Hybrid human-machine intelligent multi-disciplinary consultation method and system based on a Delphi matrix

Abstract

The application provides a kind of man-machine hybrid intelligent multidisciplinary consultation method and system based on research matrix, it is related to medical artificial intelligence and man-machine collaborative decision-making technical field, the method comprises: based on the target disease of consultation patient, collection multi-source heterogeneous clinical data;Multi-source heterogeneous clinical data include the dialogue data stream of consultation expert and the clinical data stream of consultation patient;Based on multi-source heterogeneous clinical data, structured speech text in dialogue data stream is mapped to the initial consultation research matrix pre-constructed, generates target consultation research matrix;Based on target consultation research matrix and monte carlo tree search algorithm, generate consultation guide path;Consultation guide path is used to guide consultation expert to carry out consultation discussion to target disease.The man-machine hybrid intelligent multidisciplinary consultation method and system based on research matrix provided by the application can correct the divergent trajectory of expert group in real time, ensure that research converges to key decision point, improve multidisciplinary expert consultation efficiency.

Description

Technical Field

[0001] This invention relates to the field of medical artificial intelligence and human-machine collaborative decision-making technology, and in particular to a human-machine hybrid intelligent multidisciplinary consultation method and system based on a discussion matrix. Background Technology

[0002] Multidisciplinary team (MDT) consultation is a standard treatment model for complex diseases such as malignant tumors and rare diseases. It aims to integrate fragmented clinical information through multidisciplinary collaboration to form the best decision.

[0003] Current MDT practices suffer from the following shortcomings: First, clinical data is heterogeneous and fragmented. Vital signs, test results, imaging features, and pathological molecular data belong to independent systems such as the Hospital Information System (HIS), Laboratory Information System (LIS), and Picture Archiving and Communication System (PACS), and are presented in fragmented ways such as oral reports and screen sharing, lacking a unified semantic integration framework, making it difficult for experts to conduct a comprehensive analysis within a limited time. Second, there are significant differences in the cognitive frameworks of different specialties. For example, radiology, pathology, and clinical departments make judgments based on professional terminology systems such as imaging features, histological morphology, and clinical symptoms, and the interpretation of the same clinical problem often contains implicit contradictions, but the existing process lacks a systematic conflict detection and prompting mechanism. Third, high-value ideas emerging in interdisciplinary discussions (such as rare disease associations and potential drug interaction risks) are easily overlooked due to the linear process and cannot be incorporated into decision-making. Fourth, the consultation process lacks intelligent convergence guidance, and discussions are prone to excessive divergence or jumping on non-critical issues, resulting in insufficient argumentation of core decisions. Due to the aforementioned shortcomings, existing MDTs are inefficient. Summary of the Invention

[0004] This invention provides a human-machine hybrid intelligent multidisciplinary consultation method and system based on a discussion matrix, which solves the technical problems of low efficiency in multidisciplinary expert consultations caused by the difficulties in integrating clinical data, the difficulty in aligning cross-specialty cognition, the tendency for discussions to diverge, the easy loss of key insights, and the lack of systematic assurance of decision credibility in existing technologies.

[0005] This invention provides a human-machine hybrid intelligent multidisciplinary consultation method based on a discussion matrix, comprising the following steps: Based on the target disease of the consulted patient, multi-source heterogeneous clinical data is collected; the multi-source heterogeneous clinical data includes the dialogue data stream of the consulted experts and the clinical data stream of the consulted patient; Based on the aforementioned multi-source heterogeneous clinical data, the structured speech text in the dialogue data stream is mapped to a pre-constructed initial consultation and discussion matrix to generate a target consultation and discussion matrix. The structured speech text includes the speech timestamp, speaker number, department, and original text corresponding to the consultation expert. The structured speech text is generated by audio conversion of the dialogue data stream. The original text consists of the speech statements corresponding to the consultation expert. The initial consultation and discussion matrix is ​​constructed based on multiple treatment stages of the target disease and multiple clinical tasks corresponding to each treatment stage. Based on the target consultation discussion matrix and the Monte Carlo tree search algorithm, a consultation guidance path is generated; the consultation guidance path is used to guide the consultation experts to conduct consultation and discussion on the target disease.

[0006] The present invention provides a human-machine hybrid intelligent multidisciplinary consultation method based on a discussion matrix, which further includes a step of constructing the initial consultation discussion matrix, the construction step further including: An initial horizontal axis is constructed using multiple stages of diagnosis and treatment of the target disease as horizontal axis elements; An initial vertical axis is constructed using multiple clinical tasks corresponding to each stage of diagnosis and treatment as vertical axis elements; Based on the initial horizontal axis and the initial vertical axis, the initial consultation and discussion matrix is ​​constructed.

[0007] According to the present invention, a human-machine hybrid intelligent multidisciplinary consultation method based on a discussion matrix is ​​provided, wherein the step of mapping structured speech text in the dialogue data stream to a pre-constructed initial consultation discussion matrix based on the multi-source heterogeneous clinical data to generate a target consultation discussion matrix includes: Obtain a brief task description corresponding to the clinical task; The brief task description is expanded into a detailed task description using a large language model and preset prompts; the large language model includes any one of DeepSeek-V2, DeepSeek-V3, GPT-4, Claude 3.5, or Tongyi Qianwen 2.5; the preset prompts are used to indicate the expansion requirements corresponding to the brief task description. Based on the detailed task description, a task description vector corresponding to the clinical task is generated; Based on the structured speech text, corresponding medical entities, clinical intentions, and medical semantic vectors are generated; Based on the similarity between the medical semantic vector and the task description vector and a first threshold, one or more candidate tasks are selected from the clinical tasks; the first threshold is a preset fixed value. Based on the detailed task description, medical entity, and clinical intent corresponding to the candidate task, semantic matching is performed to map the structured speech text to the relevant task cells of the initial consultation and discussion matrix, thereby generating the target consultation and discussion matrix; the relevant task cells are the structured functional units corresponding to the candidate task in the initial consultation and discussion matrix; the structured functional units are the cells formed by the intersection of the initial horizontal axis and the initial vertical axis.

[0008] According to the present invention, a human-machine hybrid intelligent multidisciplinary consultation method based on a discussion matrix is ​​provided, wherein generating a consultation guidance path based on the target consultation discussion matrix and a Monte Carlo tree search algorithm includes: The search state is a complete snapshot of the target consultation and discussion matrix at the current moment; the complete snapshot is the collection of all state information of the target consultation and discussion matrix at the current moment. The theoretically optimal path is obtained by calculating based on the search state using the Monte Carlo tree search algorithm. Based on the structured speech text in the target consultation discussion matrix, determine the current discussion trajectory of the consultation experts; The theoretically optimal path and the current discussion trajectory are compared to generate the consultation guidance path.

[0009] According to the present invention, a human-machine hybrid intelligent multidisciplinary consultation method based on a discussion matrix is ​​provided, wherein, when the complete snapshot includes the number of conflicts, the method further includes: Based on the aforementioned clinical data stream, a medical fact base is constructed; Based on the structured speech text, a temporary fact base is constructed; The data in the temporary fact base and the medical fact base are compared to obtain a contradiction value; Based on the number of conflict values, the number of conflicts is marked in the target consultation and discussion matrix.

[0010] According to the present invention, a human-machine hybrid intelligent multidisciplinary consultation method based on a discussion matrix, wherein when the complete snapshot includes a dynamic number of minutes, the method further includes: Based on the process sequence of the diagnosis and treatment stages, the candidate tasks are divided into current tasks, near-term tasks, and long-term tasks according to the process sequence from front to back; the current tasks are the candidate tasks corresponding to the first a diagnosis and treatment stages in the process sequence; the near-term tasks are the candidate tasks corresponding to the (a+1)th to (a+b)th diagnosis and treatment stages in the process sequence; the long-term tasks are the candidate tasks corresponding to the (a+b+1)th to (a+b+c)th diagnosis and treatment stages in the process sequence; where a, b, and c are all integers greater than or equal to 1; Determine the first degree of correlation between the structured speech text and the current task; Determine a second degree of correlation between the structured speech text and the recent task; Determine the third degree of correlation between the structured speech text and the long-term task; When both the first correlation degree and the second correlation degree are less than the second threshold, and the third correlation degree is greater than or equal to the second threshold, a core summary corresponding to the structured speech text is generated; the second threshold is a preset fixed value. Based on the number of core summaries, the number of dynamic minutes is marked in the target consultation and discussion matrix.

[0011] This invention also provides a human-machine hybrid intelligent multidisciplinary consultation system based on a discussion matrix, comprising the following modules: A multi-source data acquisition module is used to collect multi-source heterogeneous clinical data based on the target disease of the consulted patient; the multi-source heterogeneous clinical data includes the dialogue data stream of the consulting experts and the clinical data stream of the consulted patient; The discussion matrix construction module is used to map the structured speech text in the dialogue data stream to a pre-constructed initial consultation discussion matrix based on the multi-source heterogeneous clinical data, thereby generating a target consultation discussion matrix. The structured speech text includes the speech timestamp, speaker number, department, and original text corresponding to the consultation expert. The structured speech text is generated by audio conversion of the dialogue data stream. The original text is the speech statement corresponding to the consultation expert. The initial consultation discussion matrix is ​​constructed based on multiple treatment stages of the target disease and multiple clinical tasks corresponding to each treatment stage. The guidance path generation module is used to generate a consultation guidance path based on the target consultation discussion matrix and the Monte Carlo tree search algorithm; the consultation guidance path is used to guide the consultation experts to conduct consultation and discussion on the target disease.

[0012] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the human-machine hybrid intelligent multidisciplinary consultation method based on the discussion matrix as described above.

[0013] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the human-machine hybrid intelligent multidisciplinary consultation method based on the discussion matrix as described above.

[0014] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the human-machine hybrid intelligent multidisciplinary consultation method based on the discussion matrix as described above.

[0015] The present invention provides a human-machine hybrid intelligent multidisciplinary consultation method and system based on a discussion matrix. This method collects multi-source heterogeneous clinical data based on the target disease of the consulted patient. This multi-source heterogeneous clinical data includes the dialogue data stream of the consulting experts and the clinical data stream of the consulted patient. By integrating the expert dialogue stream and the patient's cross-system clinical data stream, it solves the problem of data fragmentation in traditional MDT (Multidisciplinary Team) consultations, providing a unified and complete data foundation for subsequent structured analysis and intelligent guidance. Based on the multi-source heterogeneous clinical data, the structured speech text in the dialogue data stream is mapped to a pre-constructed initial consultation discussion matrix to generate the target consultation discussion. The system employs a matrix to anchor unstructured expert statements to matrix task cells, thereby structuring the discussion content, eliminating ambiguity in natural language, and clarifying viewpoint attribution. Based on the target consultation discussion matrix and the Monte Carlo tree search algorithm, a consultation guidance path is generated. This path guides the experts in discussing the target disease, modeling the MDT process as a sequential decision problem and dynamically deducing the optimal discussion path using Monte Carlo tree search. This generates the consultation guidance path, corrects the divergent trajectory of the expert group in real time, and ensures that the discussion converges efficiently to the key decision point, thus improving the efficiency of multidisciplinary expert consultations. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 This is a flowchart illustrating the human-machine hybrid intelligent multidisciplinary consultation method based on a discussion matrix provided by the present invention.

[0018] Figure 2 This is a schematic diagram of the structure of the human-machine hybrid intelligent multidisciplinary consultation system based on the discussion matrix provided by the present invention.

[0019] Figure 3 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0021] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element. Furthermore, it should be noted that the scope of the methods and systems in the embodiments of this application is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

[0022] In this invention, the term "and / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. The character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0023] In this application's embodiments, "determine B based on A" means that factor A must be considered when determining B. It is not limited to "B can be determined based solely on A," but should also include: "determine B based on A and C," "determine B based on A, C, and E," "determine C based on A, and further determine B based on C," etc. Additionally, it can include using A as a condition for determining B, for example, "when A meets the first condition, determine B using the first method"; another example, "when A meets the second condition, determine B," etc.; another example, "when A meets the third condition, determine B based on the first parameter," etc. Of course, it can also be a condition where A is a factor in determining B, for example, "when A meets the first condition, determine C using the first method, and further determine B based on C," etc.

[0024] It should also be noted that the terms "target," "first," and "second" in this invention are used to distinguish similar objects, not to describe a specific order or sequence. It should be understood that such terms can be used interchangeably where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first" and "second" are generally of the same class, without limiting the number of objects; for example, the first object can be one or more.

[0025] In this invention, the term "multiple" refers to two or more, and other quantifiers are similar.

[0026] The following is combined with Figures 1 to 3 This invention describes a human-machine hybrid intelligent multidisciplinary consultation method and system based on a discussion matrix.

[0027] Figure 1 This is a flowchart illustrating the human-machine hybrid intelligent multidisciplinary consultation method based on a discussion matrix provided by the present invention, as shown below. Figure 1 As shown, the method includes the following steps: Step 101: Based on the target disease of the consulted patient, collect multi-source heterogeneous clinical data; the multi-source heterogeneous clinical data includes the dialogue data stream of the consulted experts and the clinical data stream of the consulted patient.

[0028] Step 102: Based on the multi-source heterogeneous clinical data, map the structured speech text in the dialogue data stream to a pre-constructed initial consultation discussion matrix to generate a target consultation discussion matrix; the structured speech text includes the speech timestamp, speaker number, department, and original text corresponding to the consultation expert; the structured speech text is generated by audio conversion of the dialogue data stream; the original text is the speech statement corresponding to the consultation expert; the initial consultation discussion matrix is ​​constructed based on multiple treatment stages of the target disease and multiple clinical tasks corresponding to each treatment stage.

[0029] Specifically, through multi-channel audio acquisition and speaker recognition technology, the audio of consultation experts from various departments is acquired in real time to obtain a dialogue data stream. Then, using a medical-adapted Automatic Speech Recognition (ASR) engine, the dialogue data stream is transformed into structured text containing {speech timestamp, speaker number, department, and original text} corresponding to the consultation experts.

[0030] For example, in one embodiment, each participating expert in the MDT consultation room is equipped with an independent microphone channel. The dialogue manager collects the audio signals from each channel in real time to obtain the dialogue data stream. The speech recognition engine uses an ASR model finely tuned with a medical terminology corpus (including Chinese clinical terms, drug names, surgical procedure names, anatomical terms, etc.) to improve the recognition accuracy of medical professional dialogues. At the same time, each speech segment in the dialogue data stream is automatically identified based on voiceprint features, and the final output structured text format is {speech timestamp: "2024-05-15 14:32:18", speaker ID: "DR003", department: "thoracic surgery", name: "Director Zhang", original text: "Based on the CT findings, this left upper lung nodule has obvious spiculation and pleural traction, and enlarged ipsilateral hilar lymph nodes are visible in the mediastinal window. I think it should be T2aN1 based on imaging staging, but pathological confirmation of the lymph node nature is needed"}.

[0031] By interfacing with multiple systems such as HIS, LIS, PACS, and Electronic Medical Record (EMR) systems, the system automatically extracts basic information, vital signs, laboratory test results (including values, units, and reference ranges), key measurements and imaging feature descriptions from imaging reports, histological diagnoses and molecular testing results from pathology reports, as well as medication records and allergy history of the patients being consulted, thus obtaining a comprehensive clinical data stream for the patients being consulted.

[0032] For example, in one embodiment, a data interface is established with various clinical information systems through the hospital integration platform (ESB / HL7). (1) Obtain basic patient information from HIS: age 65 years, gender male, admission date, chief complaint "left lung mass found during physical examination 1 week ago". (2) Obtain laboratory test results from LIS: each result contains structured fields {test item name, result value, unit, lower limit of reference range, upper limit of reference range, abnormality marker, sampling time}, such as {white blood cell count, 7.2, ×10 9 / L, 3.5, 9.5, normal, 2024-05-14 08:00}, {hemoglobin, 125, g / L, 130, 175, ↓, 2024-05-14 08:00}. (3) Obtain structured data of imaging reports from PACS: extract {examination site, examination method, key findings description, measurement value list}, such as {chest, CT enhanced scan, "an irregular soft tissue density nodule was seen in the anterior segment of the left upper lobe, about 3.2cm×2.5cm in size, with spiky edges, adjacent pleura traction depression, and uneven enhancement on enhanced scan, and an enlarged lymph node was seen in the left hilum of the mediastinum window, with a short diameter of about 1.2cm", [{"lesion long diameter":3.2,"unit":"cm","level":"arterial phase"}, {"lymph node short diameter":1.2,"unit":"cm","location":"left hilum"}, {"SUV max ":"Primary 8.5","Examination":"PET-CT"}]}. (4) Obtain the medical history summary, physical examination record, medication order list (amlodipine 5mg qd po) and allergy history (no known drug allergies) from the EMR system.

[0033] In some embodiments, the collected multi-source heterogeneous clinical data can also be standardized, including medical terminology normalization, unit standardization of test values, and outlier labeling, to improve the accuracy and standardization of the data.

[0034] For example, in one embodiment, a standardized processing procedure is performed on the aforementioned multi-source heterogeneous clinical data. Medical entity normalization: Synonyms used by different doctors (e.g., "left upper lobe lesion," "left upper lobe nodule," "left upper lobe mass," "left upper lobe tumor") are mapped to a unified standard term ("left upper lobe tumor") and its corresponding SNOMED CT code. Test value unit standardization: Test values ​​of the same type but with different units are converted to a unified standard unit and anomalies are marked (e.g., values ​​above the upper limit of the reference range are marked ↑, and values ​​below the lower limit are marked ↓). Drug information normalization: Standard representations of drug generic names, Anatomical Therapeutic Chemical Classification System (ATCS) codes, dosages, and routes of administration are extracted.

[0035] This invention addresses the problem of data fragmentation by simultaneously acquiring the dialogue data stream of consulting experts and the full-dimensional clinical data stream of consulting patients through dual channels. It enables real-time correlation between expert opinions and patient data, avoiding the decision-making omissions caused by data dispersion in traditional MDT.

[0036] By using medical semantic understanding and two-stage mapping, the structured speech text in the dialogue data stream is accurately anchored to the relevant task cells of the pre-built initial consultation and discussion matrix, generating the target consultation and discussion matrix and realizing the structured accumulation of the discussion content.

[0037] Furthermore, it also includes a step of constructing the initial consultation and discussion matrix, the construction step further including: An initial horizontal axis is constructed using multiple stages of diagnosis and treatment of the target disease as horizontal axis elements; An initial vertical axis is constructed using multiple clinical tasks corresponding to each stage of diagnosis and treatment as vertical axis elements; Based on the initial horizontal axis and the initial vertical axis, the initial consultation and discussion matrix is ​​constructed.

[0038] Specifically, before the start of the MDT consultation, the expert in charge of the consultation defines the structure of the initial consultation discussion matrix on a visual interface based on the case type of the target disease and the consultation objectives.

[0039] The horizontal axis elements represent multiple stages of the diagnosis and treatment process (such as: clinical data summary and medical history review, imaging evaluation, pathological and molecular diagnostic evaluation, clinical staging confirmation, treatment plan discussion, treatment risk assessment, and follow-up plan development).

[0040] The vertical axis elements represent the specific clinical tasks at each stage of diagnosis and treatment (such as "assessing the constraints of lung function on surgical procedures", "confirming the integrity of the evidence chain for TNM staging", and "screening for interactions between targeted therapies and concomitant medications").

[0041] For example, in one embodiment, before the MDT consultation begins, the consulting experts (such as the head of oncology) define the structure of the initial consultation discussion matrix on the system's visual interface based on the characteristics of the lung cancer case. The treatment stages (horizontal axis elements) defined in this embodiment include seven stages: clinical data summary and medical history review, imaging assessment, pathological and molecular diagnostic assessment, clinical staging confirmation (TNM staging), treatment plan discussion, treatment risk assessment and management, and follow-up plan development. The vertical axis elements represent the specific clinical tasks under each stage. For example, tasks defined under the "Imaging Assessment" stage include: assessing the imaging characteristics of the primary lesion (size, density, margins, enhancement pattern), assessing imaging evidence of lymph node metastasis, assessing the presence of distant metastasis, and recording key imaging measurements and labeling the sources of evidence. Tasks defined under the "Treatment Plan Discussion" stage include: assessing the feasibility of surgical resection and the choice of surgical method, assessing the indications for neoadjuvant therapy, discussing targeted therapy regimens (based on EGFR mutation status), discussing adjuvant chemotherapy regimens, and discussing the timing and target area of ​​radiotherapy intervention.

[0042] This invention provides a clear discussion guide for consulting experts by constructing an initial consultation and discussion matrix, ensuring coverage of all diagnosis and treatment stages and avoiding omission of key decision points.

[0043] Further, the step of mapping the structured spoken text in the dialogue data stream to a pre-constructed initial consultation discussion matrix based on the multi-source heterogeneous clinical data to generate a target consultation discussion matrix includes: Obtain a brief task description corresponding to the clinical task; The brief task description is expanded into a detailed task description using a large language model and preset prompts; the large language model includes any one of DeepSeek-V2, DeepSeek-V3, GPT-4, Claude 3.5, or Tongyi Qianwen 2.5; the preset prompts are used to indicate the expansion requirements corresponding to the brief task description. Based on the detailed task description, a task description vector corresponding to the clinical task is generated; Based on the structured speech text, corresponding medical entities, clinical intentions, and medical semantic vectors are generated; Based on the similarity between the medical semantic vector and the task description vector and a first threshold, one or more candidate tasks are selected from the clinical tasks; the first threshold is a preset fixed value. Based on the detailed task description, medical entity, and clinical intent corresponding to the candidate task, semantic matching is performed to map the structured speech text to the relevant task cells of the initial consultation and discussion matrix, thereby generating the target consultation and discussion matrix; the relevant task cells are the structured functional units corresponding to the candidate task in the initial consultation and discussion matrix; the structured functional units are the cells formed by the intersection of the initial horizontal axis and the initial vertical axis.

[0044] Specifically, the consulting experts input a brief task description for each clinical task, such as "confirm whether the patient's lung function can tolerate lobectomy".

[0045] An agent invokes a large language model and inputs preset prompts to expand a brief task description into a logically complete, essential, and unambiguous detailed task description. After confirmation by consulting experts, a pre-trained language model in the medical field is used to calculate a task description vector for each task description and construct a vector index library for subsequent retrieval of task description vectors. The large language model can be any model such as DeepSeek-V2, DeepSeek-V3, GPT-4, Claude 3.5, or Tongyi Qianwen Qwen2.5.

[0046] For example, in one embodiment, after receiving a brief task description input by a consulting expert, the Agent calls the Large Language Model (LLM), combines clinical diagnostic knowledge to write and input preset prompt words, and automatically expands the brief task description into a logically clear and complete detailed task description. For example, for the clinical task of "assessing the feasibility of surgical resection and the choice of surgical approach", the Agent's expanded detailed clinical description is as follows: "Based on imaging assessment results (tumor location, size, relationship with surrounding major blood vessels and bronchi), pulmonary function test results (forced expiratory volume in 1 second (FEV1), diffusing capacity of the lung for carbon monoxide (DLCO)), cardiovascular function assessment (echocardiography, electrocardiogram), age of the consulting patient and comorbidities (severity of chronic obstructive pulmonary disease (COPD), hypertension control status), comprehensively assess: (1) whether the tumor is anatomically resectable (the possibility of R0 resection); (2) whether the consulting patient's pulmonary function reserve can tolerate lobectomy or whether sublobar resection (segmentectomy / wedge resection) should be considered; (3) whether the status of mediastinal lymph node metastasis requires neoadjuvant therapy first. Refer to the pre-set criteria for resectableness." Finally, this detailed task description is presented to the consulting experts for confirmation. After consultation and confirmation, a word embedding model is used to convert the detailed task description of each clinical task into a high-dimensional semantic vector (i.e., task description vector), and a vector index library is built to provide a foundation for subsequent real-time semantic matching.

[0047] It should be noted that the aforementioned Agent, through writing and inputting preset prompts, assigns specific roles and rules to the large language model, thereby expanding the brief task description input by the consulting expert into a detailed task description that meets the expansion requirements. The preset prompts can also include medical terminology, output format constraints, and chain-of-thought guidance. Expansion requirements can include various criteria such as logical completeness, completeness of elements, or unambiguity.

[0048] Furthermore, the large language model and clinical knowledge base in the medical field used by the Agent in this embodiment of the invention need to be updated regularly to reflect the latest clinical practice guidelines, information on newly approved drugs, and drug interaction data. The update frequency is consistent with the update cycle of major clinical guidelines.

[0049] The Agent performs deep medical semantic analysis on each structured speech text and performs the following two tasks in parallel: (1) extracting medical entities such as disease diagnosis, symptoms and signs, drug names, test items, anatomical locations, and surgical procedures from the structured speech and mapping them to standardized medical coding systems (such as ICD-11, SNOMED CT, LOINC, RxNorm, ATC) to identify medical entities and generate corresponding medical entities; (2) determining the clinical intent type corresponding to the structured speech (proposing diagnostic hypotheses, recommending treatment plans, indicating risks / contraindications, supplementing clinical information, questioning / refuting existing views, confirming / summarizing interim conclusions) and generating corresponding clinical intents.

[0050] For example, in one embodiment, after receiving a new structured speech text, the Agent first runs a Medical Named Entity Recognition (Medical NER) model to identify and classify the following medical entities from the structured speech text and map them to a standard coding system: (i) Disease / Diagnosis → ICD-11 coding; (ii) Symptoms / Signs → SNOMED CT coding; (iii) Drugs / Compounds → RxNorm / ATC coding; (iv) Test Items → LOINC coding; (v) Anatomical Location → SNOMEDCT Anatomy coding; (vi) Surgery / Procedure → ICD-10-PCS or Surgical Procedure Classification Coding; (vii) Staging Information → TNM Staging System Coding. For example, for the speech "Based on the imaging, it should be T2aN1, but pathological confirmation of the lymph node's nature is needed," the system identifies: {Stage: T2a, TNM_T_stage}, {Stage: N1, TNM_N_stage}, {Anatomical Location: Lymph Node, SNOMED: XXX}, {Examination: Pathological Examination, LOINC: XXX}.

[0051] For example, in one embodiment, the Agent invokes the intent classification model in parallel, combining the context of the structured speech text to determine the type of clinical intent of the structured speech text: (i) to propose a diagnostic hypothesis (e.g., "I think this should be stage IIIA"); (ii) to recommend a treatment plan (e.g., "I suggest two cycles of neoadjuvant chemotherapy"); (iii) to indicate risks or contraindications (e.g., "Pay attention to the impact of the patient's COPD on postoperative lung function"); (iv) to supplement clinical information (e.g., "The patient also has a history of hypertension"); (v) to question or raise different opinions (e.g., "The SUV value of the enlarged lymph nodes is only 4.2, which does not necessarily indicate metastasis"); and (vi) to confirm or summarize interim conclusions. The intent classification model can be any of the following: a lightweight sentence vector classifier based on a linear model (e.g., FastText / Logistic Regression / Embedding), a deep semantic classification model based on a Transformer encoder (e.g., BERT / RoBERTa / ClinicalBERT), or a cue word routing classifier based on a large language model.

[0052] Based on the above embodiments, the following two-stage mapping is performed: In the first stage, the cosine similarity is calculated between the medical semantic vector corresponding to the structured speech text and the task description vectors corresponding to all clinical tasks in the initial consultation and discussion matrix. Task description vectors with similarity higher than a preset threshold (i.e., the first threshold) are selected, and the clinical tasks corresponding to these task description vectors are used as candidate tasks, resulting in one or more candidate tasks, forming a candidate task set. In the second stage, the Agent integrates the structured speech text, the identified medical entities and clinical intentions, and the detailed task descriptions of the candidate tasks to perform deep semantic reasoning. Finally, it determines that the structured speech text is accurately mapped to the discussion records of one or more relevant task cells in the initial consultation and discussion matrix, thus converting the initial consultation and discussion matrix into a target consultation and discussion matrix. Here, the relevant task cells are the structured functional units corresponding to the candidate tasks in the initial consultation and discussion matrix; the structured functional units are the cells formed by the intersection of the initial horizontal axis and the initial vertical axis.

[0053] For example, in one embodiment, the currently collected structured speech text is input into a vector representation model, and the corresponding medical semantic vector is extracted as its medical semantic representation. The cosine similarity between this medical semantic vector and all task description vectors in the vector index is calculated. Let the medical semantic vector be... The task description vector is Then the similarity Filter out those with similarity scores higher than the upper limit threshold. All clinical tasks constitute the candidate task set At the same time, retain similarity values ​​between the lower threshold. and upper limit threshold The candidate tasks between them constitute the alternative task set. This rapidly narrows the semantic search space. The vector representation model can be any of the following: a static word vector model (such as Word2Vec / GloVe), a Transformer-based dynamic context representation model (such as BERT / ClinicalBERT), or a contrastive learning-based sentence vector representation model (such as Sentence-BERT / SimCSE).

[0054] For example, in one embodiment, the Agent takes the following four pieces of information as input for comprehensive reasoning: (a) structured speech text; (b) identified medical entities and their standardized codes; (c) the type of identified clinical intent; and (d) a set of candidate tasks. and backup mission set The Agent calls LLM to perform multi-level semantic reasoning. The judgment strategy includes: (1) If the clinical intent is "supplementing clinical information" and the identified medical entity is highly consistent with the evaluation elements clearly listed in the detailed task description of a candidate task, then a one-to-one precise mapping is performed. For example, in the statement "The patient's FEV1 is 55% of the predicted value, and according to the GOLD classification, it belongs to moderate COPD", the medical entity {FEV1, 55%, predicted value} and the clinical intent {supplementing clinical information} are identified. The Agent judges that it is precisely matched with the candidate task "evaluate the feasibility of surgical resection and the choice of surgical procedure" (because the detailed task description clearly lists the need to evaluate lung function parameters such as FEV1), and maps the structured speech text to the relevant task cell. (2) If the content of the structured speech text (such as a comprehensive diagnosis and treatment suggestion) affects multiple candidate tasks at the same time, then a one-to-many mapping is performed. The Agent calls LLM to perform summary splitting of the structured speech text, generate sub-summaries for different candidate tasks, and map them to the corresponding relevant task cells. For example, an oncology expert stated, "Based on the results of EGFR 19del mutation and PD-L1 TPS 60%, I recommend osimertinib as first-line treatment. However, considering the COPD issue, the risk of interstitial pneumonia needs to be closely monitored during treatment." The agent judged that this statement involved two candidate tasks: "discussing targeted therapy options" and "treatment risk assessment and management," and split it into a treatment option section and a risk warning section for mapping.

[0055] This invention generates a target consultation and discussion matrix by mapping structured spoken text in the dialogue data stream to a pre-constructed initial consultation and discussion matrix, thereby eliminating ambiguity in natural language, ensuring accurate classification of expert opinions, and providing a structured foundation for subsequent path guidance.

[0056] Step 103: Based on the target consultation and discussion matrix and the Monte Carlo tree search algorithm, generate a consultation guidance path; the consultation guidance path is used to guide the consultation experts to conduct consultation and discussion on the target disease.

[0057] Specifically, this invention employs a clinical pathway optimization and convergence guidance system based on the Monte Carlo Tree Search (MCTS) algorithm. To enable the agent to provide evidence-based consultation guidance, this invention models the MDT (Multidisciplinary Team) consultation process as a Monte Carlo Tree Search problem. Specifically, the search state is defined based on a complete snapshot of the target consultation discussion matrix. Task transitions are used as decisions, and the optimal discussion path is simulated and deduced through MCTS to generate a consultation guidance path. This dynamically corrects divergent trajectories during the consultation process, guides discussion convergence, avoids excessive entanglement in non-critical issues, and improves the efficiency of multidisciplinary expert consultations.

[0058] The present invention provides a human-machine hybrid intelligent multidisciplinary consultation method based on a discussion matrix. This method collects multi-source heterogeneous clinical data based on the target disease of the consulted patient. This multi-source heterogeneous clinical data includes the dialogue data stream of the consulting experts and the clinical data stream of the consulted patient. By integrating the expert dialogue stream and the patient's cross-system clinical data stream, it solves the problem of data fragmentation in traditional MDT (Multidisciplinary Team) consultations, providing a unified and complete data foundation for subsequent structured analysis and intelligent guidance. Based on the multi-source heterogeneous clinical data, the structured speech text in the dialogue data stream is mapped to a pre-constructed initial consultation discussion matrix to generate a target consultation discussion matrix. This approach anchors unstructured expert statements to matrix task cells, enabling structured accumulation of discussion content, eliminating ambiguity in natural language, and clarifying viewpoint attribution. Based on the target consultation discussion matrix and Monte Carlo tree search algorithm, a consultation guidance path is generated. This path guides the consultation experts to discuss the target disease. By modeling the MDT process as a sequential decision problem and dynamically deducing the optimal discussion path based on Monte Carlo tree search, the consultation guidance path is generated, correcting the divergent trajectory of the expert group in real time, ensuring that the discussion converges efficiently to the key decision point, and thus improving the efficiency of multidisciplinary expert consultations.

[0059] Furthermore, the step of generating a consultation guidance path based on the target consultation discussion matrix and the Monte Carlo tree search algorithm includes: The search state is a complete snapshot of the target consultation and discussion matrix at the current moment; the complete snapshot is the collection of all state information of the target consultation and discussion matrix at the current moment. The Monte Carlo tree search algorithm is used to perform calculations based on the search state to explore the path and obtain the theoretically optimal path. Based on the structured speech text in the target consultation discussion matrix, determine the current discussion trajectory of the consultation experts; The theoretically optimal path and the current discussion trajectory are compared to generate the consultation guidance path.

[0060] Specifically, the MDT consultation process is modeled as a sequential decision problem, and the optimal discussion path is dynamically planned through the MCTS algorithm to provide a quantitative decision basis for the Agent's guidance behavior. Specifically, (1) State definition: An MCTS state (search state) is a complete snapshot of the target consultation discussion matrix at a certain moment, including the completion status of each task (not started / in progress / completed), the discussion depth under each task (mapped number of speeches), the number of conflicts (unresolved / resolved), the number of dynamic minutes, the clinical tasks currently focused on by the consultation expert group, and the statistical distribution of the group's clinical intentions. (2) Action definition: An MCTS action is for the Agent to perform a guidance behavior, guiding the expert group's attention to the next optimal unfinished clinical task. (3) Design process reward signals: Positive rewards include the confirmation of completion of key clinical tasks, the effective resolution of previously marked medical conflicts, the generation of new high-value clinical insights from discussions (such as the identification of previously unnoticed drug interaction risks), and the discussion order conforming to the logical dependencies of clinical diagnosis and treatment (such as confirming the staging before discussing specific plans); Negative rewards include the discussion trajectory frequently jumping between unrelated tasks, lingering on completed tasks, skipping key prerequisite tasks (such as discussing the dosage of chemotherapy drugs excreted through the kidneys without fully assessing renal function), and the group intention deviating significantly from the current stage clinical goals. (4) MCTS execution process: The Agent continuously runs the MCTS algorithm in the background, dynamically identifying the theoretically optimal path in the current state through a large number of iterations in four steps: selection (based on UCB balance exploration and utilization) → expansion (creating new task nodes) → simulation (rapid deduction based on clinical pathway heuristic model, simulating the probability of each task being successfully completed and its clinical benefits) → backtracking (cumulative rewards propagating backward along the path). (5) Guiding decision generation: The Agent compares the current discussion trajectory of the expert group with the theoretical optimal path calculated by MCTS in real time. If the current discussion trajectory deviates significantly (e.g., the expert group skips a task marked as a key prerequisite on the optimal path), the guidance module is triggered to call the large language model to generate natural and neutral guidance discourse (i.e., consultation guidance path), which will guide the consultation experts’ attention to the next clinical task suggested by the optimal path, while providing clinical logical basis.

[0061] For example, in one embodiment, the MCTS model of the MDT consultation process is as follows: State Definition: An MCTS state S is defined as a complete snapshot of the target consultation and discussion matrix at time t, that is, the set of all state information of the target consultation and discussion matrix at time t, which includes the following feature vectors: Completion status of each candidate task ∈{Not started (0), In progress (1), Completed (2)}, ( (Total number of tasks) In-depth discussion of each candidate task = The number of speeches mapped to this candidate task (i.e., the number of corresponding structured speech texts); Number of unresolved conflicts for each candidate task: ; Number of dynamic minutes for each candidate task: ; Current key tasks for the expert consultation group: ; Distribution of clinical intentions in the population (percentage of each type of clinical intention) The expression is: in, , , , , , These represent the proportions of clinical intentions, namely diagnostic hypothesis, treatment recommendation, risk warning, information supplementation, questioning and rebuttal, and summary confirmation, among all clinical intentions.

[0062] Proportion of time spent on consultation ,in, Indicates the current elapsed time. Indicates the total planned consultation time; Action definition: An MCTS action Defined as an agent performing a guiding action, specifically suggesting that the expert group shift the focus of discussion to a particular unfinished task. In state Below, all tasks that meet the following conditions constitute a set of optional actions. (i) Task Completion status (ii) Incomplete There is a clinically logical reachability relationship with the current focus task (such as adjacent tasks within the same phase, the first task in the next phase after the closing task of the current phase is completed, etc.); (iii) The prerequisite tasks (if any) are already completed.

[0063] Design process reward signal: Continuously calculate the reward signal during the consultation process, corresponding to the process reward function. R The design is as follows: in, (Progress Reward): Positive rewards are given when a new task is marked as "completed," with the weighting related to the clinical importance of the task. Higher reward weights are given for the completion of key decision-making tasks (such as "confirming TNM staging" or "determining treatment plan"). (Quality Awards) include the following sub-items: (i) awards are given when medical conflicts are effectively resolved (e.g., experts discuss and reach a consensus on alerts from conflict detection); (ii) awards are given when dynamic minutes of discussions are cited in subsequent stages of discussion and have a substantial impact; (iii) awards are given when treatment regimens have no alerts in drug interaction detection or when alerts are properly handled; and (iv) awards are given when the order of discussions conforms to clinical logical dependencies (e.g., staging confirmation is completed before specific regimen discussions are carried out). (Deviation Penalty): Calculated based on the entropy value of the discussion trajectory. If the expert group frequently jumps between unrelated tasks (e.g., jumping to "radiotherapy target discussion" before completing "radiological assessment"), the trajectory entropy increases, resulting in a negative reward. Additionally, an extra penalty is incurred when the KL divergence between the group's intention distribution and the expected intention distribution at the current stage exceeds a threshold. (Stagnation Penalty): A penalty for continuing to spend time on a completed task; or a penalty for prolonged discussion on a task without producing new insights. Weighting coefficient. The consultation can be dynamically adjusted according to the type and stage of the consultation, such as in tumor MDT. and Higher settings can be used to improve decision-making efficiency; MDT for rare disease diagnosis It can be set higher to encourage exploratory discussions.

[0064] MCTS Execution Flow: The agent continuously and rapidly runs the MCTS algorithm in the background, using the search status as input to find the optimal guidance action for the current consultation status and generate the theoretically optimal path. The specific calculation process of the MCTS algorithm includes: Selection: Starting from the root node representing the current consultation status, recursively apply the tree strategy. The tree search strategy that balances exploration and exploitation, namely the Upper Confidence Bound (UCB) formula, is expressed as follows: in, This represents the cumulative average reward. This indicates the MCTS status during the current consultation. Indicates MCTS action; represents the number of visits, and 'c' represents the exploration parameter. Select child nodes until a node that is not fully expanded is reached (i.e., its number of child nodes is less than the size of the set of available actions).

[0065] Extension: For a selected incomplete node, randomly select an action that has not yet been tried from its set of available actions (i.e., guide to a task node that has not yet been expanded), and create a new child node representing the expected new state after the action is executed (assuming that the action is adopted and the discussion of the target task has made initial progress).

[0066] Simulation: Starting with the newly created child node, a quick, lightweight "simulation workshop" is conducted. This simulation process is based on a clinical pathway heuristic model: each incomplete task... Assigned an expected probability of completion and the number of discussion steps required to complete . Depending on the clinical difficulty of the task, the depth of existing discussions, and the existence of unresolved conflicts, the simulation rapidly progresses to the end of the consultation (all tasks are marked as completed or have reached the total number of steps): In each simulation step, based on the current state of each selectable task... Value-weighted sampling selects the next focus of discussion, based on Randomly determine whether the task is "completed" in the current step, and press [button] upon completion. Rewards are then applied. If a "conflict" occurs in this step (simulated randomly with a low probability), a "conflict resolution" subprocess is simulated, and the reward value is accumulated throughout the simulation.

[0067] Backtracking: Accumulated rewards obtained during the simulation phase The process propagates backward along the selected path from the root node to the simulated node. This propagation occurs at each visited node along the path. Value and If all values ​​are updated, then the updated value will be... Value and The corresponding value represents the new average reward. and number of new visits : in, This represents the old average reward; This indicates the number of previous visits.

[0068] The generation of guiding decisions: After thousands of selections → expansions → simulations → backtracking iterations, starting from the root node of the MCTS tree, the distribution of the number of visits to each child node reflects the statistical merits of each optional action (guided to different tasks) in the current consultation state.

[0069] The Agent executes the following steps to generate specific consultation guidance behavior: Theoretical optimal path extraction: Starting from the root node, select the MCTS action corresponding to the child node with the most visits, which is the optimal guidance direction that MCTS believes is optimal in the current state—directing expert attention to the optimal next task. At the same time, the child node with the second highest number of visits is extracted as an alternative.

[0070] Bias detection: The agent maintains a sliding window to record the expert group's recent... The current discussion trajectory of secondary focus shift .Will Focus transfer sequence on the theoretically optimal path calculated by MCTS Perform a comparison and calculate the deviation. .when Exceeding the preset threshold At that time, the guidance is triggered.

[0071] Guiding Message Generation (i.e., Consultation Guidance Path): The Agent invokes the LLM, inputting the following information to generate the guiding message: (i) a description of the current focus task; (ii) the optimal next task suggested by the MCTS. Detailed description; (iii) (iv) Explanation of deviations. The LLM generates a neutral, clinically logical guiding statement. Example output: "Dear consulting experts, we have thoroughly discussed the details of the imaging assessment. Considering the integrity of the clinical pathway, the MCTS model suggests that before proceeding to the specific treatment plan discussion, we might first confirm the driver gene testing results in the 'Pathology and Molecular Diagnostic Assessment' stage, which will directly determine the direction of targeted therapy." This guiding statement does not contain any mandatory instructions but is presented as a suggestion, explaining its clinical logical basis, which the consulting experts decide whether to adopt. The Agent will not repeatedly initiate similar guiding statements in a short period of time to avoid interference.

[0072] This invention generates a guidance strategy by comparing the current discussion trajectory with the theoretically optimal path in real time. The guidance strategy is dynamically adjusted through a quantitative reward mechanism to ensure that the consultation guidance path conforms to clinical logic, while avoiding the abruptness of agent intervention. It corrects the divergent trajectory of the expert group in real time, ensuring that the discussion converges efficiently to the key decision point, thereby improving the efficiency of multidisciplinary expert consultation.

[0073] Furthermore, if the complete snapshot includes the number of collisions, the method further includes: Based on the aforementioned clinical data stream, a medical fact base is constructed; Based on the structured speech text, a temporary fact base is constructed; The data in the temporary fact base and the medical fact base are compared to obtain a contradiction value; Based on the number of conflict values, the number of conflicts is marked in the target consultation and discussion matrix.

[0074] Specifically, in this embodiment of the invention, the Agent, as the core cognitive engine, continuously performs intelligent maintenance on the matrix content, including: (1) Construction of a medical fact base: Based on the clinical data flow, a medical fact base is constructed, and the clinical factual statements (i.e., structured speech text) obtained in the entity extraction stage are structured as standardized medical fact triples of {subject, attribute, value, source of evidence, confidence level} and stored in the temporary fact base of this consultation. For example, "{left upper lobe nodule, long diameter, 3.2cm, CT enhanced scan (2024-05-01), 0.98}". (2) Conflict detection: When a new fact triple is added, it automatically detects whether it has a contradictory value with an existing record of the same subject and attribute in the medical fact base (e.g., the radiology department reports that the long diameter of the nodule is 3.2 cm while the pathology department mentions that the gross specimen measurement is 2.8 cm); at the same time, based on the drug interaction knowledge base and the clinical diagnosis and treatment guideline knowledge base, it automatically detects potential drug-drug interactions, drug-disease contraindications, and deviations between the diagnosis and treatment plan and the clinical guidelines in the medication regimen.

[0075] The fact base is maintained incrementally. Each time a new medical fact is extracted, the fact base is first checked to see if there is an existing record with the same subject and the same attribute. If it exists and the values ​​are different, a fact contradiction check is triggered; if it does not exist, the new fact is directly added to the fact base.

[0076] For example, in one embodiment, the following fact contradiction detection is performed: when a new fact triple (subject) ,property ,value When adding, query the fact base to see if the subject exists. ,property ,value )and ≠ If a discrepancy exists, it is considered a factual contradiction. A red warning icon is highlighted next to the relevant task cell in the target consultation and discussion matrix, and the sources and values ​​(number of conflicts) of the contradictory values ​​are displayed side by side. For example, if the radiology department believes the short diameter of a lymph node is 1.2cm based on a CT report, while the pathology department subsequently reports that the lymph node was measured at 0.9cm in the gross specimen, this difference will be automatically detected and an alarm will be triggered, prompting the consulting experts to pay attention and confirm which source should be used as the standard.

[0077] For example, in one embodiment, drug interaction detection is performed as follows: During the "treatment plan discussion" phase, after an expert proposes a specific medication plan, the list of drugs in the plan (including the patient's current medications and proposed new drugs) is entered into a drug interaction knowledge base (such as authoritative drug databases like DrugBank and Micromedex) for batch querying. For example, when an expert proposes using osimertinib, the system automatically detects the potential interaction between osimertinib and the patient's current amlodipine, as well as the absorption impact when used in combination with a PPI drug (such as omeprazole) mentioned by an expert in a previous discussion. A drug interaction alert is generated in the target consultation and discussion matrix, indicating that special attention is needed under the "treatment risk assessment and management" task. At the same time, the number of conflicts is counted and the corresponding number of conflicts is marked.

[0078] For example, in one embodiment, the following contraindication test is performed: the proposed treatment plan is cross-referenced with the patient's list of comorbidities and key abnormal test values. For instance, osimertinib carries a rare but serious risk of interstitial lung disease (ILD), and the patient already has moderate COPD (FEV1 55%) with reduced lung function reserve; the ILD risk would have a greater clinical impact on the patient. Simultaneously, the treatment recommendations from various experts are compared with the NCCN guidelines, marking any deviations from the guidelines and their degree for the expert panel's reference and discussion.

[0079] The statistical number of conflicts can be further used as one of the feature vectors defined in the complete snapshot, thereby improving the accuracy of MCTS modeling and generating more accurate guiding discourse for consultation discussions.

[0080] Furthermore, when the complete snapshot includes a dynamic number of minutes, the method further includes: Based on the process sequence of the diagnosis and treatment stages, the candidate tasks are divided into current tasks, near-term tasks, and long-term tasks according to the process sequence from front to back; the current tasks are the candidate tasks corresponding to the first a diagnosis and treatment stages in the process sequence; the near-term tasks are the candidate tasks corresponding to the (a+1)th to (a+b)th diagnosis and treatment stages in the process sequence; the long-term tasks are the candidate tasks corresponding to the (a+b+1)th to (a+b+c)th diagnosis and treatment stages in the process sequence; where a, b, and c are all integers greater than or equal to 1; Determine the first degree of correlation between the structured speech text and the current task; Determine a second degree of correlation between the structured speech text and the recent task; Determine the third degree of correlation between the structured speech text and the long-term task; When both the first correlation degree and the second correlation degree are less than the second threshold, and the third correlation degree is greater than or equal to the second threshold, a core summary corresponding to the structured speech text is generated; the second threshold is a preset fixed value. Based on the number of core summaries, the number of dynamic minutes is marked in the target consultation and discussion matrix.

[0081] Specifically, the workflow sequence for the multiple stages of diagnosis and treatment of the target disease, from front to back, can be: clinical data collection and medical history review, imaging evaluation, pathological and molecular diagnostic evaluation, clinical staging confirmation, treatment plan discussion, treatment risk assessment, and follow-up plan development. Each stage of diagnosis and treatment corresponds to one or more candidate tasks. Candidate tasks corresponding to the first *a* stages can be selected as the current task, candidate tasks corresponding to stages *a+1* to *a+b* can be selected as near-term tasks, and candidate tasks corresponding to stages *a+b+1* to *a+b+c* can be selected as long-term tasks. Here, *a*, *b*, and *c* are all integers greater than or equal to 1.

[0082] For example, if a=1, b=2, and c=4, then the candidate tasks corresponding to the first treatment stage (clinical data summary and medical history review) are the current tasks; the candidate tasks corresponding to the second and third treatment stages (imaging assessment, pathological and molecular diagnostic assessment) are the near-term tasks; and the candidate tasks corresponding to the fourth to seventh treatment stages (clinical staging confirmation, treatment plan discussion, treatment risk assessment, and follow-up plan development) are the long-term tasks.

[0083] If a structured speech text has low relevance to both the current and recent tasks (i.e., both the first and second relevance are less than the second threshold), but the agent determines that it is highly relevant to a future task (e.g., during the medical history review phase, a consulting expert mentions a past medication history that is crucial for the selection of subsequent treatment plans), i.e., the third relevance is greater than or equal to the second threshold, then the dynamic minutes mechanism is automatically triggered. The core summary of the structured speech text is captured and pre-placed in a specific area of ​​the cell for the future relevant task to form a dynamic minutes, and the number of dynamic minutes is indicated by a special icon.

[0084] For example, in one embodiment, if the similarity between the content of the structured speech text and all candidate tasks is below a lower threshold... (i.e., the second threshold), but if the Agent determines, based on deep semantic reasoning, that a statement is highly relevant to a candidate task in a future stage, then the dynamic summary mechanism is triggered. For example, in the "Clinical Data Summary and Medical History Review" stage, a consulting expert mentions that "the patient took omeprazole three years ago for a gastric ulcer, and it is necessary to note that the combined use of osimertinib and PPI drugs may affect absorption." This structured speech text does not highly match the candidate tasks in the current stage, but the Agent recognizes its importance to the precautions for targeted drug use in the future "Treatment Plan Discussion" stage, marks it as a "thought emergent," and triggers the dynamic summary. When the Agent determines that a structured speech text is a valuable "thought emergent," it calls LLM to generate a core summary of the speech (containing key medical entities, clinical logic points, and potential impacts), and stores it in the "dynamic summary" area of ​​the relevant future task. For example, during the "Imaging Assessment" phase, a radiology expert mentioned that "the morphological characteristics of this nodule are highly consistent with the radiomic characteristics of EGFR mutations; our team has conducted a similar retrospective analysis before." The agent marked this as a dynamic summary that is valuable for targeted therapy decisions during the "Treatment Plan Discussion" phase. When the consultation progresses to the "Treatment Plan Discussion" phase, a lightbulb icon will appear on the relevant task cell. Experts can click on it to view all previously accumulated relevant dynamic summaries, ensuring that key insights from the early stages are not forgotten.

[0085] The number of dynamic minutes counted can be further used as one of the feature vectors defined in the complete snapshot, thereby improving the accuracy of MCTS modeling and generating more accurate guiding discourse for consultation discussions.

[0086] Based on the above embodiments, the Agent automatically proposes to the consulting experts to update the completion status of the task by monitoring the conclusive statements in the dialogue data stream or the discussion silence period under a certain task, and executes the update after confirmation by the consulting experts.

[0087] For example, in one embodiment, the agent continuously monitors discussion activity under each candidate task. When a series of conclusive statements appear under a candidate task (such as "We agree on the imaging staging, classifying it as T2aN1M0, Stage IIB"), and no new dissenting statements are added under that candidate task within a subsequent period, the agent automatically suggests to the presiding expert that the candidate task be marked as "completed." After the presiding expert confirms, the background of the relevant task cell turns green. If the presiding expert rejects the suggestion, the agent records the reason for the rejection and continues monitoring.

[0088] Based on any of the above embodiments, an MDT visualization and interactive interface can be further designed to present the entire target consultation and discussion matrix and its dynamic changes to all consultation experts in real time through a graphical interface designed specifically for multidisciplinary consultation scenarios. The interface includes: (1) Global consultation matrix view: The target consultation and discussion matrix is ​​displayed in a grid layout. Each relevant task cell is marked with a background color to indicate the completion status, and information prompts such as conflict alarm icons, dynamic minutes icons, and pending conclusion icons are superimposed; (2) Clinical data integration panel: The key clinical data summary of the patient is displayed synchronously on the side of the matrix (basic information, vital sign trends, key test values ​​and their deviation from the reference range, image key frame thumbnails, and current medication list); (3) Interactive drill-down and task details view: Consulting experts can click on any relevant task cell to expand and view the detailed task description of the clinical task, real-time discussion records organized by timeline and speaker, related dynamic minutes captured from other stages, and clinical conflict details and their context detected by the Agent; (4) Guidance transparency view: When the Agent initiates guidance suggestions, the interface clearly marks them and briefly displays their clinical logic basis. Consulting experts can choose to adopt or ignore them.

[0089] For example, in one embodiment, the MDT visualization and interactive interface specifically includes: Global Consultation Matrix View: The main view displays the target consultation and discussion matrix in a clear grid layout. The horizontal axis represents the seven treatment stages arranged from left to right, and the vertical axis represents the specific clinical tasks under each stage. Each relevant task cell indicates its completion status with a background color (e.g., gray = not started, blue = in progress, green = completed). Simple icon prompts are overlaid on the cells: red exclamation mark = unresolved conflicts, yellow light bulb = dynamic minutes to be reviewed, blue checkmark = task completion proposals to be confirmed by experts. The task cells that the expert group is currently focusing on are dynamically highlighted with a highlighted border.

[0090] Clinical Data Integration Panel: The patient clinical data summary panel is permanently displayed on the right or below the matrix view. This panel contains the following areas: (1) Basic Information Area: Patient age, gender, chief complaint, allergy history; (2) Key Test Trend Area: Displays the historical changes of important test indicators (such as Hb, WBC, Cr, ALT, etc.) and their relationship with the reference range in a mini trend chart, with abnormal values ​​marked in red; (3) Summary of Key Imaging Findings: Displays the key measurement values ​​and feature descriptions of each imaging examination in structured text, and clicking on them will jump to the complete imaging report; (4) Current Medication List: Displays the generic name, dosage, and frequency of drugs, and if there is a drug interaction alarm, an alarm icon will be displayed next to the corresponding drug; (5) Key Timeline: Displays important events and examination results in the patient's diagnosis and treatment process in chronological order.

[0091] Interactive drill-up and task details view: When an expert clicks on any relevant task cell in the global consultation matrix view, the interface will expand the details view of the corresponding task in the form of a sidebar or pop-up window, which includes: (1) Detailed task description: A complete and detailed task description generated by human-computer collaboration during the initialization phase; (2) Discussion records: All speeches mapped to this task are listed in chronological order, and each speech is clearly marked with the speaker's name, department and timestamp, and supports keyword search; (3) Dynamic minutes: All dynamic minutes captured from other discussion phases and related to this task and their source context are displayed; (4) Conflict and alarm details: All clinical conflicts and risk alarms detected by the Agent that are related to this task are listed, and information comparison of the conflicting parties and alarm basis explanation are provided. Experts can discuss and mark them as "resolved" or "ignored" here; (5) Related clinical data: The most relevant clinical data for this task is automatically filtered and displayed (such as the key measurement values ​​of CT and PET-CT are automatically displayed under the "Imaging Assessment" task, and the current test values ​​and medication list are displayed under the "Treatment Plan Discussion" task).

[0092] User Feedback and Control: The interface provides necessary user control functions to establish human-machine trust. Experts can: (1) manually remap a structured speech text to a more suitable task cell; (2) add personal annotations and tags; (3) "confirm" or "reject" the task status update automatically proposed by the Agent; (4) "adopt" or "ignore" the Agent's MCTS guidance suggestions. Feedback signals are recorded for optimizing the timing and wording of subsequent guidance strategies.

[0093] All of Agent's automated operations (including entity recognition results, intent determination, mapping decisions, conflict detection results, state update proposals, and guidance suggestions) have clear traceability labels on the interface, allowing consulting experts to view the reasoning basis at any time, rather than facing a "black box".

[0094] This invention, through the design of a visual presentation and interactive interface for multidisciplinary team (MDT), enables the Agent to gradually accumulate the recognition and trust of consulting experts in its capabilities through each traceable and verifiable cognitive assistance behavior, thereby improving the degree of human-machine hybrid intelligent collaboration and ultimately enhancing the efficiency of multidisciplinary expert consultations.

[0095] The following describes the human-machine hybrid intelligent multidisciplinary consultation system based on the discussion matrix provided by the present invention. The human-machine hybrid intelligent multidisciplinary consultation system based on the discussion matrix described below can be referred to in correspondence with the human-machine hybrid intelligent multidisciplinary consultation method based on the discussion matrix described above.

[0096] Figure 2 This is a schematic diagram of the structure of the human-machine hybrid intelligent multidisciplinary consultation system based on the discussion matrix provided by the present invention, as shown below. Figure 2 As shown. The human-machine hybrid intelligent multidisciplinary consultation system based on discussion matrix provided in this embodiment of the invention includes a multi-source data acquisition module 201, a discussion matrix construction module 202, and a guidance path generation module 203, wherein: The multi-source data acquisition module 201 is used to acquire multi-source heterogeneous clinical data based on the target disease of the consulted patient; the multi-source heterogeneous clinical data includes the dialogue data stream of the consulting experts and the clinical data stream of the consulted patient; The discussion matrix construction module 202 is used to map the structured speech text in the dialogue data stream to a pre-constructed initial consultation discussion matrix based on the multi-source heterogeneous clinical data, thereby generating a target consultation discussion matrix. The structured speech text includes the speech timestamp, speaker number, department, and original text corresponding to the consultation expert. The structured speech text is generated by audio conversion of the dialogue data stream. The original text is the speech statement corresponding to the consultation expert. The initial consultation discussion matrix is ​​constructed based on multiple treatment stages of the target disease and multiple clinical tasks corresponding to each treatment stage. The guidance path generation module 203 is used to generate a consultation guidance path based on the target consultation discussion matrix and the Monte Carlo tree search algorithm; the consultation guidance path is used to guide the consultation experts to conduct consultation and discussion on the target disease.

[0097] Figure 3 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 3As shown, the electronic device may include: a processor 310, a communications interface 320, a memory 330, and a communication bus 340, wherein the processor 310, the communications interface 320, and the memory 330 communicate with each other via the communication bus 340. The processor 310 can call logical instructions in the memory 330 to execute a human-machine hybrid intelligent multidisciplinary consultation method based on a discussion matrix, which includes: Based on the target disease of the consulted patient, multi-source heterogeneous clinical data is collected; the multi-source heterogeneous clinical data includes the dialogue data stream of the consulted experts and the clinical data stream of the consulted patient; Based on the aforementioned multi-source heterogeneous clinical data, the structured speech text in the dialogue data stream is mapped to a pre-constructed initial consultation and discussion matrix to generate a target consultation and discussion matrix. The structured speech text includes the speech timestamp, speaker number, department, and original text corresponding to the consultation expert. The structured speech text is generated by audio conversion of the dialogue data stream. The original text consists of the speech statements corresponding to the consultation expert. The initial consultation and discussion matrix is ​​constructed based on multiple treatment stages of the target disease and multiple clinical tasks corresponding to each treatment stage. Based on the target consultation discussion matrix and the Monte Carlo tree search algorithm, a consultation guidance path is generated; the consultation guidance path is used to guide the consultation experts to conduct consultation and discussion on the target disease.

[0098] Furthermore, the logical instructions in the aforementioned memory 330 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0099] On the other hand, the present invention also provides a computer program product, the computer program product comprising a computer program that can be stored on a non-transitory computer-readable storage medium, wherein when the computer program is executed by a processor, the computer is able to execute the human-computer hybrid intelligent multidisciplinary consultation method based on the discussion matrix provided by the above methods, the method comprising: Based on the target disease of the consulted patient, multi-source heterogeneous clinical data is collected; the multi-source heterogeneous clinical data includes the dialogue data stream of the consulted experts and the clinical data stream of the consulted patient; Based on the aforementioned multi-source heterogeneous clinical data, the structured speech text in the dialogue data stream is mapped to a pre-constructed initial consultation and discussion matrix to generate a target consultation and discussion matrix. The structured speech text includes the speech timestamp, speaker number, department, and original text corresponding to the consultation expert. The structured speech text is generated by audio conversion of the dialogue data stream. The original text consists of the speech statements corresponding to the consultation expert. The initial consultation and discussion matrix is ​​constructed based on multiple treatment stages of the target disease and multiple clinical tasks corresponding to each treatment stage. Based on the target consultation discussion matrix and the Monte Carlo tree search algorithm, a consultation guidance path is generated; the consultation guidance path is used to guide the consultation experts to conduct consultation and discussion on the target disease.

[0100] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the human-machine hybrid intelligent multidisciplinary consultation method based on the discussion matrix provided by the above methods, the method comprising: Based on the target disease of the consulted patient, multi-source heterogeneous clinical data is collected; the multi-source heterogeneous clinical data includes the dialogue data stream of the consulted experts and the clinical data stream of the consulted patient; Based on the aforementioned multi-source heterogeneous clinical data, the structured speech text in the dialogue data stream is mapped to a pre-constructed initial consultation and discussion matrix to generate a target consultation and discussion matrix. The structured speech text includes the speech timestamp, speaker number, department, and original text corresponding to the consultation expert. The structured speech text is generated by audio conversion of the dialogue data stream. The original text consists of the speech statements corresponding to the consultation expert. The initial consultation and discussion matrix is ​​constructed based on multiple treatment stages of the target disease and multiple clinical tasks corresponding to each treatment stage. Based on the target consultation discussion matrix and the Monte Carlo tree search algorithm, a consultation guidance path is generated; the consultation guidance path is used to guide the consultation experts to conduct consultation and discussion on the target disease.

[0101] The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0102] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0103] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A human-machine hybrid intelligent multidisciplinary consultation method based on a discussion matrix, characterized in that, include: Based on the target disease of the patients being consulted, collect multi-source heterogeneous clinical data; The multi-source heterogeneous clinical data includes the dialogue data stream of the consulting experts and the clinical data stream of the consulting patients; Based on the aforementioned multi-source heterogeneous clinical data, the structured speech text in the dialogue data stream is mapped to a pre-constructed initial consultation discussion matrix to generate a target consultation discussion matrix. The structured speech text includes the speech timestamp, speaker number, department, and original text corresponding to the consultation expert. The structured speech text is generated by audio conversion of the dialogue data stream. The original text consists of the speech statements corresponding to the consultation expert. The initial consultation discussion matrix is ​​constructed based on multiple treatment stages of the target disease and multiple clinical tasks corresponding to each treatment stage. A consultation guidance path is generated based on the target consultation discussion matrix and a Monte Carlo tree search algorithm. The consultation guidance path is used to guide the consultation experts to conduct a consultation and discussion on the target disease; It also includes the step of constructing the initial consultation and discussion matrix, the construction step further including: An initial horizontal axis is constructed using multiple stages of diagnosis and treatment of the target disease as horizontal axis elements; An initial vertical axis is constructed using multiple clinical tasks corresponding to each stage of diagnosis and treatment as vertical axis elements; Based on the initial horizontal axis and the initial vertical axis, construct the initial consultation and discussion matrix; The step of mapping the structured spoken text in the dialogue data stream to a pre-constructed initial consultation discussion matrix based on the multi-source heterogeneous clinical data to generate a target consultation discussion matrix includes: Obtain a brief task description corresponding to the clinical task; The brief task description is expanded into a detailed task description using a large language model and preset prompts; the large language model includes any one of DeepSeek-V2, DeepSeek-V3, GPT-4, Claude 3.5, or Tongyi Qianwen 2.5; the preset prompts are used to indicate the expansion requirements corresponding to the brief task description. Based on the detailed task description, a task description vector corresponding to the clinical task is generated; Based on the structured speech text, corresponding medical entities, clinical intentions, and medical semantic vectors are generated; Based on the similarity between the medical semantic vector and the task description vector and a first threshold, one or more candidate tasks are selected from the clinical tasks; the first threshold is a preset fixed value. Based on the detailed task description, medical entity, and clinical intent corresponding to the candidate task, semantic matching is performed to map the structured speech text to the relevant task cells of the initial consultation and discussion matrix, thereby generating the target consultation and discussion matrix; the relevant task cells are the structured functional units corresponding to the candidate task in the initial consultation and discussion matrix; the structured functional units are the cells formed by the intersection of the initial horizontal axis and the initial vertical axis.

2. The human-machine hybrid intelligent multidisciplinary consultation method based on a discussion matrix according to claim 1, characterized in that, The process of generating a consultation guidance path based on the target consultation discussion matrix and the Monte Carlo tree search algorithm includes: The search state is a complete snapshot of the target consultation and discussion matrix at the current moment; the complete snapshot is the collection of all state information of the target consultation and discussion matrix at the current moment. The theoretically optimal path is obtained by calculating based on the search state using the Monte Carlo tree search algorithm. Based on the structured speech text in the target consultation discussion matrix, determine the current discussion trajectory of the consultation experts; The theoretically optimal path and the current discussion trajectory are compared to generate the consultation guidance path.

3. The human-machine hybrid intelligent multidisciplinary consultation method based on a discussion matrix according to claim 2, characterized in that, If the complete snapshot includes the number of collisions, the method further includes: Based on the aforementioned clinical data stream, a medical fact base is constructed; Based on the structured speech text, a temporary fact base is constructed; The data in the temporary fact base and the medical fact base are compared to obtain a contradiction value; Based on the number of conflict values, the number of conflicts is marked in the target consultation and discussion matrix.

4. The human-machine hybrid intelligent multidisciplinary consultation method based on a discussion matrix according to claim 2, characterized in that, If the complete snapshot includes a dynamic number of minutes, the method further includes: Based on the process sequence of the diagnosis and treatment stages, the candidate tasks are divided into current tasks, near-term tasks, and long-term tasks according to the process sequence from front to back; the current tasks are the candidate tasks corresponding to the first a diagnosis and treatment stages in the process sequence; the near-term tasks are the candidate tasks corresponding to the (a+1)th to (a+b)th diagnosis and treatment stages in the process sequence; the long-term tasks are the candidate tasks corresponding to the (a+b+1)th to (a+b+c)th diagnosis and treatment stages in the process sequence; where a, b, and c are all integers greater than or equal to 1; Determine the first degree of correlation between the structured speech text and the current task; Determine a second degree of correlation between the structured speech text and the recent task; Determine the third degree of correlation between the structured speech text and the long-term task; When both the first correlation degree and the second correlation degree are less than the second threshold, and the third correlation degree is greater than or equal to the second threshold, a core summary corresponding to the structured speech text is generated; the second threshold is a preset fixed value. Based on the number of core summaries, the number of dynamic minutes is marked in the target consultation and discussion matrix.

5. A human-machine hybrid intelligent multidisciplinary consultation system based on a discussion matrix, characterized in that, include: The multi-source data acquisition module is used to collect multi-source heterogeneous clinical data based on the target disease of the consulted patient. The multi-source heterogeneous clinical data includes the dialogue data stream of the consulting experts and the clinical data stream of the consulting patients; The discussion matrix construction module is used to map the structured speech text in the dialogue data stream to a pre-constructed initial consultation discussion matrix based on the multi-source heterogeneous clinical data, thereby generating a target consultation discussion matrix. The structured speech text includes the speech timestamp, speaker number, department, and original text corresponding to the consultation expert. The structured speech text is generated by audio conversion of the dialogue data stream. The original text is the speech statement corresponding to the consultation expert. The initial consultation discussion matrix is ​​constructed based on multiple treatment stages of the target disease and multiple clinical tasks corresponding to each treatment stage. The guidance path generation module is used to generate a consultation guidance path based on the target consultation discussion matrix and the Monte Carlo tree search algorithm; The consultation guidance path is used to guide the consultation experts to conduct a consultation and discussion on the target disease; Also includes: An initial horizontal axis is constructed using multiple stages of diagnosis and treatment of the target disease as horizontal axis elements; An initial vertical axis is constructed using multiple clinical tasks corresponding to each stage of diagnosis and treatment as vertical axis elements; Based on the initial horizontal axis and the initial vertical axis, construct the initial consultation and discussion matrix; The step of mapping the structured spoken text in the dialogue data stream to a pre-constructed initial consultation discussion matrix based on the multi-source heterogeneous clinical data to generate a target consultation discussion matrix includes: Obtain a brief task description corresponding to the clinical task; The brief task description is expanded into a detailed task description using a large language model and preset prompts; the large language model includes any one of DeepSeek-V2, DeepSeek-V3, GPT-4, Claude 3.5, or Tongyi Qianwen 2.5; the preset prompts are used to indicate the expansion requirements corresponding to the brief task description. Based on the detailed task description, a task description vector corresponding to the clinical task is generated; Based on the structured speech text, corresponding medical entities, clinical intentions, and medical semantic vectors are generated; Based on the similarity between the medical semantic vector and the task description vector and a first threshold, one or more candidate tasks are selected from the clinical tasks; the first threshold is a preset fixed value. Based on the detailed task description, medical entity, and clinical intent corresponding to the candidate task, semantic matching is performed to map the structured speech text to the relevant task cells of the initial consultation and discussion matrix, thereby generating the target consultation and discussion matrix; the relevant task cells are the structured functional units corresponding to the candidate task in the initial consultation and discussion matrix; the structured functional units are the cells formed by the intersection of the initial horizontal axis and the initial vertical axis.

6. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the human-machine hybrid intelligent multidisciplinary consultation method based on the discussion matrix as described in any one of claims 1 to 4.

7. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the human-machine hybrid intelligent multidisciplinary consultation method based on the discussion matrix as described in any one of claims 1 to 4.

8. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the human-machine hybrid intelligent multidisciplinary consultation method based on the discussion matrix as described in any one of claims 1 to 4.

Images