Multi-department consultation method based on multi-round collaborative decision mechanism and storage medium

Abstract

The application provides a multi-department consultation method based on a multi-round collaborative decision mechanism and a storage medium, which comprises the following steps: extracting case points from medical record information; determining a first target department participating in a first round of consultation and sharing whiteboard information; in the process of the first round of consultation, determining diagnosis and treatment reference information corresponding to each first target department according to the case points, the shared whiteboard information and a prompt template; summarizing each diagnosis and treatment reference information, updating the shared whiteboard information, determining whether to enter a next round of consultation, focus information of the next round of consultation and a second target department needing to participate in the next round of consultation according to each diagnosis and treatment reference information and the updated shared whiteboard information; and in the process of the next round of consultation, determining diagnosis and treatment reference information according to the case points, the updated shared whiteboard information, diagnosis and treatment reference information corresponding to the last round of consultation and the focus information. The consultation effect and efficiency can be improved.

Description

Technical Field

[0001] This application relates to the field of data processing technology, specifically to a multi-departmental consultation method based on a multi-round collaborative decision-making mechanism and a storage medium. Background Technology

[0002] With the widespread application of artificial intelligence (AI) technology in the medical field, intelligent diagnosis and treatment systems based on Large Language Models (LLMs) are increasingly being used for medical record analysis, assisted diagnosis, and clinical decision support. Especially against the backdrop of continuously improving structuring of electronic medical records, utilizing AI to automatically analyze medical records and output treatment suggestions has become an important direction in the current development of medical informatization.

[0003] In real clinical practice, complex and difficult cases often require multidisciplinary team (MDT) consultations to integrate the diagnostic and treatment opinions of different specialties. Collaborative decision-making across multiple departments can reduce the risk of misdiagnosis and improve the comprehensiveness and safety of treatment plans. Therefore, how to utilize artificial intelligence technology to simulate the real MDT consultation process has become a crucial issue that urgently needs to be addressed in the field of intelligent healthcare.

[0004] However, in terms of existing technology, the technical routes related to this application can be roughly summarized into the following three categories:

[0005] (I) Single-model, single-round generative diagnostic and treatment system

[0006] Most existing medical AI systems employ a single-step model inference path, directly generating diagnostic suggestions after inputting the complete medical record into a large model. These systems typically have a "single-agent decision-making structure," completing all diagnostic and treatment support analysis through a single output. While this technology improves information processing efficiency, it suffers from the following shortcomings: it lacks the ability to simulate multi-role perspectives; it cannot simulate realistic multi-departmental collaboration mechanisms; it lacks conflict identification and mediation capabilities; it lacks multi-round collaborative reasoning functionality; and it lacks a decision stability detection mechanism. Therefore, this type of technology is essentially still a single-model augmented question-answering system and cannot achieve multi-departmental collaborative decision-making.

[0007] (II) Multi-agent parallel generation and simple summarization system

[0008] Some studies attempt to introduce multiple "expert agents" to simulate experts from different departments, then concatenate or aggregate the multiple outputs through voting. These systems typically employ a parallel generation structure, with each agent operating independently, ultimately forming the result through simple merging or selection. However, this approach still suffers from the following problems: a lack of structured sharing of intermediate states among agents; the absence of a clear round-robin control mechanism; a lack of conflict focus identification capabilities; a lack of dynamic scheduling mechanisms; a lack of decision convergence logic; and a tendency for opinions to diverge or logic to repeat. This type of technology only achieves "multiple output overlay" and does not form a true collaborative decision-making algorithm framework.

[0009] (III) Multi-turn conversational medical question-and-answer system

[0010] Some systems support multi-turn question-and-answer mechanisms, progressively refining treatment suggestions by adding context. However, the multiple turns in such systems are merely contextual continuations, not structured collaborative rounds, and lack: role division mechanisms; shared state modeling mechanisms; conflict identification and coordination mechanisms; facilitator scheduling structures; and decision convergence rules. Essentially, this type of technology belongs to dialogue enhancement systems, rather than structured collaborative decision-making systems.

[0011] Based on the above technical approaches, the following shortcomings of existing technologies can be summarized:

[0012] (1) Lack of a multi-round collaborative decision-making control structure;

[0013] (2) Lack of a unified, structured, shared intermediate state;

[0014] (3) Lack of conflict identification and focus scheduling mechanisms;

[0015] (4) Lack of decision convergence determination algorithm;

[0016] (5) Lack of a closed-loop automatic assessment system for medical risks;

[0017] (6) Lack of complete traceable decision trajectory modeling.

[0018] Therefore, in the context of complex multi-system medical records, there is an urgent need for a multi-departmental consultation method and system that can simulate the real MDT consultation process, has multi-round collaborative control capabilities, conflict identification capabilities, structured shared state modeling capabilities, and decision convergence and risk assessment mechanisms, so as to further improve the consultation effect and efficiency and provide medical professionals with high-quality diagnostic and treatment references. Summary of the Invention

[0019] This application aims to provide a multi-departmental consultation method and storage medium based on a multi-round collaborative decision-making mechanism, which can improve the effectiveness and efficiency of consultations and provide high-quality reference information for medical professionals. In particular, by constructing a multi-round convergent collaborative decision-making mechanism, a structured shared whiteboard, and a host scheduling mechanism, a controllable, traceable, and evaluable multi-departmental consultation system is realized, thereby making up for the structural deficiencies of existing technologies in medical collaborative decision-making.

[0020] In a first aspect, embodiments of this application provide a multi-departmental consultation method based on a multi-round collaborative decision-making mechanism, including:

[0021] Obtain the medical record information corresponding to the case to be analyzed, and extract the key points of the medical record from the medical record information;

[0022] Based on the key points of the medical record, a first target department is determined from multiple candidate departments, and the primary attending department is determined from the first target department;

[0023] Construct initial shared whiteboard information corresponding to the medical record information, wherein the initial shared whiteboard information includes at least basic factual information corresponding to the key points of the medical record;

[0024] In the first round of consultations, for each of the first target departments, first-round diagnostic reference information is generated based on the initial shared whiteboard information;

[0025] The structured fields of each of the first round of diagnosis and treatment reference information are extracted and processed, and the processing results are written into the shared whiteboard to obtain the updated shared whiteboard information;

[0026] Based on the first round of diagnostic reference information and the updated shared whiteboard information, determine whether to proceed to the next round of consultation;

[0027] When determining whether to proceed to the next round of consultation, the focus information for the next round of consultation is generated based on the conflict detection results in the processing results, and the second target department to participate in the next round of consultation is determined based on the focus information.

[0028] In the next round of consultation, for the current second target department in the second target department, the diagnosis and treatment reference information corresponding to the current second target department in the two adjacent rounds of consultation is compared at the field level to generate difference entries, and it is determined whether to continue to the next round of consultation based on the difference entries and the consultation end conditions.

[0029] When the consultation termination conditions are met, the consultation reference results are determined based on the diagnostic and treatment reference information obtained in each round of consultations and the final shared whiteboard information.

[0030] Secondly, embodiments of this application also provide a computer-readable storage medium storing a program or instructions that cause a computer to perform the steps of the multi-departmental consultation method based on a multi-round collaborative decision-making mechanism as described in any embodiment.

[0031] In summary, this application proposes a multi-departmental consultation method based on a multi-round collaborative decision-making mechanism. Through a cyclical mechanism of "independent initial judgment by multiple departments—conflict identification—chairmanship and scheduling—difference-driven update—automatic convergence," multiple departments gradually reach a stable conclusion through multiple rounds of interaction. Compared with the single-round conclusion generation method in existing technologies, this application explicitly introduces a chairmanship and convergence determination mechanism, enabling multiple departments to conduct targeted discussions around the conflict focus in different rounds. This avoids redundant generation or invalid reasoning, improves the completeness and stability of complex case analysis, enhances consultation effectiveness and efficiency, and provides high-quality reference information for medical professionals. Attached Figure Description

[0032] Figure 1 This is a flowchart of a multi-departmental consultation method based on a multi-round collaborative decision-making mechanism provided in an embodiment of this application. Detailed Implementation

[0033] The present application will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0034] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0035] Example 1

[0036] Figure 1 This is a flowchart illustrating a multi-departmental consultation method based on a multi-round collaborative decision-making mechanism, as provided in an embodiment of this application. See also... Figure 1 Specifically, it includes the following steps:

[0037] S110. Obtain the medical record information corresponding to the case to be analyzed, and extract the key points of the medical record from the medical record information.

[0038] Optionally, the system receives medical record information text or structured fields, cleans and segments the text, divides the content into basic information segments and examination and testing segments, and extracts key medical record points that can be used for departmental classification and reasoning (such as chief complaint, symptoms, laboratory indicators, imaging descriptions, etc.).

[0039] S120. Based on the key points of the medical record, determine the primary target department from multiple candidate departments, and then determine the primary attending department from the primary target department.

[0040] Optionally, the relevance between medical record information and each candidate department can be calculated based on the key points of the medical record. Then, the candidate departments with higher relevance (such as reaching a threshold) can be selected as the primary target departments for the first round of consultations, and the department with the highest relevance can be determined as the primary attending department.

[0041] For example, based on key points of the medical record, the primary target department for participating in the first round of consultations is determined from the candidate departments, including one or more of the following:

[0042] Based on the keywords included in the medical record key points and the corresponding keywords of each candidate department, the primary target department for the first round of consultations is determined from the candidate departments. For example, if the medical record key points include keywords A, B, and C, and candidate department one corresponds to keyword A, while candidate department two corresponds to keywords B and C, then candidate department two can be determined as the primary target department for the first round of consultations. In other words, the more candidate departments share the same keywords as those included in the medical record key points, the higher the relevance between them and the medical record information. Candidate departments can be ranked according to relevance, and the candidate department with the highest relevance is determined as the primary attending department. Multiple candidate departments with relevance greater than a threshold are designated as auxiliary departments to participate in the consultation. Therefore, the primary target department for the consultation includes both the primary attending department and auxiliary departments.

[0043] Based on the coding vectors corresponding to the key points of the medical record and the coding vectors corresponding to each candidate department, the primary target department for the first round of consultations is determined from the candidate departments. For example, the distance (e.g., Euclidean distance) between the coding vectors corresponding to the key points of the medical record and the coding vectors corresponding to each candidate department can be calculated. The closer the distance, the higher the relevance. Candidate departments with a distance less than a preset value can be selected as the primary target departments for the first round of consultations. Alternatively, the cosine similarity between the coding vectors corresponding to the key points of the medical record and the coding vectors corresponding to each candidate department can be calculated. The higher the cosine similarity, the higher the relevance. Candidate departments with a cosine similarity greater than a threshold can be selected as the primary target departments for the first round of consultations.

[0044] Using a classification model, based on key points of the medical records, the primary target department for the first round of consultations is determined from candidate departments. Specifically, the key points of the medical records are input into a trained classification model to obtain the probability of each candidate department being selected as the primary target department for the first round of consultations. Departments with higher probabilities are then selected as the primary target departments for the first round of consultations. The classification model can be a logistic regression model, a gradient boosting tree model, or a neural network model. The model output is the probability distribution of each department, and the top few departments can be selected as the primary target departments for the first round of consultations based on their probabilities.

[0045] When determining the primary target department for the first round of consultations using at least two of the above methods, the intersection of the primary target departments determined by different methods can be used as the final primary target department for the first round of consultations. Alternatively, each method can be executed sequentially, for example, first selecting some reference departments from the candidate departments through keyword matching, and then further selecting the primary target department from the reference departments by calculating the distance between the encoded vectors.

[0046] In some implementations, to reduce computational load, improve computational efficiency, and ensure that the primary target departments participating in the consultation are comprehensive yet not redundant, each primary target department participating in the consultation is necessary, thereby laying the foundation for improving treatment outcomes, the following steps are taken before determining the primary target departments participating in the first round of consultation based on the key points of the medical records: extracting keywords from the key points of the medical records and filtering out departments that clearly do not meet the criteria from the candidate departments based on the keywords. The ineligible departments can be pre-defined departments associated with the keywords. For example, extracting keywords representing age from the key points of the medical records; when the age is displayed as adult, filtering out pediatrics from the candidate departments, thereby reducing the number of candidate departments participating in subsequent calculations and avoiding interference from pediatrics in the selection of other departments, thus achieving the goals of reducing computational load, improving computational efficiency, and ensuring computational effectiveness. In this scenario, the departments associated with age (18 years and older) are pediatrics and neonatology, which are ineligible departments. Similarly, for example, keywords for the location can be extracted from key points in the medical record, and irrelevant departments can be filtered out from the candidate departments based on the extracted location.

[0047] The selection of primary target departments for the first round of consultations through the above method not only avoids the problem of missing diagnostic perspectives due to insufficient department selection, but also avoids the problem of increased reasoning costs and conflict noise due to selecting too many departments, thus providing a prerequisite for obtaining high-quality diagnostic and treatment reference information.

[0048] In general, based on the key points of the medical record, the primary target department is determined from multiple candidate departments, including: extracting at least one of the following from the key points of the medical record: age information, chief complaint information, symptom and sign information, and abnormal examination and test information; conducting preliminary screening of candidate departments according to preset filtering rules to exclude candidate departments that do not match the age characteristics or disease characteristics; calculating the correlation score between the preliminary screened candidate departments and the key points of the medical record; and determining the primary target department based on the correlation score.

[0049] For each candidate department after initial screening, a relevance score is calculated between it and the key points of the medical record, including one or more of the following: calculating the relevance score based on the matching degree between symptom keywords, abnormal examination keywords in the key points of the medical record and the keyword set of the candidate department; calculating the relevance score based on the similarity between the vector of the key points of the medical record and the profile vector of the candidate department; inputting the key points of the medical record into a pre-trained department classification model to obtain the predicted score corresponding to each candidate department as the relevance score.

[0050] In this context, a department profile is a digital description of the characteristics of a clinical department, containing information such as the department's core features, areas of expertise, and typical diseases, represented in a computer-processable form (vector). For example, a traditional department description like cardiology is "a department that treats heart disease," while the department profile vector for cardiology could be: [Heart disease: 0.95, Hypertension: 0.88, Coronary artery disease: 0.92, Chest pain: 0.85, Electrocardiogram abnormalities: 0.90]. There are various ways to generate department profiles, such as manually constructing them based on medical knowledge graphs or automatically learning them from historical medical record data; this embodiment does not limit these methods.

[0051] S130. Construct initial shared whiteboard information corresponding to medical record information. The initial shared whiteboard information shall include at least the basic factual information corresponding to the key points of the medical record.

[0052] Specifically, based on the key points of the medical record, an initial shared whiteboard information is generated according to the set template. The key points of the medical record (such as chief complaint, symptoms, laboratory indicators, imaging descriptions, etc.) are written into the base layer of the shared whiteboard in a structured manner (for example, setting separate chief complaint fields and symptom fields, and then filling the corresponding chief complaint information into the corresponding position of the chief complaint field, and filling the symptom information into the corresponding position of the symptom field, so as to facilitate subsequent reading operations). This information is used for reference by each primary target department and for reasoning and calculation.

[0053] S140. In the first round of consultations, for each primary target department, the first round of diagnostic and treatment reference information is generated based on the initial shared whiteboard information.

[0054] Specifically, in the first round of consultations, for each primary target department, reference information for the first round of diagnosis and treatment is generated based on key points of the medical record, initial shared whiteboard information, and prompt templates corresponding to the primary target department.

[0055] The data in the prompt templates differs depending on the primary target department. The data in these templates includes treatment guidance information, such as the department's focus, common disease range, and recommended examination strategies. The purpose of providing these prompt templates is to enable each department to analyze medical records from its own specialty perspective.

[0056] Specifically, a corresponding intelligent agent can be configured for each primary target department. Key points of the medical record, shared whiteboard information, and prompt templates are input into the intelligent agent, which then outputs diagnostic and treatment reference information for the corresponding primary target department. The intelligent agent can be implemented based on existing large language model instances or rule-driven reasoning modules. By applying different departmental prompt templates to the same medical record input, specialized diagnostic and treatment reference information for different departments is generated. In particular, the diagnostic and treatment reference information is required to be output in the form of structured fields, such as preliminary diagnosis, differential diagnosis, examination suggestions, treatment suggestions, and risk warnings. This is to facilitate subsequent processing and calculations, ensuring the accuracy and efficiency of the processing.

[0057] During the first round of consultation, each first-target department's intelligent agent can perform inference calculations in parallel or sequentially. Each first-target department's intelligent agent independently performs inference calculations and determines the corresponding diagnostic reference information. To achieve multi-round collaborative consultations and traceability, facilitate management, and constrain the output format of the department's intelligent agents, the diagnostic reference information output by the department's intelligent agents must adopt a consultation document structure, using unified structured fields, including at least: preliminary assessment for this round, differential diagnosis, examination recommendations and priorities (tests), treatment plan, high-risk signs and emergency treatment prompts (red_flags), key evidence rationale (must cite key points from medical records or examination indicators), triage (the department's suggested triage department), is_end (the department's judgment flag for whether the consultation can be ended), and delta (the point of change and reason relative to the previous round, applicable to the second and subsequent consultation rounds).

[0058] To ensure the independence of the first round of consultations, when determining the consultation documents for each primary target department, only the key points of the medical record and the basic information of the initial whiteboard for this round were referenced, without referring to the consultation results already generated by other departments. At the same time, an independent departmental prompt template was configured for each department. The prompt template contains information such as the diagnostic and treatment focus of the corresponding department's specialty, the scope of common diseases, and recommended examination strategies, so that each department can analyze the medical record from its specialty perspective.

[0059] Since each department cannot see the diagnostic and treatment reference information of other departments during the first round of consultations, multiple independent preliminary judgments from different specialties can be formed. This avoids the problem of "aggregation-based conformity" caused by sharing information in advance in subsequent discussions, which leads to a decrease in accuracy, and provides a basis for subsequent conflict identification and focus scheduling.

[0060] In some implementations, considering the instability of the outputs of departmental agents, if intermediate results from each department's agents cannot be reliably parsed during consultation, it can lead to multiple rounds of collaborative interruptions or error propagation. To address this issue, the output structure of each department's agents can be limited, adopting a unified, structured output pattern (e.g., a set of JSON keys) to reduce parsing difficulty and improve parsing efficiency and stability. Fallback logic is included during parsing, such as attempting to parse the entire JSON, extracting the first object, and clearly defining the `think` tag. For abnormal outputs, default values ​​can be used to ensure the reasoning process proceeds normally, rather than failing entirely. Field-level validity checks can also be added.

[0061] S150. Extract and process the structured fields of each of the first round of diagnosis and treatment reference information, and write the processing results into the shared whiteboard to obtain the updated shared whiteboard information.

[0062] Specifically, structured fields are extracted from each of the first round of diagnostic reference information, and deduplication, merging, and conflict detection are performed on the extracted structured fields to obtain the processing results. The processing results are then written to the shared whiteboard to obtain the updated shared whiteboard information.

[0063] The structured fields can specifically include: preliminary assessment, differential diagnosis, tests, treatment plan, red_flags (high-risk signs and emergency treatment prompts), rationale (key evidence), triage (triage pointer), and is_end (consultation end judgment flag).

[0064] Specifically, pre-defined structured fields are extracted from each of the first round of diagnostic reference information; these pre-defined structured fields are mapped to shared whiteboard entries, which include at least an entry identifier, an entry category (e.g., a set of diagnostic assumptions H, a set of examination suggestions T, and a set of risk warnings R), an entry content, and the department of origin; semantic deduplication, merging, and conflict detection are performed on shared whiteboard entries from different departments, and the processing results are written to the shared whiteboard to obtain updated shared whiteboard information.

[0065] Semantic deduplication, merging, and conflict detection are performed on shared whiteboard entries from different departments, including: merging shared whiteboard entries with the same semantics or similarity exceeding a preset threshold; recording the source department set and the number of supporting departments for each merged shared whiteboard entry; merging different evidence from multiple departments for the same shared whiteboard entry into a supporting evidence list for the shared whiteboard entry when different departments provide different evidence; prioritizing shared whiteboard entries based on the number of supporting departments, risk level, and clinical urgency; and performing consistency comparison on preset structured fields in the diagnostic and treatment reference information corresponding to different departments to identify at least one of diagnostic conflicts, examination conflicts, risk conflicts, and triage conflicts, and generating corresponding conflict detection results.

[0066] Optionally, after each department completes the first round of consultation output, structured elements are extracted from the consultation documents generated by each department. These elements are then semantically deduplicated, merged, and conflict detected. The processing results are then written to a shared whiteboard to update the initial shared whiteboard information. The structured elements include at least a set of diagnostic hypotheses H, a set of examination suggestions T, and a set of risk warnings R. In other words, the diagnostic reference information output by each department is uniformly mapped to three core elements: the set of diagnostic hypotheses H, the set of examination suggestions T, and the set of risk warnings R. By constructing a structured shared whiteboard model containing "the set of diagnostic hypotheses H—the set of examination suggestions T—the set of risk warnings R," cross-departmental information aggregation and conflict alignment are achieved. This enables information compression, state transfer, and consensus tracking in multi-round reasoning, resolving the issues of context loss of control and reasoning drift in multi-agent collaborative consultation systems.

[0067] Before writing the extracted structured elements to the shared whiteboard, elements from different departments are deduplicated, merged, and sorted according to preset aggregation rules. Specifically, this includes:

[0068] Semantic deduplication rule: Merge elements from different departments that are semantically identical or highly similar. Source tagging rule: Record the department that proposed the element and the total number of departments that proposed it (i.e., the total number of supporting departments) for each element to reflect the degree of cross-departmental consensus. Priority ranking rule: Rank elements according to the total number of supporting departments, risk level, and clinical urgency. For example, risk warning elements with acute and critical illness characteristics have a higher priority than risk warning elements without such characteristics.

[0069] Supplementary information merging rule: When multiple departments provide different evidence for the same diagnostic hypothesis, the relevant evidence will be merged into a list of supporting evidence for that element.

[0070] Furthermore, a confidence level can be set for each element recorded on the shared whiteboard. This confidence level can be quantified based on the total number of departments supporting the element and / or the support provided by each department for the element in two consecutive rounds of consultations. For example, if a total of 5 primary target departments participate in the consultation, and 3 primary target departments output information for a certain element (meaning 3 departments support the element), then the confidence level for that element can be set to 60%. Alternatively, if a total of 5 primary target departments participate in the consultation, and 3 primary target departments output information for a certain element in both consecutive rounds of consultations (meaning 3 departments support the element), then the confidence level for that element can be set to 80%.

[0071] Setting confidence levels helps guide the consultation departments in the next round to output more accurate diagnostic and treatment reference information.

[0072] Furthermore, after uniformly mapping the diagnostic and treatment reference information output by each department into three core elements—a set of diagnostic hypotheses H, a set of examination suggestions T, and a set of risk warnings R—diagnostic and treatment conflicts between departments are determined based on the statistical core elements. For example, if Department 1 believes in diagnostic hypothesis a and Department 2 believes in diagnostic hypothesis b, and the two are inconsistent, then a diagnostic and treatment conflict (specifically, a conflict between diagnostic hypotheses) is marked between Department 1 and Department 2. The corresponding diagnostic hypotheses are then associated with each department and recorded in the form of structured fields in the shared whiteboard.

[0073] Optionally, when multiple departments have inconsistencies in their diagnostic assumptions, examination recommendations, or risk warnings, these can be identified as conflicting relationships, and corresponding conflict records can be generated and filled into the shared whiteboard.

[0074] In general, a shared whiteboard includes at least the following layers: a basic fact layer (initial information determined based on key points of the medical record, such as the patient's age and chief complaint), a diagnostic hypothesis layer (recording the diagnostic hypotheses output by each participating department), an examination suggestion layer (recording the examination suggestions output by each participating department), a risk warning layer, and a conflict layer (recording diagnostic conflicts between different departments). Each layer consists of specific entries; for example, a specific diagnostic hypothesis is a specific entry. Each entry includes an identity ID, source department, round index, evidence anchor, number of supporting departments, and a status marker (e.g., whether it was added in this round of consultation, retained from the previous round, modified, invalid, or conflicting).

[0075] The shared whiteboard proposed in this embodiment realizes cross-departmental information aggregation and conflict alignment, and serves as the input for each participating department in the next round of consultation, enabling each department to focus on the diagnostic and treatment conflicts that existed in the previous round of consultation. It realizes information compression, state transfer and consensus tracking in multi-round reasoning, and solves the problems of context loss of control and reasoning drift in multi-agent collaborative consultation systems.

[0076] S160. Based on the first round of diagnostic reference information and the updated shared whiteboard information, determine whether to proceed to the next round of consultation.

[0077] The consultation proceeds to the next round when one or more of the following conditions are met: the similarity between the first round of diagnostic reference information is less than a first threshold; the degree of change between the updated shared whiteboard information and the unupdated shared whiteboard information is greater than a second threshold; the proportion of diagnostic reference information containing preset risk warnings does not reach a third threshold; the difference in structured fields in the shared whiteboard information between two adjacent consultation rounds is calculated, and the consultation convergence index is determined by combining the weights of different structured fields; the consultation proceeds to the next round when the consultation convergence index does not meet the termination conditions.

[0078] In other words, a decision to proceed to the next round of consultations is made when one or more of the following conditions are met:

[0079] The similarity between the diagnostic reference information corresponding to each primary target department is less than the first threshold (indicating significant discrepancies in diagnosis and treatment among departments); the degree of change between the updated shared whiteboard information and the previous shared whiteboard information is greater than the second threshold (indicating significant changes in the diagnosis and treatment results in this consultation round compared to the previous round, requiring further consultation to confirm whether the results will continue to change significantly); the proportion of diagnostic reference information containing preset risk warnings does not reach the third threshold (i.e., there is no consensus among departments regarding risk warnings, requiring further consultation); the difference in structured fields in the shared whiteboard information between adjacent consultation rounds is calculated, and the consultation convergence index is determined by combining the weights of different structured fields. When the consultation convergence index meets the conditions, assigning different weights to different structured fields can highlight important diagnostic and treatment information and weaken less important information, thereby accelerating convergence without affecting the quality of the diagnosis and treatment results.

[0080] In general, if there are still unresolved high-priority conflict points, no consensus on key risk warnings, no consensus on key examination recommendations, or differences in the final triage opinions among different departments exceed a preset threshold, then need_next_round will be set to True to determine that the next round of consultation is required.

[0081] If most departments have `is_end` set to `True` (meaning the consultation can be ended), and the rate of change of the diagnostic hypothesis set in two consecutive rounds is lower than a preset threshold, and there are no high-priority unresolved conflicts, then `need_next_round` is set to `False` and the consultation is terminated. To avoid invalid loops, a duplicate decision detection rule can be set. When the focus information `focus_topics` is basically the same in two consecutive rounds and the number of new elements added to the shared whiteboard is lower than a threshold, a forced termination is triggered.

[0082] Identify the second target departments that need to participate in the next round of consultations. This includes: identifying the first target departments that are associated with the information of concern as the second target departments that need to participate in the next round of consultations; and identifying clinics that hold different viewpoints (reflected in inconsistent or contradictory elements output by different clinics) as the second target departments that need to participate in the next round of consultations.

[0083] Optionally, a second target department to participate in the next round of consultation can be determined based on the department that provides the elements related to the conflict focus, the primary attending department, and the department's contribution to information in the current round. Generally, at least the primary attending department should be retained, and departments directly related to high-priority conflict focuses, offering opposing judgments, or providing key supporting evidence should be prioritized for the next round. Departments that have not provided any new and valid information in this round and are not directly related to the current focus will not be included in the next round to reduce redundant reasoning overhead. The primary attending department is the department with the highest relevance to the key points of the medical record.

[0084] Optionally, the focus information for the next round of consultation can be determined based on the conflict records. For example, the conflict records can be identified as the focus information for the next round of consultation, or the conflict records can be semantically summarized and the semantic summary can be identified as the focus information for the next round of consultation.

[0085] S170. When determining to proceed to the next round of consultation, generate focus information for the next round of consultation based on the conflict detection results in the processing results, and determine the second target department to participate in the next round of consultation based on the focus information.

[0086] The second target department should include at least the primary outpatient department.

[0087] Specifically, the departments that are associated with the focus information (such as the two departments that raised opposing opinions) and the primary attending department will be identified as the second target departments.

[0088] S180. In the next round of consultation, for the current second target department in the second target department, the corresponding diagnosis and treatment reference information of the current second target department in the two adjacent rounds of consultation is compared at the field level to generate difference entries, and based on the difference entries and the consultation end conditions, it is determined whether to continue to the next round of consultation.

[0089] Specifically, in the next round of consultation, for the current second target department among the second target departments, based on the key points of the medical record, the updated shared whiteboard information, the corresponding diagnosis and treatment reference information of the current second target department in the previous round of consultation, the focus information, and the prompt template corresponding to the current second target department, the corresponding diagnosis and treatment reference information for the current second target department in the current round will be generated.

[0090] Unlike the first round where each department's AI agent makes an independent initial judgment, in the second and subsequent rounds of consultation, the department's AI agent no longer makes a diagnosis solely based on the key points of the original medical record. Instead, it simultaneously receives: the department's own consultation document from the previous round, information on key concerns, and updated shared whiteboard information (including conflict records). Based on this, the department's AI agent performs a difference-driven update, outputting a new structured consultation document and explicitly providing the changes relative to the previous round's conclusions and the corresponding reasons for these changes.

[0091] Optionally, in addition to requiring the department agent to explicitly provide the changes and reasons for the previous round's conclusions, an additional comparison step can be added to determine the changes in the current round's conclusions relative to the previous round. For example, the field content in the consultation documents output by the same department agent in two consecutive rounds can be compared (e.g., preliminary assessment, differential diagnosis list, examination recommendations and priorities, treatment plan, high-risk signs and emergency treatment prompts (red_flags), key basis rationale (must cite key points of the medical record or examination indicators), the triage suggested by the department, the department's judgment flag for whether the consultation can be ended (is_end), and the changes and reasons (delta) relative to the previous round). This comparison determines whether the following types of changes have occurred in each field: addition, deletion, priority adjustment, conclusion correction, or basis supplementation. When a change is detected, it is recorded as a delta entry (i.e., a change point). Each delta entry includes at least the changed field, the content before the change, the content after the change, and the reason for the change. The reasons for the changes can come from at least one or more of the following: information on the focus of attention, newly added inspection suggestions or risk warnings in the shared whiteboard, opposing opinions corresponding to the focus of conflict, and new evidence generated by the agent's reasoning in this round.

[0092] In the second and subsequent rounds of consultations, the department's intelligent agent first reads the consultation document output from the previous round as the historical state; secondly, it reads the focus of attention information and conflicting points in the shared whiteboard; then, based on the read information, it generates the consultation document for the current round. Afterwards, a field-level comparison can be performed between the consultation document output for the current round and the historical state to determine the difference delta and the reasons for the changes, and this information is written to the historical record field. Through this mechanism, the output of the department's intelligent agent in the second and subsequent rounds is no longer a repetitive generation of the original medical record, but rather a targeted correction based on conflicting points and newly added evidence, thereby achieving difference-driven multi-round collaborative discussions.

[0093] By using a shared whiteboard and conflict focus marking mechanism, the contradictions in the opinions of multiple departments are automatically identified and recorded. The next round of consultation is guided by a chairing and scheduling mechanism to discuss the contradictions, thereby improving the effectiveness of multi-round collaborative diagnosis.

[0094] Furthermore, to prevent the departmental AI agent from mechanically repeating the conclusions of the previous round in the second and subsequent rounds of consultation, this embodiment can also set difference constraint rules. For example, if the consultation document output in this round is completely identical to the consultation document output in the previous round in terms of core fields, and no basis for addition or reason for change is given, the consultation result of this round is marked as an invalid increment and used for subsequent convergence determination. Thus, a structured mechanism based on "change point - change reason" can be used to promote the gradual convergence of multiple rounds of consultation.

[0095] The system performs field-level comparisons of the diagnostic and treatment reference information for the current second target department in two adjacent rounds of consultations, generates difference entries, and determines whether to continue to the next round of consultations based on the difference entries and the consultation termination conditions.

[0096] Specifically, the structured fields corresponding to the consultations in two adjacent rounds are compared to identify at least one type of change, including additions, deletions, priority adjustments, conclusion revisions, or supplementary evidence. When at least one type of change is detected, a corresponding difference entry is generated. Each difference entry includes at least the changed field, the content before the change, the content after the change, and the reason for the change. The reason for the change comes from at least one of the following: focus information, newly added examination suggestions or risk warnings in the shared whiteboard, opposing opinions corresponding to conflict records, or new evidence generated in this round of reasoning.

[0097] For example, when the number of discrepancies is small, it means that the diagnostic reference information of the second target department is becoming stable. In order to avoid duplicate or ineffective consultations, it can be determined that the second target department does not need to continue to participate in subsequent consultation rounds, so as to avoid wasting computing resources.

[0098] Optionally, if the diagnostic and treatment reference information output by the current second target department in the current round is completely identical to the diagnostic and treatment reference information output in the previous round in terms of core structured fields, and does not contain any new basis or reason for change, the output of the current round will be marked as invalid increment. Invalid increment will be used as one of the criteria for determining the end of the consultation, and the consultation will be terminated when any of the following conditions are met:

[0099] The consultation completion flag is_end is true for most departments, and the number of new entries in the shared whiteboard is less than the preset threshold; the rate of change of the diagnostic hypothesis set in two consecutive rounds is less than the preset threshold; the focus information in two consecutive rounds is basically the same and the number of new entries in the shared whiteboard is less than the preset threshold; the preset maximum number of rounds is reached.

[0100] Optionally, the consultation may proceed to the next round if one or more of the following conditions are not met:

[0101] There are unresolved high-priority conflict records; different departments have not reached a consensus on key risk warnings; different departments have not reached a consensus on key examination recommendations; the differences in the final triage opinions of different departments exceed the preset threshold; the consultation convergence index does not meet the termination conditions; among them, the consultation convergence index is calculated based on the structured field difference degree of shared whiteboard information and / or diagnosis and treatment reference information in two adjacent rounds of consultation.

[0102] S190. When the conditions for the end of the consultation are met, the treatment reference results are determined based on the treatment reference information obtained in each round of consultations and the final shared whiteboard information.

[0103] Specifically, the diagnostic conclusion set is retrieved from the latest shared whiteboard information; candidate diagnoses are determined from the set based on risk level and the number of departments supporting the diagnosis; and candidate diagnoses that have appeared multiple times (multiple times means two or more times) in historical consultations are identified as reference results for treatment. For example, diagnoses with higher risk levels and supported by more departments are identified as candidate diagnoses.

[0104] Specifically, after the consultation concludes, the final consultation document and shared whiteboard are retrieved, and the results of multiple rounds of consultations are aggregated in a structured manner to generate a final summary (i.e., treatment reference results). The summary generation process is not a simple patchwork of the outputs from each round, but rather a stable conclusion is obtained through a comprehensive analysis of the aggregated information on the whiteboard and the consultation documents from multiple rounds.

[0105] Specifically, the final set of diagnostic hypotheses, recommended examinations, and risk warnings are retrieved from the shared whiteboard and sorted according to the number of supporting departments, risk level, and clinical priority to obtain a set of candidate conclusions. Then, combining this with multi-round consultation documents from historical records, a support index is calculated for each candidate conclusion. The support index includes at least the number of supporting departments, cross-round stability, and the presence of conflicting records. When a conclusion remains stable and without conflicting associations in two consecutive rounds, that conclusion is determined as the final diagnostic tendency or final recommendation item.

[0106] When unresolved conflicts are detected between different departments, decisions are made based on the strength of supporting evidence, the level of risk involved, and the priority of the conflict focus. Specifically, one or a combination of the following methods may be used: prioritize retaining conclusions that support more departments and have a higher risk level; retain multiple possible diagnoses in the summary and mark them as differential diagnoses; based on key examination suggestions in the shared whiteboard, transform the conflict into further examination suggestions, such as differential diagnosis through imaging or laboratory tests.

[0107] After completing the above aggregation and conflict resolution, a structured consultation summary is generated according to a preset template. The preset template includes at least the following parts:

[0108] Doctor's Summary: Includes diagnostic tendencies, differential diagnoses, recommended examinations, treatment plans, critical value alerts, and follow-up points. Patient's Summary: Translates the doctor's conclusions into plain language and provides daily precautions and suggestions. Final Triage Recommendations: Generates recommended departments based on the primary attending department and risk alerts, indicating whether emergency room visits or further referrals are advised.

[0109] When generating final triage recommendations, priority is given to the primary attending department and high-priority risk alerts on the shared whiteboard. When a risk alert related to acute and critical illness (which is a high-priority risk alert) is detected, an emergency indication flag can be automatically triggered, and a recommendation to prioritize emergency care will be given in the summary.

[0110] Through the above-mentioned summary generation mechanism, information from multiple rounds of consultations can be aggregated into stable and structured diagnostic conclusions, while retaining reasonable differential diagnostic paths when there are disagreements, thereby ensuring that the final summary is both consistent and clinically interpretable.

[0111] The multi-departmental consultation method based on a multi-round collaborative decision-making mechanism provided in this application uses a cyclical mechanism of "independent initial judgment by multiple departments—conflict identification—chairmanship and scheduling—difference-driven update—automatic convergence" to enable multiple departments to gradually form stable conclusions through multiple rounds of interaction. Compared with the single-round conclusion generation method in the prior art, this application explicitly introduces a chairmanship and convergence determination mechanism, enabling multiple departments to conduct targeted discussions around the focus of conflict in different rounds, thereby avoiding repeated generation or invalid reasoning, improving the completeness and stability of complex case analysis, enhancing consultation effectiveness and efficiency, and providing high-quality reference for medical professionals.

[0112] Example 2

[0113] Based on the above embodiments, this embodiment further adds a risk warning step. After obtaining the diagnostic reference results, a risk warning is issued when one or more of the following situations are detected, in order to assist medical staff in raising their vigilance, ensuring the accuracy of diagnosis and treatment, and significantly improving the ability to identify medical risks.

[0114] The latest shared whiteboard information does not include the diagnostic reference results (there may be errors in information extraction, aggregation, or other processing); the diagnostic reference results do not include the examination information associated with the key points of the medical record (there may be diagnostic errors).

[0115] Specifically, after obtaining the diagnostic and treatment reference results, a multi-dimensional risk assessment is conducted on these results. The assessment process can be based on a pre-defined medical rule base, a set of risk warnings, and aggregated information from a shared whiteboard, performing consistency checks and safety analyses on the diagnostic and treatment reference results. For example, it may include at least four categories: error risk assessment, omission risk assessment, potential harm risk assessment, and bias risk assessment.

[0116] Regarding error risk assessment, the diagnostic bias, examination recommendations, and treatment plans in the final treatment reference results are compared for consistency with the key points of the medical record and the critical evidence recorded on the shared whiteboard. When the treatment reference results significantly mismatch the core symptoms or key examination results, or when the treatment plan logically contradicts the diagnostic assumption, this situation is marked as a potential error risk. For example, if the medical record shows obvious characteristics of hypoxia or respiratory distress but the treatment reference results do not reflect relevant treatment recommendations, an error risk warning will be triggered.

[0117] Regarding the risk assessment of omissions, a comparative analysis is performed on key points in the medical record, the risk warning set in the shared whiteboard, and the diagnostic and treatment reference results to determine whether any critical examinations or key diagnostic and treatment steps were not included in the final diagnostic and treatment reference results. When high-risk symptoms or signs are detected in the medical record text, but the final diagnostic and treatment reference results do not suggest corresponding examinations or risk warnings, it will be recorded as an omission risk. For example, if a case presents with persistent high fever and elevated infection indicators, but the diagnostic and treatment reference results do not suggest infection-related examinations or anti-infection assessments, the system will mark it as a missed critical examination.

[0118] In terms of potential harm risk assessment, the risk of harm from potential medical procedures is evaluated based on the diagnostic and treatment reference results and risk warnings. A medical safety rule base can be used to determine the risk level of potential delays in treatment, misdiagnosis, or dangerous procedures that may result from inappropriate advice.

[0119] In terms of bias risk assessment, the examination checks whether the clinical reference results have potential bias in specific populations or scenarios with missing information. Bias detection rules include age-related bias, gender-related bias, and overly certain judgments under conditions of insufficient information. When medical record information is missing or there are special population characteristics (such as elderly patients, children, or pregnant women), the examination checks whether the clinical reference results contain corresponding uncertainty warnings or supplementary examination suggestions; if not, it is marked as a risk of bias.

[0120] After completing the above assessments, structured scoring results are generated based on different risk dimensions. The scoring results include at least an overall safety score, scores for each risk type, and corresponding explanations, and are output in structured form to the results storage module for subsequent quality control analysis and system optimization.

[0121] Through the above mechanism, the results of multiple rounds of consultations can be automatically assessed for safety, thereby forming a closed-loop medical decision-making quality control process.

[0122] Furthermore, the consultation process can be output in real time in the form of events by the event flow output module, including events such as round start, department incremental output, department end, chair decision, consultation end and final summary; at the same time, the traceable storage module persistently saves the consultation status, shared whiteboard snapshot, chair decision, final summary and scoring results, realizing full process auditability and traceability.

[0123] Compared with existing single-model, single-round diagnosis and treatment systems, multi-agent simple aggregation systems, and multi-round dialogic systems, this application achieves significant technical improvements in diagnostic completeness, system stability, resource efficiency, risk control capabilities, and traceability by constructing a multi-round convergent collaborative decision-making architecture, a structured shared whiteboard, and a host scheduling and convergence control mechanism. The specific improvements are as follows:

[0124] (1) Significantly improve the diagnostic completeness of complex cases

[0125] By employing a multi-round mechanism of "independent initial assessment by multiple departments + discussion driven by differences," the problem of information omission caused by decision-making from a single perspective was avoided.

[0126] Comparative experiments were conducted on a test dataset (sample size N=500) containing cases of multi-system comorbidity. The results showed that the key diagnostic point coverage of the single-round model system was approximately 68%; that of the multi-agent model parallel aggregation system was approximately 74%; and that of the multi-round convergent system proposed in this application reached 88%–92%. Compared with the single-round system, the key point omission rate was reduced by approximately 25%–35%.

[0127] (2) Effectively improve conflict identification and coordination consistency

[0128] This application utilizes a shared whiteboard and conflict focus marking mechanism to automatically identify and record contradictory points in opinions from multiple departments, and guides focused discussions through a moderation mechanism. Test results show that the automatic conflict identification rate of traditional multi-agent models is approximately 40%; the conflict identification rate of this application reaches over 93%; and the stability rate of the diagnostic hypothesis reaches over 85% after two consecutive rounds. Therefore, this application significantly improves the consistency and controllability of multi-agent collaborative decision-making.

[0129] (3) Achieve automatic decision convergence and reduce resource consumption.

[0130] By introducing convergence criteria (including diagnostic hypothesis change thresholds, duplicate decision detection, and maximum round count control), invalid rounds can be automatically terminated. Experimental data shows that the average number of rounds in the non-convergent control system is 3.8 rounds; the average number of rounds in the proposed system is controlled at 2.1–2.4 rounds; average inference resource consumption is reduced by approximately 32%–45%; and average latency is reduced by approximately 28%. These technical results demonstrate that the proposed solution improves computational efficiency while maintaining decision quality.

[0131] (4) Significantly improve the ability to identify medical risks

[0132] By using a pre-set medical rule base, a set of risk warnings, and aggregated information from a shared whiteboard, consistency testing and safety analysis of diagnostic and treatment reference results can be performed, effectively identifying high-risk factors.

[0133] This enhanced the system's medical safety assurance capabilities.

[0134] (5) Construct a complete and traceable decision-making trajectory

[0135] This application will structure and store the departmental opinions, conflict points, decision-making process, shared whiteboard evolution process and final results for each round, forming a complete decision-making trajectory.

[0136] Compared to existing systems that only store the final results, this application has: full-cycle traceability; the ability to locate conflict generation paths; the ability to analyze the contribution ratio of each department; and support for medical liability audits and quality reviews. Traceability completeness reaches 100%.

[0137] (6) Enhance the realism of multidisciplinary consultation scenarios

[0138] To avoid the uncertainty of subjective human scoring, this application uses a large-parameter language model (with a parameter scale of over 10 billion) as a third-party evaluation model to comprehensively score the consultation results generated by intelligent agents from different departments based on structural consistency, logical integrity, and medical rationality.

[0139] Under a unified test dataset (sample size N=300), the evaluation results using the same prompt template are as follows:

[0140] The average score for a single-model system is 6.1 out of 10; the score for a simple summary system of multi-agent models is 6.8; and the score for the collaborative system proposed in this application is 8.4.

[0141] In addition to the methods described above, embodiments of this application may also be computer program products, which include computer program instructions. When executed by a processor, the computer program instructions cause the processor to perform the steps of the multi-departmental consultation method based on a multi-round collaborative decision-making mechanism provided in any embodiment of this application.

[0142] Computer program products can be written in any combination of one or more programming languages ​​to perform the operations of the embodiments of this application. The programming languages ​​include object-oriented programming languages ​​such as Java and C++, as well as conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the user's computing device, partially on the user's computing device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0143] Furthermore, embodiments of this application may also be computer-readable storage media storing computer program instructions, which, when executed by a processor, cause the processor to perform the steps of the multi-departmental consultation method based on a multi-round collaborative decision-making mechanism provided in any embodiment of this application.

[0144] Computer-readable storage media may take the form of any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may, for example, include, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or devices, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: electrical connections having one or more wires, portable disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0145] It should be noted that the terminology used in this application is for the purpose of describing specific embodiments only and is not intended to limit the scope of this application. As shown in the specification and claims of this application, unless the context clearly indicates otherwise, words such as "a," "an," "an," and / or "the" do not specifically refer to the singular and may also include the plural. The terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, or apparatus that includes the element.

[0146] It should also be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation on this application. Unless otherwise expressly specified and limited, the terms "installed," "connected," "linked," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication between two elements. For those skilled in the art, the specific meaning of the above terms in this application can be understood according to the specific circumstances.

[0147] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this application. The above are only preferred embodiments of this application. It should be noted that due to the limitations of textual expression, while there are objectively infinite specific structures, those skilled in the art can make several improvements, modifications, or changes without departing from the principles of this invention, and can also combine the above technical features in an appropriate manner. These improvements, modifications, changes, or combinations, or the direct application of the inventive concept and technical solution to other situations without modification, should all be considered within the scope of protection of this application.

Claims

1. A multi-departmental consultation method based on a multi-round collaborative decision-making mechanism, characterized in that, include: Obtain the medical record information corresponding to the case to be analyzed, and extract the key points of the medical record from the medical record information; Based on the key points of the medical record, a first target department is determined from multiple candidate departments, and the primary attending department is determined from the first target department; Construct initial shared whiteboard information corresponding to the medical record information, wherein the initial shared whiteboard information includes at least basic factual information corresponding to the key points of the medical record; In the first round of consultations, for each of the first target departments, first-round diagnostic reference information is generated based on the initial shared whiteboard information; The structured fields of each of the first round of diagnosis and treatment reference information are extracted and processed, and the processing results are written into the shared whiteboard to obtain the updated shared whiteboard information; Based on the first round of diagnostic reference information and the updated shared whiteboard information, determine whether to proceed to the next round of consultation; When determining whether to proceed to the next round of consultation, the focus information for the next round of consultation is generated based on the conflict detection results in the processing results, and the second target department to participate in the next round of consultation is determined based on the focus information. In the next round of consultation, for the current second target department in the second target department, the diagnosis and treatment reference information corresponding to the current second target department in the two adjacent rounds of consultation is compared at the field level to generate difference entries, and it is determined whether to continue to the next round of consultation based on the difference entries and the consultation end conditions. When the consultation termination conditions are met, the consultation reference results are determined based on the diagnostic and treatment reference information obtained in each round of consultations and the final shared whiteboard information.

2. The multi-departmental consultation method based on a multi-round collaborative decision-making mechanism according to claim 1, characterized in that, The process of determining the first target department from multiple candidate departments based on the key points of the medical record includes: Extract at least one of the following from the key points of the medical record: age information, chief complaint information, symptom and sign information, and abnormal examination and test information, and obtain the extraction results; Based on the extraction results and preset filtering rules, the candidate departments are initially screened to exclude those that do not match the extraction results; the relevance scores between the initially screened candidate departments and the key points of the medical records are calculated; and the first target department is determined based on the relevance scores.

3. The multi-departmental consultation method based on a multi-round collaborative decision-making mechanism according to claim 2, characterized in that, The relevance score between the candidate departments after initial screening and the key points of the medical records is calculated, including one or more of the following: A relevance score is calculated based on the degree of matching between symptom keywords, abnormal examination keywords in the medical record key points and the candidate department keyword set; The relevance score is calculated based on the similarity between the medical record key point vector and the candidate department profile vector; The key points of the medical record are input into a pre-trained department classification model to obtain the predicted score for each candidate department as the relevance score.

4. The multi-departmental consultation method based on a multi-round collaborative decision-making mechanism according to claim 1, characterized in that, The step of extracting and processing structured fields from each of the first round of diagnostic reference information, and writing the processing results into the shared whiteboard to obtain updated shared whiteboard information includes: Extract pre-defined structured fields from each of the first round of diagnostic reference information; The preset structured fields are mapped to shared whiteboard entries, and the shared whiteboard entries include at least an entry identifier, an entry category, an entry content, and a source department; Semantic deduplication, merging, and conflict detection are performed on shared whiteboard entries from different departments, and the processing results are written to the shared whiteboard to obtain the updated shared whiteboard information.

5. The multi-departmental consultation method based on a multi-round collaborative decision-making mechanism according to claim 4, characterized in that, The semantic deduplication, merging, and conflict detection processing of shared whiteboard entries from different departments includes: Shared whiteboard entries with the same semantics or similarity exceeding a preset threshold are merged; Record the source department set and the number of supporting departments for each merged shared whiteboard entry; When multiple departments provide different evidence for the same shared whiteboard item, the different evidence will be combined into a list of supporting evidence for the shared whiteboard item. The shared whiteboard items are prioritized based on the number of supporting departments, risk level, and clinical urgency. The system performs consistency comparisons on preset structured fields in the diagnostic and treatment reference information corresponding to different departments to identify at least one of diagnostic conflicts, examination conflicts, risk conflicts, and triage conflicts, and generates corresponding conflict detection and processing results.

6. The multi-departmental consultation method based on a multi-round collaborative decision-making mechanism according to claim 1, characterized in that, The step of determining whether to proceed to the next round of consultation based on the first round of diagnostic reference information and the updated shared whiteboard information includes: determining to proceed to the next round of consultation when one or more of the following conditions are met: The similarity between each of the first round of diagnostic reference information is less than a first threshold; The degree of change in the updated shared whiteboard information compared to the previous shared whiteboard information is greater than the second threshold; The proportion of diagnostic and treatment reference information that includes preset risk warnings has not reached the third threshold; Calculate the difference in structured fields in the shared whiteboard information between two adjacent consultation rounds, and determine the consultation convergence index by combining the weights of different structured fields. When the consultation convergence index does not meet the termination condition, the consultation is terminated.

7. The multi-departmental consultation method based on a multi-round collaborative decision-making mechanism according to claim 1, characterized in that, The process of determining the second target department to participate in the next round of consultations based on the aforementioned focus information includes: The source department associated with the information of concern and the primary attending department are identified as the second target department.

8. The multi-departmental consultation method based on a multi-round collaborative decision-making mechanism according to claim 1, characterized in that, The process of determining the diagnostic reference results based on the diagnostic reference information obtained from each round of consultations and the final shared whiteboard information includes: Read the set of diagnostic conclusions from the latest shared whiteboard information; Candidate diagnostic conclusions are determined from the set of diagnostic conclusions based on the risk level and the number of departments supporting the diagnostic conclusion. Candidate diagnostic conclusions that have appeared multiple times in previous rounds of consultations will be identified as reference results for diagnosis and treatment.

9. The multi-departmental consultation method based on a multi-round collaborative decision-making mechanism according to claim 1, characterized in that, The process of comparing the diagnostic and treatment reference information of the current second target department in two adjacent rounds of consultations at the field level to generate difference entries includes: Compare the corresponding structured fields in two adjacent rounds of consultation to identify at least one type of change, including addition, deletion, priority adjustment, conclusion correction, or basis supplementation. When at least one type of change is detected, a corresponding difference entry is generated; Each difference entry includes at least the changed field, the content before the change, the content after the change, and the reason for the change. The reason for the change comes from at least one of the following: focus information, newly added inspection suggestions or risk warnings in the shared whiteboard, opposing opinions corresponding to conflict records, and new evidence generated in this round of reasoning.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program or instructions that cause a computer to perform the steps of the multidisciplinary consultation method based on a multi-round collaborative decision-making mechanism as described in any one of claims 1 to 9.

Images