents that matches the complexity of the case information.
[0095] One possible implementation is to select a target agent that matches the complexity of the case information based on the first correspondence. The first correspondence represents either a preset correspondence between complexity and an agent, or a pre-defined correspondence between a complexity range and an agent.
[0096] It should be understood that matching the complexity of the target intelligent agent with the case information means that it possesses more suitable algorithms, models, and knowledge reserves to handle the case. For complex cases, it can take into account more influencing factors, thereby generating more accurate and personalized treatment plans and improving treatment outcomes.
[0097] In some embodiments, the complexity, from low to high, includes a first complexity, a second complexity, and a third complexity; the multiple intelligent agents include: a first intelligent agent, a second intelligent agent, and a third intelligent agent; the first intelligent agent is used to process case information with a complexity of the first complexity; the second intelligent agent is used to process case information with a complexity of the second complexity; and the third intelligent agent is used to process case information with a complexity of the third complexity.
[0098] Among them, the first intelligent agent, the second intelligent agent, and the third intelligent agent satisfy at least one of the following relationships:
[0099] The reasoning speed of the first agent is greater than that of the second agent, and the reasoning speed of the second agent is greater than that of the third agent; the reasoning ability of the first agent is weaker than that of the second agent, and the reasoning ability of the second agent is weaker than that of the third agent.
[0100] S603. Generate a treatment plan using case information through the target intelligent agent.
[0101] Therefore, by identifying the complexity of case information, cases can be assigned to the most suitable target agent, giving full play to the strengths of each agent and improving the quality and efficiency of treatment plan generation.
[0102] In some embodiments, different agents may employ different algorithms, models, or knowledge systems. Having them generate candidate treatment plans independently allows for full utilization of this diversity, considering the problem from multiple perspectives and approaches, and increasing the likelihood of finding a better treatment plan. Figure 7 As shown, therefore, the above S502 also includes the following steps:
[0103] S701: Generate corresponding candidate treatment plans by using case information through multiple intelligent agents.
[0104] As one possible implementation, each agent independently generates candidate treatment plans based on its own algorithm and model, as well as the acquired case information. For example, case information is input into four agents, and each agent generates a treatment plan.
[0105] S702. Determine a treatment plan from the candidate treatment plans generated by various agents.
[0106] One possible approach is to first identify multiple indicators that influence the treatment plan, and then evaluate the candidate treatment plans generated by each agent using quantitative metrics. Evaluation indicators include tumor radiation dose, radiation dose to organs at risk, predicted treatment side effects, and treatment costs.
[0107] In one possible implementation, the various evaluation indicators can be comprehensively calculated according to preset weights to obtain a score for each candidate treatment plan, and the treatment plan with the highest score can be used as the final treatment plan.
[0108] Therefore, by integrating candidate plans generated by multiple agents from different perspectives, and determining the treatment plan from multiple candidate treatment plans, the one-sidedness and limitations that may occur with a single agent can be avoided, thereby improving the quality of the treatment plan.
[0109] In some embodiments, such as Figure 8 As shown, the goal of tumor treatment is to effectively control the tumor while minimizing damage to endangered organs and normal tissues. However, for highly complex case information, a treatment plan generated by the agent in a single step may not meet treatment expectations. When the predicted treatment effect does not meet expectations, the treatment plan needs to be optimized. Therefore, the above methods also include:
[0110] S801. Evaluate the treatment plan and determine the predicted treatment effect of the treatment plan.
[0111] The assessment includes dose assessment and / or motion assessment of the multi-leaf grating blades. Dose assessment is used to determine whether the dose required for tumor radiotherapy is greater than the preset target area threshold and whether the dose to organs at risk is less than the preset organ at risk threshold. Motion assessment of the multi-leaf grating blades is used to determine whether the matching degree between the irradiation field formed by the multi-leaf grating blades and the tumor target area is greater than the preset matching degree threshold.
[0112] As one possible approach, the dose distribution received by the tumor target area can be calculated, and the calculated dose required for tumor radiotherapy can be compared with a preset target area threshold. If the dose is greater than the preset target area threshold, it indicates that the treatment plan may achieve the expected effect in tumor control and can effectively kill tumor cells.
[0113] Similarly, the dose distribution received by organs at risk (such as the heart, lungs, liver, spinal cord, and other vital organs) is obtained. These doses are compared with their respective preset thresholds for organs at risk. Only when the doses to organs at risk are all less than the preset thresholds can the treatment plan be considered to provide adequate protection for the organs at risk, reducing damage to vital organ function during treatment and decreasing the probability of complications.
[0114] As one possible implementation method, multi-leaf collimators (MLCs) are important devices for forming conformal radiation fields in tumor treatment. After the treatment plan is determined, the motion trajectory of the MLC blades and the shape of the final radiation field are analyzed and precisely compared with the actual shape and location of the tumor target area to obtain the matching degree. The calculated matching degree is compared with a preset matching degree threshold. If it is greater than the preset matching degree threshold, it indicates that the radiation field formed by the MLC blades can effectively cover the tumor target area while minimizing the irradiation of surrounding normal tissues, so that the radiotherapy dose can be accurately concentrated at the tumor site, improving the precision of treatment.
[0115] S802. If the predicted treatment effect of the treatment plan fails to meet the expected effect, the candidate treatment plans in the candidate treatment plan set shall be optimized multiple times until the preset stopping conditions are met.
[0116] The preset stopping conditions include at least one of the following: the predicted treatment effect of the optimized treatment plan reaches the expected effect; the duration of the optimization process is greater than the preset time threshold.
[0117] As one possible implementation, in each optimization process, the candidate treatment plans in the first treatment plan set are input into the target agent, and the target agent optimizes the candidate treatment plans to obtain the optimized treatment plan. The first treatment plan set is the set of candidate treatment plans remaining after the treatment plan corresponding to the target agent.
[0118] For example, if the treatment plan does not achieve the expected results, the optimization results of different agents are used as initial values and then given to another agent for optimization. This process is repeated iteratively until the expected results are achieved or a preset time threshold is exceeded, at which point optimization stops and is handed over to physical revision (i.e., technical adjustments to the plan parameters or equipment configuration made by a medical physicist based on physical principles and equipment limitations).
[0119] Therefore, the candidate treatment plans gathered through a pool of candidate treatment plans may each have their own advantages and disadvantages in different aspects. By optimizing them, the potential of these treatment plans can be fully explored, making the final treatment plan closer to the ideal treatment plan, thereby improving the quality of the treatment plan.
[0120] The foregoing mainly describes the solutions provided by the embodiments of this application from a methodological perspective. To achieve the above functions, the treatment plan determination device includes hardware structures and / or software modules corresponding to the execution of each function. Those skilled in the art should readily recognize that, based on the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein, this application can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed in hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.
[0121] This application embodiment can, according to the above method, exemplarily divide the treatment plan determination device into functional modules. For example, the treatment plan determination device may include functional modules corresponding to each functional division, or two or more functions may be integrated into one processing module. The integrated module can be implemented in hardware or as a software functional module. It should be noted that the module division in this application embodiment is illustrative and only represents one logical functional division; in actual implementation, there may be other division methods.
[0122] Figure 9 This is a schematic diagram of a treatment plan determination device provided in an embodiment of this application. (Refer to...) Figure 9 The treatment plan determination device 900 includes: an acquisition unit 910 and a processing unit 920.
[0123] Acquisition unit 910 is used to acquire patient medical records.
[0124] The processing unit 920 is used to coordinate the work of multiple intelligent agents based on the patient's medical records to generate a treatment plan for the patient; wherein the multiple intelligent agents have different structures.
[0125] In some embodiments, the various agents differ in at least one of the following: the model structure adopted by the agent, the reasoning ability of the agent, the reasoning speed of the agent, the hardware requirements of the agent, the functions of the agent, and the complexity of the information that the agent supports processing.
[0126] In some embodiments, the multiple intelligent agents include at least one of the following: a first intelligent agent, a second intelligent agent, a third intelligent agent, and a fourth intelligent agent; wherein, the first intelligent agent is an intelligent agent constructed based on a constraint network; the second intelligent agent is an intelligent agent constructed based on a covariance matrix adaptive evolution algorithm; the third intelligent agent is an intelligent agent constructed based on a graph neural network; and the fourth intelligent agent is an intelligent agent constructed based on a large language model.
[0127] In some embodiments, the processing unit 920 is specifically configured to identify the complexity of case information, the complexity being used to characterize the level of conflict between the patient's required dose and protection of organs at risk; determine a target agent from a variety of agents that matches the complexity of the case information; and generate a treatment plan using the case information through the target agent.
[0128] In some embodiments, the complexity, from low to high, includes a first complexity, a second complexity, and a third complexity; the multiple intelligent agents include: a first intelligent agent, a second intelligent agent, and a third intelligent agent; the first intelligent agent is used to process case information with a complexity of the first complexity; the second intelligent agent is used to process case information with a complexity of the second complexity; and the third intelligent agent is used to process case information with a complexity of the third complexity; wherein the first intelligent agent, the second intelligent agent, and the third intelligent agent satisfy at least one of the following relationships: the reasoning speed of the first intelligent agent is greater than the reasoning speed of the second intelligent agent, and the reasoning speed of the second intelligent agent is greater than the reasoning speed of the third intelligent agent; the reasoning ability of the first intelligent agent is weaker than the reasoning ability of the second intelligent agent, and the reasoning ability of the second intelligent agent is weaker than the reasoning ability of the third intelligent agent.
[0129] In some embodiments, the processing unit 920 is specifically configured to generate corresponding candidate treatment plans using case information through multiple agents; and to determine a treatment plan from the candidate treatment plans generated by the multiple agents.
[0130] In some embodiments, the processing unit 920 is further configured to evaluate the treatment plan and determine the predicted treatment effect of the treatment plan; if the predicted treatment effect of the treatment plan does not meet the expected effect, the candidate treatment plan in the candidate treatment plan set is optimized multiple times until a preset stopping condition is met; wherein the preset stopping condition includes at least one of the following: the predicted treatment effect of the optimized treatment plan meets the expected effect; the duration of the optimization process is greater than a preset time threshold.
[0131] In some embodiments, during each optimization process, candidate treatment plans from the first treatment plan set are input into the target agent, and the target agent optimizes the candidate treatment plans to obtain optimized treatment plans. The first treatment plan set is the set of candidate treatment plans remaining in the candidate treatment plan set excluding the treatment plan corresponding to the target agent.
[0132] In some embodiments, the assessment includes: dose assessment and / or motion assessment of the multileaf grating blades, wherein the dose assessment is used to determine whether the dose required for tumor radiotherapy is greater than a preset target area threshold and whether the dose to organs at risk is less than a preset organ at risk threshold, and the motion assessment of the multileaf grating blades is used to determine whether the matching degree between the irradiation field formed by the multileaf grating blades and the tumor target area is greater than a preset matching degree threshold.
[0133] Figure 10 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 10 As shown, the electronic device 1000 includes, but is not limited to, a processor 1001 and a memory 1002.
[0134] The memory 1002 described above is used to store executable instructions of the processor 1001. It is understood that the processor 1001 is configured to execute instructions to implement the treatment plan determination method in the above embodiments.
[0135] It should be noted that those skilled in the art will understand that Figure 10 The electronic device structure shown does not constitute a limitation on the electronic device; the electronic device may include, but is not limited to, other electronic devices. Figure 10 This may indicate more or fewer components, or combinations of certain components, or different component arrangements.
[0136] The processor 1001 is the control center of the electronic device. It connects various parts of the electronic device via various interfaces and lines. By running or executing software programs and / or modules stored in the memory 1002, and by calling data stored in the memory 1002, it performs various functions and processes data, thereby providing overall monitoring of the electronic device. The processor 1001 may include one or more processing units. Optionally, the processor 1001 may integrate an application processor and a modem processor. The application processor mainly handles the operating system, user interface, and applications, while the modem processor mainly handles wireless communication. It is understood that the modem processor may not be integrated into the processor 1001.
[0137] The memory 1002 can be used to store software programs and various data. The memory 1002 may primarily include a program storage area and a data storage area. The program storage area may store the operating system, application programs required by at least one functional module (such as a determination unit, processing unit, etc.), etc. Furthermore, the memory 1002 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device.
[0138] In an exemplary embodiment, a computer-readable storage medium including instructions is also provided, such as a memory 1002 including instructions, which can be executed by a processor 1001 of an electronic device 1000 to implement the methods in the above embodiments.
[0139] In actual implementation, Figure 9 Both the acquisition unit 910 and the processing unit 920 can be derived from... Figure 10 The processor 1001 calls the computer program stored in the memory 1002 to implement the process. The specific execution process can be found in the description of the method section in the previous embodiment, and will not be repeated here.
[0140] Optionally, the computer-readable storage medium may be a non-transitory computer-readable storage medium, such as a read-only memory (ROM), random access memory (RAM), CD-ROM, magnetic tape, floppy disk, and optical data storage device.
[0141] In an exemplary embodiment, this application also provides a computer program product including one or more instructions, which can be executed by the processor 1001 of an electronic device to perform the methods described above.
[0142] It should be noted that when one or more instructions in the computer-readable storage medium or computer program product are executed by the processor of the electronic device, they implement the various processes of the above method embodiments and achieve the same technical effect as the above method. To avoid repetition, they will not be described again here.
[0143] Through the above description of the embodiments, those skilled in the art can clearly understand that, for the sake of convenience and brevity, only the division of the above functional modules is used as an example. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above.
[0144] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another apparatus, or some features may be ignored or not executed. Furthermore, the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.
[0145] The units described as separate components may or may not be physically separate. A component shown as a unit can be one or more physical units; that is, it can be located in one place or distributed in multiple different locations. Some or all of the constituent units can be selected to achieve the purpose of this embodiment, depending on actual needs.
[0146] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.
[0147] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a readable storage medium. Based on this understanding, the technical solution of the embodiments of this application, essentially, or the part that contributes to the prior art, or a complete or partial classification of the technical solution, can be embodied in the form of a software product. This software product is stored in a storage medium and includes several instructions to cause a device (which may be a microcontroller, chip, etc.) or processor to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, ROM, RAM, magnetic disks, or optical disks.
[0148] In the description of the embodiments of this application, specific features, structures, materials or characteristics may be combined in any suitable manner in one or more embodiments or examples.
[0149] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.
Claims
1. A method for determining a treatment plan, characterized in that, The method includes: Obtain the patient's medical records; Based on the patient's medical records, multiple intelligent agents work in coordination to generate a treatment plan for the patient; wherein the multiple intelligent agents have different structures.
2. The method according to claim 1, characterized in that, The various intelligent agents differ in at least one of the following: the model structure adopted by the intelligent agent, the reasoning ability of the intelligent agent, the reasoning speed of the intelligent agent, the hardware requirements of the intelligent agent, the functions of the intelligent agent, and the complexity of the information that the intelligent agent supports processing.
3. The method according to claim 1, characterized in that, The plurality of intelligent agents includes at least one of the following: a first intelligent agent, a second intelligent agent, a third intelligent agent, and a fourth intelligent agent; The first intelligent agent is an intelligent agent constructed based on a constraint network; The second agent is an agent constructed based on the covariance matrix adaptive evolution algorithm; The third intelligent agent is an intelligent agent constructed based on a graph neural network; The fourth intelligent agent is an intelligent agent built based on a large language model.
4. The method according to claim 1, characterized in that, The process of generating a treatment plan for the patient based on the patient's medical information, through the coordinated work of multiple intelligent agents, includes: The complexity of the case information is identified, which characterizes the level of conflict between the patient's required dose and protection of organs at risk. Identify a target agent whose complexity matches that of the case information from among the various agents; The target intelligent agent generates the treatment plan using the case information.
5. The method according to claim 4, characterized in that, The complexity levels, from low to high, include first complexity, second complexity, and third complexity; the multiple intelligent agents include: first intelligent agent, second intelligent agent, and third intelligent agent; The first intelligent agent is used to process case information with a complexity of the first complexity; The second intelligent agent is used to process case information with a complexity of the second complexity. The third intelligent agent is used to process case information with a complexity of the third complexity. Wherein, the first intelligent agent, the second intelligent agent, and the third intelligent agent satisfy at least one of the following relationships: The reasoning speed of the first agent is greater than that of the second agent, and the reasoning speed of the second agent is greater than that of the third agent. The reasoning ability of the first intelligent agent is weaker than that of the second intelligent agent, and the reasoning ability of the second intelligent agent is weaker than that of the third intelligent agent.
6. The method according to claim 1, characterized in that, The process of generating a treatment plan for the patient based on the patient's medical information, through the coordinated work of multiple intelligent agents, includes: The various intelligent agents respectively utilize the case information to generate corresponding candidate treatment plans; The treatment plan is determined from the candidate treatment plans generated by the various intelligent agents.
7. The method according to any one of claims 1 to 6, characterized in that, The method further includes: The treatment plan is evaluated to determine its predicted therapeutic effect; If the predicted treatment effect of the treatment plan fails to achieve the expected effect, the candidate treatment plans in the candidate treatment plan set are optimized multiple times until a preset stopping condition is met; wherein, the preset stopping condition includes at least one of the following: The optimized treatment plan predicted that the treatment effect would meet expectations. The duration of the optimization process is greater than a preset time threshold.
8. The method according to claim 7, characterized in that, In each optimization process, candidate treatment plans from the first treatment plan set are input into the target agent, which optimizes the candidate treatment plans to obtain optimized treatment plans. The first treatment plan set is the set of candidate treatment plans remaining in the candidate treatment plan set excluding the treatment plan corresponding to the target agent.
9. The method according to claim 7, characterized in that, The assessment includes dose assessment and / or motion assessment of the multileaf grating blades. The dose assessment is used to determine whether the dose required for the tumor radiotherapy is greater than a preset target area threshold and whether the dose to the organs at risk is less than a preset organ at risk threshold. The motion assessment of the multileaf grating blades is used to determine whether the matching degree between the irradiation field formed by the multileaf grating blades and the tumor target area is greater than a preset matching degree threshold.
10. A system for determining a treatment plan, characterized in that, The system includes a master control agent and multiple agents for determining treatment plans; The main control agent is configured to: acquire patient medical records; Based on the patient's medical records, the various intelligent agents work in coordination to generate a treatment plan for the patient.
11. The system according to claim 10, characterized in that, The various intelligent agents differ in at least one of the following: the model structure adopted by the intelligent agent, the reasoning ability of the intelligent agent, the reasoning speed of the intelligent agent, the hardware requirements of the intelligent agent, the functions of the intelligent agent, and the complexity of the information that the intelligent agent supports processing.
12. The system according to claim 10, characterized in that, The plurality of intelligent agents includes at least one of the following: a first intelligent agent, a second intelligent agent, a third intelligent agent, and a fourth intelligent agent; The first intelligent agent is an intelligent agent constructed based on a constraint network; The second agent is an agent constructed based on the covariance matrix adaptive evolution algorithm; The third intelligent agent is an intelligent agent constructed based on a graph neural network; The fourth intelligent agent is an intelligent agent built based on a large language model.
13. The system according to claim 10, characterized in that, The multiple intelligent agents may be deployed on one electronic device, or the multiple intelligent agents may be deployed on different electronic devices.
14. An electronic device, characterized in that, The electronic device includes: processor; A memory configured to store processor-executable instructions; The processor is configured to execute the instructions to implement the method as described in any one of claims 1 to 9.
15. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that, when executed on a computer, cause the computer to perform the method as described in any one of claims 1 to 9.
16. A computer program product, characterized in that, The computer program product includes computer instructions that, when executed on an electronic device, cause the electronic device to perform the method as described in any one of claims 1 to 9.
