Medical artificial intelligence system and method based on situation intelligence and multi-agent cooperation

Abstract

The invention particularly relates to a medical artificial intelligence system and method based on situation intelligence and multi-agent cooperation, and relates to the technical field of medical artificial intelligence. Performing context analysis and task planning; selecting and scheduling an expert agent; and performing collaborative execution and result convergence. According to the invention, multiple modules cooperate to greatly optimize the clinical working process, and the diagnosis and treatment efficiency and decision accuracy are improved; the unified context bus can aggregate multi-source heterogeneous data such as patient diagnosis information, allergy history and examination results in real time, and automatically standardizes the data into structured context objects, so that the tedious operation of manually arranging and dispersing data by medical staff is avoided, and a large amount of data preprocessing time is saved; the context awareness and intelligent orchestration engine accurately schedules an adaptive agent by means of the logic reasoning ability of a large language model, and avoids the logic vulnerability of artificial task decomposition.

Description

Technical Field

[0001] This invention relates to the field of medical artificial intelligence technology, and in particular to a medical artificial intelligence system and method based on contextual intelligence and multi-agent collaboration. Background Technology

[0002] Artificial intelligence (AI) technology, especially since the deep learning wave, has been explored in the medical field for more than a decade and has achieved many remarkable results. From early medical image-assisted diagnosis to rule-based clinical decision support, and more recently, medical knowledge question answering and medical record generation driven by large language models, AI technology is gradually penetrating all aspects of medical services.

[0003] However, despite continuous breakthroughs in individual technologies, the overall application model of medical AI currently remains largely at a rudimentary stage, resembling a "functional plug-in." Whether it's a standalone image diagnostic AI software, a fixed medical record quality control module, or a general-purpose medical question-and-answer robot, they mostly exist as isolated tools outside or on the periphery of complex clinical workflows. This model has revealed its profound and fundamental limitations, becoming a core bottleneck hindering AI from realizing greater value in the medical field: A lack of deep understanding of clinical "context": Clinical work is inherently highly context-dependent. The same examination result can have drastically different clinical significance for patients with different diagnoses, ages, and disease stages. The same medical task performed by healthcare professionals with different roles and seniority will have vastly different focuses and operational procedures. "Plug-in" AI tools, lacking the ability to perceive this dynamic, multi-dimensional "context," often provide rigid and impersonal assistance. For example, a CDSS (Consciousness, Diagnosis, and Surgery) tool that pops up the same guidelines for all doctors regardless of the scenario will quickly be ignored by clinicians due to "alarm fatigue," significantly diminishing its value. This phenomenon is academically known as AI "contextual errors," where the model's output, while seemingly reasonable on the surface, is flawed because it fails to consider crucial, context-specific information. Fragmented Application Experience and Disrupted Workflow: Because AI functions exist as isolated plugins, healthcare professionals must frequently switch between their core systems (such as HIS / EMR) and multiple AI tools, disrupting workflows and resulting in an extremely disjointed user experience. For example, a doctor might need to first view images in a PACS system, then switch to AI software for analysis, and then manually copy and paste the AI ​​analysis results into an EMR report. This fragmented experience not only fails to improve efficiency but may also increase operational complexity and cognitive load. Lack of collaborative capabilities: Real clinical work is a complex collaborative process involving multiple roles and tasks. For example, successful preoperative preparation requires the effective integration and collaborative decision-making of information from multiple parties, including surgeons, anesthesiologists, nurses, and radiologists. However, existing AI toolchains only address one single problem at a time, and they cannot effectively communicate and collaborate with each other to form a cohesive force that can assist in the entire complex collaborative task. Poor scalability and adaptability: The healthcare environment is constantly changing, with new diseases, guidelines, technologies, and workflows emerging frequently. "Plug-in" AI applications are often hard-coded for specific tasks, making it difficult to adapt quickly to these changes. Whenever a new requirement arises, redevelopment or complex system modifications are often necessary, resulting in extremely poor scalability and adaptability.

[0004] Recent research from world-leading research institutions such as Harvard Medical School clearly indicates that to unleash the full potential of AI in the real world, its design philosophy must be fundamentally changed, shifting from building static, functionally fixed models to building intelligent systems that can dynamically adapt to new situations. This means that the core competitiveness of next-generation medical AI will no longer be the accuracy of a single algorithm, but rather the entire system's ability to perceive, understand, and respond to complex and ever-changing clinical "contexts"—the so-called "contextual intelligence."

[0005] Based on this profound insight, there is an urgent need in this field for a novel system architecture and implementation method that can deeply and seamlessly integrate AI capabilities into every aspect of clinical work, much like the kernel services of an operating system. This architecture should proactively perceive and understand context, and intelligently allocate the most appropriate capabilities to collaborate efficiently with medical staff. This is precisely the core technical problem that this invention aims to solve. Summary of the Invention

[0006] The purpose of this invention is to propose a medical artificial intelligence system and method based on contextual intelligence and multi-agent collaboration in order to solve the above-mentioned problems.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: Medical AI methods based on contextual intelligence and multi-agent collaboration include: Real-time context building: Aggregate multi-source heterogeneous data related to specific clinical scenarios in real time and continuously, and standardize the data to build context objects; context objects include at least patient context, user context, and task context; Context analysis and task planning: A context-aware and intelligent orchestration engine analyzes contextual objects in real time to identify the current core clinical scenario and user intent; based on the identification results, the macro-level clinical task is broken down into a plan containing multiple sub-tasks; Expert agent selection and scheduling: Based on the characteristics of each subtask in the execution plan, select an expert agent to execute the subtask; and assign task instructions and relevant context object information to the selected expert agent. Collaborative execution and result aggregation: After receiving instructions, the selected expert agent executes its assigned subtasks and returns structured output containing results, confidence scores, and execution process logs; the orchestration engine aggregates the outputs of the subtasks and uses them as input for subsequent subtasks or for the generation of the final result. Context-adaptive interaction: The key information of the final result or intermediate process is presented to the user through a user interface that can dynamically adjust its presentation content and layout according to the context object, and the user's interaction commands are received to trigger a new cycle of context building and analysis.

[0008] Preferably, in the real-time context construction step, the patient context includes the patient's diagnostic information, allergy history, key examination and test results, and vital sign trends; User context includes the operator's role, title, and specialty; The task context includes the name of the task the user is currently executing and the location of the system interface.

[0009] Preferably, the expert agent includes at least the following three types: Basic capability intelligent agents are responsible for performing atomized general AI tasks, including medical report text parsing, knowledge retrieval, or quality control of medical record content; Role-playing intelligent agents are responsible for simulating specific medical roles and interacting or generating content in a manner that matches the tone, style, and knowledge level of that role. Specialty knowledge agent, responsible for encapsulating in-depth knowledge and reasoning logic of specific clinical specialties.

[0010] Preferably, the generation of the execution plan utilizes the thought chain reasoning ability of the large language model to decompose the macro task into a series of logically dependent, specific sub-task sequences that can be executed by expert intelligent agents.

[0011] Preferably, the method further includes: Agent optimization: Record user behaviors such as adoption, modification, or rejection of results generated by expert agents, and use these interactions as feedback signals to iteratively optimize the configuration or model of expert agents.

[0012] Medical AI systems based on contextual intelligence and multi-agent collaboration include: The unified context bus module is configured to aggregate heterogeneous data from multiple healthcare-related data sources in real time and standardize it into structured, dynamically updated context objects. AI capability module, configured to form a pool containing multiple expert agents, each of which is configured to perform one or a class of specific medical-related AI tasks; The context-aware and intelligent orchestration engine module is configured as the intelligent hub of the system, which connects the unified context bus and the AI ​​capability layer. It analyzes context objects in real time, automatically decomposes macro-clinical tasks into sub-task sequences, and selects and schedules appropriate agents from the pool of expert agents to execute the sub-tasks. The application and interaction module is configured to present the final results or intermediate process information processed by the orchestration engine to the user through a context-adaptive user interface and to receive user interaction commands.

[0013] Preferably, the specialty knowledge agent in the AI ​​capability module is constructed by fine-tuning a large language model on a pre-defined clinical specialty dataset.

[0014] Preferably, the context-aware and intelligent orchestration engine module includes a task planner that utilizes the logical reasoning capabilities of a large language model to generate an execution graph containing agent call sequences and data dependencies based on the current context and a preset clinical workflow template.

[0015] Preferably, the user interface of the application and interaction module provides a traceability mechanism, allowing users to trace the original data source or knowledge base entry on which the content generated by any expert agent is based when viewing it.

[0016] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are: 1. This invention significantly optimizes clinical workflows through multi-module collaboration, improving diagnostic and treatment efficiency and decision-making accuracy; the unified context bus can aggregate multi-source heterogeneous data such as patient diagnostic information, allergy history, and test results in real time, automatically standardizing them into structured context objects, avoiding the tedious operation of medical staff manually sorting scattered data and saving a lot of data preprocessing time; the context-aware and intelligent orchestration engine, with the help of the logical reasoning capabilities of large language models, accurately schedules and adapts intelligent agents, avoiding logical loopholes in human task decomposition.

[0017] 2. This invention ensures that diagnostic and treatment decisions are based on complete and real-time patient context through dynamic aggregation of sub-task results and data dependency management, reducing the risk of misdiagnosis or missed diagnosis due to incomplete information or processing delays, making clinical work more efficient and decision-making more reliable. Attached Figure Description

[0018] Further details, features, and advantages of this application are disclosed in the following description of exemplary embodiments in conjunction with the accompanying drawings, in which: Figure 1 This is a structural diagram of the method of the present invention; Figure 2 This is a system flowchart of the present invention. Detailed Implementation

[0019] Several embodiments of this application will now be described in more detail with reference to the accompanying drawings to enable those skilled in the art to implement this application. This application may be embodied in many different forms and for various purposes and should not be limited to the embodiments set forth herein. These embodiments are provided to make this application thorough and complete, and to fully convey the scope of this application to those skilled in the art. The embodiments described do not limit this application.

[0020] Unless otherwise defined, all terms used herein (including technical and scientific terms) shall have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. It will be further understood that terms such as those defined in commonly used dictionaries shall be interpreted as having a meaning consistent with their meaning in the relevant field and / or the context of this specification, and shall not be interpreted in an idealized or overly formal sense unless expressly defined herein.

[0021] Example 1 Its specific implementation method is combined with the appendix Figure 1 and attached Figure 2 Please provide a detailed explanation.

[0022] Appendix Figure 1 The flowchart of the medical artificial intelligence method based on contextual intelligence and multi-agent collaboration provided for the embodiments of the present invention illustrates the complete steps from real-time context construction to agent optimization.

[0023] Appendix Figure 2 The block diagram of a medical artificial intelligence system based on contextual intelligence and multi-agent collaboration provided in this embodiment of the invention shows the connection relationship between the unified context bus module and the application and interaction module, and marks the main functional interaction flow of each module.

[0024] In this embodiment, it includes: Real-time context building: From one or more medical-related data sources, multi-source heterogeneous data related to a specific clinical scenario are aggregated in real time and continuously, and the data is standardized to build a structured and dynamically updated context object that describes the current clinical scenario; the context object includes at least patient context, user context, and task context; Patient context includes the patient's diagnostic information, allergy history, key test results, and vital sign trends; User context includes the operator's role, title, and specialty; The task context includes the name of the task the user is currently executing and the location of the system interface.

[0025] Context analysis and task planning: A context-aware and intelligent orchestration engine analyzes contextual objects in real time to identify the current core clinical scenario and user intent; based on the identification results, a macro-level clinical task is dynamically and automatically decomposed into an ordered execution plan containing multiple sub-tasks; Expert agent selection and scheduling: Based on the characteristics of each subtask in the execution plan, one or more of the most suitable expert agents are selected from a pre-set pool containing multiple expert agents to execute the subtask; and task instructions and related context object information are assigned to the selected expert agents; wherein, the selection of the most suitable expert agent is based on a predefined mapping rule base, which establishes a correspondence between clinical subtask types and expert agent capability labels; Expert agents include at least the following three types: Basic capability intelligent agents are responsible for performing atomized general AI tasks, including medical report text parsing, knowledge retrieval, or quality control of medical record content; Role-playing intelligent agents are responsible for simulating specific medical roles and interacting or generating content in a manner that matches the tone, style, and knowledge level of that role. Specialty knowledge agent, responsible for encapsulating in-depth knowledge and reasoning logic of specific clinical specialties.

[0026] Collaborative execution and result aggregation: After receiving instructions, the selected expert agents independently or collaboratively execute their assigned subtasks and return structured outputs containing results, confidence scores, and execution logs; the orchestration engine aggregates the outputs of one or more subtasks and uses them as inputs for subsequent subtasks or for the generation of the final result. The execution plan is generated by leveraging the chain-of-thought reasoning ability of the large language model to decompose the macro task into a series of logically dependent, specific sub-task sequences that can be executed by expert agents.

[0027] Context-adaptive interaction: The key information of the final result or intermediate process is presented to the user through a user interface that can dynamically adjust its presentation content and layout according to the context object, and receive user interaction commands to trigger a new context construction and analysis cycle; the user interface is a dynamic and modular interface, and its displayed components and information flow are rendered by the context-aware and intelligent orchestration engine in real time according to the current context. Agent optimization: Record user behaviors such as adoption, modification, or rejection of results generated by expert agents, and use these interactions as feedback signals to iteratively optimize the configuration or model of expert agents.

[0028] Example 2 A medical artificial intelligence system based on contextual intelligence and multi-agent collaboration includes the following components: The unified context bus module is configured to aggregate heterogeneous data from multiple healthcare-related data sources in real time and standardize it into structured, dynamically updated context objects. Establish secure and efficient connections with multiple core information systems within the hospital through pre-built, configurable adapters. These systems include, but are not limited to: Clinical business systems, such as hospital information systems (HIS) and electronic medical record systems (EMR), provide patients with core clinical information such as diagnoses, medical orders, and medical records.

[0029] Medical technology systems, such as Laboratory Information Systems (LIS) and Picture Archiving and Communication Systems (PACS), provide test reports and image reports.

[0030] Operations management systems, such as human resource planning (HRP) and supply chain management (SCM) systems, provide operators with identity, permissions, and material information.

[0031] Internet of Things (IoT) platform: Connects bedside monitors, infusion pumps, and other devices to provide real-time vital signs data streams.

[0032] External knowledge bases: Connect to external medical knowledge bases, clinical guideline databases, or public health information sources via API.

[0033] Data standardization and context object generation: The aggregated raw data is heterogeneous and fragmented. The core of this step is to transform it into a machine-understandable, structured "context object".

[0034] Data cleaning and standardization: The raw data undergoes missing value processing, outlier detection, and format standardization. More importantly, machine learning-assisted semantic mapping technology (as detailed in the aforementioned patent embodiments, based on BERT word embedding and cosine similarity methods) is used to map heterogeneous terms from different hospital areas or systems (such as "myocardial infarction" vs. "acute myocardial infarction") onto a unified standardized data dictionary for the group, ensuring semantic consistency.

[0035] Context object construction: The system dynamically populates a "context object" instance with standardized data based on a predefined, extensible JSON schema. This object contains at least four core modules: Patient Context: Encapsulates all clinical information related to the current patient.

[0036] User Context: Encapsulates all relevant information about the current user.

[0037] TaskContext: Encapsulates a description of the task that the user is currently executing.

[0038] Environment Context: Encapsulates information about the physical and system environment in which a task occurs.

[0039] Context service-oriented architecture: The constructed "context object" is exposed to the upper-layer system through a high-performance internal API service (preferably using gRPC for low latency and strong type definition). The upper-layer "orchestration engine" can obtain the most complete and up-to-date context snapshot for any interaction event at any time through the getContext(contextId) request.

[0040] AI capability module, configured to form a pool containing multiple expert agents, each of which is configured to perform one or a class of specific medical-related AI tasks; Specialty knowledge agents are constructed by fine-tuning large language models on high-quality datasets of pre-defined clinical specialties. Intelligent agents are classified and defined as follows: Intelligent agents in the system are divided into different types to address different levels of needs: Foundational Agents: These are reusable "building blocks" that perform atomized AI tasks.

[0041] For example, the core of the "Report Parsing Body" is a finely tuned Named Entity Recognition (NER) and Relation Extraction (RE) model used to extract structured information from unstructured reports; the core of the "Knowledge Retrieval Body" is based on the RAG (Retrieval Enhanced Generation) architecture, which first uses a vector database (such as Milvus) for efficient similarity retrieval, and then feeds the retrieved content as context to the LLM to generate accurate and verifiable answers.

[0042] Role-playing agents: At their core, these agents imbue LLMs with specific "personalities" and communication styles through carefully engineered prompts. For example, the System Prompt for a "Patient Communicator" might include instructions like, "You are a compassionate healthcare professional skilled in using metaphors. Please answer patient questions in plain and easy-to-understand language..." while the Prompt for a "Doctor Assistant" might be, "You are a meticulous and experienced clinician. Please use professional medical terminology and draft the following document with clear logic..."

[0043] Specialty Agents: These are the system's "trump cards," representing deep expertise in specific clinical domains. Their construction typically requires fine-tuning a base model on high-quality specialty datasets. For example, building an "oncology agent" requires collecting a large amount of anonymized cancer records, authoritative guidelines such as NCCN, and the latest oncology literature. Continuous training or domain-adaptive fine-tuning on this data allows the model to significantly outperform general-purpose models in tasks such as TNM staging and chemotherapy regimen recommendation.

[0044] Agent Registration and Discovery Mechanism: All developed agents need to be registered in a central agent registry (which can be a database or service discovery component such as Consul). Registration information includes: the agent's unique ID, capability description (natural language description of its functions), capability tags (e.g., #parsing#, #imaging#, #oncology#), input / output JSON schema definitions, and estimated call costs and response times. This allows the upper-layer "orchestration engine" to dynamically discover and select the agents it needs, much like searching for apps in an app store.

[0045] The context-aware and intelligent orchestration engine module is configured as the intelligent hub of the system, connecting the unified context bus and the AI ​​capability layer. It analyzes context objects in real time, automatically decomposes macro-clinical tasks into sub-task sequences, and selects and schedules appropriate agents from the pool of expert agents to execute the sub-tasks. This includes a task planner that uses the logical reasoning capabilities of a large language model to generate an execution plan containing agent call sequences and data dependencies based on the current context and a preset clinical workflow template. Context analysis, task planning, and agent scheduling steps: Contextual Analysis and Intent Recognition: The core loop of the engine begins with the analysis of "contextual objects" provided by the "Unified Context Bus". It first uses a lightweight classification model (such as logistic regression or a small Transformer) or a single LLM call to quickly classify the "contextual objects" to identify the current core clinical scenario (such as "outpatient prescription", "ward rounds", "preoperative assessment") and the user's initial intent.

[0046] Dynamic Task Planning and Chain-of-Thought: Once the macro-level scenario and intent are identified, the engine utilizes a powerful LLM trained on complex task planning to generate a multi-step, logically clear execution plan. This process is an advanced application of the LLM's chain-of-thought capabilities; the LLM, like a project manager, breaks down a vague, macro-level task into a series of specific sub-tasks that can be executed by expert agents.

[0047] Expert Agent Selection and Scheduling: For each subtask in the execution plan, the orchestration engine needs to select the most suitable agent from the "expert agent pool." The selection mechanism is a hybrid strategy. Rule / tag-based fast matching: For a specific task (such as "parse this pathology report"), the engine can directly find the most matching "pathology report parser" in the registry based on the task's tags (#parse#, #pathology#).

[0048] LLM-based fuzzy matching and decision-making: For more complex tasks (such as "providing further treatment suggestions for this heart failure patient"), the engine may send the task description and patient context summary to a "meta-agent" (which is itself an LLM), which will then determine whether to call the "cardiology agent" to obtain professional solutions, or the "knowledge retrieval agent" to find the latest guidelines, or both.

[0049] Collaborative Execution and Result Convergence: Based on the plan, the orchestration engine calls the selected expert agents sequentially or in parallel through standard interfaces and collects their structured outputs. These outputs are stored in a temporary "scratchpad" and can be used as input for subsequent subtasks, thereby enabling information transfer and collaborative work between agents.

[0050] The application and interaction module is configured to present the final results or intermediate process information processed by the orchestration engine to the user through a context-adaptive user interface and to receive user interaction commands.

[0051] The user interface provides a traceability mechanism that allows users to trace the original data source or knowledge base entry on which content generated by any expert agent is based when viewing it. Context-adaptive interaction steps: Dynamic UI Rendering: The interface design philosophy at this layer is "interface as data, driven by AI." For example, the HISAI Assistant's interface is not hardcoded in the front end, but rather a library composed of multiple reusable UI components (such as information cards, buttons, and charts). After completing task planning and execution, the context-aware and intelligent orchestration engine outputs not only the business results but also an "interface rendering instruction" (e.g., a JSON object describing the interface layout and content). Upon receiving this instruction, the front-end application dynamically and in real-time renders the interface best suited to the current context.

[0052] Explainability and controllability of interaction design: Information sourcing: All important content generated by AI provides an interactive element on the interface indicating its "source" or "basis." When the user clicks on it, a window pops up clearly showing which original data (e.g., a highlighted excerpt from the EMR) or which guideline in the knowledge base the conclusion is based on.

[0053] Interactive editing: Users can directly edit the AI-generated text drafts on the original text. The system records these modifications, using them both as the final version and as valuable feedback data for future fine-tuning and optimization of the corresponding AI agent.

[0054] Multimodal interaction integration: The system supports seamless switching between multiple input methods. For example, when doctors are making rounds, they can use the voice input function on the mobile device to dictate the ward round record, and the "nurse assistant intelligent agent" or "doctor assistant intelligent agent" in the background will convert it into structured text; after returning to the office, they can edit and confirm this text on the PC.

[0055] Example 3 Admission process for a complex patient Scenario: A 68-year-old male patient arrives at the emergency room complaining of chest pain and difficulty breathing, and is subsequently admitted to the cardiology ward. Resident physician Xiao Wang needs to complete the admission record and preliminary treatment plan for him.

[0056] System workflow: 1. Context Establishment: Xiao Wang opens the patient's medical record in EMR. The Unified Context Bus is immediately activated, aggregating all the patient's information in the emergency room (vital signs, electrocardiogram, emergency room medical record) and their historical medical records to form an initial "context object".

[0057] 2. Orchestration Engine Startup: The context-aware and intelligent orchestration engine analyzes the current context as {user:"Resident Physician", task:"Processing Newly Admitted Patients"}. An execution plan is immediately generated.

[0058] 3. Collaborative execution by intelligent agents: Step 1: The engine schedules the "Report Parser" to automatically parse the emergency ECG and the recently returned troponin test report. The parser returns: "ECG indicates acute inferior myocardial infarction" and "High-sensitivity troponin I is significantly elevated".

[0059] Step 2: The engine updates the above results in the "context object" and analyzes it again. It determines that the current core task is "drafting the admission record". Therefore, it dispatches the "doctor assistant agent" and sends it the instruction: "Please draft the 'chief complaint', 'present illness' and 'preliminary diagnosis' sections of the 'admission record' based on the current patient's context." Step 3: The doctor's assistant AI generated a draft document and presented it on Xiao Wang's EMR interface.

[0060] Step 4: While reviewing and revising the draft, Xiao Wang wanted to prescribe anticoagulation therapy. When she entered "heparin," the programming engine detected this action and analyzed the patient's test results, finding information about "elevated creatinine."

[0061] Step 5: The engine immediately dispatches the "cardiology agent" and the "drug knowledge base retrieval agent" and asks them: "What are the precautions for using heparin in an acute myocardial infarction patient with renal insufficiency?" Step 6: The agents returned the conclusion that "the dosage of heparin needs to be adjusted according to renal function, and low molecular weight heparin may be a safer option for this patient."

[0062] Step 7: The application and interaction layer immediately displays the above suggestion next to Xiao Wang's medical order entry interface with a non-intrusive highlight.

[0063] Results: With the seamless collaboration of AI throughout the process, Xiao Wang not only completed the medical record writing quickly and with high quality, but also obtained a crucial clinical decision support regarding medication safety in a timely manner. The whole process was efficient, smooth, and safe.

[0064] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0065] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

[0066] It should be noted that, in this document, the use of relational terms such as "first" and "second" is merely for distinguishing one entity or operation from another, and does not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0067] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0068] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0069] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0070] 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 units can be selected to achieve the purpose of this embodiment according to actual needs.

[0071] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0072] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0073] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A medical artificial intelligence method based on situational intelligence and multi-agent collaboration, characterized in that, Comprise: Real-time context construction: Real-time aggregation of multi-source heterogeneous data related to a specific clinical scenario, and standardized processing of the data to construct a context object; The context object at least contains patient context, user context and task context; Context analysis and task planning: A context perception and intelligent arrangement engine analyzes the context object in real time to identify the current core clinical scenario and user intent; Based on the identification result, the macro clinical task is decomposed into a plan containing multiple sub-tasks; Expert agent selection and scheduling: According to the characteristics of each sub-task in the execution plan, select an expert agent to execute the sub-task; And, assign task instructions and related context object information to the selected expert agent; Collaborative execution and result aggregation: The selected expert agent receives instructions, executes the assigned sub-task, and returns structured output containing results, confidence and execution process log; The arrangement engine aggregates the output results of the sub-tasks and uses them as input for subsequent sub-tasks or for the generation of final results; Context adaptive interaction: Present the final result or key information of the intermediate process to the user through a user interface that can dynamically adjust its presentation content and layout according to the context object, and receive the user's interaction instructions to trigger a new context construction and analysis cycle. 2.The context-based intelligent and multi-agent collaborative medical artificial intelligence method according to claim 1, characterized in that, In the real-time context construction step, the patient context includes the patient's diagnosis information, allergy history, key examination test results and vital sign trend; User context includes the operator's role, title and affiliated specialty; Task context includes the name of the task the user is currently performing and the location of the system interface. 3.The context-aware intelligent and multi-agent collaborative medical artificial intelligence method according to claim 1, characterized in that, Expert agents at least include the following three types: Basic capability agent, responsible for executing atomized general AI tasks, including medical report text analysis, knowledge retrieval or medical record content quality control; Role-playing agent, responsible for simulating a specific medical role to interact or generate content in accordance with the role's tone, style and knowledge depth; Specialty knowledge agent, responsible for encapsulating deep knowledge and reasoning logic of a specific clinical specialty. 4.The method of claim 1, wherein, The generation of the execution plan uses the chain-of-thought reasoning capability of the large language model to decompose the macro task into a sequence of specific sub-tasks that can be executed by expert agents, with logical dependencies. 5.The context-aware intelligent and multi-agent collaborative medical artificial intelligence method according to claim 1, wherein, Also includes: Agent optimization: Record user's adoption, modification or rejection of expert agent generated results, and use these interaction behaviors as feedback signals for iterative optimization of expert agent configuration or model.

6. The medical artificial intelligence system based on situational intelligence and multi-agent collaboration according to any one of claims 1-5, characterized in that, Comprise: Unified context bus module, configured to aggregate heterogeneous data from multiple medical-related data sources in real time, and standardize it into structured, dynamically updated context objects; AI capability module, configured to form a pool containing multiple expert agents, each of which is configured to perform one or a type of specific medical-related AI task; The context-aware and intelligent orchestration engine module is configured for the intelligent hub of the system, for connecting the unified context bus and the AI capability layer, which analyzes situational objects in real time, automatically decomposes macro clinical tasks into sub-task sequences, and selects and dispatches appropriate agents from the pool of expert agents to execute the sub-tasks; The application and interaction module is configured for presenting the final results or intermediate process information processed by the orchestration engine to the user through a context-adaptive user interface, and receiving the user's interaction instructions. 7.The context-aware intelligent and multi-agent collaborative medical artificial intelligence system according to claim 6, wherein, The specialty knowledge agents in the AI capability module are constructed by fine-tuning large language models on preset clinical specialty data sets. 8.The context-aware intelligent and multi-agent collaborative medical artificial intelligence system according to claim 6, wherein, The context-aware and intelligent orchestration engine module includes a task planner that uses the logical reasoning capabilities of large language models to generate an execution graph containing a sequence of agent calls and data dependencies based on the current context and preset clinical workflow templates. 9.The context-aware intelligent and multi-agent collaborative medical artificial intelligence system according to claim 6, wherein, The user interface of the application and interaction module provides a traceability mechanism that allows users to trace the original data sources or knowledge base entries on which any generated content by an expert agent is based when viewing the content.

Images