A Method and System for Generating Multidisciplinary Diagnosis and Treatment Recommendations for Lung Cancer Based on a Large Language Model

Abstract

This disclosure provides a method and system for generating multidisciplinary treatment recommendations for lung cancer based on a large language model, relating to the field of medical artificial intelligence technology. Specifically, the method for generating multidisciplinary treatment recommendations for lung cancer based on a large language model includes: acquiring the medical record information of a lung cancer patient; using a diagnostic layer agent to determine diagnostic information based on the medical record information, the diagnostic information including at least one of staging information, pathological information, and metabolic assessment information; using a treatment layer agent to generate preliminary treatment recommendations based on the diagnostic information, the preliminary treatment recommendations including local treatment recommendations and / or systemic treatment recommendations; and using a coordination agent to adjust the preliminary treatment recommendations based on preset conflict resolution rules, and generate multidisciplinary treatment recommendations.

Description

Technical Field

[0001] This specification relates to the field of medical artificial intelligence technology, and in particular to a method, system, computer device, and computer program product for generating multidisciplinary diagnosis and treatment suggestions for lung cancer based on a large language model. Background Technology

[0002] Multidisciplinary team (MDT) care refers to a relatively fixed team of medical experts from multiple related disciplines who conduct regular, scheduled clinical discussions on a specific disease or patient. Based on the combined opinions of all disciplines, they develop a personalized and effective treatment plan for the patient, which is then implemented by the relevant disciplines individually or collaboratively. However, in some regions, expert resources are scarce, and the process of multidisciplinary care is time-consuming, making it difficult to meet the needs of emergency situations or rapid decision-making.

[0003] In view of this, some embodiments of this specification provide a method, system, computer device, and computer program product for generating multidisciplinary treatment recommendations for lung cancer based on a large language model, which aims to simulate the real lung cancer MDT consultation process and improve the efficiency of lung cancer MDT. Summary of the Invention

[0004] This specification provides one or more embodiments of a method for generating multidisciplinary treatment recommendations for lung cancer based on a large language model. The method includes: acquiring the medical record information of a lung cancer patient; using a diagnostic layer agent to determine diagnostic information based on the medical record information, wherein the diagnostic information includes at least one of staging information, pathological information, and metabolic assessment information; using a treatment layer agent to generate preliminary treatment recommendations based on the diagnostic information, wherein the preliminary treatment recommendations include local treatment recommendations and / or systemic treatment recommendations; and using a coordination agent to adjust the preliminary treatment recommendations based on preset conflict resolution rules, and generate multidisciplinary treatment recommendations.

[0005] According to one or more embodiments of this specification, a method is provided to adjust preliminary treatment recommendations using a coordinating agent based on preset conflict resolution rules and generate multidisciplinary treatment recommendations, including: using a coordinating agent to identify agents with conflicts based on preset conflict resolution rules; using agents with conflicts to conduct a new round of analysis based on medical record information and / or diagnostic information to adjust the preliminary treatment recommendations; and using a coordinating agent to generate multidisciplinary treatment recommendations based on the adjusted preliminary treatment recommendations.

[0006] According to one or more embodiments of this specification, the method includes a preset conflict resolution rule, which includes a preset conflict type. The method involves using a coordinating agent to determine the agent with conflict based on the preset conflict resolution rule. The method includes: using a coordinating agent to determine the agent with conflict based on the preset conflict type. The preset conflict type includes at least one of the following types: disagreement on treatment direction, dispute on molecular subtyping and staging, dispute on perioperative timing, dispute on drug resistance strategy, dispute on oligometastasis strategy, and impact on economic accessibility.

[0007] According to one or more embodiments of this specification, a method is provided to utilize conflicting agents to perform a new round of analysis based on medical record information and / or diagnostic information, and to adjust preliminary treatment recommendations. This includes: utilizing conflicting agents to perform iterative analysis based on medical record information and / or diagnostic information; stopping the iterative analysis and adjusting the preliminary treatment recommendations when preset conditions are met; wherein the preset conditions include at least one of the following: reaching an upper limit on the number of iterations, wherein the preset conflict resolution rules include preset conflict types, and the upper limit on the number of iterations is determined based on the severity level corresponding to the preset conflict type; and the conflicting agents reaching a consensus.

[0008] According to one or more embodiments of this specification, a safety review agent adjusts preliminary treatment recommendations based on preset safety rules, wherein the adjustment priority of the preset safety rules is higher than that of preset conflict resolution rules; the preset safety rules include at least one of the following rules: rules related to treatment contraindications, rules related to drug interactions, and rules related to risks in special populations.

[0009] According to the method provided in one or more embodiments of this specification, the medical record information includes at least one of the following: basic information of the lung cancer patient, clinical manifestation information, imaging report information, pathological and molecular test results.

[0010] According to one or more embodiments of this specification, the diagnostic layer intelligent agent includes at least one of radiology intelligent agent, pathology intelligent agent, and nuclear medicine intelligent agent; the diagnostic layer intelligent agent determines diagnostic information based on medical record information, including at least one of the following processes: using the radiology intelligent agent to determine staging information based on image report information; using the pathology intelligent agent to determine pathological information based on pathological and molecular test results; and using the nuclear medicine intelligent agent to determine metabolic assessment information based on image report information.

[0011] According to one or more embodiments of this specification, the treatment layer intelligent agent includes at least one of an oncology intelligent agent, a surgical intelligent agent, and a radiation oncology intelligent agent; local treatment recommendations include surgical treatment recommendations and / or radiation therapy recommendations; the treatment layer intelligent agent generates preliminary treatment recommendations based on diagnostic information, including at least one of the following processes: using the oncology intelligent agent to generate systemic treatment recommendations based on diagnostic information, the systemic treatment recommendations including treatment goals and medication regimens; using the surgical intelligent agent to determine surgical treatment recommendations based on staging information and / or pathological information, the surgical treatment recommendations including resectability grading information and surgical details; using the radiation oncology intelligent agent to determine radiation therapy recommendations based on diagnostic information, the radiation therapy recommendations including radiation therapy indication assessment information and radiation therapy details.

[0012] According to one or more embodiments of this specification, the systemic treatment recommendation further includes economic classification information, and the method further includes: using a treatment cost agent to determine economic stratification information corresponding to the medication regimen based on the drug type corresponding to the medication regimen, wherein the economic stratification information includes economic classification and treatment cost.

[0013] The method provided according to one or more embodiments of this specification further includes: using a follow-up agent to determine follow-up recommendations based on at least one of medical record information, diagnostic information, and preliminary treatment recommendations, wherein the follow-up recommendations include follow-up time and a list of examination items; generating multidisciplinary treatment recommendations includes: generating multidisciplinary treatment recommendations based on the follow-up recommendations.

[0014] According to one or more embodiments of this specification, the method further includes: using a clinical research agent to determine clinical research information matching a lung cancer patient based on a preset clinical research database, preset matching dimensions, and at least one of the following: medical record information and diagnostic information; wherein the preset matching dimensions include at least one of the following dimensions: histological type, molecular subtype, number of lines of treatment, and clinical stage; generating multidisciplinary treatment recommendations includes: generating multidisciplinary treatment recommendations based on clinical research information.

[0015] One or more embodiments of this specification also provide a multidisciplinary treatment suggestion generation system for lung cancer based on a large language model. The system includes: an acquisition module for acquiring medical record information of lung cancer patients; a diagnostic information determination module for determining diagnostic information based on the medical record information using a diagnostic layer agent, wherein the diagnostic information includes at least one of staging information, pathological information, and metabolic assessment information; a treatment suggestion generation module for generating preliminary treatment suggestions based on the diagnostic information using a treatment layer agent, wherein the preliminary treatment suggestions include local treatment suggestions and / or systemic treatment suggestions; and a multidisciplinary treatment suggestion generation module for adjusting the preliminary treatment suggestions based on preset conflict resolution rules using a coordination agent and generating multidisciplinary treatment suggestions.

[0016] One or more embodiments of this specification also provide a computer device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it is able to implement the multidisciplinary diagnostic and treatment suggestion generation method described in some embodiments of this specification.

[0017] One or more embodiments of this specification also provide a computer program product, including a computer program that, when at least a portion of the computer program is executed by a processor, enables the multidisciplinary diagnostic and treatment recommendation generation method described in some embodiments of this specification. Attached Figure Description

[0018] This specification will be further described by way of exemplary embodiments, which will be described in detail with reference to the accompanying drawings. The same numbers in the drawings denote the same structures or steps.

[0019] Figure 1 This is an exemplary flowchart illustrating a method for generating multidisciplinary diagnosis and treatment recommendations for lung cancer based on a large language model, according to some embodiments of this specification.

[0020] Figure 2 This is a schematic diagram of a framework for generating multidisciplinary treatment recommendations for lung cancer, as illustrated in some embodiments of this specification.

[0021] Figure 3 This is a flowchart illustrating a method for generating multidisciplinary diagnostic and treatment recommendations based on some embodiments of this specification.

[0022] Figure 4 This is an exemplary block diagram of a lung cancer multidisciplinary diagnosis and treatment suggestion generation system based on a large language model, as shown in some embodiments of this specification. Detailed Implementation

[0023] To more clearly illustrate the technical solutions of the embodiments in this specification, the embodiments will be described in detail below with reference to the accompanying drawings. Obviously, the content described below are some examples or embodiments of this specification. For those skilled in the art, without creative effort, the technical solutions or means disclosed in this specification can be applied to other scenarios based on this technical content.

[0024] It should be understood that the terms "system," "device," "unit," and / or "module" used in this specification are a method of distinguishing different components, elements, parts, sections, or assemblies at different levels. However, if other terms can achieve the same purpose, they may be replaced by other expressions.

[0025] Unless otherwise specified, the technical terms used to describe components, elements, etc. in this specification are not singular but may include plural. Generally speaking, terms such as "comprising" or "including" only indicate that explicitly identified steps, elements, or components are included, and these steps, elements, and components do not constitute an exclusive list, as the described method or apparatus may also include other steps or components.

[0026] This specification uses flowcharts to illustrate the operational steps performed by the apparatus or system of related embodiments. However, unless otherwise specified, the order in which these steps are described should not be construed as a limitation on the order of execution. Those skilled in the art can adjust the order of these steps based on the knowledge and information conveyed by the embodiments in this specification. Such adjustments include, but are not limited to, reversing the order of steps, merging multiple steps, and splitting a step.

[0027] Lung cancer is a malignant tumor with extremely high incidence and mortality rates worldwide. The number of new cases and deaths from this disease remains consistently high globally each year, resulting in an extremely heavy disease burden.

[0028] Multidisciplinary team (MDT) care is an important model for the standardized diagnosis and treatment of lung cancer. MDT refers to a relatively fixed team of medical experts from multiple related disciplines who, through regular, scheduled meetings, conduct clinical discussions on a specific disease or patient. Based on the combined opinions of all disciplines, they develop a personalized and effective treatment plan for the patient, which is then implemented individually or collaboratively by the relevant disciplines. For example, in the multidisciplinary treatment of lung cancer, medical experts or physicians from multiple specialties such as oncology, radiation oncology, radiology, pathology, surgery, and nuclear medicine can collaborate to develop a personalized treatment plan based on the patient's condition. However, high-level MDT teams are mostly concentrated in large tertiary hospitals, leaving primary care hospitals and remote areas with a scarcity of expert resources and difficulty in obtaining equally high-quality treatment advice. Coordinating meeting times among experts from various disciplines is also crucial for multidisciplinary care, making the process time-consuming and unsuitable for emergency or rapid decision-making needs.

[0029] To this end, some embodiments of this specification propose a method for generating multidisciplinary treatment recommendations for lung cancer based on a large language model. This method includes: acquiring the medical record information of a lung cancer patient; using a diagnostic layer agent to determine diagnostic information based on the medical record information, the diagnostic information including at least one of staging information, pathological information, and metabolic assessment information; using a treatment layer agent to generate preliminary treatment recommendations based on the diagnostic information, the preliminary treatment recommendations including local treatment recommendations and / or systemic treatment recommendations; and using a coordination agent to adjust the preliminary treatment recommendations based on preset conflict resolution rules, and generate multidisciplinary treatment recommendations. This simulates a real lung cancer MDT consultation process, improving the efficiency of lung cancer MDT.

[0030] Figure 1 This is an exemplary flowchart illustrating a method for generating multidisciplinary diagnosis and treatment recommendations for lung cancer based on a large language model, according to some embodiments of this specification. Figure 1 The illustrated process 100 can be executed by a computer device, for example, by a lung cancer multidisciplinary diagnosis and treatment suggestion generation system 400 based on a large language model deployed on a computer device. In some embodiments, the computer device can be a server. The server can include a local server or a cloud server, and depending on different service needs, a local server corresponding to that region can be deployed in one or more regions. In some embodiments, the server can be a single computer or a computing cluster composed of multiple computers, thereby providing more powerful computing power and more efficient response to user service requests. In some embodiments, the computer device can be a terminal device, including but not limited to desktop computers, smartphones, laptops, tablets, etc. In some embodiments, a portion of process 100 can be executed by the terminal device, while another portion can be executed by the server. Figure 1 As shown, process 100 may include the following steps.

[0031] Step 110: Obtain the medical record information of the lung cancer patient. In some embodiments, step 110 can be implemented by the acquisition module 410.

[0032] In some embodiments, users (such as clinicians) can import (or upload) the original medical records of lung cancer patients to a computer device. These original medical records may include the patient's basic data, clinical manifestation data, various imaging reports, and pathology and molecular testing reports. A coordinating agent within the computer device can preprocess the original medical records. Preprocessing may include extracting character content and performing text cleaning. For example, the coordinating agent can use a large language model to extract character content from the original medical records, or it can utilize OCR (Optical Character Recognition) technology. Text cleaning may include removing page numbers, table lines, extra spaces, and standardizing line breaks to obtain plain text content. The coordinating agent can then convert this plain text content into text information in a preset format to obtain the medical record information. For example, a set of structured fields can be predefined in the coordinating agent, and a large language model (or extraction using rule-based keyword matching and regular expressions) can be called to extract the specific values ​​corresponding to each field in the structured field from the cleaned plain text content. Then, the field name and the corresponding value are filled into a preset JSON data structure to obtain the medical record information of the structured field. Figure 2 This is a schematic diagram illustrating the framework of a method for generating multidisciplinary treatment recommendations for lung cancer, based on some embodiments of this specification. Figure 2 As shown, the coordinating agent can extract structured medical record information from the original medical records.

[0033] In some embodiments, the computer device may also mark a subset of fields in the structured fields as required fields, such as the maximum diameter of the primary lesion, histological type, molecular subtype, etc. If the value of any required field is empty, the computer device generates a prompt message to notify the user (e.g., a clinician) to supplement the relevant information before generating multidisciplinary treatment recommendations, or directly interrupts the treatment recommendation generation process and returns an error message, to ensure that the subsequent agent can make a diagnosis based on complete diagnostic evidence.

[0034] In some embodiments, medical record information may include basic information about the lung cancer patient. This basic information may include the patient's gender, age, smoking history, ECOG score, etc. The ECOG score can be used to assess the patient's activities of daily living and physical condition. For example, an ECOG score of 0 indicates complete normality, with no symptoms or activity limitations; an ECOG score of 2 indicates the patient can care for themselves but cannot perform any work and has limited activity; and an ECOG score of 4 indicates complete loss of self-care ability and being bedridden.

[0035] In some embodiments, medical record information may also include clinical presentation information. Clinical presentation information may include patient complaints, comorbidities, lines of treatment, previous treatment history, best efficacy, PFS (Progression-Free Survival), reasons for discontinuation, and toxicity. Patient complaints may include cough, chest pain, hemoptysis, and dyspnea. Comorbidities may include hypertension, diabetes, heart disease, and emphysema. Lines of treatment may include initial treatment, first-line, second-line, and later-line treatment. Previous treatment history may include each line of treatment regimen, treatment duration, and best efficacy (e.g., the best treatment outcome achieved by the patient throughout a specific line of treatment). PFS may refer to the time from the start of treatment to disease progression or death from any cause. Reasons for discontinuation may include disease progression, intolerable toxicity, patient's own choice, or completion of the planned treatment cycle. Toxicity may refer to adverse reactions or side effects during anticancer treatment.

[0036] In some embodiments, medical record information may also include imaging report information. Imaging report information may include chest CT report information, cranial MRI report information, bone scan report information, abdominal CT or MRI report information, PET-CT report information, echocardiography report information, etc. Specifically, chest CT report information may include the location of the primary lesion, the maximum diameter of the primary lesion, the relationship between the primary lesion and surrounding structures (e.g., chest wall, mediastinum, trachea, etc.), whether there is lymph node metastasis, the extent of lymph node metastasis (e.g., ipsilateral metastasis, contralateral metastasis), the specific location of lymph node metastasis, whether the metastasis is single-site or multi-site, whether there is distant metastasis, the location and number of distant metastases, and whether there are any special circumstances (e.g., malignant pleural effusion or pericardial effusion, pleural nodules, etc.). Cranial MRI report information may include the presence or absence of abnormal signal foci in the brain, the location / size / number / morphology of the lesions, the presence or absence of mass effect, the presence or absence of midline shift, the presence and extent of cerebral edema, and the enhancement pattern after contrast-enhanced scanning, etc. Bone scan information may include the location, extent, and degree of uptake of abnormal radioactive concentration foci, etc. Abdominal CT or MRI reports may include information on the presence of low-density or abnormal signal lesions in the liver, masses or nodules in the adrenal region, and enlarged lymph nodes in the abdominal cavity and retroperitoneum. PET-CT reports may include the location, size, shape, and maximum standardized uptake value (SUVmax) of the primary lesion; SUVmax and metabolic activity grading (e.g., no metabolism, mild, moderate, high) of each lymph node station (e.g., ipsilateral hilum, contralateral hilum, mediastinum, etc.); location, size, and SUVmax of distant metastases (e.g., bone, liver, adrenal glands, contralateral lung, etc.); quantitative parameters such as metabolic tumor volume (MTV) and total glycolysis (TLG) of the lesion; and indications of possible false positives (e.g., inflammation, granulomas, etc.). Echocardiography reports may include left ventricular ejection fraction (LVEF), wall motion, valvular structure and function, and the presence or absence of pericardial effusion.

[0037] In some embodiments, medical record information may also include pathological and molecular testing results. Pathological and molecular testing results may include descriptive information on histological type and differentiation degree, sample type (e.g., surgical resection specimen, biopsy specimen, cytological specimen), immunohistochemical marker data (e.g., TTF-1, Napsin A, p40, Syn, CgA, Ki-67, etc.), molecular testing information (e.g., molecular testing method, detection coverage, status and subtype of each driver gene), PD-L1 TPS, TMB value, MSI status, liquid biopsy results, etc.

[0038] Step 120: Determine diagnostic information based on medical record information using a diagnostic layer agent. In some embodiments, step 120 can be implemented by a diagnostic information determination module 420.

[0039] In some embodiments, the intelligent agent can be a decision-making unit that digitally simulates the role of a specialist physician in a real clinical multidisciplinary team. It can use a large language model as its reasoning engine, and by encapsulating knowledge bases or tools, defining input / output specifications, and decision rules, it can constitute a functional entity capable of simulating the analysis and decision-making of a specialist physician. For example, the intelligent agent can be pre-configured with structured prompt templates. These templates can include role definitions (e.g., radiology expert, pathology expert, nuclear medicine expert, thoracic surgery expert, oncology expert, radiation oncology expert, clinical research expert, safety review expert, follow-up expert, treatment cost expert, CSCO guideline expert, coordination and management expert, etc.), decision rules, input / output format specifications, mandatory constraints, and relevant clinical guidelines / reference knowledge. During runtime, the intelligent agent can combine the structured prompt template with its input information to form a complete prompt, which is then input into the large language model for reasoning and result generation. The intelligent agent can also provide the large language model with tool descriptions for tools such as knowledge base retrieval, clinical trial database queries, and cost calculation models. During inference, the large language model can generate tool invocation requests based on task requirements. The intelligent agent can then invoke the corresponding tools to perform queries based on these requests and return the query results to the large language model, allowing it to continue inference or generate the final result. The large language model can be a large-scale artificial intelligence model based on deep learning, whose core tasks can include understanding, generating, converting, and reasoning about natural language. For example, the large language model can be a deep learning model based on the Transformer architecture, which can utilize self-attention mechanisms to process sequential data and understand contextual meaning.

[0040] like Figure 2 As shown, in some embodiments, the diagnostic layer agent may include a radiology agent, and the diagnostic information may include staging information. The computer device can utilize the radiology agent to determine the staging information based on the image report information. For example, the radiology agent can invoke a large language model to analyze the image report information based on staging-related reference knowledge and staging judgment considerations to determine the size, location, and whether the lung cancer has spread, and obtain staging information.

[0041] In some embodiments, staging-related reference knowledge can be based on reference knowledge determined in the AJCC Cancer Staging Manual. For example, staging-related reference knowledge may include TNM staging definitions for non-small cell lung cancer (NSCLC), TNM staging definitions for small cell lung cancer (SCLC), and the correlation between TNM staging and clinical staging (e.g., T3N2bM0 corresponds to clinical stage IIIB). Staging-related reference knowledge can be updated according to updates to the AJCC Cancer Staging Manual. Considerations for staging determination may include staging determination steps, mandatory constraints, and a checklist for staging calculation verification. For example, staging determination steps may include first determining the T stage, then the N stage, then the M stage to obtain the complete cTNM, and finally determining the clinical stage based on the complete cTNM. By limiting the staging determination steps, the accuracy of staging information can be improved. For example, mandatory constraints may include "T2 stage must be subdivided into T2a or T2b, not just T2", "T1 must be subdivided into a / b / c", etc., thus avoiding ambiguous staging data and further improving the accuracy of staging information. For example, the staging calculation verification checklist may include checks that must be checked before output: T stage has been accurate to a / b / c; N stage has been correctly determined; M stage has been correctly determined; clinical stage has been cross-validated using a combination table; the T1 subdivision and stage correspondence for early NSCLC have been verified (again emphasizing T1a→IA1, T1b→IA2, T1c→IA3). The large language model can verify the staging information according to the staging calculation verification checklist before outputting, thereby ensuring the accuracy of the output staging information. In some embodiments, staging information may include T stage, N stage, M stage, complete cTNM, clinical stage, primary lesion description information, distant metastasis site description information, etc., and staging information can serve as an important basis for determining preliminary treatment recommendations.

[0042] like Figure 2 As shown, in some embodiments, the diagnostic layer agent may include a pathology agent, and the diagnostic information may include pathological information. The computer device can utilize the pathology agent to determine the pathological information based on pathological and molecular testing results. For example, the pathology agent can invoke a large language model to analyze the pathological and molecular testing results based on histological reference knowledge to determine the histological type of lung cancer and the presence of mutated genes, thereby obtaining pathological information.

[0043] In some embodiments, histology-related reference knowledge can be based on WHO histological classification standards (e.g., the 2021 version) and CSCO lung cancer diagnosis and treatment guidelines (e.g., the 2025 version). This histology-related reference knowledge may include histological analysis standards, such as descriptive information on non-small cell lung cancer and small cell lung cancer, so that the large language model can determine the histological type and differentiation degree based on these standards. Histological type can include disease type and pathological subtype. Disease type can include small cell lung cancer and non-small cell lung cancer, and pathological subtype can include adenocarcinoma, squamous cell carcinoma, large cell carcinoma, adenosquamous carcinoma, sarcomatoid carcinoma, etc. Differentiation degree can include well-differentiated, moderately differentiated, poorly differentiated, and undifferentiated. Histology-related reference knowledge may also include a molecular marker knowledge base, which may include different types of driver genes, detection methods for driver genes, mutation types, and treatment strategies (e.g., applicable targeted drugs, treatment methods, etc.), so that the large language model can determine the molecular subtype based on this knowledge base. This molecular subtype can guide the subsequent determination of targeted / immunotherapy regimens. Histology-related reference knowledge can also include specimen requirements, so that large language models can determine the reliability of pathological and molecular test results based on these requirements. Histology-related reference knowledge can also include a mandatory test list (e.g., a list of molecular markers that must be examined), so that large language models can assess the completeness of the molecular profile based on this list.

[0044] In some embodiments, pathological information may include histological conclusions, which may be structured information. For example, histological conclusions may include histological type, degree of differentiation, and immunohistochemical marker information. The immunohistochemical marker information may include marker types, such as positive or negative, positive intensity, and percentage of positive cells. Histological conclusions can serve as an important basis for determining preliminary treatment recommendations. In some embodiments, pathological information may also include molecular subtyping. For example, molecular subtyping may include the type of driver gene, detection method, and mutation type. Pathological information may also include molecular profile completeness. When molecular profile completeness is below a preset value, the pathology agent may generate prompts for supplementary molecular testing, or generate items requiring further examination when generating multidisciplinary treatment recommendations to remind users of incomplete molecular testing. In some embodiments, pathological information may also include analytical results related to immunotherapy or targeted therapy (e.g., PD-L1 TPS analysis results, TMB analysis results, MSI analysis results, liquid biopsy results), which can be used to guide subsequent immunotherapy or targeted therapy. In some embodiments, pathological information may also include molecular resistance indications, such as C797S exhibiting resistance to third-generation TKIs (e.g., osimertinib).

[0045] like Figure 2As shown, in some embodiments, the diagnostic agent may include a nuclear medicine agent, and the diagnostic information may include metabolic assessment information. The computer device can utilize the nuclear medicine agent to determine the metabolic assessment information based on the image report information. For example, the nuclear medicine agent can invoke a large language model to analyze the PET-CT report information based on metabolic-related reference knowledge to assess the activity level of lung cancer and whether distant metastasis has occurred, and obtain metabolic assessment information. The metabolic assessment information can be used to assist in determining the accuracy of the staging information output by the radiology agent.

[0046] In some embodiments, metabolic-related reference knowledge may include definitions of metabolic parameters (e.g., SUVmax represents maximum standardized uptake, MTV represents metabolic tumor volume, and TLG represents total glycolysis in the lesion) and metabolic activity stratification criteria (e.g., SUVmax less than 2.5 indicates low metabolism, and SUVmax greater than 10 indicates very high metabolism). Metabolic-related reference knowledge may also include the sensitivity and specificity of PET-CT detection rates for different sites (e.g., mediastinal lymph nodes, adrenal glands, bone, liver, brain), as well as common scenarios leading to false negatives (e.g., interference with normal brain metabolism). This allows the large language model to assess the reliability of PET-CT detection conclusions for metastatic lesions in specific sites, avoiding misclassification of false positives as metastases or omitting false negatives due to low sensitivity.

[0047] In some embodiments, metabolic assessment information may include metabolic parameters and metabolic activity of the primary lesion. Metabolic parameters of the primary lesion may include, for example, SUVmax, MTV, and TLG. Metabolic activity may include low metabolism, moderate metabolism, high metabolism, and very high metabolism; higher metabolic activity indicates a higher degree of tumor malignancy. Metabolic assessment information may also include lymph node metabolic information at each station to assist in assessing the accuracy of N staging. Metabolic assessment information may also include information on the presence or absence of distant metastasis to assist in assessing the accuracy of M staging. Metabolic assessment information may also include a conclusion on the consistency between metabolic staging and morphological staging, used to compare the metabolic analysis conclusion with the staging information output by the radiology agent and output a conclusion on whether the two are consistent. Metabolic assessment information may also include information indicating the source of false positives or false negatives, for example, providing an explanation of the possible increased metabolism caused by inflammatory lymph nodes to remind users (such as clinicians) to further confirm suspicious lesions. In some embodiments, when the imaging report information does not include PET-CT report information, the computer device may bypass the nuclear medicine agent to save computing resources.

[0048] Step 130: Generate preliminary treatment suggestions based on diagnostic information using the treatment layer agent. In some embodiments, step 130 can be implemented by the treatment suggestion generation module 430.

[0049] like Figure 2As shown, in some embodiments, the treatment layer agent may include a surgical agent, which can be used to determine whether a patient is suitable for surgery and the specific surgical details. Preliminary treatment recommendations may include local treatment recommendations, which may include surgical treatment recommendations. The computer device can utilize the surgical agent to determine surgical treatment recommendations based on staging information and / or pathological information. In some embodiments, the computer device may also use medical record information as input information for the surgical agent, so that the surgical agent can perform a more comprehensive analysis based on staging information, pathological information, and medical record information to determine surgical treatment recommendations. For example, the surgical agent can invoke a large language model to analyze surgical-related reference knowledge, staging information (e.g., cTNM staging), pathological information (e.g., mediastinal lymph node pathological status), and medical record information (e.g., ECOG score, patient age, cardiopulmonary function description information, comorbidities, frailty score, previous thoracic surgery history, patient's surgical wishes, etc.) to obtain surgical treatment recommendations.

[0050] In some embodiments, surgical reference knowledge may be based on the CSCO Lung Cancer Diagnosis and Treatment Guidelines (e.g., the 2025 edition). This surgical reference knowledge may include criteria for determining surgical resectability. For example, patients with stage I or II non-small cell lung cancer whose cardiopulmonary function can tolerate surgery are assessed as resectable, while patients with stage IIIB / C lung cancer exhibiting N3 or T4 invasion of key structures or whose functional status cannot tolerate surgery are assessed as unresectable. Surgical reference knowledge may also include knowledge of surgical approach selection. For example, lobectomy combined with systematic lymph node dissection is applicable to the vast majority of resectable cases, with lymph node dissection extending to lung groups 10-14 and mediastinal groups 2-9. Total pneumonectomy is suitable for cases where the tumor spans multiple lobes or invades the main pulmonary artery. The specific extent of lymph node dissection can also be determined based on histological type and resection margin status in pathological information. Surgical reference knowledge may also include knowledge of perioperative treatment, such as the indications and treatment regimens for preoperative neoadjuvant therapy and postoperative adjuvant therapy. The large language model can determine the necessity and timing of neoadjuvant or adjuvant therapy based on perioperative treatment knowledge. Surgical-related reference knowledge can also include patient tolerability assessment knowledge; for example, for pulmonary function assessment, lobectomy requires an FEV1 greater than 1.5 liters, and pneumonectomy requires an FEV1 greater than 2 liters, an FEV1% greater than 40%, and a DLCO greater than 40%. The large language model can determine whether a patient can tolerate surgery and the extent of surgery they can tolerate based on patient tolerability assessment knowledge.

[0051] In some embodiments, surgical treatment recommendations include resectability grading information and surgical details. Exemplarily, resectability grading information may include resectable, potentially resectable, or unresectable. Surgical details may include recommendations for surgical approach, lymph node dissection strategy, timing of surgery, and surgical tolerance assessment results. Exemplarily, the recommended surgical approach may be lobectomy combined with systematic lymph node dissection; the recommended lymph node dissection strategy may be lymph node dissection of lung groups 10-14 and mediastinal groups 2-9; the timing of surgery may be immediate; and the surgical tolerance assessment result may be an FEV1 greater than 1.5 liters, indicating tolerance to lobectomy. Surgical treatment recommendations can serve as an important basis for subsequently developing local and systemic treatment plans.

[0052] like Figure 2 As shown, in some embodiments, the treatment layer agent may include a radiotherapy agent, and the diagnostic information may include radiotherapy recommendations. The computer device can utilize the radiotherapy agent to determine radiotherapy recommendations based on the diagnostic information. Specifically, the radiotherapy agent can be used to determine whether a lung cancer patient is suitable for radiotherapy and to formulate specific radiotherapy recommendations (e.g., irradiation site, dose, number of sessions, etc.) based on the patient's condition. In some embodiments, the computer device can also use medical record information as input information for the radiotherapy agent, so that the radiotherapy agent can perform a more comprehensive analysis based on staging information, pathological information, and medical record information to determine radiotherapy recommendations. For example, the radiotherapy agent can invoke a large language model to analyze radiotherapy-related reference knowledge, staging information (e.g., cTNM staging), pathological information (e.g., histological conclusions), and medical record information (e.g., ECOG score, cardiopulmonary function description information, spinal cord distance, previous history of chest radiotherapy, brain metastasis, proposed systemic treatment plan, etc.) to obtain radiotherapy recommendations.

[0053] In some embodiments, the radiotherapy-related reference knowledge can be based on the CSCO Lung Cancer Diagnosis and Treatment Guidelines (e.g., the 2025 edition). This reference knowledge may include criteria for determining radiotherapy indications, such as radical, palliative, and adjuvant / neoadjuvant radiotherapy indications. For example, for patients with inoperable stage I-II non-small cell lung cancer or those refusing surgery, radical radiotherapy with stereotactic body radiotherapy (SBRT) may be recommended; for patients with oligometastasis (i.e., a limited number and localized range of metastatic lesions), a combined radiotherapy regimen for the primary lesion and metastatic lesions may be recommended. The radiotherapy-related reference knowledge may also include knowledge on radiotherapy technique selection; for example, stereotactic body radiotherapy (SBRT) is suitable for early-stage inoperable patients, offering advantages such as high precision and short treatment duration. Radiation therapy-related reference knowledge can also include dosage regimen knowledge. For example, early radical stereotactic body radiotherapy can be performed using 54 Gy divided into 3 fractions or 48 Gy divided into 4 fractions. Large language models can determine the recommended dosage fractionation based on dosage regimen knowledge, combined with the results of radiotherapy indication assessment and the proposed systemic treatment regimen (such as the selection of concurrent chemoradiotherapy regimen).

[0054] In some embodiments, radiotherapy recommendations may include radiotherapy indication assessment information and radiotherapy details. For example, radiotherapy indication assessment information may include whether radiotherapy is suitable, treatment goals (e.g., radical treatment, adjuvant / neoadjuvant treatment, palliative treatment, maintenance treatment), and the rationale for that decision. Radiotherapy details may include the selected radiotherapy technique (e.g., SBRT, VMAT, IMRT, 3D-CRT, etc.), target volume description information, dose fractionation regimen, concurrent chemoradiotherapy recommendations, and patient tolerability assessment results.

[0055] like Figure 2 As shown, in some embodiments, the treatment layer agent may include an oncology agent, and preliminary treatment recommendations may include systemic treatment recommendations. The computer device can utilize the oncology agent to generate systemic treatment recommendations based on diagnostic information. In some embodiments, the computer device can also use medical record information as input to the oncology agent, so that the oncology agent can perform a more comprehensive analysis based on diagnostic and medical record information to determine systemic treatment recommendations. For example, the oncology agent can invoke a large language model to analyze relevant internal medicine reference knowledge, diagnostic information (e.g., cTNM staging, histological type, etc.), and medical record information (e.g., ECOG score, complete blood count results, comorbidities, concomitant medications, previous treatment history, patient preferences, etc.) to obtain systemic treatment recommendations.

[0056] In some embodiments, the relevant medical reference knowledge can be based on the CSCO Lung Cancer Diagnosis and Treatment Guidelines (e.g., the 2025 edition). This relevant medical reference knowledge may include knowledge of treatment goal determination. Treatment goals may include radical treatment, adjuvant / neoadjuvant treatment, palliative treatment, and maintenance treatment. For example, for patients with stage I-II operable non-small cell lung cancer, the treatment goal is adjuvant therapy. The large language model can determine the treatment goal based on the knowledge of treatment goal determination, combined with information such as TNM staging and line of treatment. The relevant medical reference knowledge may also include molecular subtyping-related reference knowledge. For example, the large language model can determine the specific driver gene mutation type based on molecular subtyping-related reference knowledge, combined with the molecular subtyping in the diagnostic information, and match the corresponding targeted drugs, dosages, and usage regimens. The large language model can also determine the immunotherapy monotherapy or immunotherapy combined with chemotherapy regimen based on molecular subtyping-related reference knowledge, combined with the histological type and PD-L1 TPS analysis results in the diagnostic information.

[0057] In some embodiments, systemic treatment recommendations may include treatment goals and medication regimens. Treatment goals may include radical, adjuvant, neoadjuvant, palliative, and maintenance therapies, and may also include the rationale for choosing that goal. Medication regimens may include information such as drug name, dosage, route of administration, cycle, duration, and supporting evidence.

[0058] In some embodiments, systemic treatment recommendations may further include economic stratification information. The computer device may also utilize a treatment cost agent to determine the economic stratification information corresponding to the medication regimen based on the drug type. The drug type corresponding to the medication regimen may include original drugs, domestically produced alternatives, and drugs covered by basic medical insurance. Economic stratification information may include economic categories and treatment costs. For example, economic categories may include original drug regimens, domestically produced alternatives, and basic medical insurance regimens. The treatment cost agent may invoke a large language model to analyze cost-related reference knowledge, systemic treatment recommendations, and medical record information (e.g., patient's medical insurance type, commercial insurance status) to obtain economic categories and corresponding treatment costs. The oncology agent may also adjust systemic treatment recommendations based on economic stratification information. For example, it may recommend a medication regimen matching the patient based on the patient's economic situation in the medical record information (e.g., whether they have medical insurance, commercial insurance, or the patient's self-reported economic situation).

[0059] In some embodiments, cost-related reference knowledge may include a cost database. This database may store drug costs for different medications, costs for different treatment methods or means, basic hospitalization costs, costs for different examinations, and costs for adverse reaction management. Cost-related reference knowledge may also include national medical insurance data, such as the latest version of the National Medical Insurance Drug Catalog and medical insurance reimbursement ratios, used to calculate medical insurance reimbursement amounts. Cost-related reference knowledge may also include commercial insurance data, such as commercial insurance reimbursement information for critical illness insurance and medical insurance, used to calculate commercial insurance reimbursement amounts. Cost-related reference knowledge may also include charitable drug donation data, such as the applicable conditions and reduction / exemption amounts for patient assistance programs of various pharmaceutical companies, used to determine the eligibility for charitable drug donations in the medication regimen and the corresponding reduction / exemption amounts for charitable drug donations. Cost-related reference knowledge may also include cost calculation rules, allowing the treatment cost agent to calculate the treatment costs corresponding to each economic category based on these rules. For example, treatment costs can include total costs, medical insurance reimbursement, commercial insurance reimbursement, the value of charitable donated drugs, and the patient's actual out-of-pocket expenses. Total costs can be the sum of drug costs, examination costs, treatment costs, hospitalization costs, and adverse reaction management costs. The patient's actual out-of-pocket expenses = total costs - medical insurance reimbursement - commercial insurance reimbursement - amount that can be reduced or exempted by charitable donated drugs.

[0060] like Figure 2 As shown, in some embodiments, the computer device may further include a follow-up agent. The computer device may also utilize the follow-up agent to determine follow-up recommendations based on at least one of medical record information, diagnostic information, and preliminary treatment recommendations. For example, the follow-up agent may invoke a large language model to analyze the follow-up reference knowledge, medical record information, diagnostic information, and preliminary treatment recommendations, and generate follow-up recommendations. These follow-up recommendations may include follow-up times and a list of examination items.

[0061] In some embodiments, follow-up reference knowledge can be reference knowledge determined based on the CSCO Lung Cancer Diagnosis and Treatment Guidelines or preset follow-up rules. For example, follow-up reference knowledge may include follow-up times and a list of examinations corresponding to different treatment goals (e.g., radical treatment, adjuvant / neoadjuvant treatment, palliative treatment, maintenance treatment). Follow-up reference knowledge may also include follow-up times and a list of examinations corresponding to different treatment regimens. For example, immunotherapy requires regular checks of thyroid function, liver function, blood glucose, myocardial enzymes, etc.; follow-up reference knowledge may also include high-risk recurrence characteristics corresponding to TNM staging, histological type, molecular subtype, etc., and the frequency of imaging examinations corresponding to different high-risk recurrence characteristics. In some embodiments, when generating multidisciplinary treatment recommendations, the coordinating agent can generate multidisciplinary treatment recommendations based on follow-up recommendations.

[0062] like Figure 2As shown, in some embodiments, the computer device may further include a clinical research agent. The computer device may also utilize the clinical research agent to determine clinical research information matching lung cancer patients based on a preset clinical research database, preset matching dimensions, and at least one of the following: medical record information and diagnostic information. The preset matching dimensions may include at least one of the following dimensions: histological type, molecular subtype, number of lines of treatment, and clinical stage. The preset clinical research database may store clinical research projects that can be matched with lung cancer patients. Each clinical research project may include the registration number of each trial, the investigational drug or treatment regimen, applicable patient conditions (e.g., suitable histological type, molecular subtype, stage information, number of lines of treatment, etc.), recruitment status, and other information. For example, the clinical research agent may extract key fields corresponding to the preset matching dimensions from the medical record information and / or diagnostic information, and then invoke a clinical trial database query tool to search the preset clinical research database based on the key fields to determine matching candidate clinical research projects. In some embodiments, when preset conditions are met, the clinical research agent may also invoke a real-time web retrieval tool to perform a real-time query on a preset webpage based on the preset matching dimensions to obtain matching candidate clinical research projects. The preset conditions may include at least one of the following: the number of candidate clinical research projects matched from the preset clinical research database is less than a preset threshold; the lung cancer patient has special conditions (e.g., rare driver gene mutations, brain metastases, autoimmune diseases, organ dysfunction, etc.); the preset clinical research database has not been updated for a preset period of time. Furthermore, the clinical research agent can perform a weighted summation calculation based on the base scores and weights corresponding to each preset matching dimension to obtain a matching score for each clinical research project. For example, if a lung cancer patient's histological type, molecular subtype, and stage information all match a certain clinical research project, but the number of treatment lines for the lung cancer patient does not match the clinical research project, then the matching score corresponding to the clinical research project = histological type base score × histological type weight + molecular subtype base score × molecular subtype weight + stage information base score × stage information weight. In some embodiments, the weights corresponding to the preset matching dimensions, in descending order, can be: histological type weight > molecular subtype weight > clinical subtype weight > number of treatment lines weight. The clinical research agent can screen clinical research projects according to the different importance levels of each preset matching dimension. When a lung cancer patient has a special condition, that special condition can also be used as an additional correction factor to adjust the matching score. For example, for clinical research projects that explicitly exclude the special condition, the additional correction factor can be used to deduct points from the matching score, while for clinical research projects specifically targeting the special condition, the additional correction factor can be used to add points to the matching score.After obtaining the matching scores for each clinical research project, the clinical research agent can rank the projects according to their matching scores from highest to lowest, filter out projects ranked below a preset number, and obtain the final target clinical research projects, outputting clinical research information. This information may include the name, registration number, research title, research stage, investigational drug or intervention, control protocol, matching score, fulfillment of inclusion criteria, expected benefits and opportunity costs, and supplementary examinations for each target clinical research project. In some embodiments, when generating multidisciplinary treatment recommendations, the coordinating agent can generate such recommendations based on the clinical research information.

[0063] In some embodiments, the computer device may further include a treatment guideline agent, which can be used to generate treatment pathways based on medical record information. For example, the treatment guideline agent can invoke a large language model to determine the treatment pathway based on the CSCO lung cancer treatment guideline knowledge base and medical record information. The CSCO lung cancer treatment guideline knowledge base can pre-set multiple decision nodes and a node judgment order. The large language model can determine the judgment results of each decision node based on the medical record information and output the judgment results according to the node judgment order. Decision nodes may include histological type judgment, staging information judgment, molecular subtyping judgment, treatment line judgment, and treatment plan judgment (e.g., determining the optimal treatment plan as surgery, radiotherapy, or systemic medication). For example, for the decision node "histological type judgment," this decision node can be used to determine the histological type of the lung cancer patient based on the medical record information. If it is small cell lung cancer, it outputs "small cell lung cancer"; if it is adenocarcinoma, it outputs "non-small cell lung cancer → adenocarcinoma" (adenocarcinoma belongs to non-small cell lung cancer). For example, a treatment pathway could be: "Non-small cell lung cancer → squamous cell carcinoma → Stage IV → Stage IVB → No driver gene → First-line treatment → High PD-L1 expression (>50%) → Grade I recommendation: Immunotherapy monotherapy or immunotherapy combined with chemotherapy." This pathway can intuitively show clinicians the current disease stage and molecular characteristics of the lung cancer patient, as well as the standardized treatment process recommended by the CSCO guidelines. In some embodiments, the coordinating agent can also output this treatment pathway so that users (such as clinicians) can refer to the treatment pathway recommended by the CSCO guidelines to read multidisciplinary treatment recommendations.

[0064] Step 140: The coordinating agent adjusts the preliminary treatment recommendations based on preset conflict resolution rules and generates multidisciplinary treatment recommendations. In some embodiments, step 140 can be implemented by the multidisciplinary treatment recommendation generation module 440.

[0065] In some embodiments, the coordinating agent can aggregate information generated by other agents, identify conflicting agents based on preset conflict resolution rules, and feed back the conflicting information to the conflicting agents. Upon receiving the conflicting information, each conflicting agent can conduct a new round of analysis based on medical record information and / or diagnostic information. The coordinating agent can adjust the preliminary treatment recommendations based on the results of this new round of analysis and generate multidisciplinary treatment recommendations based on the adjusted preliminary treatment recommendations. For a detailed explanation of this part, please refer to the description in flowchart 300 below.

[0066] Figure 3 This is a flowchart illustrating a method for generating multidisciplinary diagnostic and treatment recommendations based on some embodiments of this specification. Figure 3 The illustrated process 300 can be executed by a computer device, for example, by the multidisciplinary treatment suggestion generation module 440 in a large language model-based multidisciplinary treatment suggestion generation system for lung cancer deployed on a computer device. Figure 3 As shown, in some embodiments, process 300 may include the following steps.

[0067] Step 310: Use the coordinating agent to determine the agents with conflicts based on preset conflict resolution rules.

[0068] In some embodiments, preset conflict resolution rules may include preset conflict types. A computer device can utilize a coordinating agent to determine which agents are in conflict based on these preset conflict types. Preset conflict types may include at least one of the following: treatment direction disagreement, molecular subtyping and staging dispute, perioperative timing dispute, drug resistance strategy dispute, oligometastatic strategy dispute, and economic accessibility impact. Treatment direction disagreement can refer to a conflict between treatment-layer agents regarding the treatment direction for lung cancer patients. For example, the coordinating agent can examine the information output by the oncology agent, surgical agent, and radiotherapy agent based on the treatment direction disagreement. If the surgical agent outputs "resectable" while the oncology agent outputs systemic treatment recommendations, indicating a conflict, then the surgical agent and the oncology agent are identified as conflicting agents. Molecular subtyping and staging dispute can refer to a conflict between the staging information output by the radiology agent or the molecular subtyping output by the pathology agent and the information output by other agents. For example, the staging information output by the radiology agent might show M0 for stage M, indicating no distant metastasis, while the metabolic assessment information output by the nuclear medicine agent suggests distant metastasis, creating a conflict. Since errors in diagnostic information can affect initial treatment recommendations, when the conflicting agent includes diagnostic agents, the treatment agent can also be identified as a conflicting agent. Perioperative timing disputes can refer to disagreements regarding the timing of surgery after neoadjuvant therapy versus the timing of adjuvant therapy. For example, the surgical agent might recommend immediate surgery, while the oncology agent recommends completing four cycles of neoadjuvant therapy before surgery, creating a conflict. Resistance strategy disputes can arise from conflicts between molecular resistance indications from the pathology agent and the medication regimens from the oncology agent. For instance, the pathology agent might indicate resistance to third-generation TKIs (such as osimertinib), but the oncology agent's medication regimen recommends osimertinib. Oligometastatic strategy controversies can refer to disagreements between the radiotherapy agent and the medical oncology agent regarding treatment strategies when a lung cancer patient is in an oligometastatic state. For example, the radiotherapy agent might recommend radiotherapy in its assessment of radiotherapy indications, arguing that a combined radiotherapy regimen for the primary tumor and metastases is preferred for oligometastatic patients. However, the medical oncology agent might recommend systemic therapy (such as targeted therapy, immunotherapy, or chemotherapy) as the primary treatment, with local treatment of oligometastatic lesions as a follow-up consolidation measure or not administered at this time. In such cases, the coordinating agent can determine that a conflict exists between the medical oncology agent and the radiotherapy agent. Economic accessibility impacts can refer to discrepancies between the treatment plan recommended by the medical oncology agent and the patient's actual economic situation. For instance, the medical record might explicitly state that the patient has a poor economic situation, but the medical oncology agent's recommended treatment plan prioritizes original drugs, indicating a potential error in the calculation of treatment costs by either the agent or the medical oncology agent, resulting in a conflict.

[0069] In some embodiments, each agent may output conflict warning information when outputting its own information. For example, when the radiology agent outputs staging information, it may warn that the staging information may conflict with the metabolic assessment information output by the nuclear medicine agent. The coordinating agent may determine whether there is a conflict between the staging information and the metabolic assessment information based on the conflict warning information. If there is a conflict, the radiology agent and the nuclear medicine agent are identified as the agents with the conflict.

[0070] Step 320: Utilize conflicting agents to conduct a new round of analysis based on medical record information and / or diagnostic information, and adjust the initial treatment recommendations.

[0071] In some embodiments, a computer device can utilize conflicting agents to perform iterative analysis based on medical record information and / or diagnostic information. When preset conditions are met, the iterative analysis stops, and the initial treatment recommendations are adjusted. For example, a coordinating agent can send conflict notifications to each conflicting agent. Upon receiving a conflict notification, each conflicting agent can re-analyze based on medical record information and / or diagnostic information, select a preset stance, and output that stance. Preset stances can include maintaining the original viewpoint, revising the viewpoint, or proposing a compromise. When choosing to maintain the original viewpoint, the agent needs to output evidence and reasons for maintaining it; when choosing to revise the viewpoint, the agent needs to explain any omissions or errors in the original output; when choosing to propose a compromise, each conflicting agent needs to provide supporting evidence. In some embodiments, conflicting agents can determine whether to modify the original output based on their chosen stance, thereby adjusting the initial treatment recommendations. For example, choosing to maintain the original viewpoint means no modification to the original output; choosing to revise the viewpoint or propose a compromise means modification of the original output. In some embodiments, the explicit selection of a position and output of corresponding content by a conflicting agent from a preset set of positions can be achieved through conditional constraints on the agent. For example, the agent's conditional constraints may include: upon receiving a conflict notification, the agent must explicitly select one of three positions, and each position must be accompanied by specific reasons and a weighted reassessment process of evidence.

[0072] In some embodiments, the preset conditions may include at least one of the following: reaching the maximum number of iterations, or the conflicting agents reaching a consensus. The maximum number of iterations may be determined based on the severity level corresponding to a preset conflict type. In some embodiments, a fundamental disagreement on treatment direction can be classified as "severe," with a maximum of three iterations; a difference in protocol details can be classified as "moderate," with a maximum of two iterations; and a fine-tuning of dosage or timing can be classified as "mild," with a maximum of one iteration. For example, disagreements on treatment direction and disputes over molecular subtyping and staging can be classified as "severe," with a maximum of three iterations; disputes over drug resistance strategies and oligometastasis strategies can be classified as "moderate," with a maximum of two iterations; and disputes over perioperative timing and economic accessibility can be classified as "mild," with a maximum of one iteration.

[0073] In some embodiments, each agent can identify conflict-related content when outputting its own information. For example, when outputting surgical treatment recommendations, the surgical agent can mark resectability classification information (e.g., resectable, potentially resectable, or inresectable) as key decision points. The coordinating agent can determine whether a conflict exists based on the key decision points output by each agent, and perform a weighted calculation based on the weights corresponding to each key decision point to determine the consensus score among the conflicting agents. When the consensus score is greater than a preset threshold, it is determined that the conflicting agents have reached a consensus.

[0074] In some embodiments, if the conflicting agents still fail to reach a consensus after reaching the maximum number of iterations, the coordinating agent may not force the output of a single treatment plan, but instead output a "serious disagreement" flag, along with the output results of each agent and the iterative discussion record (such as the chosen position, supporting evidence, consensus score, etc.) to prompt the clinician to manually review the information.

[0075] In some embodiments, the computer device can also utilize a security review agent to adjust preliminary treatment recommendations based on preset security rules. These preset security rules include at least one of the following: rules related to treatment contraindications, rules related to drug interactions, and rules related to risks in specific populations. For example, a rule related to treatment contraindications might prohibit surgical treatment in cases of distant metastasis (e.g., M stage M1a-c), contralateral mediastinal lymph node metastasis (e.g., N stage N3), or severe cardiopulmonary insufficiency (e.g., FEV1 < 0.8L, DLCO < 30%). A rule related to drug interactions might require that two or more drugs should be avoided from being taken simultaneously. For example, a rule related to drug interactions might include prohibiting the simultaneous use of drug A and drug B, and if the patient's past treatment history (from medical records) shows that the patient is currently taking drug B, then the rule related to drug interactions is violated if the systemic treatment plan output by the oncology agent includes drug A. A rule related to risks in specific populations might include treatment contraindications or precautions related to specific populations (e.g., the elderly, pregnant / lactating women, organ transplant recipients, etc.). For example, all anti-tumor drugs are contraindicated in pregnant women.

[0076] In some embodiments, the adjustment priority of preset safety rules can be higher than that of preset conflict resolution rules. For example, according to preset conflict resolution rules, the surgical agent outputs "resectable" and the oncology agent outputs "systemic treatment recommendation." The oncology agent and the surgical agent disagree on the treatment direction, and the patient is pregnant. According to the rule related to the risk of special populations, pregnant patients are prohibited from using all anti-tumor drugs, so the systemic treatment recommendation can be directly discarded.

[0077] In some embodiments, when there are no conflicting agents, the coordinating agent can also use each agent to conduct a new round of analysis based on medical record information and / or diagnostic information to review the conclusions output by each agent, confirm whether the conclusions output by each agent are accurate and whether there is any missing information, and determine whether to modify the output conclusions based on the review results, and output the modified content in the new round of analysis.

[0078] Step 330: The coordinating agent generates multidisciplinary treatment recommendations based on the adjusted preliminary treatment suggestions.

[0079] In some embodiments, when conflicting agents reach a consensus, each agent can modify its output information based on the consensus result. For example, after iterative analysis, the surgical agent selects a "corrected viewpoint," changes "operable" to "inoperable," deletes surgical details, and outputs the reasons and supporting evidence for the modification. The coordinating agent can integrate the information output by each agent and perform structured processing to output multidisciplinary treatment recommendations. For example, the multidisciplinary treatment recommendations may include the content output by each agent in the first round of analysis (e.g., preliminary treatment recommendations, diagnostic information, follow-up recommendations, patient-matched clinical research information, iterative discussion records, treatment pathways, medical record summaries, etc.) and the iterative analysis results (e.g., the adjusted preliminary treatment recommendations, conflicting agents, the positions chosen by the conflicting agents, their reasons, supporting evidence, etc.). The coordinating agent can also generate consensus summary information based on the adjusted preliminary treatment recommendations, follow-up recommendations, clinical research information, etc. For example, the consensus summary information could be "surgical resection + postoperative observation and follow-up."

[0080] For example, multidisciplinary treatment recommendations may include input medical record information and system output. Input medical record information may include: the patient is a 68-year-old male with an ECOG score of 1 and a smoking history of 40 pack-years. Imaging shows a 5.5cm mass in the left hilum with mediastinal lymph node metastasis; PET-CT shows bone and adrenal metastases. The pathological type is squamous cell carcinoma. Molecular testing results show EGFR wild-type, ALK negative, and PD-L1 TPS 85%.

[0081] The system output can include information from each agent in the first round, information from the second round of review, information from the third round of comprehensive confirmation, and a consensus summary. Specifically, the first round of agent output can include: Treatment pathway: NSCLC → squamous cell carcinoma → Stage IV → Stage IVB → No driver gene → First-line treatment → High PD-L1 expression (>50%) → Grade I recommendation: Immunotherapy monotherapy or immunotherapy combined with chemotherapy. Radiology agent output: cT3N2M1c, Stage IVB (bone metastasis + adrenal metastasis). Pathology agent output: Squamous cell carcinoma, no driver gene, PD-L1 TPS 85%. Surgical agent output: Unresectable (M1c with multiple distant metastases), no surgical indication. Medical oncology agent output: Immunotherapy monotherapy or combination chemotherapy (Evoxicillinab monotherapy approximately 30,000 RMB / year (medical insurance); Pembrolizumab monotherapy approximately 70,000 RMB / year (donation gift)). Clinical trial agent output: Recommend enrollment in ChiCTR2600119193 (a real-world study of edoximumab monotherapy as first-line treatment for PD-L1-high expression advanced NSCLC, **** Hospital). Safety review agent output: ECOG score 1, 40 pack-years of smoking history (high risk of ILD), tolerable treatment, infection screening + HRCT required. Follow-up agent output: CT evaluation every 6-8 weeks, bone scan every 6 months or when symptoms appear, continued use of denosumab 120mg q4w for bone protection. Second round of review information may include: consistent agent outputs, medical supplementation for bone metastases requires concurrent denosumab bone protection. Third round of comprehensive confirmation information may include: high consensus. Consensus summary information may include: immunotherapy monotherapy or combination chemotherapy (PD-L1 TPS high expression 85%), concurrent denosumab bone protection, recommended enrollment in clinical trial ChiCTR2600119193.

[0082] This manual also provides a system for generating multidisciplinary diagnosis and treatment recommendations for lung cancer based on a large language model. Figure 4 This is an exemplary block diagram of a lung cancer multidisciplinary treatment suggestion generation system based on a large language model, according to some embodiments of this specification. In some embodiments, the lung cancer multidisciplinary treatment suggestion generation system 400 based on a large language model can be deployed on a computer device. Figure 4 As shown, in some embodiments, the lung cancer multidisciplinary diagnosis and treatment suggestion generation system 400 based on a large language model may include an acquisition module 410, a diagnosis information determination module 420, a treatment suggestion generation module 430, and a multidisciplinary diagnosis and treatment suggestion generation module 440.

[0083] The acquisition module 410 can be used to acquire medical record information of lung cancer patients.

[0084] The diagnostic information determination module 420 can be used to determine diagnostic information based on medical record information using a diagnostic layer intelligent agent. The diagnostic information includes at least one of staging information, pathological information, and metabolic assessment information.

[0085] The treatment suggestion generation module 430 can be used to generate preliminary treatment suggestions based on diagnostic information using the treatment layer agent. The preliminary treatment suggestions include local treatment suggestions and / or systemic treatment suggestions.

[0086] The multidisciplinary treatment suggestion generation module 440 can be used to adjust preliminary treatment suggestions based on preset conflict resolution rules using a coordinating agent, and generate multidisciplinary treatment suggestions.

[0087] In some optional embodiments, the multidisciplinary treatment suggestion generation module 440 can also be used to: use a coordinating agent to determine the agent with conflict based on preset conflict resolution rules; use the agent with conflict to conduct a new round of analysis based on medical record information and / or diagnostic information to adjust the preliminary treatment suggestion; and use the coordinating agent to generate a multidisciplinary treatment suggestion based on the adjusted preliminary treatment suggestion.

[0088] In some optional embodiments, the preset conflict resolution rules may include preset conflict types. The multidisciplinary treatment suggestion generation module 440 may also be used to use a coordinating agent to determine the agent with conflict based on the preset conflict types. The preset conflict types include at least one of the following types: treatment direction disagreement, molecular subtyping and staging dispute, perioperative timing dispute, drug resistance strategy dispute, oligometastasis strategy dispute, and economic accessibility impact.

[0089] In some optional embodiments, the multidisciplinary treatment suggestion generation module 440 can also be used to: utilize conflicting agents to perform iterative analysis based on medical record information and / or diagnostic information; stop the iterative analysis when preset conditions are met, and adjust the preliminary treatment suggestions; wherein the preset conditions include at least one of the following conditions: reaching the upper limit of the number of iterations, wherein the preset conflict resolution rules include preset conflict types, and the upper limit of the number of iterations is determined based on the severity level corresponding to the preset conflict type; and the conflicting agents reach a consensus.

[0090] In some optional embodiments, the multidisciplinary treatment recommendation generation module 440 can also be used to adjust the preliminary treatment recommendations based on preset safety rules using a safety review agent, wherein the adjustment priority of the preset safety rules is higher than that of preset conflict resolution rules; the preset safety rules include at least one of the following rules: rules related to treatment contraindications, rules related to drug interactions, and rules related to risks in special populations.

[0091] In some optional embodiments, the diagnostic layer agent may include at least one of a radiology agent, a pathology agent, and a nuclear medicine agent; the diagnostic information determination module 420 may also be used to determine diagnostic information based on medical record information using the diagnostic layer agent, including at least one of the following processes: determining staging information based on image report information using the radiology agent; determining pathological information based on pathological and molecular test results using the pathology agent; and determining metabolic assessment information based on image report information using the nuclear medicine agent.

[0092] In some optional embodiments, the treatment layer agent includes at least one of an oncology agent, a surgical agent, and a radiation oncology agent; local treatment recommendations include surgical treatment recommendations and / or radiation therapy recommendations; the treatment recommendation generation module 430 can also be used to generate preliminary treatment recommendations based on diagnostic information using the treatment layer agent, including at least one of the following processes: generating systemic treatment recommendations based on diagnostic information using the oncology agent, the systemic treatment recommendations including treatment goals and medication regimens; determining surgical treatment recommendations based on staging information and / or pathological information using the surgical agent, the surgical treatment recommendations including resectability grading information and surgical details; and determining radiation therapy recommendations based on diagnostic information using the radiation oncology agent, the radiation therapy recommendations including radiation therapy indication assessment information and radiation therapy details.

[0093] In some optional embodiments, the systemic treatment recommendations may also include economic classification information. The lung cancer multidisciplinary diagnosis and treatment recommendation generation system 400 based on a large language model may also include an economic stratification information determination module 450, which is used to determine the economic stratification information corresponding to the medication regimen based on the drug type corresponding to the medication regimen using a treatment cost agent. The economic stratification information includes economic classification and treatment cost.

[0094] In some optional embodiments, the lung cancer multidisciplinary diagnosis and treatment suggestion generation system 400 based on a large language model may further include a follow-up suggestion generation module 460, which is used to determine follow-up suggestions based on at least one of medical record information, diagnostic information, and preliminary treatment suggestions using a follow-up intelligent agent. The follow-up suggestions include follow-up time and a list of examination items. Generating multidisciplinary diagnosis and treatment suggestions includes generating multidisciplinary diagnosis and treatment suggestions based on follow-up suggestions.

[0095] In some optional embodiments, the lung cancer multidisciplinary treatment suggestion generation system 400 based on a large language model may further include a clinical research information generation module 470, used to determine clinical research information matching lung cancer patients using a clinical research agent based on a preset clinical research database, preset matching dimensions, and at least one of the following: medical record information and diagnostic information; wherein, the preset matching dimensions include at least one of the following dimensions: histological type, molecular subtype, number of lines of treatment, and clinical stage; generating multidisciplinary treatment suggestions includes: generating multidisciplinary treatment suggestions based on clinical research information.

[0096] For more information on each module, please refer to [link / reference]. Figures 1-3 The relevant explanations will not be repeated here. It should be understood that... Figure 4 The apparatus, system, and modules illustrated can be implemented in various ways. For example, in some embodiments, they can be implemented by hardware, software, or a combination of both. The hardware portion can be implemented using dedicated logic; the software portion can be stored in memory and executed by an appropriate instruction execution system, such as a microprocessor or dedicated-design hardware. Those skilled in the art will understand that the methods, apparatus, and systems described above can be implemented using computer-executable instructions and / or included in the control code of a processor, such as code provided in the memory of a programmable device on a media such as a disk, CD, or DVD-ROM. The apparatus and modules described in this specification can be implemented not only by hardware circuitry such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips or transistors, or programmable hardware devices such as field-programmable gate arrays or programmable logic devices, but also by software, for example, executed by various types of processors, or by a combination of the aforementioned hardware circuitry and software (e.g., firmware).

[0097] It should be noted that the above descriptions of the devices, systems, and modules are for convenience only and should not be construed as limiting this specification to the embodiments described. It is understood that those skilled in the art, after understanding the principle of the device, can arbitrarily combine the various modules without departing from this principle to form sub-devices connected to other modules. Alternatively, some modules can be split to obtain more modules or multiple units under a single module. Such modifications are all within the scope of this specification.

[0098] Some embodiments of this specification also provide a computer device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it can implement this specification. Figures 1-3 The method shown.

[0099] Some embodiments of this specification also provide a computer-readable storage medium storing computer instructions that, when executed by a processor, can implement this specification. Figures 1-3 The method shown.

[0100] Some embodiments of this specification also provide a computer program product, including a computer program that, when at least a portion of the computer program is executed by a processor, can implement this specification. Figures 1-3 The method is illustrated. In some embodiments, the computer program product may refer only to a computer program, which may be carried on a storage medium or a processing device. In other embodiments, the computer program product may also be a storage medium or a processing device containing the aforementioned computer program. The processing device may include one or more processors, and the storage medium.

[0101] In some embodiments, the processor may be a combination of one or more of the following processors: central processing unit (CPU), application-specific integrated circuit (ASIC), application-specific instruction set processor (ASIP), graphics processing unit (GPU), physical processing unit (PPU), digital signal processor (DSP), field-programmable gate array (FPGA), programmable logic device (PLD), programmable logic controller (PLC), reduced instruction set computer (RISC), and microprocessor.

[0102] In some embodiments, the storage medium may include one or more combinations of the following: mass storage, removable storage, volatile read-write memory, and read-only memory (ROM). Exemplary mass storage may include disks, optical disks, solid-state drives, etc. Exemplary removable storage may include flash drives, floppy disks, optical disks, memory cards, compressed hard disks, magnetic tapes, etc. Exemplary volatile read-write memory may include random access memory (RAM). Exemplary RAM may include dynamic random access memory (DRAM), dual data rate synchronous dynamic random access memory (DDRSDRAM), static random access memory (SRAM), silicon controlled retrieval memory (T-RAM), and zero-capacitance memory (Z-RAM), etc. Exemplary read-only memory may include masked read-only memory (MROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), compressed hard disk read-only memory (CD-ROM), and digital multifunction hard disk read-only memory, etc.

[0103] The beneficial effects that the embodiments of this specification may bring include, but are not limited to: by acquiring the medical record information of lung cancer patients, using a diagnostic layer intelligent agent to determine diagnostic information based on the medical record information, using a treatment layer intelligent agent to generate preliminary treatment suggestions based on the diagnostic information, using a coordination intelligent agent to adjust the preliminary treatment suggestions based on preset conflict resolution rules, and generating multidisciplinary treatment suggestions, thus simulating the real lung cancer MDT consultation process, realizing full-link automated reasoning, and improving the efficiency of lung cancer MDT; by identifying the intelligent agents with conflicts, and conducting a new round of analysis and adjusting the preliminary treatment suggestions based on the conflicting intelligent agents, generating multidisciplinary treatment suggestions based on the adjusted suggestions, thereby introducing a dynamic conflict feedback mechanism, avoiding the simple stacking or forced merging of conclusions from different disciplines, and improving the accuracy of MDT consultation output results; by matching clinical research information, providing patients with suitable clinical research enrollment opportunities and providing innovative treatment options; and by using a treatment cost intelligent agent to determine the economic stratification information corresponding to the medication regimen based on the drug type corresponding to the medication regimen, thereby providing patients with treatment plans that take into account both efficacy and economic accessibility. It should be noted that different embodiments may produce different beneficial effects. In different embodiments, the beneficial effects may be any one or a combination of the above, or any other possible beneficial effects.

[0104] The basic concepts have been described above. It is obvious that the detailed disclosure above is merely illustrative and does not constitute a limitation of this specification. Although not explicitly stated herein, various modifications, improvements, and corrections may be made to this specification by those skilled in the art. Such modifications, improvements, and corrections are taught in this specification and therefore remain within the spirit and scope of the exemplary embodiments described herein.

Claims

1. A method for generating multidisciplinary diagnosis and treatment suggestions for lung cancer based on a large language model, characterized in that, The method includes: Obtain medical records of lung cancer patients; The diagnostic layer intelligent agent determines diagnostic information based on the medical record information, wherein the diagnostic information includes at least one of staging information, pathological information, and metabolic assessment information. The treatment layer agent generates preliminary treatment recommendations based on the diagnostic information, the preliminary treatment recommendations including local treatment recommendations and / or systemic treatment recommendations; The preliminary treatment recommendations are adjusted based on preset conflict resolution rules by a coordinating intelligent agent, and multidisciplinary treatment recommendations are generated.

2. The method according to claim 1, characterized in that, The process of using a coordinating agent to adjust the initial treatment recommendations based on preset conflict resolution rules and generating multidisciplinary treatment recommendations includes: The coordinating agent is used to determine the agents in conflict based on the preset conflict resolution rules; The conflicting agent performs a new round of analysis based on the medical record information and / or the diagnostic information to adjust the initial treatment recommendations; The coordinating agent generates the multidisciplinary treatment recommendations based on the adjusted preliminary treatment suggestions.

3. The method according to claim 2, characterized in that, The preset conflict resolution rules include preset conflict types, and the step of using a coordinating agent to determine the agents with conflicts based on the preset conflict resolution rules includes: The coordinating agent is used to determine the agents in conflict based on the preset conflict type; The preset conflict type includes at least one of the following types: Disagreements over treatment direction, molecular subtyping and staging, perioperative timing, drug resistance strategies, oligometastasis strategies, and economic accessibility.

4. The method according to claim 2, characterized in that, The step of utilizing the conflicting agent to perform a new round of analysis based on the medical record information and / or the diagnostic information, and adjusting the preliminary treatment recommendations, includes: The conflicting agent performs iterative analysis based on the medical record information and / or the diagnostic information. The iterative analysis is stopped when the preset conditions are met, and the preliminary treatment recommendations are adjusted. The preset conditions include at least one of the following conditions: The maximum number of iterations is reached, wherein the preset conflict resolution rules include preset conflict types, and the maximum number of iterations is determined based on the severity level corresponding to the preset conflict type; The conflicting agents reached a consensus.

5. The method according to claim 1, characterized in that, The method further includes: The preliminary treatment recommendations are adjusted using a security review agent based on preset security rules, wherein the adjustment priority of the preset security rules is higher than that of the preset conflict resolution rules; The preset security rules include at least one of the following rules: Rules related to contraindications to treatment, rules related to drug interactions, and rules related to risks in special populations.

6. The method according to claim 1, characterized in that, The medical record information includes at least one of the following: The basic information, clinical manifestations, imaging reports, pathological and molecular test results of the lung cancer patients.

7. The method according to claim 6, characterized in that, The diagnostic layer intelligent agent includes at least one of the following: radiology intelligent agent, pathology intelligent agent, and nuclear medicine intelligent agent; The process of using a diagnostic layer agent to determine diagnostic information based on the medical record information includes at least one of the following processes: The imaging intelligence agent determines the staging information based on the image report information; The pathological information is determined using the aforementioned pathology department intelligent agent based on the pathological and molecular detection results; The metabolic assessment information is determined based on the image report information using the nuclear medicine intelligent agent.

8. The method according to claim 1, characterized in that, The treatment layer intelligent agent includes at least one of an oncology intelligent agent, a surgical intelligent agent, and a radiation therapy intelligent agent; the local treatment recommendations include surgical treatment recommendations and / or radiation therapy recommendations. The process of generating preliminary treatment recommendations based on the diagnostic information using a treatment layer agent includes at least one of the following processes: The oncology intelligence agent generates systemic treatment recommendations based on the diagnostic information, the systemic treatment recommendations including treatment goals and medication regimens; The surgical agent uses the surgical intelligence to determine the surgical treatment recommendation based on the staging information and / or the pathological information, the surgical treatment recommendation including resectability grading information and surgical details; The radiotherapy agent uses the diagnostic information to determine the radiotherapy recommendation, which includes radiotherapy indication assessment information and radiotherapy details.

9. The method according to claim 8, characterized in that, The systemic treatment recommendations also include economic classification information, and the method further includes: The treatment cost agent determines the economic stratification information corresponding to the medication regimen based on the drug type corresponding to the medication regimen. The economic stratification information includes economic classification and treatment cost.

10. The method according to claim 1, characterized in that, The method further includes: The follow-up agent determines follow-up recommendations based on at least one of the medical record information, the diagnostic information, and the preliminary treatment recommendations. The follow-up recommendations include follow-up time and a list of examination items. The generated multidisciplinary treatment recommendations include: The multidisciplinary treatment recommendations are generated based on the follow-up recommendations.

11. The method according to claim 1, characterized in that, The method further includes: The clinical research agent uses a preset clinical research database, preset matching dimensions, and at least one of the following to determine the clinical research information that matches the lung cancer patient: the medical record information and the diagnostic information. The preset matching dimensions include at least one of the following dimensions: histological type, molecular subtype, number of treatment lines, and clinical stage; The generated multidisciplinary treatment recommendations include: The multidisciplinary treatment recommendations are generated based on the clinical research information.

12. A multidisciplinary diagnostic and treatment suggestion generation system based on a large language model, characterized in that, The system includes: The acquisition module is used to retrieve medical record information of lung cancer patients; A diagnostic information determination module is used to determine diagnostic information based on the medical record information using a diagnostic layer intelligent agent. The diagnostic information includes at least one of staging information, pathological information, and metabolic assessment information. A treatment suggestion generation module is used to generate preliminary treatment suggestions based on the diagnostic information using a treatment layer agent. The preliminary treatment suggestions include local treatment suggestions and / or systemic treatment suggestions. A multidisciplinary treatment suggestion generation module is used to adjust the preliminary treatment suggestions based on preset conflict resolution rules using a coordinating intelligent agent, and generate multidisciplinary treatment suggestions.

13. A computer device, characterized in that, The computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, it is able to implement the method as described in any one of claims 1 to 11.

14. A computer program product, characterized in that, It includes a computer program that, when at least a portion of the computer program is executed by a processor, enables the implementation of the method as described in any one of claims 1 to 11.

Images