Iterative generation and feedback-based treatment regimen generation system and method
By introducing a multi-dimensional clinical assessment system and a dynamic early cessation strategy, the reliability and safety of treatment plan generation in existing technologies have been addressed, resulting in high-quality treatment plan generation and improved alignment with clinical standards and computational efficiency.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- TSINGHUA UNIVERSITY
- Filing Date
- 2026-03-23
- Publication Date
- 2026-06-09
AI Technical Summary
In existing technologies, the iterative optimization mechanism for generating treatment plans based on large language models is imperfect, the evaluation system is not sufficiently aligned with clinical standards, and there is a lack of guidance mechanisms based on successful experiences, resulting in insufficient reliability and safety of the generated treatment plans.
The system introduces a multi-dimensional clinical assessment system, a score-aware memory retrieval mechanism, and a dynamic early cessation strategy. The planner generates treatment plans, the evaluator performs multi-dimensional clinical assessments, and stores the assessment results in the memory to guide the next iteration until the preset conditions are met to output the highest-scoring plan.
It enables the accurate, comprehensive, and safe generation of treatment plans, significantly improving the quality of the plans, enhancing consistency with the judgment of clinical experts, and demonstrating high efficiency in the control of computational resources.
Smart Images
Figure CN122177418A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of medical decision support technology, and in particular to a treatment plan generation system and method based on iterative generation and feedback. Background Technology
[0002] In the field of clinical medical decision support, the automated generation of treatment plans is a key technological step in transforming diagnostic conclusions into specific, personalized treatment plans. In recent years, with the advancement of large language model technology, it has shown potential in medical text understanding and generation tasks, leading several studies to attempt to apply it to assist in the generation of treatment plans.
[0003] Among related technologies, the technical solutions for generating auxiliary treatment plans based on large language models mainly have the following shortcomings: the iterative optimization mechanism is imperfect, and the evaluation system is not aligned enough with clinical standards, making it difficult to comprehensively measure the quality of treatment plans on multidimensional clinical standards; it is also unable to simulate the cognitive process of clinicians learning from successful cases.
[0004] Therefore, there is an urgent need for a treatment plan generation scheme to improve the reliability and security of auxiliary instruction scheme generation. Summary of the Invention
[0005] The purpose of this application is to provide a treatment plan generation system and method based on iterative generation and feedback. By introducing a multi-dimensional clinical assessment system, a score-aware memory retrieval mechanism, and a dynamic early cessation strategy, the system achieves accurate, comprehensive, and safe generation of treatment plans.
[0006] Firstly, this application provides a treatment plan generation method based on iterative generation and feedback, comprising: The system comprises a planner, an evaluator, and a memory. The planner generates a corresponding treatment plan in each iteration based on a given patient case and historical information retrieved from the memory. The evaluator performs multi-dimensional clinical evaluations of the treatment plans generated by the planner in each iteration and outputs corresponding evaluation scores and reasons. The memory stores the treatment plans generated in each iteration, their corresponding evaluation scores, and evaluation reasons to guide the planner in the next iteration. When a preset stopping condition is met, the system selects the treatment plan with the highest evaluation score from the treatment plans generated over multiple iterations as the final treatment plan for output.
[0007] Optionally, the preset stopping conditions include: reaching the maximum number of iterations, or the evaluation scores of multiple consecutive iterations exceeding a preset threshold; the patient case includes at least one of the following: basic clinical information, symptoms and clinical findings, and a confirmed diagnosis.
[0008] Optionally, the planner, when retrieving historical information from the memory, selectively retrieves information based on the evaluation score, prioritizing the return of treatment plans with higher historical scores and corresponding evaluation reasons.
[0009] Optionally, the evaluator is specifically used to obtain relevant clinical evidence from an external medical knowledge base through retrieval enhancement, and / or to provide high-quality treatment plan references through small sample examples to stabilize the evaluation behavior; the evaluator is also specifically used to perform multi-dimensional clinical evaluation of the treatment plan generated by the planner in each iteration based on the relevant clinical evidence, and / or the small sample examples, using a multi-dimensional scoring system, and output the corresponding evaluation score and evaluation reason.
[0010] Optionally, the multi-dimensional scoring system includes at least one of the following dimensions: scientific consensus compliance, plan completeness, contextual relevance, evidence-based consistency, harmfulness, information accuracy, and medical content bias.
[0011] Secondly, this application also provides a treatment plan generation device based on iterative generation and feedback, comprising: In each iteration, the planner generates a treatment plan based on the given patient case and historical information retrieved from the memory. The evaluator performs a multi-dimensional clinical evaluation of the acquired treatment plan and outputs the corresponding evaluation score and evaluation reason. The treatment plan, corresponding evaluation score and evaluation reason generated in each iteration are stored in the memory to guide the planner in the next iteration. When the preset stopping condition is met, the plan with the highest evaluation score is selected from the treatment plans generated in multiple iterations as the final treatment plan output.
[0012] Optionally, the treatment plan generated by the planner based on a given patient case and historical information retrieved from the memory includes: when the planner retrieves historical information from the memory, it selectively retrieves information based on the evaluation score, prioritizing the return of treatment plans with higher historical scores and corresponding evaluation reasons.
[0013] Optionally, the step of using an evaluator to perform multi-dimensional clinical evaluation of the acquired treatment plan and outputting corresponding evaluation scores and evaluation reasons includes: obtaining relevant clinical evidence from an external medical knowledge base through enhanced retrieval, and / or providing high-quality treatment plan references through small sample examples to stabilize evaluation behavior; based on the relevant clinical evidence, and / or the small sample examples, using a multi-dimensional scoring system to perform multi-dimensional clinical evaluation of the treatment plan generated by the planner in each iteration, and outputting corresponding evaluation scores and evaluation reasons.
[0014] Thirdly, this application also provides a treatment plan generation device based on iterative generation and feedback, comprising: The treatment plan generation module is used to obtain the treatment plan generated by the planner based on the given patient case and historical information retrieved from the memory during each iteration. The treatment plan evaluation module is used to perform multi-dimensional clinical evaluation of the obtained treatment plan using the evaluator and output the corresponding evaluation score and evaluation reason. The information storage module is used to store the treatment plan generated in each iteration, the corresponding evaluation score, and the evaluation reason in the memory to guide the planner in the next iteration. The treatment plan output module is used to select the treatment plan with the highest evaluation score from the treatment plans generated in multiple iterations as the final treatment plan output when the preset stopping condition is met.
[0015] Optionally, the plan generation module is specifically used to selectively retrieve historical information from the memory based on the evaluation score, prioritizing the return of treatment plans with higher historical scores and corresponding evaluation reasons.
[0016] Optionally, the treatment plan evaluation module is specifically used to obtain relevant clinical evidence from an external medical knowledge base through enhanced retrieval, and / or to provide high-quality treatment plan references through small sample examples to stabilize evaluation behavior; the treatment plan evaluation module is also specifically used to perform multi-dimensional clinical evaluation of the treatment plan generated by the planner in each iteration based on the relevant clinical evidence, and / or the small sample examples, using a multi-dimensional scoring system, and output the corresponding evaluation score and evaluation reasons.
[0017] This application also provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of the treatment plan generation method based on iterative generation and feedback as described above.
[0018] This application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the treatment plan generation method based on iterative generation and feedback as described above.
[0019] This application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the treatment plan generation method based on iterative generation and feedback as described above.
[0020] This application provides a treatment plan generation system and method based on iterative generation and feedback, comprising: a planner, an evaluator, and a memory; the planner, in each iteration, generates a corresponding treatment plan based on a given patient case and historical information retrieved from the memory; the evaluator performs multi-dimensional clinical evaluation of the treatment plans generated by the planner in each iteration and outputs corresponding evaluation scores and evaluation reasons; the memory stores the treatment plans generated in each iteration, the corresponding evaluation scores, and the evaluation reasons to guide the planner in the next iteration; wherein, when a preset stopping condition is met, the system selects the plan with the highest evaluation score from the treatment plans generated in multiple iterations as the final treatment plan output. Thus, by introducing a multi-dimensional clinical evaluation system, a score-aware memory retrieval mechanism, and a dynamic early stopping strategy, accurate, comprehensive, and safe generation of treatment plans is achieved. Attached Figure Description
[0021] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0022] Figure 1 This is a schematic diagram of the architecture of the treatment plan generation system based on iterative generation and feedback provided in this application; Figure 2 This is a schematic diagram of the iterative process of the treatment plan generation system based on iterative generation and feedback provided in this application; Figure 3 This is a flowchart illustrating the treatment plan generation method based on iterative generation and feedback provided in this application. Figure 4 This is a schematic diagram of the treatment plan generation device based on iterative generation and feedback provided in this application; Figure 5 This is a schematic diagram of the structure of the electronic device provided in this application. Detailed Implementation
[0023] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0024] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class and the number of objects is not limited; for example, a first object can be one or more. Furthermore, "and / or" in the specification and claims indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship. All actions involving the acquisition of signal information or data in this application are performed in accordance with the relevant data protection laws and policies of the country where the application is located and with authorization from the owner of the relevant device.
[0025] In related technologies, the technical solutions for generating treatment plans based on large language models mainly include the following categories: The first type directly uses a general-purpose large language model or a model fine-tuned with a medical corpus to generate a treatment plan in one go based on the input patient information. This type of method is essentially end-to-end text generation, lacking explicit verification and iterative optimization mechanisms for the generated results, and is prone to producing outputs with incomplete information, clinical inaccuracies, or security risks.
[0026] The second category involves introducing Retrieval Enhanced Generation (RAG) or multi-stage processing workflows on top of single-generation to improve the relevance and structure of the solution. For example, iFlytek Medical's patent application (CN202511326908.6) standardizes the terminology of the preliminary treatment plan and makes adjustments by combining clinical pathway knowledge graphs and patient profiles. Although this method introduces a correction step, its optimization process depends on the quality of the external knowledge graph construction, and the correction mechanism is mainly based on rule-based medical contradiction detection, and has not yet formed a standardized multi-dimensional quality assessment and iterative feedback loop.
[0027] The third category consists of iterative medical content generation frameworks based on intelligent agent architecture that have emerged in recent years, such as FRAME (Feedback-Refined Agent Methodology) and MeWM (Medical World Model). The FRAME framework builds a three-party intelligent agent architecture of "generation-evaluation-reflection" to improve content quality through an indicator-driven feedback loop; MeWM constructs an automated optimization loop of "solution generation-simulation deduction-survival assessment".
[0028] However, the above-mentioned technical solutions in the related technologies still have the following drawbacks: Defect 1: Imperfect iterative optimization mechanism Current technologies have not yet established a truly meaningful "generation-evaluation-memory-optimization" closed loop. The FRAME framework's evaluation is mainly driven by statistical indicators and lacks a professional, multi-dimensional evaluation system for treatment plans; MeWM's evaluation dimensions are limited to a single survival risk indicator, making it difficult to comprehensively measure the quality of treatment plans across multiple clinical standards; and iFlytek's patented correction mechanism is relatively rule-based and lacks the ability to learn and guide based on historical success experiences.
[0029] Deficiency 2: Insufficient alignment between the assessment system and clinical standards Existing assessment methods are difficult to align with the judgment standards of clinical experts. Traditional lexical indicators (such as BLEU) cannot measure the medical rationality of treatment plans; the general LLM score lacks alignment and validation with professional clinical standards. Although FRAME proposes combining statistical indicators with artificial benchmarks, it has not yet formed a quantifiable, structured scoring system that conforms to multidimensional clinical standards; MeWM survival risk assessment is only applicable to cancer treatment scenarios and lacks universality.
[0030] Deficiency 3: Lack of guidance mechanisms based on successful experiences Existing technologies often employ independent, "memoryless" iterations during their iterative optimization processes, failing to effectively utilize successful experiences from historical generation. iFlytek's knowledge graph retrieval is based on semantic relevance rather than the quality of historical solutions; neither FRAME nor MeWM mentions a mechanism for selective memory retrieval based on historical scores, making it difficult to simulate the cognitive process of clinicians drawing upon successful cases.
[0031] Defect 4: Lack of a mechanism to balance iterative efficiency and quality Existing technologies mostly involve fixed-round iterative generation or lack a clear stopping mechanism. The FRAME framework does not have an early stopping mechanism based on quality targets; the exploration process of MeWM is mainly based on risk value comparison, and no general quality threshold early stopping scheme has been formed, which may lead to waste of computing resources or termination before the quality targets are met.
[0032] To address the aforementioned technical problems in related technologies, this application provides a treatment plan generation system and method based on iterative generation and feedback. By introducing a multi-dimensional clinical evaluation system, a score-aware memory retrieval mechanism, and a dynamic early cessation strategy, a closed loop of "generation-evaluation-memory-optimization" of treatment plans is achieved, generating more accurate, comprehensive, and safe personalized treatment plans, while effectively controlling computational overhead while ensuring output quality.
[0033] The treatment plan generation method based on iterative generation and feedback provided in this application will be described in detail below with reference to the accompanying drawings, through specific embodiments and application scenarios.
[0034] like Figure 1As shown in the embodiment of this application, a treatment plan generation system based on iterative generation and feedback is provided. The system includes a planner, an evaluator, and a memory unit.
[0035] The planner is used to generate a corresponding treatment plan based on a given patient case and historical information retrieved from the memory during each iteration. The evaluator is used to perform multi-dimensional clinical evaluation on the treatment plan generated by the planner in each iteration and output the corresponding evaluation score and evaluation reason. The memory is used to store the treatment plan generated in each iteration, the corresponding evaluation score, and the evaluation reason to guide the planner in the next iteration.
[0036] Specifically, when a preset stopping condition is met, the system selects the treatment plan with the highest evaluation score from the treatment plans generated through multiple iterations as the final treatment plan output. The preset stopping conditions include: reaching the maximum number of iterations, or the evaluation scores of multiple consecutive iterations exceeding a preset threshold; the patient case includes at least one of the following: basic clinical information, symptoms and clinical findings, and a confirmed diagnosis.
[0037] For example, the treatment plan generation system based on iterative reasoning and feedback provided in this application is centered on an intelligent agent workflow comprising three components: a planner, an evaluator (TheraJudge), and a memorizer, achieving a closed loop of "generation-evaluation-reflection-optimization". Given a patient case P=(d,s,y) (where d is basic clinical information, s is symptoms and clinical findings, and y is the confirmed diagnosis), the system goal is to generate a high-quality treatment plan T.
[0038] like Figure 2 The diagram illustrates the iterative workflow of the system. In each iteration k: First, the planner combines the patient case P with historical information retrieved from the memory (e.g., memories stored in the previous round). Generate treatment plan Next, the evaluator (TheraJudge) evaluates... Perform multidimensional clinical assessments and generate scores. Then, the current plan score and the reasons for the assessment As a memory of this round The data is stored in the memorizer. After the memorizer is updated, its contents will be used to guide the planner in the next (k+1) iterations, generating an improved plan. This process repeats until a stopping condition is met (such as reaching the maximum number of iterations or triggering an early stopping mechanism). Finally, the system selects the highest-scoring plan from the last few iterations. As output.
[0039] Specifically, the planner is used to selectively retrieve historical information from the memory based on the evaluation score, prioritizing the return of treatment plans with higher historical scores and corresponding evaluation reasons.
[0040] Specifically, the evaluator is used to obtain relevant clinical evidence from an external medical knowledge base through retrieval enhancement, and / or to provide high-quality treatment plan references through small sample examples to stabilize the evaluation behavior; the evaluator is also used to perform multi-dimensional clinical evaluation of the treatment plan generated by the planner in each iteration based on the relevant clinical evidence, and / or the small sample examples, using a multi-dimensional scoring system, and output the corresponding evaluation score and evaluation reason.
[0041] The multi-dimensional scoring system includes at least one of the following dimensions: scientific consensus compliance, plan completeness, contextual relevance, evidence-based consistency, harmfulness, information accuracy, and medical content bias.
[0042] For example, the system workflow includes: 1. Initialization: The Memorizer is empty. The Planner receives patient case P.
[0043] 2. Iterative generation and optimization (performing k=1,2,...,N rounds of iteration): 2.1 Planning: The planner generates the treatment plan for the current iteration based on the current patient case P and the Top-N relevant historical plans retrieved from memory, along with their evaluation feedback (constituting contextual information). .
[0044] 2.2 Evaluation (Judge): The evaluator (TheraJudge) evaluates... Perform multidimensional clinical assessments. The assessor integrates the following optional components to provide accurate scoring, specifically including: 2.2.1 Retrieval Enhancement Generation (RAG): Retrieve relevant paragraphs from external medical knowledge bases (such as clinical guidelines) as the basis for evaluation.
[0045] 2.2.2 Few-shots: Provide a small number of high-quality treatment options and scoring examples to stabilize the assessment of behavior.
[0046] 2.2.3 Multi-dimensional scoring system: The plan is quantitatively scored (e.g., 0-100 points) from seven dimensions: scientific consensus compliance, plan completeness, contextual relevance, evidence-based consistency, harmfulness, information accuracy, and medical content bias, and an overall score is calculated. .
[0047] 2.3 Memory & Reflection: This involves reviewing the currently generated treatment plan. Assessment score and the reasons for the assessment The entries are stored in memory. These memory entries will be retrieved by the planner as context in subsequent iterations, guiding it to generate a better plan.
[0048] 3. Output: After multiple iterations, the system selects the plan with the highest evaluation score from the plans generated in the last L iterations. As the final output. To improve efficiency, the system can be configured with an early stopping mechanism: if the score of three consecutive iterations exceeds a preset threshold τ, the iteration is terminated early.
[0049] It should be noted that the treatment plan generation system based on iterative generation and feedback provided in this application has the following characteristics: 1. Iterative Generation-Evaluation-Optimization Framework: The generation of treatment plans is defined as an iterative workflow that includes a planner, a judge, and a memorizer, and the plan is self-improved through multiple rounds of iteration.
[0050] 2. Domain-Specific Assessment Module (TheraJudge): A module specifically designed for assessing treatment plans. It generates assessment feedback that is highly consistent with expert judgment by combining search enhancement (RAG), small sample examples, and multi-dimensional clinical standard scores.
[0051] 3. Multi-dimensional quantitative evaluation method for treatment plans: Define and adopt a scoring system that includes multiple dimensions such as scientific consensus compliance, plan integrity, situational relevance, and hazard control to conduct a refined and structured quantitative evaluation of treatment plans.
[0052] 4. Memory and guidance mechanism based on evaluation feedback: The system stores the treatment plans generated in each iteration and their multidimensional evaluation scores in the memory, and selectively retrieves historical information based on these scores in subsequent iterations, thereby guiding the planner to generate better plans in this context.
[0053] 5. Dynamic early stop output strategy: The system monitors the changes in the evaluation score during the iteration process. When the score continuously reaches or exceeds the preset threshold, the iteration is automatically stopped, and the plan with the highest score in the most recent iterations is selected as the final output.
[0054] Compared with existing technical solutions, the treatment plan generation system based on iterative generation and feedback provided in this application has the following beneficial effects: 1. Significantly Improved Treatment Plan Quality: By introducing an iterative optimization mechanism, the system can progressively correct errors in the initial draft, supplement missing information, and optimize safety measures, thereby generating treatment plans that are significantly superior to single-generation models in terms of accuracy, completeness, and safety. In the HealthBench benchmark test, this invention outperformed the second-place model by 0.88 points in "accuracy" and 3.33 points in "completeness," two key dimensions.
[0055] 2. Clinical Applicability and Superiority: In blinded evaluations conducted by licensed medical experts, the treatment plans generated by this invention achieved an 86% success rate compared to plans written by human doctors. It is particularly superior in terms of specificity (69% success rate) and completeness (71% success rate), demonstrating its ability to generate more personalized and comprehensive plans.
[0056] 3. Provides reliable self-optimization signals: The evaluation results of TheraJudge, the core component of this invention, show a high degree of consistency with the official evaluation of the authoritative benchmark HealthBench (Pearson correlation coefficient 0.71), far superior to traditional lexical indicators and general LLM scores. This proves that TheraJudge can provide effective, clinically compliant feedback signals, driving the system to make correct self-improvements, rather than introducing noise.
[0057] 4. Enhanced Model Robustness and Generalization: This framework, acting as a "plugin," can adapt to and improve the performance of various LLM platforms (such as GPT-4o, DeepSeek-R1, etc.) in treatment plan generation tasks. In tests across different medical specialties (endocrinology, digestive, neurology, and respiratory), it delivers an average performance improvement of 8.6% to 14.6%.
[0058] 5. Efficiency and quality balance: Through score-aware memory retrieval and early stopping mechanism, the system can effectively control unnecessary computational overhead while ensuring high-quality output plans, thus improving the practicality of the method.
[0059] It should be noted that, in the embodiments of this application, the three components of the evaluator (TheraJudge)—RAG, Few-shots, and multi-dimensional scoring—can be enabled or disabled according to actual resources and needs. The planner and evaluator can be based on the same large language model backbone (such as DeepSeek-R1), or different models can be used to perform these functions respectively. The specific dimensions of the multi-dimensional scoring can be adjusted and reduced according to the characteristics of different clinical departments.
[0060] The treatment plan generation system based on iterative generation and feedback provided in this application includes: a planner, an evaluator, and a memory. The planner generates a corresponding treatment plan in each iteration based on a given patient case and historical information retrieved from the memory. The evaluator performs multi-dimensional clinical evaluation on the treatment plans generated by the planner in each iteration and outputs corresponding evaluation scores and evaluation reasons. The memory stores the treatment plans generated in each iteration, their corresponding evaluation scores, and evaluation reasons to guide the planner in the next iteration. When a preset stopping condition is met, the system selects the treatment plan with the highest evaluation score from the treatment plans generated in multiple iterations as the final treatment plan output. Thus, by introducing a multi-dimensional clinical evaluation system, a score-aware memory retrieval mechanism, and a dynamic early stopping strategy, accurate, comprehensive, and safe generation of treatment plans is achieved.
[0061] The treatment plan generation method based on iterative generation and feedback provided in this application will be described in detail below with reference to the accompanying drawings, through specific embodiments and application scenarios.
[0062] like Figure 3 As shown in the embodiment of this application, a treatment plan generation method based on iterative generation and feedback is provided. This method may include the following steps 301 to 303: Step 301: In each iteration, obtain the treatment plan generated by the planner based on the given patient case and historical information retrieved from the memory, and use the evaluator to perform multi-dimensional clinical evaluation of the obtained treatment plan, outputting the corresponding evaluation score and evaluation reason.
[0063] Step 302: Store the treatment plan, corresponding evaluation score, and evaluation reason generated in each iteration into the memory to guide the planner of the next iteration.
[0064] Step 303: When the preset stopping condition is met, select the treatment plan with the highest evaluation score from the treatment plans generated in multiple iterations as the final treatment plan output.
[0065] The preset stopping conditions include: reaching the maximum number of iterations, or the evaluation scores of multiple consecutive iterations exceeding a preset threshold; the patient case includes at least one of the following: basic clinical information, symptoms and clinical findings, and confirmed diagnosis.
[0066] Specifically, step 301 above, the step of obtaining the treatment plan generated by the planner based on the given patient case and historical information retrieved from the memory, may further include the following step 301a: Step 301a: When the planner retrieves historical information from the memory, it performs selective retrieval based on the evaluation score, prioritizing the return of treatment plans with higher historical scores and corresponding evaluation reasons.
[0067] Specifically, step 301 above, which involves using an evaluator to perform a multi-dimensional clinical evaluation of the acquired treatment plan and outputting the corresponding evaluation score and evaluation reason, may further include the following steps 301b1 and 301b2: Step 301b1: Enhance the generation of relevant clinical evidence from external medical knowledge bases through retrieval, and / or provide high-quality treatment options for stable assessment behavior through small sample examples.
[0068] Step 301b2: Based on the relevant clinical evidence and / or the small sample example, use a multi-dimensional scoring system to perform a multi-dimensional clinical evaluation of the treatment plan generated by the planner in each iteration, and output the corresponding evaluation score and evaluation reason.
[0069] It should be noted that each step in the above-mentioned treatment plan generation method based on iterative generation and feedback has been described in detail in the embodiments of the treatment plan generation system based on iterative generation and feedback. To avoid redundancy, it will not be repeated here.
[0070] The treatment plan generation method based on iterative generation and feedback provided in this application first obtains a given patient case. Then, in each iteration, a treatment plan generated by a planner based on the given patient case and historical information retrieved from a memory is obtained. An evaluator performs a multi-dimensional clinical evaluation of the obtained treatment plan, outputting the corresponding evaluation score and evaluation reason. The treatment plan generated in each iteration, the corresponding evaluation score, and the evaluation reason are stored in the memory to guide the planner in the next iteration. Finally, when a preset stopping condition is met, the treatment plan with the highest evaluation score from the treatment plans generated in multiple iterations is selected as the final treatment plan. Thus, by introducing a multi-dimensional clinical evaluation system, a score-aware memory retrieval mechanism, and a dynamic early stopping strategy, accurate, comprehensive, and safe generation of treatment plans is achieved.
[0071] It should be noted that the treatment plan generation method based on iterative generation and feedback provided in this application embodiment can be executed by a treatment plan generation device based on iterative generation and feedback, or by a control module within that device for executing the treatment plan generation method based on iterative generation and feedback. This application embodiment uses the execution of the treatment plan generation method based on iterative generation and feedback by a treatment plan generation device based on iterative generation and feedback as an example to illustrate the treatment plan generation device based on iterative generation and feedback provided in this application embodiment.
[0072] It should be noted that, in the embodiments of this application, the treatment plan generation methods based on iterative generation and feedback shown in the accompanying drawings are all illustrated using one accompanying drawing from an embodiment of this application as an example. In specific implementation, the treatment plan generation methods based on iterative generation and feedback shown in the accompanying drawings of the above methods can also be implemented in conjunction with any other accompanying drawings illustrated in the above embodiments, which will not be elaborated here.
[0073] The treatment plan generation apparatus based on iterative generation and feedback provided in this application is described below. The treatment plan generation method based on iterative generation and feedback described below can be referred to in correspondence with the treatment plan generation method described above.
[0074] Figure 4 This is a schematic diagram of the treatment plan generation device based on iterative generation and feedback provided in the embodiments of this application, as shown below. Figure 4 As shown, it specifically includes: The treatment plan generation module 401 is used to obtain the treatment plan generated by the planner based on the given patient case and historical information retrieved from the memory during each iteration. The treatment plan evaluation module 402 is used to perform multi-dimensional clinical evaluation of the obtained treatment plan using the evaluator and output the corresponding evaluation score and evaluation reason. The information storage module 403 is used to store the treatment plan generated in each iteration, the corresponding evaluation score, and the evaluation reason in the memory to guide the planner in the next iteration. The treatment plan output module 404 is used to select the treatment plan with the highest evaluation score from the treatment plans generated in multiple iterations as the final treatment plan output when the preset stopping condition is met.
[0075] Optionally, the scheme generation module 401 is specifically used to selectively retrieve historical information from the memory based on the evaluation score when the planner retrieves historical information, and to prioritize returning treatment schemes with higher historical scores and corresponding evaluation reasons.
[0076] Optionally, the scheme evaluation module 402 is specifically used to obtain relevant clinical evidence from an external medical knowledge base through enhanced retrieval, and / or to provide high-quality treatment scheme references through small sample examples to stabilize evaluation behavior; the scheme evaluation module 402 is also specifically used to perform multi-dimensional clinical evaluation of the treatment schemes generated by the planner in each iteration based on the relevant clinical evidence, and / or the small sample examples, using a multi-dimensional scoring system, and output the corresponding evaluation score and evaluation reasons.
[0077] The treatment plan generation device based on iterative generation and feedback provided in this application first acquires a given patient case. Then, in each iteration, it acquires the treatment plan generated by the planner based on the given patient case and historical information retrieved from memory. An evaluator performs a multi-dimensional clinical evaluation of the acquired treatment plan, outputting the corresponding evaluation score and evaluation reason. The treatment plan generated in each iteration, the corresponding evaluation score, and the evaluation reason are stored in memory to guide the planner in the next iteration. Finally, when a preset stopping condition is met, the plan with the highest evaluation score from the treatment plans generated in multiple iterations is selected as the final treatment plan output. Thus, by introducing a multi-dimensional clinical evaluation system, a score-aware memory retrieval mechanism, and a dynamic early stopping strategy, accurate, comprehensive, and safe generation of treatment plans is achieved.
[0078] Figure 5 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 5 As shown, the electronic device may include a processor 510, a communications interface 520, a memory 530, and a communication bus 540, wherein the processor 510, communications interface 520, and memory 530 communicate with each other via the communication bus 540. The processor 510 can call logical instructions in the memory 530 to execute a treatment plan generation method based on iterative generation and feedback. This method includes: first, acquiring a given patient case; then, in each iteration, acquiring the treatment plan generated by the planner based on the given patient case and historical information retrieved from the memory, and using an evaluator to perform a multi-dimensional clinical evaluation of the acquired treatment plan, outputting the corresponding evaluation score and evaluation reasons; storing the treatment plan generated in each iteration, the corresponding evaluation score, and the evaluation reasons in the memory to guide the planner in the next iteration; finally, when a preset stopping condition is met, selecting the plan with the highest evaluation score from the treatment plans generated in multiple iterations as the final treatment plan output. Thus, by introducing a multi-dimensional clinical assessment system, a score-aware memory retrieval mechanism, and a dynamic early cessation strategy, the treatment plan can be generated accurately, comprehensively, and safely.
[0079] Furthermore, the logical instructions in the aforementioned memory 530 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in 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, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0080] On the other hand, this application also provides a computer program product, which includes a computer program stored on a computer-readable storage medium. The computer program includes program instructions, and when the program instructions are executed by the computer, the computer can execute the treatment plan generation method based on iterative generation and feedback provided by the above methods. The method includes: first, obtaining a given patient case; then, in each iteration, obtaining the treatment plan generated by the planner based on the given patient case and historical information retrieved from the memory, and using an evaluator to perform multi-dimensional clinical evaluation on the obtained treatment plan, outputting the corresponding evaluation score and evaluation reasons; storing the treatment plan generated in each iteration, the corresponding evaluation score, and the evaluation reasons in the memory to guide the planner in the next iteration; finally, when a preset stopping condition is met, selecting the plan with the highest evaluation score from the treatment plans generated in multiple iterations as the final treatment plan output. Thus, by introducing a multi-dimensional clinical evaluation system, a score-aware memory retrieval mechanism, and a dynamic early stopping strategy, accurate, comprehensive, and safe generation of treatment plans is achieved.
[0081] Furthermore, this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, performs the aforementioned treatment plan generation methods based on iterative generation and feedback. This method includes: first, acquiring a given patient case; then, in each iteration, acquiring a treatment plan generated by a planner based on the given patient case and historical information retrieved from a memory, and using an evaluator to perform a multi-dimensional clinical evaluation of the acquired treatment plan, outputting corresponding evaluation scores and evaluation reasons; storing the treatment plan generated in each iteration, the corresponding evaluation score, and the evaluation reasons in the memory to guide the planner in the next iteration; finally, when a preset stopping condition is met, selecting the plan with the highest evaluation score from the treatment plans generated in multiple iterations as the final treatment plan output. Thus, by introducing a multi-dimensional clinical evaluation system, a score-aware memory retrieval mechanism, and a dynamic early stopping strategy, accurate, comprehensive, and safe generation of treatment plans is achieved.
[0082] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.
[0083] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.
[0084] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.
Claims
1. A treatment plan generation system based on iterative generation and feedback, characterized in that, include: Planner, evaluator, and memory; The planner is used to generate a corresponding treatment plan in each iteration based on a given patient case and historical information retrieved from the memory. The evaluator is used to perform multi-dimensional clinical evaluation of the treatment plan generated by the planner in each iteration, and output the corresponding evaluation score and evaluation reason; The memory is used to store the treatment plan, corresponding evaluation score, and evaluation reason generated in each iteration, so as to guide the planner of the next iteration; When a preset stopping condition is met, the system selects the treatment plan with the highest evaluation score from the treatment plans generated in multiple iterations as the final treatment plan output.
2. The system according to claim 1, characterized in that, The preset stopping conditions include: reaching the maximum number of iterations, or the evaluation scores of multiple consecutive iterations exceeding a preset threshold; the patient case includes at least one of the following: basic clinical information, symptoms and clinical findings, and confirmed diagnosis.
3. The system according to claim 1, characterized in that, The planner, specifically used to retrieve historical information from the memory, selectively retrieves information based on the evaluation score, prioritizing the return of treatment plans with higher historical scores and corresponding evaluation reasons.
4. The system according to claim 1, characterized in that, The evaluator is specifically used to obtain relevant clinical evidence from external medical knowledge bases through retrieval enhancement, and / or to provide high-quality treatment plan references through small sample examples to stabilize evaluation behavior; The evaluator is further configured to perform multi-dimensional clinical evaluation of the treatment plan generated by the planner in each iteration based on the relevant clinical evidence and / or the small sample examples, using a multi-dimensional scoring system, and output the corresponding evaluation score and evaluation reason.
5. The system according to claim 4, characterized in that, The multi-dimensional scoring system includes at least one of the following dimensions: scientific consensus compliance, plan completeness, contextual relevance, evidence-based consistency, harmfulness, information accuracy, and medical content bias.
6. A treatment plan generation method based on iterative generation and feedback, characterized in that, The method, applied to the treatment plan generation system based on iterative generation and feedback as described in any one of claims 1 to 5, comprises: In each iteration, the planner generates a treatment plan based on the given patient case and historical information retrieved from the memory, and the evaluator performs a multi-dimensional clinical evaluation of the obtained treatment plan, outputting the corresponding evaluation score and evaluation reason. The treatment plan, corresponding evaluation score, and evaluation reason generated in each iteration are stored in the memory to guide the planner of the next iteration. When the preset stopping condition is met, the treatment plan with the highest evaluation score is selected from the treatment plans generated in multiple iterations as the final treatment plan output.
7. The method according to claim 6, characterized in that, The preset stopping conditions include: reaching the maximum number of iterations, or the evaluation scores of multiple consecutive iterations exceeding a preset threshold; the patient case includes at least one of the following: basic clinical information, symptoms and clinical findings, and confirmed diagnosis.
8. The method according to claim 6, characterized in that, The treatment plan generated by the acquisition planner based on a given patient case and historical information retrieved from the memory includes: When the planner retrieves historical information from the memory, it performs selective retrieval based on the evaluation score, prioritizing the return of treatment plans with higher historical scores and corresponding evaluation reasons.
9. The method according to claim 6, characterized in that, The evaluation system utilizes an evaluator to perform a multi-dimensional clinical assessment of the acquired treatment plan, outputting corresponding evaluation scores and evaluation rationale, including: By enhancing the generation of relevant clinical evidence from external medical knowledge bases through retrieval, and / or by providing high-quality treatment options through small sample examples to stabilize the assessment behavior; Based on the relevant clinical evidence and / or the small sample examples, a multi-dimensional scoring system is used to perform multi-dimensional clinical evaluation of the treatment plan generated by the planner in each iteration, and the corresponding evaluation score and evaluation reason are output.
10. A treatment plan generation device based on iterative generation and feedback, characterized in that, The device includes: The treatment plan generation module is used to obtain the treatment plan generated by the planner based on the given patient case and historical information retrieved from the memory during each iteration. The treatment plan evaluation module is used to perform multi-dimensional clinical evaluation of the acquired treatment plan using an evaluator, and output the corresponding evaluation score and evaluation reason; The information storage module is used to store the treatment plan, corresponding evaluation score, and evaluation reason generated in each iteration into the memory to guide the planner of the next iteration. The treatment plan output module is used to select the treatment plan with the highest evaluation score from the treatment plans generated in multiple iterations as the final treatment plan output when the preset stopping conditions are met.